archive.org

Full text of "UNDERSTANDING READING - FRANK SMITH"

See other formats


Understanding 
READING 
Frank Smith 
SIXTH EDITION 
UNDERSTANDING 
READING 
A Psycholinguistic Analysis of 
Reading and Learning to Read 
This page intentionally left blank 
SIXTH EDITION 
UNDERSTANDING 
READING 
A Psycholinguistic Analysis of 
Reading and Learning to Read 
FRANK SMITH 
li& 
2004 
LAWRENCE ERLBAUM ASSOCIATES, PUBLISHERS 
Mahwah, New Jersey London 
Copyright © 2004 by Frank Smith. 
All rights reserved. No part of this book may be reproduced 
in any form, by photostat, microform, retrieval system, or 
any other means, without prior written permission of the 
author. 
Lawrence Erlbaum Associates, Inc., Publishers 
10 Industrial Avenue 
Mahwah, New Jersey 07430 
Cover design by Kathryn Houghtaling Lacey 
Library of Congress Cataloging- in-Publication Data 
Smith, Frank, 1928- 
Understanding reading / Frank Smith. — 6th ed. 
p. cm. 
Includes bibliographical references and index. 
ISBN 0-8058-4711-1 (cloth : alk. paper) 
ISBN 0-8058-4712-X (pbk. : alk. paper) 
1. Reading. 2. Learning, Psychology of. 3. Written communi- 
cation. I. Title. 
LB1050.S574 2003 
372.4— dc22 2003060718 
C1P 
Books published by Lawrence Erlbaum Associates are printed 
on acid-free paper, and their bindings are chosen for strength 
and durability. 
Printed in the United States of America 
10 98765432 1 
Contents 
Preface to the Sixth Edition 
vii 
Chapter 1 
The Essence of Reading 
1 
Chapter 2 
Comprehension and Knowledge 
12 
Chapter 3 
Spoken and Written Language 
31 
Chapter 4 
Information and Experience 
55 
Chapter 5 
Between Eye and Brain 
72 
Chapter 6 
Bottlenecks of Memory 
95 
Chapter 7 
Letter Identification 
110 
Chapter 8 
Word Identification 
125 
Chapter 9 
Phonics and Mediated Word Identification 
138 
VI 
CONTENTS 
Chapter 10 
The Identification of Meaning 
156 
Chapter 1 1 
Reading, Writing, and Thinking 
178 
Chapter 1 2 
Learning About the World 
194 
Chapter 13 
Learning About Written Language 
212 
Notes 
233 
Glossary 
325 
References 
333 
Author Index 
361 
Subject Index 
369 
Preface 
to the Sixth Edition 
Here are extracts from the reports of two anonymous reviewers of 
the fifth edition of Understanding Reading: 
Reviewer 1 : “Frank Smith's research summarizes a generation of in- 
vestigations across disciplines .... In one succinct, readable volume 
it comprises what I consider a thorough and incisive summation of 
core research, theory and interpretation. It represents a new main- 
stream in progressive reading research .... Its major strength is its 
straightforward, compelling presentation of approaches to reading 
and writing that are meaningful and salient to children. ” 
Reviewer 2: “7 his volume contains partial truths, contradictions, and 
cites only references that support the author’s view. Either the author 
is notfamiliar with the current research literature, or he deliberately 
avoids citing evidence that is contrary to his point of view .... This 
book is no recipe for improving reading skills of children, especially 
beginning readers and poor readers: it is a recipe for disaster.” 
Both reviewers are professors at university schools of education, 
experts in the field, with access to the same professional literature 
and with the same professional concerns. Yet a chasm separates 
viii UNDERSTANDING READING 
their points of view, one of which is expressed with vehemence and 
indignation. 
This head-on clash of attitudes currently permeates every as- 
pect of theory and research into reading and reading instruction, 
among practitioners, politicians, and the general public. It has be- 
come a focus of legislation and litigation. One has to turn to reli- 
gious fundamentalism to find another issue that arouses such 
bitter controversy. There is no other academic discipline where so 
many people claim sole possession of truth and declare those with 
a different point of view “unscientific.” 
The first edition of this book, in 1971, set out to be an objective 
(and scientific) review of every field of study that had anything rele- 
vant to say about reading and about learning to read, with the un- 
complicated aim of “understanding reading." Every edition, 
including the present one, has steadfastly resisted giving teachers 
a recipe for teaching reading while aiming to help them make their 
own decisions, based on research about reading, which is accessi- 
ble to anyone, and their experience and personal knowledge of 
their students, which only they possess. 
But it is impossible to write a book about reading, however de- 
tached the intention, without being caught in the cross fire of how 
reading should be taught. I have never professed any doctrinal alle- 
giance, yet people who believe one thing accept this book as sup- 
port for their point of view while those who take an opposing 
position anathematize it. 
Who is right? I don’t think that’s a useful question at present. 
Each side believes it has the monopoly on truth, and few of the ma- 
jor protagonists would even consider the possibility of being 
wrong. If a crucial experiment or unanswerable argument existed, 
one side would have disappeared from the scene years ago. 
A better question might be, what constitutes the grounds for 
people on both sides to feel so sure they are right and the other side 
wrong? That is the issue I focus on as I endeavor to bring the latest 
edition of Understanding Reading up to date. 
But because of the clamor of the controversy it is not enough for 
me to lay out the facts of reading and learning to read as I see them . 
The controversy itself must be examined, to ascertain why such ex- 
treme divergences of opinion can come about. All teachers of read- 
ing, and ultimately all parents and other interested observers, must 
make up their own minds about why these conflicting points of view 
exist, unless they blindly submit to the assertions of the people who 
shout loudest or wield the biggest sticks (or carrots). That is the rea- 
son I include a brief statement of “issues” at the end of every chapter. 
PREFACE TO THE SIXTH EDITION 
IX 
THE GREAT DEBATE 
The fifth edition of Understanding Reading, published in 1994, 
included a lengthy summary of the “Great Debate” then raging be- 
tween proponents of “whole language” and those of “direct instruc- 
tion.” I commented that it would be comforting not to have to 
include the section in a sixth edition, but that the portents were not 
auspicious. 
My prognostication was correct, but not quite for the reasons I 
anticipated. If the great debate no longer has the same intensity, it 
is because direct instruction carried the day. Whole language has 
been sidelined rather than vanquished. The direct instructional 
view, with its assumption that reading is a matter of decoding 
letters to sounds, has been taken for granted by those arguing for 
accountability, standardization, high-stakes testing, and external 
control of classroom instruction and teacher education, from the 
highest political levels down. In many parts of the English-speak- 
ing world — particularly the United States, Britain, Canada, 
Australia, and New Zealand — whole language (or “real reading”) 
has become a subversive, underground movement. Resistence to 
the control exercised by mandated tests and curriculums has 
become more of a confrontation over teachers’ freedom in 
classrooms than over technical and philosophical issues related to 
reading instruction. Proponents of direct instruction concede that 
there must be a place — usually late in the course of instruction — 
for meaningful reading and for “teaching comprehension,” and 
whole -language teachers and theorists deny that they ever 
proposed abolishing all reference to phonics in teaching reading, 
only against giving phonics priority and predominance. The old 
schisms have been papered over, and a new dispute — it can hardly 
be called a debate — has arisen over political issues. 
WHAT'S NEW? 
What justifies a sixth edition of Understanding Reading? The facts 
are the same, but the perspectives are different. Many of the basic 
facts about the nature of reading have been known for at least a 
century but, as Edmund Burke Huey observed in 1908, disputes 
over them seem always fresh. 
My regular excursions through mountains of related literature 
reveal that little of substance has changed since the first edition of 
Understanding Reading was published in 1971 . The main contro- 
versies persist (what it means to be a reader, how written words are 
X 
UNDERSTANDING READING 
recognized, and how reading should be taught), and the only con- 
sequence of continued research seems to be a hardening of posi- 
tions. It is unnerving to discover that what were once seen as fresh 
approaches are now ancient dogmas, adopted or rejected by waves 
of new champions, sometimes with immoderate enthusiasm or in- 
temperate scorn. No wonder teachers and students can be con- 
fused, if they expect either side of controversies to be proved right 
or to admit to being wrong. 
This is clearly not a matter of “facts." There is more evidence 
than anyone knows what to do with, but it comes into educational 
theorizing and policy as raw material, to be analyzed in different 
ways for different purposes. What matters is how facts are gath- 
ered, which facts are considered relevant, and how they are inter- 
preted — all subject to personal predilection. There are no "pure 
facts” — their place in any theory is always determined by broader 
theoretical and political intentions. One issue that has particu- 
larly attracted my attention is the assertion, which I regard as om- 
inous, that there is something unnatural about reading. I give this 
early attention. 
Substantial portions of the present volume have remained un- 
changed through all six editions. I have again examined every word 
carefully and consider that these portions are indeed unchallenge- 
able, at least without argument at the level of “Of course the world 
is flat. It’s obvious, isn’t it?” Other portions are, I think, equally 
supportable, but give rise to opposing points of view because of 
perceived instructional implications. I don’t want to claim absolute 
truth, because I am sure others will find better and more insightful 
ways of understanding reading in the future, although not, I sus- 
pect, in the way some current participants in the controversy vocif- 
erously claim that they are indubitably and everlastingly “right.” 
I have added about 220 new references (having scrutinized per- 
haps 10 times that number) and removed about 500 of the older 
references. But I have refrained from replacing the old with the new 
just for the sake of “updating.” Where earlier studies remain 
unique or have been copied but not improved on, I have let them 
stand — they have stood the test of time. I don’t support claims that 
the “latest” research is always the most significant or the most reli- 
able. Often it simply reflects current enthusiasms, special inter- 
ests, and the predispositions of funding sources. 
Apart from the introduction of a new chapter 1 , 1 have not found 
it necessary to alter the order of chapters or the general thrust of 
the discussion. Most of the major changes have been to the notes 
rather than to the chapters. Writing remains a major topic that has 
PREFACE TO THE SIXTH EDITION 
XI 
not been covered to the extent that it warrants, primarily because I 
discuss the subject at length in a companion volume, Writing and 
the Writer (Smith, 1994a). 
I have continued to resist the tendency of new editions to put on 
weight with age, and the final product is 25,000 words slimmer 
than its predecessor. In the notes (starting on page 233) there is 
an explanation for why I have retained the word psycholinguistic 
in the subtitle to this book, despite considerable changes in its 
use since I first employed it. There is also an important acknowl- 
edgment. 
TEACHERS MUST DECIDE 
In acknowledging alternative points of view, I have not concealed 
my own position, and I certainly cannot claim to give opposing be- 
liefs a comprehensive hearing in this book (although there are 
plenty of other books around that do that). But when it comes to 
the point, teachers and students must make up their own minds, 
which is something I said way back in the first edition. 
We live and learn in a world where no final answers are guaran- 
teed, and must make profound decisions for ourselves (even if only 
to accept unquestioningly the opinions or decisions of someone 
else). Throughout their professional lives, teachers are confronted 
by conflicting points of view, frequently urged with compelling au- 
thority and conviction, and they must be able to take a position. 
The first responsibility and right of all teachers and students must 
be to exercise independent thought — although in their own educa- 
tion they are often denied that opportunity with rationalizations 
that they “aren’t ready,” “shouldn’t be confused," or “lack thinking 
experience” (Smith, 1990, 1993). 
Reading is complex, but so also are walking, talking, and mak- 
ing sense of the world in general — and children are capable of 
achieving all of these, provided the environmental circumstances 
are appropriate. What is difficult to describe is not necessarily dif- 
ficult to learn. One consideration that this book emphasizes is that 
children are not as helpless in the face of learning to read as often is 
thought. 
Because an understanding of reading requires acquaintance with 
research in a variety of disciplines, more than half of the book is de- 
voted to such general topics as language, memory, learning, the de- 
velopment of spoken language ability, and the physiology of the eye 
and brain. The aim is to make these topics comprehensible, with the 
assumption that many readers will have neither the time nor the ex- 
XII 
UNDERSTANDING READING 
perience to undertake deep or specialized study in these areas. At 
the risk of offending specialists, diverse subject areas have been cov- 
ered only to the extent that they are relevant to reading. 
Not all readers automatically consult notes that are mentioned 
in the text, and few probably go on to plough through the mass of 
notes at the end. But I have preferred to keep the main part of the 
book compact and coherent, without lengthy digressions. The 
notes remain supplementary resources. 
In general, this book is designed to serve as a handbook for lan- 
guage arts teachers, a college text for a basic course on the psychology 
of reading, a guide to relevant research literature on reading, and an 
introduction to reading as an aspect of thinking and learning. 
— Frank Smith 
Notes to the Preface begin on page 233 covering: 
Psycholinguistics and cognitive science 
Research 
Acknowledgment 
The Essence of Reading 
Proponents of direct, intensive, and early phonics training for 
teaching reading (like Reviewer 2 in the preface) partly justify their 
beliefs by asserting that unlike learning spoken language, learning 
to read is not “natural” and that reading itself is an unnatural activ- 
ity. This book takes a contrary position. 
READING THE WORLD 
I’ll start my discussion of reading with a psychological point. Noth- 
ing is unnatural in the eyes of infants. Everything they encounter in 
the world is natural, even if they find it aversive. The arbitrary divi- 
sion of the world into what nature once provided and what people 
have subsequently done to it is something that has to be learned. 
Other creatures never make such a distinction. I doubt whether 
crows have different categories for cars and houses than they do 
for rocks and trees. Deer are unlikely to think “Here's where nature 
ends” when they cross from forest glade to cement highway. “Un- 
natural” is a concept that doesn’t exist outside language. 
So what is written language? For a child, print is just another 
facet of the world, not yet comprehended perhaps, but not different 
from all the complex sights, sounds, smells, tastes, and textures in 
the environment — not especially mysterious or intimidating. 
1 
2 
UNDERSTANDING READING 
And what do infants do when they are born into this wholly natu- 
ral world? They do as they will for the rest of their lives: They try to 
make sense of it, to discover how it relates to everything else that 
they know, to understand its relationship to them, its “meaning." 
Trying to make sense of any facet of the environment, including 
print, is a natural activity. 
How exactly do infants (and adults) strive continually to make 
sense of everything they encounter in the world? They read it. 
Reading is the most natural activity in the world. 
I am not taking liberties with language here. The word “reading” is 
properly employed for all manner of activities when we endeavor to 
make sense of circumstances; its original meaning was “interpreta- 
tion.” We read the weather, the state of the tides, people’s feelings 
and intentions, stock market trends, animal tracks, maps, signals, 
signs, symbols, hands, tea leaves, the law, music, mathematics, 
minds, body language, between the lines, and above all — a point I 
must come back to — we read faces. “Reading," when employed to re- 
fer to interpretation of a piece of writing, is just a special use of the 
term. We have been reading — interpreting experience — constantly 
since birth and we all continue to do so. 
What is this basic reading or "making sense" that we all engage 
in? I don’t think it needs to be explained, or even can be explained. 
It is what we are. Anyone who didn't try continually to make sense 
of the world could not be considered a functioning human being. 
Making sense is a matter of interpreting, relating the situation you 
are in to everything you know already. Not to part of what you know, 
but everything, because all our knowledge hangs together. Our un- 
derstanding of the world, all of the world, is coherent, consistent, 
and immediate. Once you know that a flame burns, you don’t have 
to say to yourself, “That is a flame, therefore it burns.” You know 
that flames burn. Once you can recognize a truck, you don’t have to 
say to yourself “That is a truck" and consult some inner encyclope- 
dia. Once you can read the written word “dog,” you don’t have to say 
to yourself, “That word says dog, I must look up what it means.” 
You know what it means. 
What do children do when they encounter a dog? They don’t say 
“I recognize that animal with a particular juxtaposition of wet nose, 
sad eyes, and floppy ears as a certain kind of dog," nor do they say 
“There’s a dog" to themselves and look up its meaning in a library 
in the brain. They certainly don’t wait to hear the animal bark to de- 
cide what it is. Recognition, whether of dogs and cats or written 
words, is not a matter of breaking something down to its compo- 
nents, but of integrating it into a larger context. 
1 . THE ESSENCE OF READING 
3 
All learning and comprehension is interpretation, understand- 
ing an event from its context ( or putting the event into a context ) . All 
reading of print is interpretation, making sense of print. You don’t 
worry about specific letters or even words when you read, any 
more than you care particularly about headlights and tires when 
you identify a car. 
The best strategy for determining the identity of meaning of an 
unfamiliar word is to work out what it is from context. As we shall 
see, this happens very quickly. An equally good way in different cir- 
cumstances is simply to ask someone what it is. Often we don’t 
have to ask. A very poor strategy is to try to “sound it out.” 
Some people seem to believe that learning to read is a particu- 
larly challenging undertaking — despite the ease with which many 
children accomplish it, and despite how much children have 
learned in other contexts. Learning to read is not rocket science. 
No one could catalogue all the things a human being, even a 
young child, has been able to make sense of in the world; it would 
be an impossible task. We live in an enormously complex and com- 
plicated world, but the times when individuals are actually con- 
fused, even babies, are remarkably few. Children aren’t usually 
confused by written language — until someone tries to instruct 
them on how to read. When people help children to read, by read- 
ing to them and with them, there is rarely confusion. It is not read- 
ing that many children find difficult, but the instruction. 
Most of our learning is unsuspected. Perhaps the most complex 
learning of all involves the human face. Researcher Daniel McNeill 
(1998) explained how 22 pairs of facial muscles are constantly or- 
chestrated to display at least four thousand different expressions, 
all produced and universally understood without any instruction 
at all. Some basic expressions of emotion — like fear, anger, sur- 
prise, disgust, sadness, and enjoyment — may be instinctive, but 
the majority are learned early in life. These expressions, involving 
the entire face from the corners of the mouth to the eyebrows, with 
each element operating individually, communicate not just physi- 
cal states, but agreement, disagreement, encouragement, puzzle- 
ment, disbelief, collusion, threat, challenge — and of course 
interest and desire. When was anyone taught to interpret all this, to 
read faces? (Or to write on faces, for that matter.) 
It is natural for children, and adults, to strive always to make 
sense of the world, to interpret what everything must mean. So why 
should language written in an alphabetic script be particularly diffi- 
cult? The answer is that it isn’t. Reading print is no more complex 
than reading faces, and other things in the world. Making sense of 
4 
UNDERSTANDING READING 
print can’t be more complicated than making sense of speech, which 
begins much earlier. Written words and spoken words share the 
same kind of grammar, meanings, and other structures. If we can 
make sense of all the words of spoken language that we know, we 
can do the same for written words. The actual numbers involved 
fade before the vast numbers of faces, places, objects, events, ex- 
pressions, and relationships that we can make sense of in the world. 
Memory is hardly a problem. Written words are actually easier to 
discriminate than speech — we can mishear what someone says, or 
be unable to recover from a lapse in concentration; in writing we can 
always check back. Some written words are easier to discriminate 
than the objects they refer to. Participants in a scientific experiment 
could identify words flashed on a screen faster than they could iden- 
tify drawings of the objects the words referred to (like house, dog, 
flower, and so forth), even after extensive practice on the limited set 
of alternative words and pictures that were presented. 
There is nothing unnatural about any of this, as I have main- 
tained. Written language is no more opaque or impenetrable than 
anything else in the world, once we have made sense of it (because 
we have encountered it in circumstances that make sense to us). 
So why do some people have so much trouble learning to read? 
The first reason might be that they are confronted by reading when 
it is not the best time for them to learn, just as not everyone learns 
to play the piano, to swim, or to play chess at the same time. They 
may be too involved in other things, or trying to recover from some 
trauma. Learning to read is not necessarily a problem at any age — 
unless there are years of reading confusion and failure in the past. 
Which leads to the second reason why some people have so much 
trouble learning to read. They’ve been confused. Instead of being 
helped, they’ve been handicapped. 
People can be confused by anything. Difficulty in learning to read 
doesn’t mean that it is unnatural (unless everything else that hu- 
mans do that is not instinctual is regarded as unnatural). 
Allusions to “scientific” studies don’t prove a thing. If phonics is 
an impossible system, even for computers, then any experimental 
study claiming to show that phonic drills have helped children to 
read must have been looking at something else. In fact, many stud- 
ies of phonics and phonemic awareness acknowledge that they are 
looking at something else. Instead of looking at reading as a matter 
of making sense of text, they look at how well children can put 
sounds to isolated words, and even to meaningless sequences of 
letters, to confirm that they use the alphabetic code. This is like ty- 
ing children’s feet together to prove they must jump before walking. 
1 . THE ESSENCE OF READING 
5 
References to mythical brain disabilities (diagnosed circularly 
in relation to perceived reading difficulties) explain nothing. Such 
phantasms are conjured up in the absence of understanding or co- 
herent theory. And even if there were rare brain malfunctions that 
make it difficult for a few children and adults to read, that doesn’t 
mean that such individuals should be subjected to regimes of un- 
natural treatment. Such individuals must still be helped to make 
sense of print — but it will take more time and patience. Calling 
them disabled is hardly likely to help. 
Reading print is as natural as reading faces. Learning to read 
should be as natural as any other comprehensible aspect of exis- 
tence. How reading is naturally accomplished, and what can go 
wrong, are the twin concerns of this book. 
DISENTANGLING THE UNDERGROWTH 
To clear the ground for the rest of the book, I must deal with several 
matters that in my view contribute to confusions or misconcep- 
tions about the nature of reading. They concern ( 1 ) the alphabet, 
(2) language, and (3) the brain. I raise these issues now because to 
some extent they contradict what often seems obvious, and there is 
no point in trying to understand reading without first examining 
critically what many people may take for granted. The remainder 
of the book will develop the arguments. 
The Alphabet 
Ever since an alphabetic writing system was invented by the 
Greeks over two thousand years ago, the 26 or so letters have had a 
profound influence on human thought. Many people through the 
centuries have been fascinated by the letters that make up words, 
and the putative relationships of these letters to the sounds of 
speech. They cannot imagine reading without a central role for the 
letters that make up individual words. Reading instruction from 
Greek and Roman times has focused on letters and sounds, de- 
spite continual efforts by critics to emphasize the vital role of 
meaning in reading (Mathews, 1966) and to demonstrate that let- 
ters play only a small, redundant, and often confusing part. Letters 
have become a fetish. People transfixed by the alphabet ask incred- 
ulously what the purpose of letters might be if not to make it possi- 
ble for readers to read. 
But the alphabet was never designed to help readers. It was not 
invented or developed for that purpose. Nor was it intended to be of 
6 
UNDERSTANDING READING 
any particular help to writers. The alphabet’s true function has al- 
ways been to help people cope with technical problems of repro- 
ducing written language, for scribes, copyists, inscribers, and 
printers. I’ll call them transcribers. Tolchinsky (2003) provided an 
excellent summary of this, adding that a particular motivation for 
trying to make writing reflect sound was so that people’s names 
would appear consistent in print (pp. 42-44). 
The prime importance of the alphabet is that it enables people to 
make marks on paper (and other surfaces) in a simple and consis- 
tent manner, so that to speakers of a language, the written words 
will always look the same. In a sense, the 26 letters are convenient 
alternatives to thousands of drawings. They are building blocks for 
the construction of visible words, like the wooden tablets used in 
many board games. “Decoding to sound” has nothing to do with it. 
Readers have coped with nonalphabetic languages like Chinese for 
centuries, and continue to do so. Learning to read an ideographic 
script has never been a particularly complicated or traumatic pro- 
cess. Even in alphabetic cultures today we all understand a multi- 
tude of symbols that don’t decompose into individual sounds, like 
the ubiquitous ® (“don’t even think about it"), the icons on wash- 
room doors, dashboards, and laundry machines, numerical sym- 
bols like 1 , 2, 3, and so forth, and such characters as (a # $ % ~ & 
*() + =? on keyboards. They have names, but they can’t be de- 
coded into sounds. Nor has the alphabet anything to do with en- 
coding, for that matter. Letters correspond to sounds only 
coincidentally; they are guidelines that keep transcribers from rep- 
resenting words in an idiosyncratic and arbitrary manner. Letters 
cut down on arguments. No one can claim that C-O-W is a better 
way of writing “horse” than H-O-R-S-E. But this was far more im- 
portant for the transcriber than for the writer. In fact, it was not un- 
til after the Gutenberg revolution, when texts began to be mass 
produced, that printers began to worry particularly about consis- 
tency. They didn’t want spellings that sounded right, just ones that 
weren’t contentious. 
The alphabet is a construction kit for putting words together — 
much like the set used by a person who constantly changes the bill- 
boards for movie theaters or supermarkets, assembling one letter 
at a time from a stock (for the English language) of 26 alternatives. 
This is an enormous advantage. From just 26 basic shapes, a 
unique visual representation of every word in the language can be 
produced. The sign writer doesn’t even need to be literate, as he 
copies one letter at a time from his script. And contrast the cost-ef- 
fectiveness of having 26 basic shapes from which to build words, 
1 . THE ESSENCE OF READING 
7 
compared with the complexity of Chinese script, which for formal 
purposes has to be drawn by an artist. (A standardized alphabetic 
form of Chinese became imperative with the advent of keyboards 
for typewriters and computers.) 
The second advantage of the alphabet is that each of these 
shapes, and their variants, has been given a name — Ay, Bee, Sea, 
etc. — so that the illiterate sign writer can be told how to construct 
every word in the language. Instead of “Use a circle, a zigzag, and a 
right- angle” he can be told to put up an O, a W, and an L. 
When a child asks “How do I write cat?” we don’t have to say 
“There’s an open circle at the beginning, then a closed circle with a 
tail, and finally a ...” (I can’t even think how to describe a “T”), we 
simply say “Cat is written C A T.” We can do that for every word in 
the language. 
This far from exhausts the utility of the alphabet. The 26 letters 
have been assigned a conventional order, so that every word in the 
language, including names, can be put into easily sorted, easily 
searched, sequences. Think of the utility of alphabetical order in 
dictionaries, directories, libraries, and other information storage 
and retrieval systems. Imagine the organizational chaos if alpha- 
betical order didn’t exist. (How could I construct an author and 
subject index for this book?) 
So the alphabet earns its keep; it is one of our most useful inven- 
tions. But it is not essential. We could have visible language without 
it. People can learn to read without a phonetic alphabet without 
great difficulty. Chapter 9 examines why the sounds associated 
with letters are largely irrelevant and frequently misleading for 
readers and writers. But here’s a quick demonstration of that fact. 
Computer programs that “read” by producing sounds from text 
that is keyboarded in, and that “write” by transforming speech in- 
put into text, don’t use phonics. The programs won’t work at the 
letter-sound level. And as for “phonemic awareness,” the detection 
of distinct sounds in spoken language that are supposed to corre- 
spond to letters, computers can't do it at all. Computers do best 
with words, especially when grouped in meaningful sequences. 
Language 
I don’t propose to enter into a lengthy disquisition on the nature of 
language, or on its uses in society, communication, and expres- 
sion. I just want to focus here on one narrow aspect of language, 
which has a considerable impact on the way everyone thinks. I 
want to consider how language creates worlds, objects and rela- 
8 
UNDERSTANDING READING 
tionships, which in no other sense exist. Language makes us think 
something is there when it isn’t. It deceives us. 
The human race is always prone to give names to aspects of ex- 
perience, and then to take for granted that whatever corresponds 
to those names exists. Give something a name (like intelligence, or 
perseverance, or wickedness), and many people will think that it 
exists, not as a kind of behavior that fits a certain description, but 
as the cause or underpinning of the behavior. Thus for example 
reading, which in general is easily identifiable behavior, has be- 
come transmuted into the reading process, which is assumed (by 
many) to actually exist within the human brain (which is also sup- 
posed to contain a writing process, a grammatical process, and a 
phonemic awareness process). 
Learning and comprehension are particularly interesting exam- 
ples of this drive to construct fictitious entities. Both are widely re- 
garded as skills, reflecting learning and comprehension processes 
in the brain. Instructional programs are devised to augment these 
processes, and standardized tests to calibrate their effectiveness. 
But a different point of view can be taken that learning and compre- 
hension are simply states of the human organism. They are neither 
skills nor processes, but a consequence of being alive. Their pres- 
ence in human beings doesn't have to be explained, only their ab- 
sence, or rather the consequences of their suppression. Any human 
in a position of being unable to learn is bored. No one would claim 
that boredom was a process; it is the opposite of learning, an alter- 
native state. Similarly absence of comprehension is not a lack of 
skills, nor the shutting down of a process; it is a state, to which we 
normally give the name of confusion. It might be tempting to con- 
sider confusion as a chaotic disorganization of certain structures of 
the brain, but it is not. It is simply a state that is the opposite of com- 
prehension. I should perhaps note that both boredom and confu- 
sion are aversive; they are not natural states to be in. All human 
beings strive naturally to be in a continuous state of learning and 
comprehending, just as they continually strive to breathe. 
So there are two problems with language. It enables us to think 
about things that don’t actually exist, and then to devise unseen 
processes that bring these things into being. 
I don’t want to belittle language in any way; it has many beautiful 
and useful characteristics. It enables us to think and to create. 
Language is particularly useful for description — a few well-chosen 
words can give a powerful image of people or objects, and of many 
associated characteristics (whether or not the person or object ac- 
tually exists). But language is distinctly lacking in explanatory 
1 . THE ESSENCE OF READING 
9 
power. Whether we try to explain a person, a group, or an activity 
like reading, we quickly fall back on fiction and metaphor. Rather 
than describe the circumstances in which individuals demonstrate 
literacy, or learn to become literate, we invent explanations. We put 
mechanisms and processes into their heads (see “Making a Mys- 
tery Out of a Marvel,” in Smith, 2003, chap. 2). 
Little of this explanation-through-fantasy would matter if we put 
two little words into our diagnostic and referential statements — 
the words as if. To say the human brain sometimes functions as if 
it were a computer is altogether less misleading than to say that the 
brain is a computer. The statement that some people read as if they 
employ knowledge of letter-sound correspondences is easier to 
comprehend, and much easier to discuss rationally, than the blunt 
assertion that people can read because they employ phonic skills. 
Language can be used for the careful dissection and analysis of 
complex human behavior — but not if wielded like a blunt instrument. 
The Brain 
My point about the brain is simply stated. Despite extravagant 
claims — usually by people who are not neurologists — no one 
knows how the brain is ultimately related to anything we see or do 
in the world. Brain research tells us nothing about anything except 
the brain itself. We may point to various irregularities in the brain 
to try to account for why things occasionally go wrong, but we can- 
not use the architecture of the brain to explain why anything we do 
or think goes right, or even why it occurs. 
We can take the brain apart and see how all the bits are joined to- 
gether. We can poke and prod at parts of the living brain and see 
how a person reacts. We can see what goes wrong when bits are 
missing. We can take various kinds of pictures of the brain and see 
how it heats up as people engage in various kinds of activity or as 
they think about particular things. But none of this explains why 
we have the kinds of thoughts or sensations that we do. If a blow to 
a particular part of my brain makes me see stars, or hear a sym- 
phony, I have to tell someone of that fact. No neuroscientist can 
look inside my brain and say, “He’s hearing Mahler’s Second right 
now.” No neuroscientist can explain why I see green, or taste salt, 
or experience the scent of a rose. Neuroscientists might claim to 
have produced a complete wiring diagram of the parts of the brain 
that seem to be involved in anything I do or experience, but they can 
never say why I have that particular experience. They can’t find a 
10 
UNDERSTANDING READING 
map in my head if I claim to know my way around the university, 
nor can they find why I might decide to walk from one part of cam- 
pus to another. 
The neuroscientist's situation is no different from that of the tele- 
vision technician who can tell you how the various components of 
your system work together, and who can explain why sometimes you 
don’t get sound or a picture, but who has no way of explaining why 
the various electromagnetic events that take place on the screen 
should make you laugh, cry, or switch channels in abject boredom. 
We’d think it absurd if our technician told us he had detected a 
sitcom-sensitive area in the television receiver, or that a particular 
module accounted for the upbeat attitudes of weather reporters, 
even though a misdirected screwdriver could certainly interfere 
with both. Yet neuroscientists have no trouble labeling a reading 
process in the brain. They will draw diagrams of the inside of the 
brain with arrows and little boxes labeled < input > < output > 
< phonemic processing> <memory> and even <understanding>. 
But they can’t explain what goes on in those boxes, or the nature of 
the "information” assumed to pass along the routes indicated by 
the arrows. 
No neuroscientist has ever been able to find any of the 26 letters 
of the alphabet in the brain, nor the connections they are supposed 
to have with particular sounds, although we are assured that such 
structures exist (although not in people afflicted with “dyslexia”). 
The alphabet is doubtless a consequence of something in the 
brain, but not anything that could be regarded as specialized and 
dedicated for the purpose of producing and making use of letters. 
The parts of the brain involved in handling letters could very well 
be the same parts that are involved in identifying birds or cooking 
omelettes. I see no reason why they should not be. The same ap- 
plies to phonics skills and phonemic awareness. Even if these 
things have any real existence in the brain (which I doubt), I have 
encountered no evidence that they would be any different from pro- 
cesses that enable us to listen to jazz or enjoy a movie. 
And because there is absolutely no evidence of how any neuro- 
logical or chemical processes in the brain might produce reading, 
it makes no sense to say that there are specialized centers or pro- 
cesses in the brain responsible for reading. Obviously there are 
parts of the brain involved in reading, and a good number of other 
activities too, but that is no reason to claim that these areas are for 
reading, any more than you can say that one part of an automobile 
engine is responsible for getting you to the supermarket and an- 
other part for driving to the beach. 
1 . THE ESSENCE OF READING 
11 
I have to admit that "brain” used to be one of my favorite words; I 
held it responsible for almost everything we do. I still regard the 
brain as an astonishing instrument, far more remarkable than it is 
often given credit for, but it makes sense much of the time to refer 
to the whole person rather than to the individual’s cranial con- 
tents, although it may sound less scientific. Reading, for example, 
is best regarded as something done by people rather than by 
brains. To say the brain “looks,” “thinks,” or “remembers” is about 
as appropriate as saying that the stomach enjoys a good meal. I 
have not been able to remove the brain from chapter 5 (“Between 
Eye and Brain”) because that is what the chapter is about, although 
I have tried to keep the discussion on as literal a level as possible. 
The brain reappears in later chapters but always when I am refer- 
ring specifically to the organ, not to the person as a whole. My prob- 
lem in trying to avoid talking about the brain as if it “learns” or 
“makes decisions” emphasizes the metaphorical nature of the lan- 
guage we must use when trying to relate physiological structures to 
feelings or behavior (see chap. 2 notes.) 
ISSUES 
Just about every statement I make in this chapter can be chal- 
lenged by some people, especially those who prefer to believe that 
learning to read is simply a matter of subjection to a rigorous re- 
gime of phonics instruction. The statements will not necessarily be 
challenged because they are refutable or inappropriate, but be- 
cause they conflict with preconceptions about how reading should 
be taught. 
SUMMARY 
The alphabet was designed as a technology to simplify making lan- 
guage visible, not as an aid to readers or writers. Reading is as nat- 
ural as recognizing and interpreting facial expressions. 
Understanding reading becomes complicated when certain meta- 
phorical states of affairs are taken as being literally true. 
Notes to chapter 1 begin on page 238 covering: 
Matters of interpretation 
Alphabets 
Writing systems 
2 
Comprehension 
and Knowledge 
Most people would say they know what the word comprehension 
means, at least in a general sense, although it is not a term that oc- 
curs often in everyday speech. In fact, it is almost exclusively found 
in the context of reading. In everyday speech we are much more 
likely to use the term understanding (as I have done in the title of 
this book) or even my preferred alternative of making sense. The 
word comprehension was rarely used in the research literature on 
reading before the 1950s, when systems analysts and behavioral 
engineers were first recruited to design reading programs (Smith, 
1998, p. 116). 
In other words, comprehension is a kind of up-market synonym 
for understanding in discussions that are (or are intended to ap- 
pear) technical and scientific. In such contexts the word frequently 
doesn’t appear alone, but in such combinations as comprehension 
skills or the comprehension process, even by people who would 
never use expressions like understanding skills or the under- 
standing process. 
I can’t avoid the word comprehension in this book — it is too well 
established in the world of reading to be ignored. But I don't regard 
comprehension as some kind of special or unusual process. As I 
12 
2. COMPREHENSION AND KNOWLEDGE 
13 
said in chapter 1 , 1 see comprehension as a state rather than a set 
of skills or a process. 
Comprehension may be regarded as relating aspects of the 
world around us — including what we read — to the knowledge, in- 
tentions, and expectations we already have in our head. It is clearly 
the purpose of reading and of learning to read. What is the point of 
any activity that causes confusion? 
We don’t have to know something in advance in order to compre- 
hend it. But we must be able to relate new things to what we already 
know if we are to comprehend them. And relating something new 
to what we already know is of course learning. We learn to read, 
and we learn through reading, by elaborating what we know al- 
ready. This is natural. 
Thus, comprehension and learning are fundamentally the same, 
relating the new to the already known. To understand all this, we 
must begin by considering what it is that “we already have in our 
heads” that enables us to understand the world. We must begin by 
comprehending comprehension. 
Cognitive Structure 
Several terms may be used to refer to the knowledge we carry 
around in our heads all the time. Prior knowledge and “nonvisual 
information ” are synonyms for the mental resources that enable us 
to make sense of "visual information” arriving through the eyes. 
Long-term memory is our permanent source of understanding of 
the world. Cognitive structure and theory of the world are two 
other terms that I am about to introduce. But the italicized terms 
do not refer to different things; they are synonymous. The knowl- 
edge we must already possess in order to understand written lan- 
guage (like the knowledge we need for understanding speech) must 
be part of our long-term memory. And remembrance of the sense 
we have made of past experience is the foundation of all new under- 
standing of language and the world. In more general contexts, this 
basis of understanding is referred to by psychologists as cognitive 
structure. The term is apt because “cognitive" means “knowledge” 
and “structure” implies organization, and that indeed is what we 
possess — an organization of knowledge. 
Certainly, it would be simplistic to suggest that what we carry 
around in our heads is just “memories.” The brain is not filled with 
an assortment of snapshots, videos, and recordings of bits of the 
past. At the very least we would have to say that all our memories 
have a meaning: they are related to everything else that we know. 
14 
UNDERSTANDING READING 
Cognitive structure is more like a summary of past experience. I 
don’t want to remember that on 1 6 July I sat on a chair, on 1 7 July I 
sat on a chair, and on 18 July I sat on a chair. I want to remember 
that chairs are for sitting on, a summary of my experience. We re- 
member specific events only when they Eire exceptions to our sum- 
mary rules or when they have some particularly dramatic or 
emotional significance. And even then our memories, when we “re- 
call” them, turn out to be highly colored by our present intentions 
and perspectives about the world (Bartlett, 1932). Specific memo- 
ries that can’t be related to our summary, to our present general 
understanding, will make little sense, which may be the reason we 
can recall so little of our childhood. 
But it would also be an oversimplification to suggest that our 
heads are filled with an accumulation of facts and summaries. The 
brain is not like a library where useful information is filed away un- 
der appropriate headings for possible future reference. And it is 
certainly not like a bank in which we save nuggets of information 
deposited by teachers and textbooks. Instead, the knowledge we 
possess is organized into an intricate and internally consistent 
working model of the world, built up through our imagination and 
our experiences in the world, and integrated into a coherent whole. 
We know far more than we were ever taught. 
THEORIES OF THE WORLD 
Everything that we know and believe is organized into a personal 
theory of what the world is like, a theory that is the basis of all our 
perceptions and understanding of the world, the root of all learn- 
ing, the source of hopes and fears, motives and expectancies, rea- 
soning and creativity. And this theory is all we have. If we can make 
sense of the world at all, it is by interpreting our experience with 
the world in the light of our theory. The theory is our shield against 
bewilderment. 
As I look around my world, I distinguish a multiplicity of mean- 
ingful objects that have all kinds of complicated relationships with 
each other and with me. But neither these objects nor their interre- 
lations are self-evident. A chair does not announce itself to me as a 
chair; I have to recognize it as such. Chairs are a part of my theory. I 
recognize a chair when I decide that a chair is what I am looking at. 
A chair does not tell me that I can sit on it, or put my coat or books 
or feet on it, or stand on it to reach a high shelf, or wedge it against a 
door that I do not wish to be opened. All this is also part of my the- 
ory. I can only make sense of the world in terms of what I know al- 
ready. All of the order and complexity that I perceive in the world 
2. COMPREHENSION AND KNOWLEDGE 
15 
around me must reflect an order and complexity in my own mind. 
Anything I can’t relate to my theory of the world will not make sense 
to me. I am bewildered. 
The fact that bewilderment is an unusual condition for most of 
us despite the complexity of our lives is a clear indication that our 
theory of the world is very efficient. The reason we are usually not 
aware of the theory is that it works so well. Just as we take the air 
we breathe for granted until deprived of it, so we become aware of 
our dependence on our theory only when it proves inadequate, and 
the world fails to make sense. That we can occasionally be bewil- 
dered only serves to demonstrate how efficiently our theory usually 
functions. When were you last bewildered by something that you 
heard or read? Our theory of the world seems ready even to make 
sense of almost everything we are likely to experience in spoken 
and written language — a powerful theory indeed. 
And yet, when was the last time you saw a bewildered baby? In- 
fants have theories of the world too, not as complex as those of 
adults, but then children have not had as much time to make their 
theories complex. But children’s theories seem to work very well 
for their needs. Even the smallest children seem able most of the 
time to make sense of their world in their own terms; they rarely 
appear confused or uncertain. The first time many children run 
into a situation that they cannot possibly relate to anything they 
know already is when they arrive at school, a time when they may 
be consistently bewildered if they are confronted by situations that 
make no sense to them. Children are often denied credit for know- 
ing very much . But, in fact, most of our knowledge of the world — of 
the kinds of objects it contains and the way they can be re- 
lated — and most of our knowledge of language have been organized 
before we arrive at school. At age 5 or 6 the framework is there, and 
the rest is mainly a matter of filling in details. 
In chapter 121 look more closely at three quite remarkable char- 
acteristics of our theory of the word: that it is coherent (it all hangs 
together), consistent (it doesn’t need radical changes every day), 
and consensual (it is largely compatible with other people’s the- 
ory). For the remainder of this chapter, I talk a little more about 
how the theory is organized and then discuss how it is used so that 
we can comprehend the world. 
THE STRUCTURE OF KNOWLEDGE 
The system of knowledge that is our theory of the world may be re- 
garded as a structure just like any other theory or system of orga- 
nizing information, such as a library or an encyclopedia. 
16 
UNDERSTANDING READING 
Information systems have three basic components — a set of cate- 
gories, some rules for specifying membership of the categories, 
and a network of interrelations among the categories. I briefly ex- 
amine each component in turn. 
Categories 
To categorize means to treat some objects or events as the same yet 
as different from other objects or events. All human beings catego- 
rize, instinctively, starting at birth. There is nothing remarkable 
about this innate propensity to categorize, because living organ- 
isms could not survive if they did not treat some objects or events 
as the same yet as different from other objects and events. 
No living organism could survive if it treated everything in its ex- 
perience as the same; there would be no basis for differentiation 
and therefore no basis for learning. There would be no possibility 
of being systematic. Just as a librarian can’t treat all books as the 
same when putting them on the stacks, so all human beings must 
differentiate throughout their lives. In our culture at least, every- 
one is expected to be able to distinguish dogs from cats, tables 
from chairs, and the letter A from the letter B. 
But similarly, no living organism could survive if it treated every- 
thing in its experience as different. If there is no basis for similarity, 
there is still no basis for learning. Thus the librarian must treat 
some books as the same in some senses — so that all chemistry 
books are stacked in the same area — even though these books may 
differ in size, color, and author's name. In the same way everyone, 
in our culture at least, is expected to ignore many differences in or- 
der to treat all dogs as the same, all cats as the same, and many dif- 
ferent shapes like A, A, a, and a as the letter "a.” 
In other words, the basis of survival and of learning is the ability 
to ignore many potential differences so that certain objects 1 will be 
treated as the same yet as different from other objects. All objects 
that belong to one category are treated as the same yet as different 
from objects belonging to other categories. 
The categories that we all observe, which are part of our theories 
of the world, are quite arbitrary; they are not generally imposed on 
us by the world itself. The world doesn't force us to categorize ani- 
mals into dogs and cats and so forth — we could divide them up in 
Vrom this point on, I refrain from the cumbersome practice of talking all the time 
about "objects or events." But every reference to “objects" applies in general to "events" 
as well. 
2. COMPREHENSION AND KNOWLEDGE 
17 
other ways, for example, treating all green-eyed animals as the 
same, in contrast to those with other eye colors, or differentiating 
those over 15 inches in height from those under 15 inches. The li- 
brarian could organize books on the basis of the color of their cov- 
ers, or their size, or the number of pages. But we can’t usually 
invent categories for ourselves — hence the qualification “in our 
culture at least” in previous paragraphs. The reason we divide ani- 
mals on a cat and dog basis and not on the basis of size or eye color 
is that the categories we have are part of our culture. Categories are 
conventions. To share a culture means in part to share the same 
categorical basis for organizing experience. Language reflects the 
way a culture organizes experience, which is why many of the 
words in our language are a clue to the categories in our shared the- 
ories of the world. We have the words dog and cat but not a word 
for animals with green eyes or less than 1 5 inches in height. When 
we have to learn new categories, the existence of a name in the lan- 
guage is often the first clue that a category exists. 
Not that words are prerequisites for the establishment of catego- 
ries. Quite the reverse — categories can exist for which we have no 
names. I can easily distinguish certain mottled brown and grey 
birds that come to my garden every morning, but I do not know a 
name for them. To know a name without an understanding of the 
category that it labels is meaningless. In fact, the existence of a cate- 
gory is a prerequisite for learning how to use words, because 
words label categories rather than specific objects. What we call a 
dog is any individual animal that we put in the category with the 
name dog. 
The category system that is part of our theory of the world is es- 
sential for making sense of the world. Any situation that we cannot 
relate to a category does not make sense; we are bewildered. Our 
categories, in other words, are the basis of our perception of the 
world. Perception must be regarded as decision making. We “see” 
what we decide we are looking at, which means the category to 
which an experience is allocated. If I see a chair in front of me, then 
I must have a category for chairs in my theory of the world, and I 
must have decided that what I am looking at belongs in that cate- 
gory. If I can see the word cat when I read, then I must have a cate- 
gory for that word quite independent of my knowledge of its name 
or possible meanings, just as I must have categories for the letters 
c, a, and t if I can distinguish those letters in the word. Interest- 
ingly, we cannot see things in more than one category at a time; it is 
not possible to see the letters c, a, and t and the word cat simulta- 
neously in the visual configuration cat, which is why children may 
18 
UNDERSTANDING READING 
find learning to read more difficult if they are required to concen- 
trate on the individual letters in words. Usually you only see what 
you are looking for and remain quite unaware of other possibili- 
ties. If I ask you to read the address 410 LION STREET, you will 
probably not notice that the numerals IO in 410 are the same char- 
acters as the letters IO in LION. When you look for the category of 
numerals you see numerals, and when you look for the category of 
letters you see letters. Even now that you are aware of what I am do- 
ing, you cannot look at IO and see both letters and numerals simul- 
taneously, any more than you can see the faces and the vase 
simultaneously in the ambiguous illustration of Fig. 2.1. We can 
make decisions about only one category at a time in relation to a 
single visual configuration (although we could see the face and vase 
in Fig. 2 . 1 simultaneously if they did not share a common contour ) . 
And if there is no category to which we can relate an object to which 
we are exposed, we can make no decision at all; the world will not 
make sense to us. We must have categories in order to make deci- 
sions, categories that embrace not only sights and sounds but also 
tastes, odors, feelings, and sensations, as well as many kinds of 
events, patterns, and relationships. 
Rules for Category Membership 
Categories in themselves are not enough. The category “chemistry 
books” is useless if a librarian has no way of recognizing a chemis- 
try book when encountering one, just as a child can make no use of 
the information that there are cats and dogs in the world without 
some notion of how to distinguish one from the other. A child who 
can recite the alphabet has established a set of 26 categories but 
may not be able to recognize a single letter. For every category that 
we employ there must be at least one way of recognizing members 
of that category. Every category must have at least one set of rules, a 
FIG. 2.1. Ambiguous visual information. 
2. COMPREHENSION AND KNOWLEDGE 
19 
specification, that determines whether an object belongs in that 
category. Sometimes a single category may have more than one set 
of rules — we can distinguish an object as an orange by its appear- 
ance, feel, smell, and taste. We can recognize the letter a in a num- 
ber of different guises. But just as we must have a category for every 
object we can distinguish in the world, so we must have at least one 
set of rules — a list of significant attributes or distinctive features — 
for allocating that object to a particular category. These are not 
usually rules that we can put into words. Knowledge of this kind is 
implicit — we can only infer that we have the categories or rules by 
the fact that we can make use of them. 
The question of what constitutes the rules that differentiate the 
various categories that we employ in reading and language gener- 
ally demands a good deal of attention in later chapters — especially 
when we see that “teaching” is often little more than telling children 
that a category exists, leaving them to discover for themselves what 
the rules are. 
Category Interrelations 
Rules permit the categories in a system to be used, but they don’t 
ensure that the system makes sense. A library doesn’t make sense 
simply because all the chemistry books are stacked in one place 
and all the poetry books in another. What makes a library a system 
is the way in which the various categories are related to each other, 
and this is the way our personal systems of knowledge make sense 
as well. 
It is impossible to list all the different interrelations among the 
categories in our theory of the world. To do so would be to document 
the complexity of the world as we perceive it. Everything that we 
know is directly or indirectly related to everything else, and any at- 
tempt to illustrate these relationships risks becoming interminable. 
For example, consider an onion. We know what that particular 
object is called — in more than one language perhaps — and also the 
names of several kinds of onion. All these are relations of the par- 
ticular object to language. We also know what an onion looks, feels, 
smells, and tastes like, again perhaps in more than one way. We 
know where an onion comes from — how it is grown — and we prob- 
ably have a good idea about how it gets to the place where we can 
buy one. We know roughly what onions cost. We know how an on- 
ion can be used in cooking and we probably know some other uses 
as well. We may know half a dozen different ways of cooking onions 
(with different names), and we certainly know a number of things 
20 
UNDERSTANDING READING 
that can be eaten with onions. We know a number of instruments 
for dealing with onions — knives, graters, and blenders, for exam- 
ple. We not only know what we can do with onions, we also know 
what onions can do to us. We know people who love onions and 
people who hate them; people who can cook them and people who 
can’t. We may even know something about the role of onions in his- 
tory. One enormous ramification of our knowledge of onions is re- 
lated to the fact that we can call them by more than one name. An 
onion can also be called a vegetable, which means that everything 
we know about vegetables in general applies to onions in particu- 
lar. Indeed, every time we relate an onion to something else — to a 
knife, a frying pan, or a particular person — then we discover that 
what we know about onions is part of what we know about knives, 
frying pans, and people. There is no end. 
Many cognitive interrelations pertain to the system of language 
that is such an important part of everyone's theory of the world. One 
complex set of interrelations is called syntax and describes how ele- 
ments of language should be related to each other in speech and 
writing. Syntax enables us to put words together in ways that are 
“grammatical” (although people whose language is often character- 
ized as “ungrammatical” don’t normally lack syntactic knowledge; 
they observe different conventions). Another set of interrelations is 
called semantics, concerned with the way language is related to the 
world at large or rather, to our perception of the world. The seman- 
tic richness of words indicates to some extent the complexity we per- 
ceive in the world (Whorf, 1956), or at least how easily we can talk 
about it (Hunt & Agnoli, 1991). Knowledge of language must also in- 
clude extensive understanding of the conventional ways in which 
language and other systems of communication are used on particu- 
lar occasions, sometimes referred to as pragmatics or as semiotics. 
And a good deal of our knowledge of the world is actually held in the 
form of language, in verbal descriptions of things that we know. Our 
heads can also contain a host of propositions, ranging from simple 
facts (Paris is the capital of France, two times two is four), through 
proverbs and other compact bundles of ideas or common sense, to 
complex verbal formulas and even entire segments of prose or po- 
etry. All of this verbal prior knowledge can become available to us at 
relevant times to help us to comprehend and even to bring about 
particular sets of circumstances. 
Schemes, Scenarios, and Stories 
Many important sets of cognitive relations relate to places and 
scenes with which we are familiar. We know the spatial organiza- 
2. COMPREHENSION AND KNOWLEDGE 
21 
tion of familiar landscapes and locations — the beach where we 
played as children, the family living room, or our first classroom. 
We are quickly aware if something in a familiar setting has changed 
(even if we cannot immediately determine what exactly is differ- 
ent). In addition, our theory also contains many more symbolic 
representations, such as maps (which is the primary way most of 
us understand the geography of the world) and diagrams. 
Our theories also contain (or can construct) extensive represen- 
tations of more general patterns or regularities that occur in our 
experience. These representations are metaphorically called 
schemes (or occasionally schemas or schemata). Most of us have 
a complex generic scheme for what classrooms are like, for exam- 
ple. We can recognize and make sense of classrooms we have never 
been in before, just because they contain familiar arrangements of 
familiar elements. Our cognitive structures similarly include 
schemes of department stores and restaurants, for example, 
which enable us to make sense of new experiences and to behave 
appropriately. Many experiments have demonstrated that our abil- 
ity to recognize scenes and to remember them depends on the ex- 
tent to which they conform to our expectations of what such scenes 
should be like, to the schemes that we already possess. 
Readers develop and require a large number of spatially orga- 
nized schemes related to the way in which books and other kinds 
of written texts are organized. Among such schemes are those of 
specific genres — newspapers are not set out in the way that maga- 
zines, novels, or textbooks are. All of these schemes, or specifica- 
tions for various kinds of text, are conventional. The appearance 
and organization of a book or a newspaper can vary considerably 
from one community or culture to another, and their schemes 
have to be known to us if we are to make sense of them. Other con- 
ventional rules of written discourse structure include organiza- 
tion into paragraphs, chapters, or sections, with titles and other 
kinds of headings, which readers as well as writers have to ob- 
serve and expect. 
The examples of schemes I have so far given have all been spa- 
tial, the way things are laid out, primarily for comprehension visu- 
ally. But we also possess innumerable schemes for other sense 
modalities — for arrangements of sounds, tastes, smells, and a va- 
riety of tactile sensations, many of them closely related to each 
other and to patterns of events in the visual world. 
Many of our most important schemes are laid out in time: they 
have a serial or temporal organization. Time and change are essen- 
tial aspects of the way we perceive the world — how otherwise could 
22 
UNDERSTANDING READING 
we understand language, music, or a football game? Schemes that 
have a temporal as well as a spatial basis Eire often referred to as 
scenarios or scripts. A depEirtment store script sets out expected 
and conventional patterns of behavior for ourselves and for others 
when we go out to shop, even when we are purchasing unfamiliar 
items in stores we have never been to before. An absence or mis- 
match of scripts can result in confusion, embarrassment, Eind mis- 
understanding. Collectively, scripts, scenarios, and schemes are 
sometimes referred to as event knowledge (Nelson, 1986). 
Knowledge of relevant schemes is obviously essential if we are to 
read any kind of text with comprehension. A child who doesn’t 
have a scenario about farming is unlikely to understand a story 
about farming or references to farming in a textbook. But there are 
special kinds of language schemes that readers particularly re- 
quire. If we are readers, or if we hope to become readers, our theo- 
ries of the world must include story schemes , specifications of 
how stories are organized and how they unfold. We must know that 
stories comprise particular kinds of plots, characters, and epi- 
sodes. How well a story is understood and remembered depends 
on how well it conforms to conventional schemes for stories and on 
how well the reader is familiar with those schemes. 
The complexity of cognitive structure is indeed astounding. Our 
prior knowledge resists all efforts to catalog it or to reduce it to a 
few simple categories. Attempts to “simplify” its organization or 
operation can only mislead, especially if made the basis of instruc- 
tionsil or diagnostic practices in education. The enormous cogni- 
tive power of every individual is frequently overlooked if there is an 
emphasis on “needs” or “disabilities.” Our ability to make sense of 
the world, like our ability to remember events, to act appropriately, 
and to predict the future, is determined by the complexity of the 
knowledge we already possess. 
THE DYNAMICS OF COGNITIVE STRUCTURE 
Cognitive structure, our personal theory of the world, may so far 
have seemed rather a crowded and static place, not very different 
in essence from a collection of facts and procedures. But the theory 
is dynamic, Eind not just in the sense that it is constantly being 
added to and modified, particularly during that lively period of in- 
tense exploration and learning we call childhood. We can do much 
more with the theory of the world than use it to make sense of expe- 
rience. We csin live in the theory, in worlds that exist only in the 
imagination. Within the theory we ceui imagine and create, testing 
2. COMPREHENSION AND KNOWLEDGE 
23 
provisional solutions to problems and examining the conse- 
quences of possible behaviors. We can explore new worlds of our 
own and can be led into other worlds by writers and artists. 
But the aspect of imagination with which we will be most con- 
cerned is more mundane, although at first encounter it may sound 
quite exotic. We can use the theory of the world to predict the fu- 
ture. This ability to predict is both pervasive and profound, be- 
cause it is the basis of our comprehension of the world, including 
our understanding of spoken and written language. Reading de- 
pends on prediction. 
The Pervasiveness of Prediction 
Everyone predicts — including children — all the time. Our lives 
would be impossible, we would be reluctant even to leave our beds 
in the morning, if we had no expectation about what the day might 
bring. We would never go through a door if we had no idea of what 
might be on the other side. And all our expectations, our predic- 
tions, can be derived from only one source, our theory of the world. 
We are generally unaware of our constant state of anticipation 
for the simple reason once again that our theory of the world works 
so well. Our theory is so efficient that when our predictions fail, we 
are surprised. We don’t go through life predicting that anything 
might happen — indeed, that would be contrary to prediction, and 
in that case rhinoceros could surprise us. The fact that something 
always could take us by surprise — like the word rhinoceros a few 
words ago — is evidence that indeed we always predict and that our 
predictions are usually accurate. It is always possible that we could 
be surprised, yet our predictions are usually so appropriate that 
surprise is a very rare occurrence. When was the last time you were 
surprised? 
We drive through a town we have never visited before, and noth- 
ing we see surprises us. There is nothing surprising about the 
buses and cars and pedestrians in the main street; they are pre- 
dictable. But we don’t predict that we might see anything — we 
would be surprised to see camels or submarines in the main street. 
Not that there is anything very surprising or unpredictable about 
camels or submarines in themselves — we would not be surprised 
to see camels if we were visiting a zoo or to see submarines at a na- 
val base. In other words, our predictions are very specific to situa- 
tions. We don’t predict that anything will happen, nor do we predict 
that something is bound to happen if it is only likely to happen (we 
are no more surprised by the absence of a bus than we are by the 
24 
UNDERSTANDING READING 
presence of one), and we predict that many things are unlikely to 
happen. Our predictions are remarkably accurate — and so are 
those of children. It is rare to see a child who is surprised. 
The Need for Prediction 
Why should we predict? Why not expect that anything could hap- 
pen all the time, and thus free ourselves from any possibility of 
surprise? I can think of three reasons. The first reason is that our 
position in the world in which we live changes constantly, and we 
are usually far more concerned with what is likely to happen in 
the near and distant future than we are with what is actually hap- 
pening right now. An important difference between an experi- 
enced driver and a novice is that the experienced driver is able to 
project the car into the future and the novice’s mind is more 
closely anchored to where the car is now — when it may be too late 
to avoid accidents. The same difference tends to distinguish expe- 
rienced readers from beginners, or from anyone having difficulty 
with a particular piece of reading. In fluent reading aloud, the eye 
is always ahead of the voice, checking for possible obstacles to a 
particular understanding. Readers concerned with the word di- 
rectly in front of their noses will have trouble predicting, and they 
will have trouble comprehending. 
The second reason for prediction is that there is too much am- 
biguity in the world; there are too many ways of interpreting just 
about anything that confronts us. Unless we exclude some alter- 
natives in advance, we are likely to be overwhelmed with possibil- 
ities. Of the many things I know about onions, I don’t want to be 
concerned with the fact that they are pulled from the ground or 
that they bring my cousin George out in spots if all I want is gar- 
nish for a hamburger. What I see is related to what I am looking 
for, not to all possible interpretations. Words have many mean- 
ings — table can be several kinds of verb as well as several kinds of 
noun — but there is only one meaning that I am concerned with, 
that I predict, if someone tells me to put my books on the table . All 
the everyday words of our language have many meanings and of- 
ten several grammatical functions — table, chair, house, shoe, 
time, walk, open, close — but by predicting the range of possibili- 
ties that a word is likely to be, we are just not aware of the poten- 
tial ambiguities. 
The final reason for prediction is that there would otherwise be 
far too many alternatives from which to choose. It takes time to 
make decisions about what the eyes are looking at, and the time 
2. COMPREHENSION AND KNOWLEDGE 
25 
that is required depends on the number of alternatives. We take 
longer to decide that we are looking at the letter A when it could be 
any one of the 26 letters of the alphabet than when we are told in 
advance that it is a vowel. We take longer to identify a word in isola- 
tion compared with a word in a meaningful sentence. The fewer the 
alternatives, the quicker is the recognition. If there are too many al- 
ternatives confronting the eyes, then it is much harder to see or to 
comprehend. 
Prediction is the core of reading. All of our schemes, scripts, and 
scenarios — our prior knowledge of places and situations, of written 
discourse, genres, and stories — enable us to predict when we read 
and thus to comprehend, experience, and enjoy what we read. Pre- 
diction brings potential meaning to texts, reducing ambiguity and 
eliminating in advance irrelevant alternatives. Thus, we are able to 
generate comprehensible experience from inert pages of print. 
Prediction is not reckless guessing, nor does it involve taking 
chances by betting everything on the most likely outcome. We don't 
go through life saying “I’ll see a bus round the next corner,” or “The 
next word I read will be rhinoceros.” We predict by opening our 
minds to the probable and by disregarding the unlikely. Here is a 
formal definition: Prediction is the prior elimination of unlikely 
alternatives. It is the projection of possibilities. We predict to re- 
duce any uncertainty we might have, and therefore to reduce the 
amount of external information that we require. Our theory of the 
world tells us the most probable occurrences, and we decide 
among those remaining alternatives until uncertainty is reduced to 
zero. And we are so good at predicting only the most likely alterna- 
tives that we are rarely surprised. 
Put more informally, prediction is a matter of asking specific 
questions. We don’t ask "What is that object over there?” but “Is that 
something I can put my books on?” or whatever we want to do. We 
don’t look at a page of print with no expectation about what we shall 
read next; instead we ask, “What is the hero going to do?” “Where is 
the villain going to hide?” and "Will there be an explosion when liq- 
uid A is mixed with powder B?” And provided the answer lies within 
the expected range of alternatives — which it usually does if we are 
reading with comprehension — then we are not aware of any doubt 
or ambiguity. We are neither bewildered nor surprised. 
Prediction and Comprehension Related 
Now at last prediction and comprehension can be tied together. 
Prediction means asking questions, and comprehension means 
26 
UNDERSTANDING READING 
being able to get some of the questions answered. Comprehension, 
as I have said, is the absence of confusion. As we read, as we listen 
to someone talking, as we go through life, we are constantly asking 
implicit questions, and if we are able to find answers to those ques- 
tions, then we comprehend. The person who doesn’t comprehend 
how to repair a bicycle is the one who can’t ask and find answers to 
such questions as “Which of these nuts and bolts goes where?” at 
appropriate times. And the person who doesn’t comprehend a 
book or newspaper article is the one who can’t find relevant ques- 
tions and answers concerning the next part of the text. There is a 
flow to comprehension, with new questions constantly being gen- 
erated from the answers that are sought. This flow, especially in the 
imagination, is a significant part of what is usually regarded as 
thought. 
Such a view of comprehension differs from the way the word is 
often used in school. So-called comprehension tests in school are 
usually given after a book has been read and as a consequence are 
more like tests of memory. (And because the effort to memorize can 
drastically interfere with comprehension, the test may destroy 
what it sets out to measure.) If I say that I comprehended a certain 
book, it doesn’t make sense to give me a test and argue that I didn't 
understand it, although I may have understood it differently from 
the test constructor. And a high score on a test certainly would not 
convince me that I had really understood a book or a speaker if my 
feeling is that I did not. 
The very notion that comprehension is relative, that it depends 
on the questions that an individual happens to ask, is not one that 
all educators find easy to accept. Some want to argue that you may 
not have understood a book even if you have no unanswered ques- 
tions at the end. They will ask, "But did you understand that the 
spy’s failure to steal the secret plans was really a symbol of human- 
ity’s ineluctable helplessness in the face of manifest destiny?” And 
if you say “No, I just thought it was a jolly good story,” they will tell 
you that you didn't really comprehend what the story was about. 
But basically what they are saying is that you were not asking the 
kind of questions they think you should have asked. 
THINKING AND "META-THINKING" 
Thinking has become a focus of attention for many educators. In 
part, this concentration of interest has been generated by cognitive 
scientists trying to develop models of thought that might serve in 
the development of “thinking computers,” and also by cognitive 
2. COMPREHENSION AND KNOWLEDGE 
27 
psychologists involved with human thinking who nevertheless 
want to simulate or test their theories on computers. As a result, 
thinking has tended to be fragmented into distinct clusters of "in- 
formation-processing” procedures, more appropriate to the pro- 
grammed sequential operations of electronic technology than to 
humans whose flow of thought and actions are based primarily on 
their intentions, interests, and values. 
It could also be argued that another reason for the sudden con- 
cern about thinking is the tendency to fragment reading and read- 
ing instruction into packages of decontextualized “basic skills,” 
none of which particularly engage thinking. 
Yet reading cannot be separated from thinking. Reading is a 
thought-full activity. There is no difference between reading and 
any other kind of thought, except that with reading, thought is en- 
gendered by a written text. Reading might be defined as thought 
stimulated and directed by written language. This entire book 
could be considered to be a disquisition on thinking from a reading 
point of view. 
Particular characteristics of the thinking ideally engaged in by 
readers must be separated into two categories, not always clearly 
recognized. The first is the thinking involved in the act of read- 
ing — such as drawing appropriate inferences in order to compre- 
hend — and the other is thinking that is a consequence of reading, 
that might transpire in concurrent or subsequent reflection. Read- 
ing involves no special kind of thought that is not already displayed 
in other aspects of mental life. 
Thinking should not be regarded as a set of specialized pro- 
cesses that are superimposed on the organization of knowledge, 
our theory of the world discussed earlier in this chapter. Think- 
ing — including reading — is not a distinct faculty or set of skills, dif- 
ferent from comprehension, prediction, or imagination; rather, it 
is our theory of the world in action (Smith. 1 990). The theory of the 
world constantly modifies itself in relation to our current concerns 
and state of affairs. Thinking is the normal operation of our dy- 
namic theory of the world. The flow of thought is powered by our 
intentions and expectations, guided by the consequent experience. 
It is creative and constructive, not passive and reactive. 
All aspects of thought that language distinguishes can be seen as 
the dynamic interplay of cognitive relationships. Reasoning usu- 
ally refers to relationships within a series of statements or states of 
affairs: the way one thing follows another. Inference involves rela- 
tionships between particular statements or states of affairs and 
some more general circumstances, and problem solving relates 
28 
UNDERSTANDING READING 
existing states of affairs to desired states. Classification, categori- 
zation, concept formation, and other manifestations of what are 
sometimes called higher order or abstract thinking all impose and 
examine relationships among statements or states of affairs. The 
terms I have italicized are just words, not different kinds of think- 
ing. We are not doing different things when we reason, draw infer- 
ences, or solve problems; they only appear different because of the 
context in which they are done or the consequences of doing them. 
We are constantly engaged in relational thought, in our everyday 
transactions of comprehending and learning about the world 
around us. What differs among individuals is not so much the gen- 
eral ability to think as the possibility of demonstrating aspects of 
thought on particular occasions. Three constraints bear on how 
well individuals can appear to think on particular occasions, none 
of them dependent on the acquisition of specialized or exotic skills. 
All of us on occasion find ourselves in situations in which we are 
unable to think — especially in “educational” contexts — but this is 
not because no one has taught us specific “thinking skills." 
Constraints on Thinking 
The first constraint on thinking is prior knowledge. Like language, 
thought always has a subject. And just as we cannot talk or write 
competently if we don’t know what we are talking about, so it is not 
possible to demonstrate thought in any way if we don't understand 
what we are expected to think about. If I have difficulty understand- 
ing an article on nuclear physics, it is not because I'm unable to 
draw conclusions, make inferences, follow arguments, or solve 
problems, but because I don’t know enough about nuclear physics. 
The second constraint on thinking is disposition. Philosopher 
John McPeck (1981) asserts that the “judicious suspension of be- 
lief,” which is his definition of critical thinking, is a disposition 
rather than a skill. Whether or not we take something for granted, 
whether we challenge other people's assertions or question our 
own opinions in the light of new evidence, depends on individual 
propensities to behave in those ways, not on the acquisition of 
abilities that can be developed through instruction or even prac- 
tice. Dispositions can be innate, aspects of personality we are 
born with, or they can be the result of experience — “once bitten, 
twice shy.” 
And finally, whether or not anyone will exercise thought, particu- 
larly of the critical variety, depends on possession of the authority 
to do so. Challenging conventional thought or other people's opin- 
2. COMPREHENSION AND KNOWLEDGE 
29 
ions, or even drawing one’s own conclusions, is not something ev- 
eryone is in a position to do, certainly not in every situation. In 
many institutions and in many patterns of personal relations, the 
authority for engaging in thought of a significant nature (as op- 
posed to accepting or providing “right answers”) is not distributed 
equally. Thinking can upset applecarts. 
Metacognition 
One other topic related to thinking that receives considerable at- 
tention from educational researchers is metacognition — literally 
“cognition about cognition” or thought about our own thought. 
Metacognitive processes are presumed to take place when we 
think about our own thinking, for example, when we reflect on 
whether we know something, whether we are learning, or whether 
we have made a mistake. 
Some researchers are inclined to regard metacognitive pro- 
cesses as yet another special set of skills which have to be taught 
and learned. On the other hand, children learn many things, in- 
cluding talking and much of literacy, without awareness of learn- 
ing. And we are usually aware when we are confused by 
something, or when we don’t know something at a time when 
some knowledge is personally relevant and important to us. Meta- 
cognition could be regarded as a newfangled label for the old- 
fashioned concept of “reflection.” 
ISSUES 
There is a widespread assumption that thinking can be broken 
down into sets of skills — like comprehension skills or information 
acquisition skills — that must be taught, rather than being a natural 
human capacity. This is the instructional issue again, with propo- 
nents of the skills point of view arguing that children won’t be able 
to think without specific instruction. Their opponents argue that 
children (and adults) simply need experience of seeing others exer- 
cising thought in various circumstances and opportunity to do the 
same themselves. The role of prediction in reading has also been 
disputed (see Notes). 
The metaphorical nature of terms like scheme and script is of- 
ten overlooked. The use of computers as an analogy for the brain in 
educational theorizing has led to a belief that schemes, scripts, and 
cognitive structure itself are “data” or “programs" that people must 
“acquire" through instruction. The alternative view is that the dis- 
30 
UNDERSTANDING READING 
dilation of experience into forms that might be conceptualized as 
schemes or scripts is natural for humans of any age. 
SUMMARY 
(Terms printed in italics in the summaries are key terms that can 
be found in the glossary. ) 
Nonuisual information, long-term memory, and prior knowl- 
edge are alternative terms for describing cognitive structure, each 
individual’s theory oj the world. The theory includes schemes, or 
generalized representations of familiar settings and situations, es- 
sential in all understanding and remembering. The theory of the 
world is the source of comprehension, as we continually generate 
and examine possibilities about situations in real and imaginary 
worlds. The basis of comprehension is prediction, the prior elimi- 
nation of unlikely alternatives. Predictions are quesdons that we 
ask, and comprehension is receiving relevant answers to those 
quesdons. If we cannot predict, we are confused. If our predictions 
fail, we are surprised. And if we have no interest or uncertainty as a 
basis for prediction, we are bored. Thinking — including metacog- 
nition, or “thinking about thinking” — is not a special set of skills 
but constant reflective activity, subject only to constraints of indi- 
vidual prior knowledge, disposition, and authority. 
Notes to chapter 2 begin on page 240 covering: 
Knowledge and constructivism 
Prediction 
Categories 
Schemes 
The narrative basis of thought 
Thinking 
Spoken and Written 
Language 
Language, naturally, constitutes a substantial part of any person’s 
theory of the world. It obviously plays a central role in reading. The 
present chapter is concerned with language from a number of per- 
spectives, including the relationships between the sounds of speech 
and their meaning, between the printed marks of written language 
and their meaning, between productive aspects of language (talking 
and writing) and receptive aspects (listening and reading), and be- 
tween spoken and written language. The chapter also refers briefly 
to grammar and to other conventions of language. 
All of these aspects of language are relevant for an understand- 
ing of reading, yet each is a complex area of study in its own right. It 
isn’t possible here to study any topic to the same theoretical depth 
as the professional linguist or cognitive psychologist would, but 
fortunately such detail is also unnecessary. The basic insights that 
a student of reading must grasp are relatively few and easy to ex- 
plain and to demonstrate. These insights, however, are not always 
part of the general awareness of educators in the field of reading; 
they are widely disregarded in many instructional programs and 
materials and in a good deal of reading research, so that they may 
appear to be new and even unfamiliar ideas. 
31 
32 
UNDERSTANDING READING 
For example, one basic but neglected insight is that the state- 
ments that people utter or write do not convey meaning in any sim- 
ple fashion. Meaning is not contained within the sounds of speech 
or the printed marks of writing, conveniently waiting to be ex- 
tracted or decoded, but rather must be constructed by the listener 
or reader. As a consequence, an understanding of reading requires 
a more complex theory of comprehension than one that simplisti- 
cally assumes that meaning takes care of itself if a reader names in- 
dividual words correctly. Most of this chapter is concerned with the 
fundamental issue of how language is comprehended. 
TWO ASPECTS OF LANGUAGE 
Surface Structure and Deep Structure 
There are two quite different ways of talking about language, 
whether spoken or written. On the one hand, you can talk about its 
physical aspect, about characteristics that can be measured, such 
as the loudness, duration, or pitch of the sounds of speech, or the 
number, size, or contrast of the printed marks of writing. All of these 
observable characteristics of language that exist in the world around 
us may be called surface structure. They are the part of language ac- 
cessible through the ears and eyes. Surface structure is a useful 
term because it is not restricted to a particular form of language, ei- 
ther spoken or written. Surface structure is the “visual information” 
of written language — the source of information that is lost to the 
reader when the lights go out — but it is also a part of spoken lan- 
guage — the part that is lost when a telephone connection is broken. 
On the other hand, there is a part of language that can neither be 
directly observed nor measured, and that is meaning. In contrast to 
surface structure, the meaning of language, whether spoken or writ- 
ten, can be referred to as deep structure. The term is apt. Meanings 
do not lie at the surface of language but far more profoundly in the 
minds of users of language, in the intentions of speakers and writers 
and in the interpretations of listeners or readers. 
These two different aspects of language, the physical surface 
structure and the meaningful deep structure, can in fact be com- 
pletely separated; it is quite possible to talk about one without ref- 
erence to the other. We can say that someone is talking loudly or 
softly, or fast or slowly, without reference to what is being said. We 
can say that a line of print is five inches wide, or has eight charac- 
ters to the inch, without fear that someone will contradict us by 
saying that we haven’t understood the meaning of the text. Con- 
3 . SPOKEN AN D WRITTEN LANGUAG E 
33 
versely, meaning is not directly affected by the form of the surface 
structure. If we are told that a certain city will host the next Olym- 
pic Games, we can't reply that it depends on whether the speaker’s 
source of information was spoken or written. The truth of an utter- 
ance is not related to its loudness or the number of repetitions. 
All of this may seem obvious, trite even, but the distinction be- 
tween the surface and the deep structure of language is crucial for 
an understanding of reading for one simple reason: the two aspects 
of language are separated by a chasm. Surface and deep structures 
are not opposite sides of the same coin; they are not mirror reflec- 
tions of each other. They are not directly and unambiguously re- 
lated. In technical jargon, there is no one-to-one correspondence 
between the surface structure of language and meaning. Meaning 
lies beyond the mere sounds or printed marks of language and 
can’t be derived from surface structure by any simple or mechanis- 
tic process. 
One way of exemplifying this absence of a one-to-one relation- 
ship between the two aspects of language is by showing that dif- 
ferences can occur in surface structure that make no difference 
to meaning and that there can be differences in meaning that are 
not represented in surface structure (Miller, 1965). For exam- 
ple, here are some radically different surface structures that 
don’t correspond to radical differences in meaning: (a) the cat is 
chasing a bird ; (b) a bird is chased by the cat ; (c) a warm- 
blooded feathered vertebrate is pursued by the domesticated 
feline quadruped ; (d) le chat chasse un oiseau. Four quite dif- 
ferent sequences of marks on paper, but all represent (in general 
terms at least) the same meaning. When we try to say what words 
mean, all we can do is offer other words (a synonym or a para- 
phrase) that reflect the same meaning. The actual meaning al- 
ways lies beyond words. It makes sense to say that bachelor 
means (or conveys the same meaning as) unmarried man, but it 
doesn’t make sense to ask what the meaning is that bachelor and 
unmarried man have in common. Alternative verbal definitions 
or descriptions simply compound the problem. They are addi- 
tional surface structures. 
On the other hand, it isn't difficult to find individual surface 
structures that have at least two possible meanings or interpreta- 
tions — for example, flying planes can be dangerous; visiting pro- 
fessors may be tedious; the chicken was too hot to eat; she runs 
through the spray and waves; he enjoys talking with old men 
and women (all women?); Cleopatra was rarely prone to talk 
(and Mark Antony wasn't inclined to argue). 
34 
UNDERSTANDING READING 
The examples just quoted represent a particular kind of ambi- 
guity, namely puns. Puns are often difficult to comprehend imme- 
diately — you may not have seen at first glance the alternative 
meanings in all of the previous examples — and therein lies an im- 
portant theoretical issue: Why are we so rarely aware of the poten- 
tial ambiguity of language? It’s not just puns but every possible 
sequence of words in our language, and just about every individual 
word for that matter, that is a source of potential misinterpreta- 
tion. To understand why we are so rarely aware of the multiple 
meanings that might be attributed to surface structure of our lan- 
guage, we must look at a more basic question. If there is this chasm 
between surface structure and deep structure, how then is lan- 
guage comprehended in the first place? The question is of consid- 
erable relevance to reading, because if meaning isn’t immediately 
and unambiguously given by the surface structure of speech, then 
there’s no point in expecting a reader to “decode” written language 
to speech in order for comprehension to occur. Speech itself needs 
to be comprehended, and print can’t be read aloud in a compre- 
hensible way unless it is comprehended in the first place. Written 
language doesn’t require decoding to sound in order to be compre- 
hended: the manner in which we bring meaning to print is just as 
direct as the manner in which we understand speech. Comprehen- 
sion is the same for all aspects of language. Reading aloud is more 
complex, and therefore more demanding, than silent reading. 
The Trouble With Words 
How then is language understood, whether spoken or written? The 
answer is not that we put together the meaning of individual words 
and thereby understand entire sentences. For a start, it seems 
doubtful whether words can be said to exist in spoken language at 
all. Scientific instruments can’t isolate the beginning and ending of 
many sounds — or even words — that we hear as quite separate. The 
actual flow of speech is relatively continuous and smoothly chang- 
ing, like the colors of a rainbow, and the segmentation into distinct 
sounds and words is largely something that listeners contribute. 
You can get some indication of this by listening carefully while you 
repeat the two words west end. You will probably find that if you in- 
troduce any pause at all in the utterance, it will be between the /s/ 
and the /t / — that actually you are saying “wes tend” rather than 
“west end. ” Of course, English speakers would never think that 
you really said wes tend. But that is only because they speak the 
language and are able to work out — and hear — the sounds you 
3. SPOKEN AND WRITTEN LANGUAGE 
35 
thought you were producing. The fact that you need to know a lan- 
guage in order to be able to hear it properly becomes apparent 
when you listen to a foreign language. Not only can you not distin- 
guish what the distinctive sounds of the language are, you can't 
even distinguish the number of words in an utterance. Speakers of 
other languages have exactly the same trouble with English. 
The very existence of words may be an artifact of the writing sys- 
tem. At least in writing we can provide a definition of a word — as 
something with a space on either side. Children learning to talk ei- 
ther produce groups of words that they use as one long word — 
allgone, drinkamilk, gowalk , alot — or else they use single words 
as entire sentences — drink, tired, no. Beginning readers often can- 
not say how many words are in a sentence, either spoken or writ- 
ten. They need to be experienced readers to understand the 
question. 
Words and Meanings 
Another reason why it is difficult to argue that the meanings of sen- 
tences are made up of the meanings of words is that it would ap- 
pear that words often get meaning by virtue of occurring in 
sentences. In fact, it is difficult to see what meaning a word in isola- 
tion might have. Even nouns, which might seem the easiest class of 
words to account for, present difficulties. It is certainly far from 
true that every object has one name and every word one meaning. 
Every object has more than one name. The family pet. for example, 
can be called an animal, a canine, a dog, a retriever, “Rover, ” and 
a variety of other titles, including, of course, “family pet” and. “ that 
slavering brute. ” What is the “real name” of the animal? There isn't 
one. The appropriate name for the speaker to use depends on the 
listener and the prior knowledge of the listener. In talking to a 
member of the family, the name “Rover” is adequate, or simply “the 
dog"; on other occasions no single word would be adequate, and 
the name would have to be qualified as “that brown dog over 
there ” or “the large retriever.” Everything depends on the knowl- 
edge of the listener or reader and the alternatives from among 
which Rover has to be distinguished. The same animal will be de- 
scribed in different ways to the same person depending on the 
characteristics of other dogs that are around. What then does a 
word like dog mean? The dictionary tells us that it is “any of a large 
and varied group of domesticated animals related to the fox, wolf, 
and jackal.” But that surely is not the meaning of dog in the sen- 
tence “ Beware of the dog, ” let alone such expressions as hot dog, 
36 
UNDERSTANDING READING 
top dog, putting on the dog, dirty dog, dog-eared, dog tired, or go- 
ing to the dogs. 
All the common words of our language have a multiplicity of 
meanings, with the most common words being the most ambigu- 
ous. To test what I say. just look up a few words in the dictionary. 
Words that come most immediately to mind — the everyday words 
like table, chair, shoe, sock, dog, field, file, take, look, go, run, 
raise, narrow — require many inches and even columns of "defini- 
tion.” Less familiar words lik e frugal, gossamer, or tergiversation 
are disposed of in a crisp line or two. Prepositions, which are among 
the most common words of our language, have so many different 
senses that they are sometimes maligned as having “function” rather 
than "content.” But it makes a difference whether something is in 
the box rather than on the box; prepositions have meanings — in 
great number. The linguist Fries (1945), for example, located in the 
Oxford Dictionary no fewer than 39 separate senses for at and by, 
40 each for in and with, and 63 for of. You would surely have no diffi- 
culty in understanding my saying that I found the book by Charles 
Dickens by the tree by chance; I shall return it by mail by Fri- 
day— but it would be difficult for you to tell me the meaning (or 
meanings) of the word by on all or any of its five occurrences. Prepo- 
sitions in context seem full of meaning, but in isolation it is impossi- 
ble to say what the meaning might be. That is why it is so difficult to 
translate prepositions from one language to another. 
It’s not necessary to pursue the argument about the nature of 
words or their meaning, because it is quite clear that sentences 
aren’t understood by trying to put together meanings of individual 
words. The man ate the fish and The fish ate the man comprise 
exactly the same words, yet they have quite different meanings. A 
Maltese cross is not the same as a cross Maltese, nor is a Venetian 
blind the same as a blind Venetian. A house that is pretty ugly is 
not exactly ugly but is certainly not pretty. Obviously, the words in 
all these examples don't combine in any simple fashion to form the 
meaning of the whole. 
Perhaps then word order is the key — the word cross has one 
meaning before Maltese and another meaning after it. But words 
in the same position can represent different meanings — compare 
the final words of She went down the drive and She went for a 
drive — and words in different positions in a sentence may reflect 
the same meaning. Words that often seem to have a similar mean- 
ing, such as look and see may suddenly acquire quite different 
meanings when an identical prefix is added, as in overlook and 
oversee. 
3. SPOKEN AND WRITTEN LANGUAGE 
37 
A common explanation is that grammar makes the difference; 
syntax (word order) is the bridge between the surface structure of 
language and its deep structure. But the problem with this point of 
view is that often it is impossible to say what a word’s grammatical 
function is before the sentence in which it occurs is understood. 
Formal grammar, of the kind often taught in school, is a descrip- 
tive grammar. It never helps anyone to say anything or to under- 
stand what anyone else is saying. It is simply a way of talking about 
surface structure. Grammar, in other words, doesn’t reveal mean- 
ing; meaning must precede grammatical analysis. Consider again 
the familiar words that I have been citing like table, chair, shoe, 
sock, file, dog, field, take, go, and so forth; these are all words that 
not only have a multiplicity of meanings but also a variety of gram- 
matical functions. To ask anyone to identify such words when they 
are written in isolation is pointless because they can commonly be 
both noun and adjective, or noun and verb, or adjective and verb, 
or perhaps all three. How do we understand a simple statement 
like Open the empty bottle ? It is not by taking into account the fact 
that open is a verb and empty an adjective, because in the equally 
comprehensible sentence Empty the open bottle, the two words 
switch grammatical roles without any difference in surface struc- 
ture. This complicated ambiguity of language is one reason it is dif- 
ficult to program computers to translate language or make 
abstracts, even when they are equipped with a “dictionary” and a 
“grammar.” Computers lack the knowledge of the world that is re- 
quired to make sense of language. Thus, a computer is befuddled 
by over a dozen different possible meanings of a simple expression 
like time flies. Is time a noun, or a verb (as in time the racehorses ), 
or an adjective (like the word fruit in fruit flies)? Is flies a noun or a 
verb? A computer is said to have interpreted out of sight, out of 
mind as invisible and insane. 
Not only is it impossible to state the grammatical function of in- 
dividual words outside of a meaningful context, it can also be im- 
possible to state the grammatical structure of entire sentences 
without prior understanding of their meaning. Most English teach- 
ers would parse The onions are planted by the farmer as a passive 
sentence, because it contains the three grammatical markers of the 
passive form — the auxiliary are, the participle ending -ed, and the 
preposition by. But the sentence The onions are planted by the 
tree is not a passive sentence, although its surface structure would 
appear to contain the appropriate three grammatical markers. 
Meaning determines the grammatical structure of these sentences, 
not the surface structure markers. In fact. The onions are planted 
38 
UNDERSTANDING READING 
by the farmer need not be a passive sentence, because it is just as 
ambiguous grammatically as She was seated by the minister- the 
grammar depends on the meaning. 
In other words — and this must be the answer to the question at 
the beginning of this section — there is only one way in which lan- 
guage can be understood, that print can be comprehended, and 
that is by having meaning brought to it. 
Comprehension Through Prediction 
The statement that language is understood by having meaning 
brought to it shouldn’t be taken to imply that any particular utter- 
ance or sentence can mean anything. Usually there would be some 
broad general agreement about the main implications of state- 
ments. at least when they are made in real-world situations. If 
someone in an elevator remarks “It’s raining outside," few people 
would want to claim that it could mean that the streets are dry. And 
by the same argument, the meanings that listeners and readers 
bring to language can’t be wild guesses; the usual broad general 
agreement about implications makes the reckless attribution of 
meaning unlikely as well. If most people seem to be in agreement 
about the kind of meaning that can be attributed to a particular se- 
quence of words, then some explanation must be found as to why 
such agreement exists. 
The explanation that can be offered should not be unfamiliar. 
Language tends to be understood in the same way on similar occa- 
sions because listeners or readers must have a pretty good idea 
about the meaning that was intended in the first place. To be more 
precise, meaning is brought to language through prediction, which 
you will remember from the previous chapter means the prior 
elimination of unlikely alternatives. Prediction doesn’t mean stak- 
ing everything on one wild guess (which would indeed run the risk 
of frequent error), nor does it mean that the precise meaning is 
known in advance (which would of course make attention to lan- 
guage unnecessary in the first place). Prediction simply means that 
uncertainty in the listener or reader is limited to a few probable al- 
ternatives, and provided that information can be found in the sur- 
face structure of the utterance to dispose of the remaining doubt — 
to indicate which predicted alternative is appropriate — then com- 
prehension occurs. 
Prediction is the reason we aren't normally overwhelmed by the 
possible number of alternatives in language; there are actually very 
few alternatives in our minds at any time that we are comprehend- 
3. SPOKEN AND WRITTEN LANGUAGE 
39 
ing what is being said. And prediction is the reason we are so rarely 
aware of ambiguity: We expect what the writer or speaker is likely 
to say and just don’t contemplate alternative interpretations. We in- 
terpret The thieves decided to head for the bank in one way if we 
know they were sitting in a car and in another way if they were 
swimming in a river. When language is comprehended, the recipi- 
ent is usually no more aware of possible ambiguity than the pro- 
ducer. The first interpretation that comes to us is the one that 
makes the most sense to us at the particular time, and alternative 
and less likely interpretations will not be considered unless subse- 
quent interpretations fail to be consistent or to make sense, in 
which case we realize our probable error and try to recapitulate. 
One interpretation usually satisfies us, provided it makes sense, so 
we don’t waste time looking for a second. This is the reason that 
puns may be difficult to see, and also why they may be mildly irri- 
tating. We don’t expect to find more than one meaning for the same 
sequence of words. 
As I indicated in the previous chapter, there is nothing remark- 
able or particularly clever about this process of prediction; it goes 
on all the time. Prediction enables us to make sense of all the 
events in our daily lives. And we are no more aware of our predic- 
tions when we read than we are at any other time for the simple rea- 
son that our predictions are usually so good. We are rarely 
surprised because our predictions rarely let us down, even when 
we read a book for the first time. 
What exactly do we predict when we read? The fundamental an- 
swer is meaning, although of course we may look at particular 
words to find evidence that will confirm or disconfirm particular 
interpretations. In other words, we look for sense. A number of de- 
tailed predictions may be made and tested simultaneously — and 
constantly modified — as we make our way through the text. Every 
specific prediction, however, no matter how detailed and transient, 
will be derived from our more general expectations about where 
the text as a whole might be leading. 
Some Practical Implications 
The preceding discussion should make it clear that it is misleading 
if not inaccurate to regard reading as a matter of “following the text” 
or to say that a listener “follows" the meaning of a speaker. Lan- 
guage is understood by keeping ahead of the incoming detail. By 
having some expectation of what the speaker or writer is likely to 
say, by making use of what we know already, we protect ourselves 
40 
UNDERSTANDING READING 
against being overwhelmed by irrelevant information. We avoid the 
confusion of ambiguity and succeed in bridging the gap between 
the surface structure of the text and the writer’s intention. 
It is easy to demonstrate how we keep ahead of any words that 
we identify as we read. Ask a friend to turn out the light while you 
read aloud, so that you are suddenly deprived of visual informa- 
tion, and you will find that your voice is able to continue “reading” 
another four or five words. Your eyes were a second or more 
ahead of the point your voice had reached when the lights went 
out. This phenomenon is known as the eye-voice span, a term 
that is rather misleading because it might suggest that we need 
more than a second to organize in speech the sounds of the partic- 
ular word that we are looking at. But this is incorrect. We don’t 
need a second to identify a word; the difference in time is not so 
much a reflection of how far thought lags behind the eye as of how 
far thought is ahead of the voice. We use our eyes to scout ahead so 
that we can make decisions about meaning, and thus about indi- 
vidual words, in advance. Indeed, the eye-voice span exists only 
when we can make sense of what we read. If we read nonsense — 
dog lazy the over jumps fox quick the rather than The quick 
brown fox jumps over the lazy dog — then eye and voice tend to 
converge on the same point, and the eye-voice span disappears. 
The span, in fact, reflects rather precisely the sense that we make 
of text, because it tends to extend to the end of a meaningful 
phrase. The four- or five-word span is merely an average. If the 
lights go out just as we are about to read ... and drove off into the 
night, we are likely to continue aloud as far as off or as far as 
night, but not stop at into or the. 
It is because a reader must keep ahead of the text that it is so 
hard for children to learn to read from material that doesn’t make 
sense to them or is so disconnected and fragmentary that predic- 
tion is impossible. Reading is similarly much more difficult for 
children who have been taught that they should get the words right 
rather than try to make sense of what is being read. Not only is “get- 
ting the words right" harder and slower unless meaning is brought 
to the text in the first place, identifying each successive word on the 
line one after the other will not, in itself, give meaning. Reading is 
never a matter of decoding the surface structure of print to the sur- 
face structure of speech; the sounds will not make sense of their 
own accord. 
The difficulty of many high school “problem readers" is not that 
they have failed over the years to learn how to sound out words 
correctly, nor that they are careless about getting every word 
3. SPOKEN AND WRITTEN LANGUAGE 
41 
right, but rather that they read one word at a time as if meaning 
should be the last concern. They expect that meaning will take 
care of itself, although this is the reverse of the way in which sense 
is made of reading. 
WHY WRITTEN AND SPOKEN LANGUAGE DIFFER 
Obviously, spoken language and written language are not the same. 
It is not difficult to detect when a speaker reads from a prepared 
text or when a passage that we read is the unedited transcription of 
spontaneous talk. Speech and print aren’t different lan- 
guages — they share a common vocabulary and the same grammati- 
cal structures — but they have different conventions for using 
vocabulary and grammar. It shouldn’t be considered surprising or 
anomalous that differences exist between spoken and written lan- 
guage; they are generally used for different purposes and ad- 
dressed to different audiences. The way we talk always varies 
depending on the reason we are talking and the circumstances we 
are in, and the same variation occurs with written language. 
Written language is different from spoken language for the good 
reason that spoken language has adapted itself to being heard but 
written language is more appropriately read. Written language is 
not made more comprehensible by being translated into “speech.” 
The Specialization of Language 
To understand why such a specialized adaptation of spoken and 
written language might have come about, consider the different de- 
mands the two aspects of language make on their recipients. There 
is, for example, the obvious fact that the spoken word dies the mo- 
ment it is uttered and can only be recaptured if held in the lis- 
tener’s fallible memory or as the result of a good deal of mutual 
inconvenience as the speaker recapitulates. Even tape recording 
does little to mitigate the essential transience of speech in contrast 
to the facile way in which the eyes can move forward (and back- 
ward) through a written text. The reader has control over time, can 
decide which parts of the text to attend to, the order in which they 
will be selected, and the amount of time that will be spent on them. 
In other words, spoken language may make considerable demands 
on attention which written language does not. 
On the other hand, written language might seem to place a far 
greater burden on memory — on what we already know about lan- 
guage and the world — than our everyday speech. To bring meaning 
42 
UNDERSTANDING READING 
to spoken language, all we need consider may be the circum- 
stances in which an utterance is made. Much of our everyday spo- 
ken language is directly related to the immediate situation in which 
it is uttered. We may pay little attention to the actual words the 
speaker is using. The relevance of the utterance is as ephemeral as 
the words themselves — “Pass the salt, please” — and there is little 
involvement of memory. Written language, by contrast, generally 
depends on nothing but what we can and do remember. 
There is the question of how the meaning of language is verified: 
How do we confirm that information we are receiving is likely to be 
true, that it makes sense, or that we are understanding it correctly? 
What is the source of the predictions that can cut through all the 
ambiguity inherent in language so that we make the most reason- 
able and reliable interpretation? For the kind of everyday spoken 
language I have been talking about, the answer is simple: Look 
around. Any uncertainty we have can probably be removed by what 
we know already of the speaker's nature, interests, and likely in- 
tentions. Anyone who asks “Pass the salt" is probably looking at the 
salt. But the language of texts offers no such shortcuts. There is 
only one final recourse if we aren’t sure of what we have read, and 
that is to return to the text itself. For verification, for disambig- 
uation, and to avoid error, a difficult and possibly unfamiliar kind 
of ability is required. That is the ability to pursue a line of thought, 
looking for internal consistencies, and evaluating arguments. Both 
the source and the test of many of the changing predictions that are 
necessary for the comprehension of written language must lie in 
the text itself, informed by the more general expectations that read- 
ers bring from their prior knowledge. The text determines what the 
actual alternatives might be and whether they have been success- 
fully predicted. For that reason alone, spoken language and written 
language can rarely be the same. 
A Different Difference in Language 
The previous section began by considering some fairly obvious dif- 
ferences between spoken and written language. But it quickly be- 
came necessary to acknowledge that the general distinctions being 
made were between a particular kind of speech, the “everyday spo- 
ken language directly related to the situation in which it is uttered,” 
and a particular kind of written language, namely, that of “texts.” 
To present the complete picture I must now explain that there is an- 
other distinction that slices right across the spoken-written di- 
mension. 
3. SPOKEN AND WRITTEN LANGUAGE 
43 
The issue basically concerns how the spoken or written words 
are selected and organized in the first place. Obviously words are 
rarely produced at random. There is usually a necessity or reason 
for every word we use, related in part to the intention we want to 
fulfill and to the language with which we propose to fulfill it. Both of 
these considerations, the reason for saying something and the lin- 
guistic vehicle we select for saying it, place considerable con- 
straints on what we say and write. But there is a third important 
constraint, the one with which I am at present most concerned, 
and that is the environment in which the language is produced. To 
use language rather arbitrarily myself for the moment, I shall use 
the term situation to refer to the physical environment in which 
words are produced — the position in which you are standing when 
you say something or the location in which written words happen 
to be written or printed — and I shall use the word context to refer 
to the language environment in which spoken or written words oc- 
cur. The context for the word context in the previous sentence, for 
example, is all the other words in that sentence and in the chapter 
as a whole. A distinction must be made between situation-depend- 
ent and context-dependent language. 
Situation-dependent speech is the spoken language with which 
infants first become familiar, and it is the basis on which they begin 
all their learning about language. By situation-dependent I mean 
that the speech is directly related to the physical and social situa- 
tion in which it is uttered. If someone says “Pass the salt, please," 
then there is likely to be some salt around , a person who would like 
some salt, and another person in a position to pass it. If someone 
says “It’s raining again,” then the streets are likely to be wet. Given 
the physical situation and the speaker’s intentions, it would not be 
possible to say anything much different, like “It’s raining again,” if 
the speaker really wanted the salt passed, or vice versa. 
The fact that such speech and the situation in which it is uttered 
(including the speaker’s intentions) are closely related is the basis 
of children’s language learning. It is the way in which language is 
first comprehended and verified. Normally we think that such lan- 
guage describes the situation in which it occurs. We hear someone 
say “Pass the salt, please,” and we can construct the probable cast 
of characters and the major props, even if we can t see what is going 
on. But the clues also work in the other direction: the situation can 
make sense of the language. A child who doesn't yet understand 
what “Pass the salt, please” means can work it out from the situa- 
tion in which the language is uttered. Indeed, if someone actually 
said “It’s raining again,” and another person passed the speaker 
44 
UNDERSTANDING READING 
the salt, then the child might assume that "It’s raining again” meant 
“Pass the salt.” 
This strategy of using the situation for clues to how unfamiliar lan- 
guage works is not uncommon. We all tend to use the strategy when 
confronted by someone speaking a foreign language or in any other 
situation when we don’t understand what is being said. If the waiter 
says something incomprehensible, we look to see if we are being of- 
fered the menu, the wine, or the bill. Because most of the meaning 
and the verification of such language rests in the particular situation 
in which it is uttered, it tends to be elliptical and brief— ‘‘Coffee?” 
“Thanks" — without much evident grammar about it. 
Just as there is a good deal of situation-dependent spoken lan- 
guage in the environment of most children, who use it to make their 
first sense of speech, so there is a good deal of situation-dependent 
written language in most contemporary environments, which chil- 
dren can again employ to make sense of reading. 1 am referring 
now to the written language of signs and labels, the ubiquitous lan- 
guage that we find on every product we buy, festooned around every 
store, on every wrapper, on every street sign, and as part of every 
television commercial. We do not have a convenient word in our 
language for such situation-dependent writing, so I refer to it as 
ambient print. It functions in exactly the same way as situation-de- 
pendent speech, because it is also closely tied to the situation in 
which it occurs. The situation provides learners with a clue to its 
meaning, and it can’t be arbitrarily changed or moved without los- 
ing its sense. The word toothpaste tells the reader what is in the 
tube and the contents of the tube tell the learner what the printed 
word is likely to be. Indeed, some children think the printed word 
Crest says toothpaste just as they may think that the sign McDon- 
ald's says hamburgers. (And the advertisers of Crest and McDon- 
ald’s will tell you that these are indeed what their brand names are 
supposed to say. ) Certainly a child who finds he is brushing his 
teeth with shampoo or that she has poured herself a bowl of deter- 
gent doesn’t need an adult to point out the reading error. If you 
don’t understand what the sign or label means, look at the situa- 
tion in which it occurs. Like situation-dependent speech, ambient 
print is tied to where it occurs. The sign Exit can't be moved to the 
middle of the wall because we are tired of seeing it over the door. 
Like situation-dependent speech, such print tends to be elliptical 
and independent of grammar. The situation takes the place of com- 
plexity of language. 
Quite different from the situation-dependent written language of 
signs and labels, however, is the continuous written language of 
3. SPOKEN AND WRITTEN LANGUAGE 
45 
texts, of books, magazines, newspapers, and every kind of refer- 
ence material. This language is more complex and has to be. It 
doesn't derive or convey its meaning from the situation in which it 
occurs; its location offers no clues to its sense. If you don’t under- 
stand something in a newspaper or novel, it won’t help you to look 
at where the text is located in the room or into the face of the person 
who gave it to you. The appropriate meaning of the text remains 
constant, whether you look at it now in the room or an hour later in 
the street. It can’t be elliptical. It is grammatically complex. 
Despite this independence from the specific location in which it 
is produced and read, however, there is just as much necessity 
about the written language of texts as there is about situation-de- 
pendent writing and speech. The writer is still not free to produce 
words arbitrarily or at random. A writer can’t decide to make the 
next word rhinoceros or platitudinous just because it is a long time 
since these words were last used. Now, however, the constraints on 
the words are determined by just two things: the topic that the 
writer is talking about (what the writer wants to say) and the lan- 
guage the writer is employing (how the writer wants to say it). In 
other words, all of the constraints on what is written occur within 
the context of the language itself. That’s why I call such language 
context dependent. Not only does the intricate texture of a written 
context give every component word its meaning, but because of the 
redundancy in the text, it is usually possible to replace a word that 
has been left out or to work out the meaning of an unfamiliar word . 
For anyone learning to read, the ability to make use of contextual 
clues to meaning is crucial. But the clues embedded in the immedi- 
ate language environment of context-dependent writing are not the 
same as those in situation-dependent writing (which is one reason 
why children who have essential insights into reading from ambi- 
ent print may still have difficulty with continuous text). Nor are the 
clues of context-dependent writing those of situation-dependent 
speech (which is one reason why being read to is such an enor- 
mous advantage in learning to read). 
The particular requirements of context-dependent writing have 
so impressed some theorists (Goody & Watt, 1972; Havelock, 1976: 
Olson, 1977) that they have argued that written language has intro- 
duced a whole new mode of thought to our basic human repertoire 
of intellectual skills. But context-dependent written language is not 
unique. Not all our spoken language is of the "everyday,’’ situa- 
tionally verifiable kind that has been discussed. Some of our spoken 
language can be as abstract, argumentative, and unrelated to the cir- 
cumstances in which it is comprehended as an article in a scientific 
46 
UNDERSTANDING READING 
journal. There is also context-dependent speech. “Academic lan- 
guage” is a particular variety of this detached way of talking. Experi- 
ence in reading makes it possible for us to understand abstract 
spoken language, which in its form is more like writing than every- 
day speech. But the contrary also applies; by hearing such speech, a 
child becomes better equipped to read. 
THE ORGANIZATION OF TEXTS 
Texts are not all written in the same way, even if they are written dif- 
ferently from the way spoken language is produced. Instead, differ- 
ent kinds of text are organized and presented in distinctive and 
characteristic ways. Each kind of text has its own conventions of 
layout, typography, and style — called genre schemes — which dis- 
tinguish it from other genres or kinds of text. Novels don’t have the 
same genre schemes as textbooks, poems, newspapers, letters, or 
telephone directories. E-mails and web sites have developed their 
own genre schemes. Furthermore, various kinds of text may have 
quite different genre schemes in different cultures. Newspapers or 
novels produced in France for French readers aren’t written and 
presented in the same way as those for readers elsewhere. Often we 
can see that texts are different from culture to culture, even though 
we can’t read the language. There is nothing particularly logical or 
necessary about specific genre schemes — they could be different, 
as they usually are from one culture to another — but they have be- 
come conventional where they are employed, and they serve their 
purposes because they are conventional. 
Genre schemes help both readers and writers. Their character- 
istic forms help readers by giving them a basis for predicting what 
a text will be like, that a novel will be constructed in a particular 
way, that a scientific article will follow a certain format, that a letter 
will observe typical conventions. Readers become so accustomed 
to the genre schemes of the texts with which they are familiar that 
they assume they are natural, inevitable, and universal. A text that 
is produced differently in a different culture may be regarded as 
odd. Genre schemes also help writers (if they know them), because 
they provide a framework for organizing what writers want to say 
and more importantly for anticipating and respecting what readers 
are likely to expect. Genre schemes facilitate communication. 
Similarly, every kind of text, and every form of spoken-language 
interaction, too, has characteristic internal relationships, called 
discourse structures, which are again largely arbitrary and acci- 
dental but which serve their purposes because they are conven- 
3. SPOKEN AND WRITTEN LANGUAGE 
47 
tional. Discourse structures in conversation tell us when we may 
interrupt (at the end of a sentence) and when we may not; they pro- 
tect speakers from interruption while allowing others the opportu- 
nity to take a turn. In written language, readers can expect writers 
to observe conventional discourse structures and writers can ex- 
pect readers to understand them. The structures form another ba- 
sis for prediction. The manner in which chapters and paragraphs 
are arranged in books is a matter of discourse structure. 
Even stories have their conventions, whether they are spoken or 
written. These conventional ways of telling a story, of relating se- 
quences of events, are known as story grammars. They are the 
framework upon which various characters, plots, motives, and 
resolutions are linked in related episodes and represented in ways 
that will be intelligible. If a story makes sense to us, if it sounds like 
a story, this is not just because the story is told in an appropriate 
way but also because we know the appropriate way in which stories 
are told, at least in our culture. Stories must reflect the story 
schemes with which readers are familiar, if writers and readers are 
to connect. 
The important function served for readers by all these conven- 
tional and characteristic structures of texts is underlined by evi- 
dence that the structures are the basis of our comprehension of 
texts. If we don’t know the relevant structures, then we won’t un- 
derstand the text, or our reading of it will be distorted. Re- 
searchers have noted that readers’ comprehension of texts is 
similar to the structures in the text itself. Ask people to recapitu- 
late what they have read in a story, and they will tend to do so with 
the same structural form as the story rather than with the same 
words or even “their own” words. Novice readers have been shown 
to insert into their retelling of stories conventional aspects that 
have been omitted in the telling but that are part of their own story 
grammars. They put more into the story than was in it originally, 
because this is their way of making sense of stories. 
It is also the structures in the head rather than those in the text 
that determine our memory for texts; they are the forms in which 
texts are remembered. Discourse structures and story grammars 
are part of our own cognitive structure, part of the way we orga- 
nize our knowledge of the world (and therefore a reason that read- 
ing is important — it provides us with new frameworks for 
perceiving the world and organizing experience). The more we 
can anticipate and employ the formal structures that an author 
uses, the more we can understand and remember what we read, 
because the structures also form the basis of our understanding 
48 
UNDERSTANDING READING 
and remembering. And the more an author knows and respects 
the structures that the reader will predict, the more the text will 
be readable and memorable. 
None of this is new. The British psychologist Frederick Bartlett 
( 1932) demonstrated experimentally more than 70 years ago that 
the way stories are interpreted and remembered varies with the 
cultural backgrounds and expectations of their readers and listen- 
ers. Who would expect otherwise? But experiments and theoretical 
work into these matters demonstrate convincingly what in the past 
has perhaps been only intuitively obvious (or should have been in- 
tuitively obvious). 
Two brief qualifications must be added. First, the structures of 
texts should be seen as the basis for comprehension but not for 
comprehension itself. Some researchers assume that we have un- 
derstood a story if we can repeat large parts of it. But comprehen- 
sion is less a matter of being able to reproduce the facts in a text 
than of what one is able to do as a consequence of interacting with 
the text. You don’t prove that you have understood anything by re- 
peating it. Second, these structures that can be observed and ana- 
lyzed in the organization of texts aren’t structures that require our 
conscious attention. We don’t need to be able to talk about a partic- 
ular grammatical construction or other convention in order to un- 
derstand and use it. The knowledge that enables us to make sense 
of the world and of language is not knowledge of which we are 
aware, even if we are psychologists or linguists. 
There is no evidence that making text structures explicit im- 
proves comprehension, or that teaching such structures explicitly 
to children helps them understand. In fact, without the prior un- 
derstanding, such "explanations” are themselves meaningless. 
Children learn the structures by being helped to understand the 
texts in which the structures are employed. 
THE CONVENTIONS OF LANGUAGE 
There is one final characteristic of language, both spoken and writ- 
ten, situation-dependent and context-dependent, that I must em- 
phasize. It is that all language is conventional. Semiotics — an area 
of study that interests a number of reading researchers — is specifi- 
cally concerned with the nature of all different kinds of communi- 
cative conventions, their use, and how they develop. This is an 
enormous topic with multiple ramifications, but I must try to deal 
with it briefly, first by explaining what it means to say that all lan- 
guage is conventional, then why the statement is critical. 
3. SPOKEN AND WRITTEN LANGUAGE 
49 
All language is conventional in the sense that every aspect of lan- 
guage is a matter of chance and of mutual respect. All the various 
forms of language must work, they must fulfill a function, but the 
nature of the forms themselves is always arbitrary, a matter of his- 
torical accident; they could always be different. There is no partic- 
ular logic or necessity about the specific forms employed in any of 
the 6,000 or more different languages that exist in the world. That 
is what the word “conventional” means, arbitrary forms that could 
be different, functioning in the way they do because their form is 
mutually respected among the users of each language. 
The use of red to mean stop in traffic signs is a matter of conven- 
tion. The convention works because it is mutually accepted that red 
should mean stop (in those cultures where red means stop). But 
things could have worked out differently. Green could mean stop to- 
morrow, provided everyone agreed on the change. In some cultures, 
in certain circumstances, it is a mark of respect to remove your hat. 
In other cultures, the mark of respect is to keep it on. What makes 
removing your hat (or putting it on) a mark of respect has nothing to 
do with the act itself but with the mutual understanding that this is 
how the act is to be interpreted, the intention that it reflects. 
Every aspect of language is conventional, starting from the very 
sounds and meanings of words we use. In English, yes means yes, 
but this is only a convention. No could mean yes. In other lan- 
guages, other words mean yes. The same applies to all the words in 
every language. No one has a free choice with words. No one can 
call anything something different from everyone else in their lan- 
guage community, not if they want to be understood. 
Words are conventional, and so is grammar. Different languages 
have different grammatical structures and there is nothing more 
logical or rational or efficient about one grammar than another. All 
languages solve the same kind of problems, but they solve them in 
different ways. I have already noted that story grammars, discourse 
structures, and genre schemes are conventions. They could be dif- 
ferent, they are different in different languages and cultures, and 
they function despite their arbitrary nature because they are all mat- 
ters of mutual agreement among the people who employ them. 
There is an enormous range of conventions in language, many of 
which have not yet been mentioned. For example, there are the con- 
ventions of idiom. Language is far more than grammar and vocab- 
ulary (although a good deal of instruction in reading, in English, 
and in other languages, seems to assume that this is what language 
consists of) . Knowledge of grammar and vocabulary gives no one a 
mastery of language, either in producing or in understanding it. By 
50 
UNDERSTANDING READING 
far the greatest part of any language, the “working” part of it, is id- 
iom, the way people actually speak, and by definition idiom can’t be 
accounted for by vocabulary and grammar. Idiom is the way words 
in the vocabulary and structures in the grammar are actually used 
in a particular language community, and this usage is a complex 
and constantly changing system of conventions. Idioms usually 
can’t be translated word for word from one language to another. 
There are conventions of cohesion. Speech doesn’t consist of 
one statement after another, and paragraphs are more than a sim- 
ple succession of sentences. Statements and sentences are inter- 
locked; they cohere. I can say “J looked for John. But he had 
gone" but not “ But he had gone. I looked for John." I would have 
to change the sentences to something like “John had gone. I 
looked for him." The pronoun and the but are two of a number of 
cohesive devices that lock sentences together in English, but they 
are conventions because different languages cohere in different 
ways. You can’t change the order of sentences in any language 
without having to change the sentences themselves, at least not in 
meaningful text. (This is a useful way of finding out whether mate- 
rial prepared for beginning readers is meaningful. If the order of 
sentences can be arbitrarily changed without anyone noticing the 
difference, then they don’t make sense; they aren’t normally func- 
tioning language.) 
There are tremendously subtle and intricate conventions of 
language, both spoken and written, concerned with register. This 
term refers to the fact that you must choose and put your words 
together differently depending on the subject you are talking 
about, the person you are talking to, and the circumstances in 
which you are talking. You can’t speak a language unless you em- 
ploy the forms of vocabulary, grammar, idiom, and cohesion ap- 
propriate to the relevant register. Children quickly learn they 
must speak in one way to younger children, another way to their 
peers, another way to teachers, and another way to other adults. 
All of the differences of register are conventional; there is no in- 
trinsic logic about the particular form that comes to be appropri- 
ate at a particular time. You can’t carry your own conventions of 
language with you when you travel, not if you want to understand 
and be understood. Not even the nonverbal conventions of lan- 
guage, like how close you should stand to another person in con- 
versation or how long you should look someone directly in the 
eye, are consistent from one culture to another. 
Written language has its own substantial set of conventions. 
There are conventions of spelling, punctuation, letter formation. 
3. SPOKEN AND WRITTEN LANGUAGE 
51 
the size of handwriting or type, capitalization, paragraphing, page 
layout, and bookbinding — and in e-mails and web chat. All of these 
could be different, and all of them are, in other languages, other 
cultures. Every aspect of language is conventional. 
Why should I bother to point all this out? Because it is important 
to understand that language isn’t just vocabulary and grammar. Fa- 
miliarity with written language conventions is essential for readers 
and writers because conventions make prediction possible. The 
forms of particular conventions can’t be predicted; they vary by 
chance or historical accident from one language community to an- 
other, and they also change with time. But knowledge of what to ex- 
pect makes the conventions that will be used on particular 
occasions predictable. To be able to read a text, we must be able to 
anticipate the conventions that its writer will employ. This under- 
standing of the appropriate conventions, together with prior 
knowledge related to the subject matter, is the essential contribu- 
tion of readers to the act of reading. But the understanding must be 
mutual. To be comprehensible, the writer must anticipate and re- 
spect the conventions that the reader will predict. Conventions are 
the common currency of every language transaction. 
There is a tendency to think of language as “logical," as “ratio- 
nal," even as if it could be in our genes. But language is enor- 
mously complex, and all of its complexity is arbitrary and 
accidental. It could all be different. The implication of this is that 
no one ever learns language by sitting down and thinking about it, 
by anticipating what it will be like, or even by learning a few rules. 
Learning a language or learning to read involves learning a tre- 
mendous number of conventions. And these can't be learned by 
rule or by rote, nor are they instinctive in any way. They must be 
experienced, one at a time, in ways that are most meaningful for 
every learner. 
There is one other important implication of the conventional 
nature of language. It is that language is social in all its aspects. 
Language does things for people, and its particular conven- 
tions — the way it does things — are matters of social contract and 
social identification. We talk the way people around us talk — 
provided we can identify with that kind of person. We use lan- 
guage in the ways that it is used by the people around us, again 
provided we don’t see ourselves as different from them. Above 
and beyond all the technical aspects of reading discussed in this 
book and in many other books on the same topic, reading is a so- 
cial activity, learned (or not learned) in a social rather than an in- 
tellectual context. 
52 
UNDERSTANDING READING 
Language About Language 
An interesting question about all the complexities of language that I 
have discussed in this chapter concerns how much they need to be 
consciously known by a learner. Is it necessary for beginning read- 
ers to be instructed in the difference between surface structure and 
meaning, in the fine points of grammar, or in all the other essential 
conventions of language, both spoken and written? Should they be 
able to talk about language as well as be able to use it? 
There is a special word for language about language — the word is 
metalanguage. In a general sense, this entire chapter has been writ- 
ten in metalanguage, because it has been on the topic of language. 
More specifically, there are a number of metalinguistic terms that 
are frequently central to any discussion involving language — terms 
like noun, verb, word, syllable, phrase, and sentence. 
The word metalanguage may remind you of the word meta- 
cognition, which was introduced in chapter 2. Metacognition is 
thinking (or language) about thinking, just as metalanguage is 
language (or thinking) about language. And there is a controversy 
in psychology and educational research about how important 
ability in both metacognition and metalanguage is for learning to 
read and to write. 
Some researchers argue that children must be aware of their 
own learning processes and able to talk about specific aspects of 
spoken and written language, if they are to learn to read. Downing 
(1979), for example, asserted that children who don’t have meta- 
linguistic competence are in a state of “cognitive confusion” when 
someone tries to teach them about reading. (Downing himself uses 
technical metacognitive language here. Instead of saying that chil- 
dren maybe “in a state of cognitive confusion,” he could simply de- 
scribe them as “confused.”) Other theorists argue that children are 
obviously capable of learning without being able to talk about 
learning — how else would babies learn to talk in the first place? We 
have all learned many things in our lives without being able to talk 
about what we were learning or, indeed, being aware at the time 
that we were learning. Many people can read and write phrases and 
paragraphs without being able to provide a linguistic definition for 
them or to parse a sentence. 
As for being able to understand the language of language, it is 
also evident that many people learn to read without understanding 
the meaning of many metalinguistic terms. Indeed, terms like 
word, sentence, comma, and period have no meaning until we can 
read. They are not parts of spoken language, certainly not in any di- 
3. SPOKEN AND WRITTEN LANGUAGE 
53 
rect or conspicuous way. Like Moliere’s bourgeois gentilhomme 
we all speak prose without knowing we are doing so until someone 
points it out to us ( and explains the technical meaning of the word ) . 
Why then should knowledge of metalinguistic terminology be 
thought to be so critical in reading instruction? The explanation 
seems to be that children need to understand what teachers are 
talking about, and if teachers find it necessary to use metalin- 
guistic or metacognitive language, then children are in difficulty if 
they don’t understand such language themselves. Cognitive confu- 
sion is caused by instruction that is not comprehensible. Whether 
it is essential for classroom teachers to employ the abstract techni- 
cal language of linguistics and other specialized disciplines in 
teaching reading is another matter. 
ISSUES 
Many of the topics in this chapter are subject to endless dispute 
among specialists — including the nature of language, the role of 
grammar, and the meaning of meaning. Such issues have resisted 
solution or agreement for over 2,000 years, and there are no rea- 
sonable grounds for believing they can ever be resolved. The issues 
are abstract and need have little bearing on the teaching of reading. 
More central is the controversy over the “relationship” of written to 
spoken language, which comes up frequently in this book. Some 
people regard written language as “parasitical" on speech, or “un- 
natural” in some other way. Historical, linguistic, or philosophical 
speculation is unlikely to settle the arguments, and the convictions 
of various specialists reflect faith rather than infallibility. Another 
matter of preference rather than evidence concerns the nature of 
metalinguistic discourse (is language about language any different 
from language about anything else?) and its role in instruction. 
SUMMARY 
The sounds of language and the visual information of print are 
surface structures of language that do not represent meaning di- 
rectly. Meaning resides in the deep structure of language, in the in- 
tentions of speakers and writers, and in the interpretations of 
listeners and readers. Written language and spoken language are 
not the same, and language also differs to the extent that it is situa- 
tion-dependent or context-dependent. The basis of comprehen- 
sion is prediction, made possible by the complex conventional 
nature of language. 
54 
UNDERSTANDING READING 
Notes to chapter 3 begin on page 245 covering: 
Surface structure and deep structure 
Semiotics 
Discourse and genre 
Text organization and comprehension 
Some technical terms 
Speech, writing, and “language” 
More about words 
Information 
and Experience 
If reading is a natural activity, then literacy education should obvi- 
ously center on aspects of reading that are most natural to us. The 
most natural activity for human beings is to engage in interesting 
experience , the absence of which leads to boredom and with- 
drawal. But experience is not a topic that has much currency in ed- 
ucation, except for the absurd suggestion that some students don’t 
do as well as others because they haven’t had many experiences. In- 
stead the focus is on the information students are supposed to ac- 
quire. And the deliberate acquisition of information is not a 
particularly natural activity. People usually accumulate informa- 
tion without trying, in the course of engaging in interesting experi- 
ences. The interest is always in the experience, rather than in the 
information. The intentional acquisition of information, especially 
at the arbitrary behest of others, is one of the most tedious and un- 
natural activities anyone can engage in. 
Louise Rosenblatt made the crucial distinction as long ago as 
1978. She says there are two ways to read — for information or for 
experience — and it is easy to tell the difference between the two. 
When information is what we want, we are perfectly content to get it 
in any way we can. No one ever says, “Don’t tell me that telephone 
55 
56 
UNDERSTANDING READING 
number, I want the pleasure of looking it up for myself." But when 
we read for experience, said Rosenblatt, we are reluctant to be de- 
prived of even a moment. I can’t imagine anyone saying “I can’t be 
bothered to read the last chapter — tell me if the butler did it." Often 
we slow down as we near the end of a novel, as we might at the end 
of a good meal, to protract the experience. 
The distinction between information and experience is usually 
disregarded in education. Books or other texts that should be read 
for experience are treated only as sources of information. 
Rosenblatt said this is because it is easier to grade readers on the in- 
formation they might be expected to acquire than on the experience 
they might enjoy. She satirized such an approach in an article enti- 
tled “What Facts Does This Poem Teach You?" (Rosenblatt, 1980). 
It is often said that we live in an “information age” — but the word 
information is used very loosely. Usually it is taken to be synony- 
mous with “facts” or “data." Despite its vagueness and ambiguity, 
the word has become ubiquitous in education. Reading and learn- 
ing are both referred to as “the acquisition of information" and 
writing (and teaching) as its “transmission." But this is a vacuous 
misuse of the word information. 
Information has been given a precise technical definition that 
most people in education are not aware of, although it has enabled 
some aspects of reading to be accurately measured. I shall use the 
word in this strict technical sense when I examine how the visual 
system solves complex problems of identifying letters and words in 
print. On the other hand, the widespread use of the word in a more 
general sense in educational and psychological research — for ex- 
ample, in the characterization of the brain as an “information-pro- 
cessing device” — distorts rather than facilitates efforts to 
understand literacy and learning. And information, when the word 
is used in a more general sense, can’t be measured. 
First, I examine the technical definition of information and its 
relationship to another general term that can also be used techni- 
cally in a very precise sense — uncertainty. I also look at how infor- 
mation can be related to comprehension and to another important 
concept in reading, redundancy. I then refer to some limitations 
on the way in which individuals can make use of information and 
also on the contrast between information and experience. 
INFORMATION AND UNCERTAINTY 
We shouldn’t expect to be able to measure information the way we 
measure height and weight. Information can be found in a multi- 
4. INFORMATION AND EXPERIENCE 
57 
tude of guises — in marks on paper, facial expressions and other 
bodily gestures, the configuration of clouds, trees (and sometimes 
tea leaves), and in the sounds of speech. Obviously sources of in- 
formation do not have much in common, and neither do the chan- 
nels through which information passes. 
Consider the mutations of information when we listen to a 
broadcast recording of someone talking. What we hear begins as 
an intention in the speaker’s mind, represented in some deeply 
mysterious way in the flux of chemical and bioelectrical activities 
in the structures of the brain. This intention is then translated into 
bursts of neural energy, dispatched from the brain at different 
times, rates, and directions to the musculature of the jaw, mouth, 
lips, tongue, vocal cords, and chest, orchestrating the expulsion of 
breath in such a manner that distinctive pressure waves of con- 
trasting intensity and frequency radiate through the surrounding 
atmosphere. These fleeting disturbances in the molecules of the air 
cause the tiny diaphragm of a microphone to resonate in sympathy, 
triggering a flow of electrical energy along a wire quite unlike the 
corresponding patterns of neural energy in the nervous systems of 
the speaker or listener. Amplified and modulated, the electrical im- 
pulses from the microphone impress subtle combinations of mag- 
netic forces onto a plastic tape or etch wavy lines into a plastic disk, 
often after being “digitized.” Through further mechanical and elec- 
tronic incarnations, the information may then be diffused by radio 
transmission (perhaps diverting through the transistors of 
earth-orbiting satellites) before being reconstituted by an elec- 
tronic receiver and loudspeaker into airborne pressure waves that 
lap against the listener’s ear. And still the transformations are not 
done. The oscillations of the eardrum are conveyed to another res- 
onating membrane across a tiny bridge of three articulating 
bones — the hammer, the anvil, and the stirrup of the listener’s in- 
ner ear. And then, perhaps most bizarrely, a pressure wave pulses 
back and forth through liquid in the coiling canals of the inner ear, 
a labyrinth carved into the skull itself, where microscopic hair cells 
wave like reeds with the movement of the fluid in which they are 
contained. The roots of these fronds are the tiny beginnings of the 
mighty auditory nerve, and they generate the final relays of neural 
impulses that travel the hundreds of thousands of separate fibers 
of the auditory nerve, through half a dozen booster and trans- 
former stations in various recesses of the brain, to become at last 
subjective experiences of meaningfulness and sound. And this 
meaningfulness of acoustic events can be congruent with a subjec- 
tive meaningfulness and visual experience from perhaps the same 
58 
UNDERSTANDING READING 
words written down and reaching the brain by a completely differ- 
ent route through the eyes. How can all or even part of this com- 
plexity be identified and evaluated as “information”? 
On Making Decisions 
The technical answer is that information can be evaluated by look- 
ing at what it enables the “receiver” — the listener or the reader — to 
do. Information enables a person to make decisions, to choose 
among alternative possibilities or competing courses of action. In- 
formation can be assessed, not from its source or from the various 
forms that it can take during transmission, but from what it en- 
ables the receiver to do. Reading requires decisions, whether by a 
child striving to understand a brief story or by a scholar struggling 
to decipher an obscure medieval text. And anything that helps a 
reader to make a decision is information. 
Put into other words again, information reduces uncertainty. 
The change of focus from the facilitation of decisions to the reduc- 
tion of uncertainty may not seem to be much of a conceptual gain, 
but it permits information to be measured, or at least estimated 
comparatively. Information can’t be quantified directly, any more 
than the size or weight of a decision can be calculated directly. But 
it is possible to put a number to uncertainty and thus indirectly to 
the amount of information that eliminates or reduces that uncer- 
tainty. The trick is accomplished by defining uncertainty in terms 
of the number of alternatives confronting the decision maker. If 
you Eire confronted by a lot of alternatives, you have a great deal of 
uncertainty; there are many different decisions you could make. If 
you have fewer alternatives, it may be just as hard for you to make 
up your mind but theoretically your uncertainty is less: there are 
fewer alternative decisions you might make. The argument has 
nothing to do with the importance of the decisions to you. only with 
the number of alternatives. Theoretically, your uncertainty is the 
same whether you must decide for or against major surgery or for 
having your eggs scrambled or fried. The number of alternatives is 
the same in each case, and so therefore is your uncertainty. 
And now information can be defined more precisely: Informa- 
tion reduces uncertainty by the elimination of alternatives . Infor- 
mation, very reasonably, is anything that moves you closer to a 
decision. It is beside the point whether the decision concerns the 
identification of particular objects or events or the selection among 
various choices of action. Uncertainty and information are defined 
in terms of the number of alternative decisions that could be made 
4. INFORMATION AND EXPERIENCE 
59 
no matter what the alternatives are. However, it is easier to reach 
an understanding of these concepts if particular situations are 
taken as examples. 
Suppose that the information sought involves a single letter of 
the alphabet, say, someone’s middle initial. There are 26 letters of 
the alphabet, and the uncertainty requires a decision or choice 
among 26 alternatives. If the situation involves bidding in a bridge 
game, and the uncertainty concerns a partner’s strongest suit, then 
the number of alternatives will be 4. For the simple toss of a coin, 
the number of alternatives is 2; for the roll of a die, it is 6. Some- 
times the exact number of alternatives is not immediately appar- 
ent — for example, if a word rather than a letter is involved. But it 
may still be possible to determine when this indefinite amount of 
uncertainty has been reduced — for example, if a reader learns that 
a word begins with a particular letter or is of a particular length. Ei- 
ther of these pieces of information will reduce the number of alter- 
native possibilities of what the word might be. 
We can now return to the definition of information as the reduc- 
tion of uncertainty. Just as the measure of uncertainty is con- 
cerned with the number of alternatives among which the decision 
maker has to choose, so information is concerned with the number 
of alternatives that are eliminated. If the decision maker is able to 
eliminate all alternatives except one and thus can make a fully in- 
formed decision, then the amount of information is equal to the 
amount of uncertainty that existed. A bridge player who receives 
the information that the partner's strongest suit is red has had un- 
certainty reduced by one half; if the information is that the stron- 
gest suit is hearts, uncertainty is reduced completely. Similarly, a 
child who knows the alphabet well enough to decide that a particu- 
lar letter is a vowel has acquired information reducing uncertainty 
from 26 alternatives to 5. If the letter is correctly identified, then 
the information gained from the letter must have been equal to the 
original uncertainty. 
Some aspects of reading involve the acquisition of information in 
order to make decisions, to reduce uncertainty. For the visual identi- 
fication of letters and words and possibly some aspects of “reading 
for meaning,” uncertainty can be calculated and therefore also the 
amount of information required to make a decision. The exact num- 
ber of alternatives can be specified for letters, an approximate figure 
can be put to the number of words, but the number of alternatives 
for a meaning, if it can be estimated at all, must obviously be closely 
related both to the text being read and to the particular individual 
who is doing the reading. Examples are given in the notes. 
60 
UNDERSTANDING READING 
COMPREHENSION AND INFORMATION 
Comprehension can’t be measured in the way that some aspects of 
information can. Comprehension can’t be measured at all, despite 
constant educational efforts to do so, because it is not a quantity of 
anything. Comprehension doesn’t have dimension or weight; it is 
not incremental. Comprehension is not the opposite of uncertainty 
or even of ignorance, and therefore is not quantifiable as the accu- 
mulation of a number of facts or items of information. As I pro- 
posed in chapter 2, comprehension is the condition of relating 
whatever we are attending to in the world around us to knowledge, 
intentions, and expectations we already have in our head. 
We comprehend the situation that we are in if we are not confused 
by it, whether we are reading a book, repairing an appliance, or try- 
ing to find our way through the streets of an unfamiliar city. Absence 
of comprehension means not knowing what to do next or which way 
to turn. When we can’t comprehend, we can’t predict, we can’t ask 
questions. Absence of comprehension makes itself immediately evi- 
dent to the person involved and to anyone looking on, even if it can’t 
be measured. I don’t need a numerical test to detect confusion in 
myself or in others; bewilderment doesn’t conceal itself. If I see your 
brows furrow and your eyes glaze, then I know that all is not well 
with your comprehension. Without comprehension, there can be no 
reduction of uncertainty. The rote memorization of “facts" without 
comprehension is not uncertainty reduction. What we learn — with 
difficulty — under such conditions becomes informative to us only in 
the future, if by chance we should suddenly discover the sense it is 
supposed to make. Conversely, when uncertainty reduction is tak- 
ing place, there must be some comprehension. 
Comprehension doesn’t entail that all uncertainty is eliminated. 
As readers, we comprehend when we can relate potential answers 
to actual questions that we are asking of the text. We usually have 
unanswered questions when we read a newspaper — there wouldn’t 
be much point in reading it if we knew everything in advance. And 
we don’t need to have all our uncertainty reduced in order to com- 
prehend. In fact, as we acquire information that reduces uncer- 
tainty in some ways, we usually expand our uncertainty in other 
ways. We find new questions to ask. 
Absence of uncertainty is not a condition that we tolerate for very 
long; we find it boring. There is no “experience’’ to it. We seek un- 
certainty, provided we can keep it under control and clear of confu- 
sion. We comprehend when we can “make sense" of experience. 
Throughout this book, I usually refer to comprehension in reading 
4. INFORMATION AND EXPERIENCE 
61 
as “making sense of text," relating written language to what we 
know already and to what we want to know or experience. 
Errors and Noise 
Of course we may think we comprehend, and look as if we compre- 
hend, but nevertheless make a mistake. Comprehension doesn’t 
come with an unconditional guarantee. The way we understand 
something now may prove to be inappropriate later. To have a 
wrong idea about something is a constant possibility, but again not 
something that can be measured. And no one else can decide for us 
whether we are in a state of comprehension or confusion, though 
they can dispute whether we are in such a state for good reason and 
even help us to move from one state to the other. Comprehension 
and confusion are the consequences of how well we cope with the 
particular situation that we happen to be in, with whether or not we 
feel we know what to do next. What may be comprehensible to you 
may not be comprehensible to me. 
Similarly, what is information for you may not be information 
for me, if it doesn’t contribute to my comprehension. And such neg- 
ative information can have more than just a neutral, inconsequen- 
tial effect. It can be positively disruptive. 
A technical term for a signal or message that does not convey in- 
formation is noise. The term is not restricted to acoustic events but 
can be applied to anything that makes communication less clear or 
effective, such as poorly printed material, or inadequate illumina- 
tion, or distraction of the reader’s attention. The static that some- 
times interferes with television reception is visual noise. Any part 
of a text that a reader lacks the skill or knowledge to comprehend 
becomes noise. The present chapter offers information to readers 
who understand its language and general theme, but is noise for 
anyone else. And noise can’t easily be ignored; it is not an absence 
of information so much as interference that increases uncertainty. 
Because anything becomes noise if one lacks the familiarity or 
knowledge to understand it, reading may be intrinsically more dif- 
ficult for the novice than for the experienced reader. On the other 
hand, reading can be made so difficult for experienced readers that 
they behave no differently from beginners. 
The Relativity of Information and Comprehension 
What is commonly called information can’t always be measured. 
Facts are often called information, but the informativeness of 
62 
UNDERSTANDING READING 
facts depends on the prior knowledge of the person receiving 
them. “Paris is the capital of France” is a fact, but it is not infor- 
mative to Tom, who knows it already, nor to Dick, who doesn’t 
understand what the word “capital” means. And although the 
statement is informative to Harry, who wasn’t aware of the fact 
before, it is not possible to say how informative it is because 
Harry’s uncertainty can't be calculated . We don’t know how many 
alternative cities Harry thought might be the capital of France or 
how many countries he thought Paris might be the capital of. We 
don’t even know if he cares. Quite possibly, “Paris is the capital of 
France” is a fact with no information value to Harry when he 
learns it, although it may be useful to him later in his life. On the 
other hand, information that serves only to clutter the mind is 
really noise. 
Information exists only when it reduces uncertainty, which is 
relative to the knowledge and purposes of the individual receiv- 
ing it. And comprehension also depends on what an individual 
already knows and needs or wants to know. Comprehension 
doesn’t entail assimilating or even examining all of the informa- 
tion in a text, but rather being able to make sense of the text in 
terms of the reader’s expectations and intentions. Even fluent 
readers must read some texts more than once in order to com- 
prehend them or to remember a lot of detail. Reading always in- 
volves asking questions of a text, and comprehension ensues to 
the extent that such questions are answered. I may not compre- 
hend a particular text in the same way as you, but then I may not 
be asking the same questions. Arguments about how a novel, 
poem, or any other text is most appropriately or "correctly” com- 
prehended are usually arguments about the most relevant kind 
of questions to ask. A child who claims to have understood a 
story may not have understood it in the same way as the teacher, 
but the child was probably not asking the same questions as the 
teacher. The teacher's questions may be noise to the child. A 
large part of comprehending literature in any conventional man- 
ner is knowing the conventional questions to ask and how to find 
their answers. 
All the preceding discussion of information and comprehension 
underlines the importance in reading of what goes on behind the 
eyes, where prior knowledge, purposes, uncertainty, and ques- 
tions reside. So also do the next two major topics that are dis- 
cussed, the matter of information that is available from more than 
one source and the importance of having more than one source of 
information available. 
4. INFORMATION AND EXPERIENCE 
63 
Redundancy 
Redundancy exists whenever the same information is available 
from more than one source, when the same alternatives can be 
eliminated in more than one way. And one of the basic skills of 
reading is the selective elimination of alternatives through the use 
of redundancy. 
An obvious type of redundancy is repetition, for example, when the 
alternative sources of information are two identical successive sen- 
tences. A different means of having the same information twice would 
be its concurrent presentation to the eye and to the ear — an audiovi- 
sual or multimedia situation. Repetition is an eminently popular 
technique in advertising, especially in television commercials, exem- 
plifying one of the practical advantages of redundancy — that it re- 
duces the likelihood that recipients will unwittingly make a mistake, 
or overlook anything, in their comprehension of the message. There 
are other aspects of redundancy, however, that are not always as obvi- 
ous but that play a more important role in reading. 
The fact that the same alternatives are eliminated by two sources 
of information is often not apparent. Consider the following pair of 
sentences: 
1. The letter of the alphabet that I am thinking of is a vowel. 
2 . The letter I am thinking of is from the first half of the alphabet. 
At first glance the statements might appear to provide comple- 
mentary pieces of information telling us that the letter is a vowel in 
the first half of the alphabet. However, if we look at the alternatives 
eliminated by each of the two statements, we can see that they actu- 
ally contain a good deal of overlapping information. Statement 1 
tells us that the letter is not b, c, d,J, g, h,j, k, l, m, n, p, q, r, s, t, v, 
w, x, y, z, and statement 2 tells us that it is not n, o, p, q, r, s, t, u. v, 
w, x, y, z. Both statements tell us that the letter is not n. p, q. r, s, t, 
v, w, x, y, z, and it is to this extent (the extent to which the excluded 
sets of alternatives intersect) that the statements are redundant. In 
fact, the only new information provided by statement 2 is that the 
letter is not o or u; all the other information is already provided in 
statement 1. 
There are frequent examples of redundancy in reading. As an il- 
lustration, consider the unfinished sentence (which could perhaps 
be the bottom line of a right-hand page of a book): 
The captain ordered the mate to drop the an- 
64 
UNDERSTANDING READING 
There are four ways of reducing uncertainty about the remain- 
der of that sentence, four alternative and therefore redundant 
sources of information. First, we could turn the page and see how 
the last word finished — this would be visual information. But we 
could also make some reasonable predictions about how the sen- 
tence will continue without turning the page. For example, we 
could say that the next letter is unlikely to be b,f, j, m, p, q, r, w, or z 
because these letters just don’t occur after an in common words of 
the English language; we can therefore attribute the elimination of 
these alternatives to orthographic (or spelling) information. There 
are also some things that can be said about the entire word before 
turning the page. We know that it is most likely to be an adjective or 
a noun because other types of words such as articles, conjunc- 
tions, verbs, and prepositions, for example, are most unlikely to 
follow the word the ; the elimination of all these additional alterna- 
tives can be attributed to syntactic (or grammatical) information. 
Finally, we can continue to eliminate alternatives even if we con- 
sider as candidates for the last word only nouns or adjectives that 
begin with an plus one of the letters not eliminated by the ortho- 
graphic information already discussed. We can eliminate words 
like answer and anagram and antibody even though they are not 
excluded by our other criteria because our knowledge of the world 
tells us these Eire not the kinds of things that captains normally or- 
der mates to drop. The elimination of these alternatives can be at- 
tributed to semantic information. 
Obviously, the four alternative sources of information about the 
incomplete word in the previous example, visual, orthographic, 
syntactic, and semantic, to some extent provide overlapping infor- 
mation. We don’t need as much visual information about the next 
word as we would if it occurred in isolation because the other 
sources of information eliminate many alternatives. The four 
sources of information, therefore, are sill to some extent redun- 
dant. The skilled reader who can make use of the three other 
sources needs much less visual information than the less fluent 
reader. The more redundancy there is, the less visual information 
the skilled reader requires. In passages of continuous text, pro- 
vided that the language is familiar and the content not too difficult, 
every other letter can be eliminated from most words, or about one 
word in five omitted altogether, without making the passage too dif- 
ficult for a reader to comprehend. 
One last point. I have talked of redundancy in reading as if it ex- 
ists in the written words themselves, which of course in a sense it 
does. But in a more important sense, redundancy is information 
4. INFORMATION AND EXPERIENCE 
65 
that is available from more than one source only when one of the 
alternative sources is the reader’s own prior knowledge. Put an- 
other way, there is no utility in redundancy in the text if it doesn’t 
reflect something the reader knows already, whether it involves 
the visual, orthographic, syntactic, or semantic structure of writ- 
ten language. The reader must know that b is unlikely to follow 
an- and that anchors are ordered dropped by captains. In making 
use of redundancy, the reader makes use of prior knowledge, us- 
ing something that is already known to eliminate some alterna- 
tives and thus reduce the amount of visual information that is 
required. Redundancy represents information you don’t need be- 
cause you have it already. 
LIMITS TO THE UTILITY OF INFORMATION 
The reason for the importance of redundancy, and of prior knowl- 
edge in general, is that there are severe limits to the amount of new 
information we can cope with at any one time, whether through the 
eyes or any other sense modality. We may have questions that we 
want answered and potential answers to those questions may be in 
front of our eyes, but if our uncertainty is extensive or if we are try- 
ing to make sense of too much information, then we may not be 
able to handle all the information we need to reduce our uncer- 
tainty. We may fail to comprehend. 
In plain language, we can try so hard to understand and to re- 
member more of what we read that we succeed only in confusing 
ourselves and learning less. Limitations of the visual system and of 
memory are discussed in the following chapters. But there is an- 
other factor to be taken into account that may sound rather para- 
doxical — the more we strive to avoid error, the less likely we are to 
be right. We always have a choice about how wrong we will be. 
Hits, Misses, and Criteria 
There’s not a fixed amount of information that readers require in 
order to identify a letter or a word, no matter how much redun- 
dancy is involved. Exactly how much information a reader will 
seek before making a decision about a particular letter, word, or 
meaning depends on the difficulty of the task (which must always 
be defined with respect to a particular reader) and on the “cost” of 
making a decision. 
A useful term for the amount of information that individuals re- 
quire before coming to a decision is their criterion level. If the 
66 
UNDERSTANDING READING 
amount of information about a particular letter, word, or meaning 
meets a reader’s criterion level for making a decision, then a choice 
will be made at that point, whether or not the reader has enough in- 
formation to make a decision correctly. We see a letter or word 
when we are ready and willing to decide what it is. 
Individuals vary in the way they establish a criterion — ranging 
from a supercautious attitude requiring almost an absolute cer- 
tainty to willingness to take a chance on minimal information, even 
at the risk of making a mistake. To understand why a particular 
criterion level is established, it is necessary to understand what the 
effect of setting a high or low criterion might be. 
The concept of criterion levels for perception developed in an area 
of study called signal detection theory, which upset a number of 
venerable ideas. It is traditional to think, for example, that one ei- 
ther sees something or one does not and that there is no area of free- 
dom between within which the perceiver can choose whether 
something is seen or not. Signal detection theory, however, shows 
that in many circumstances the question of whether an object is per- 
ceived depends less on the intensity of the object — on its “clarity," if 
you like — than on the attitude of the observer. It is also traditional to 
think that there is an inverse relationship between correct re- 
sponses and errors, that the more correct responses there are on 
any particular task, the lower the number of errors must be. Signal 
detection theory, however, shows that the cost of increasing the pro- 
portion of correct responses will be an increase in the number of er- 
rors. In other words, the more often you want to be right, the more 
often you must tolerate being wrong. The paradox can be explained 
by examining in a little more detail how the theory originated. 
Signal detection theory was originally concerned with the abil- 
ity of radar operators to distinguish between the "signals" and 
“noise” on their radar screens when they wanted to identify air- 
craft presumed to be hostile. As far as the actual situation is con- 
cerned, there are only two possibilities: A particular blip on the 
screen is either a signal or noise; an aircraft is present, or it is 
not. As far as the operator is concerned, there are also only two 
possibilities: a decision that the blip on the screen is an aircraft 
or a decision that it is not. In an ideal world, the combination of 
the actual situation and the operator's decision would still per- 
mit only two possibilities: Either the blip is a signal, in which 
case the operator decides that there is an aircraft, or the blip is 
merely noise, in which case the decision is that no aircraft is in- 
volved. We may call each of these two alternatives hits in the 
sense that they are both correct identifications. However, there 
4. INFORMATION AND EXPERIENCE 
67 
are two other possibilities, of quite different kinds, that must be 
considered errors. The first type of error occurs when no air- 
craft is present but the operator decides that there is — this situa- 
tion may be called a false alarm. And the other type of error 
occurs when there is an aircraft present but the operator decides 
that there is not, that the signal is actually noise — a situation 
that can be termed a miss. 
The problem for the operator is that the numbers of hits, false 
alarms, and misses are not independent; the number of one can’t 
be changed without a change in the number of another. If the opera- 
tor is anxious to avoid false alarms and wants to get maximum in- 
formation before deciding to report an aircraft, then there will be 
more misses. If, on the other hand, the operator wants to maximize 
the number of hits , reducing the possibility of a miss by deciding in 
favor of an aircraft on less information, then there will also be more 
false alarms. 
Of course, with increased skills of discrimination radar opera- 
tors can improve their level of efficiency and increase the ratio of 
hits to false alarms, just as increased clarity of the situation will 
make the task easier. But in any given situation the choice is always 
the same between maximizing hits and minimizing false alarms. 
Always the perceiver has to make the choice, to decide where to set 
the criterion for distinguishing signal from noise, friend from foe, a 
from b. The higher the criterion, the fewer will be the false alarms 
but the fewer also will be the hits. There will be more hits if the cri- 
terion is set lower, if decisions are made on less information, but 
there will also be more false alarms. 
Now we can approach the question of the basis on which the cri- 
terion is established: What makes the perceiver decide to set a cri- 
terion high or low? The answer lies in the relative costs and 
rewards of hits, misses, and false alarms. A radar operator who is 
heavily penalized for false alarms will set the criterion high, risking 
an occasional missed identification. One who is highly rewarded 
for the identification of a possible enemy and excused for the occa- 
sional mistake will set the criterion low. 
Readers can’t afford to set a criterion level too high before mak- 
ing decisions. A reader who demands too much visual information 
will often be unable to get it fast enough to read for sense. Readi- 
ness to take chances is critical for beginning readers who may be 
forced to pay too high a price for making “errors.” The child who 
stays silent (who “misses”) rather than risk a “false alarm” may 
please the teacher but develop a habit of setting a criterion too high 
for efficient reading. Poor readers often are afraid to take a chance; 
68 
UNDERSTANDING READING 
they may be so concerned about getting words wrong that they 
miss meaning altogether. 
INFORMATION AND EXPERIENCE 
I have gone to some length to present a technical consideration of 
the nature of information and its relevance in the study of reading. 
But my qualification must be emphasized. The information-pro- 
cessing point of view is useful for thinking about decision-mak- 
ing aspects of reading , but not about reading in general. Readers 
need to make sense of the visual information in a text in order to be 
able to read that text, but reading is much more than the identifica- 
tion of visual information. In a sense, reading is what you do after 
you get visual information; the visual information is just the raw 
material. 
I am again arguing against the view that reading is the “acquisi- 
tion of information" from text or, even more specifically, that read- 
ing is a matter of receiving particular messages or facts put into a 
text by the writer. This is the common “communication model," 
which sees text as some kind of channel along which information 
passes from writers to readers. Sometimes the communication 
metaphor becomes even more specific, with writers “encoding” 
messages in texts, which readers in their turn must then “decode.” 
However, many kinds of text and considerations of reading are 
distorted if not fundamentally misperceived if the communication 
and information-processing metaphors are applied too generally. 
As Rosenblatt (1978) pointed out, there is reading done for the 
sake of experience, which is usually the case with novels and po- 
etry, and also for the stimulation and exploration of ideas. In these 
cases what the reader brings to the text, looks for in the text, and 
does as a consequence of this interaction with the text are far more 
important and relevant than being able to “identify” and recall the 
actual content of the text. Indeed, I suspect that very little reading is 
done for purely factual purposes, where information provided by 
the text is of primary importance. Such reading (outside of formal 
school tasks) is rarely “cover-to-cover,” but rather is extremely se- 
lective and localized, limited by the specific intentions of the 
reader. I am referring to the occasions on which we consult ency- 
clopedias, dictionaries, catalogs, web pages, television guides, and 
telephone directories. At other times — even with newspapers and 
magazines — we read more for the experience generated by the 
reading, for the satisfaction of the act, than for the specific informa- 
tion that the reading provides. 
4. INFORMATION AND EXPERIENCE 
69 
The information-transmission metaphor is widespread in edu- 
cation, where all aspects of literacy are likely to be categorized 
and perceived as "communication skills.” The metaphor comes, 
of course, from the ubiquitous electronic technology in our envi- 
ronment, from radio, television, telephone, and computers. But 
even in these contexts the information and communication per- 
spective is limited and narrow. For example, television is often 
seen as a source of either “information” or “entertainment.” But 
there is another alternative. Most of the so-called informational 
and entertainment programs also present the possibility of expe- 
rience, far more relative to each individual’s knowledge and pur- 
poses than either of the other two supposed categories and 
probably far more important. Indeed, the view that education is a 
matter of acquiring information leads to misconceptions not only 
about reading (and television watching) but also about learning it- 
self, culminating in the dubious belief that children will soon be 
able to do all their learning (acquire all the necessary facts) at the 
consoles of computers. This exclusive emphasis on information 
acquisition overlooks the critical importance in education, and in 
life in general, of experience and self-directed exploration. 
The decision-making part of reading is usually only a minor part 
of the act as a whole, involving the identification of occasional let- 
ters, individual words, and possibly from time to time one of a lim- 
ited range of meanings. Research has tended to concentrate on 
these restricted aspects of reading. But the information that en- 
ables you to make such identifications is not the same as the “mes- 
sage” that you interpret from the text, or the understanding that 
you bring to it, and certainly not the same as the experience that it 
might generate for you. 
It might be best to regard the information offered by texts in a 
more general sense as evidence rather than as a message, the basis 
for a response or understanding rather than the content of com- 
prehension. Information may be what the brain looks for in read- 
ing, through the eyes, but it is not the end of reading. It is the basis 
on which a meaning is interpreted, an experience constructed, or 
the exploration of an idea launched. 
In this book, I don’t use the term information in its broad and 
imprecise sense at all. The term is used fairly extensively in the 
next few chapters, but only in the strictly technical sense, in dis- 
cussions of how uncertainty related to visual “input” from the 
eyes is resolved. Despite the time I have spent discussing informa- 
tion, I don’t regard it as the greatest or most important aspect of 
reading. 
70 
UNDERSTANDING READING 
What is experience? It can’t be measured and isn’t easily defined 
(see “As-if” in the notes to chapter 1 , page 238). Perhaps experience 
doesn’t need definition. It is synonymous with being, with creating, 
exploring, and interacting with worlds — real, possible, and in- 
vented. It is engagement and participation, always involving the 
emotions and often including a deliberate quest for uncertainty. It 
is an essential condition for being human and alive. 
Reading is experience. Reading about a storm is not the same 
thing as being in a storm, but both are experiences. We respond 
emotionally to both, and can learn from both. The learning in each 
case is a by-product of the experience. We don’t live to acquire in- 
formation, but information, like knowledge, wisdom, abilities, atti- 
tudes, and values, comes with the experience of living. 
ISSUES 
The chapter has been almost entirely about one of the most crucial 
(although often unspoken) sources of contention in education — 
whether teaching and learning should focus on acquisition of infor- 
mation or quality of experience. Almost all systematic instruction 
in schools (including tertiary institutions) is based on information 
transmission, and almost all evaluation is based on information 
assimilation. Whether more specific emphasis should be placed on 
experience, both for teachers and for students, is an issue involv- 
ing values, not research. 
SUMMARY 
There are two fundamental reasons for reading — for information 
and for experience. Although it has a clearly defined meaning in a 
narrow technical sense, the word information is widely overused 
and misused. Information may be regarded as the reduction of un- 
certainty concerning the alternatives among which a reader must 
decide. How much visual information a reader will require is af- 
fected by the reader’s willingness to risk an erroneous decision. 
Readers who set too high a criterion level for information before 
making decisions will find comprehension more difficult. Because 
there are limits to how much information the brain can cope with 
in making sense of texts, readers must make use of all forms of re- 
dundancy in written language — orthographic, syntactic, and se- 
mantic. Because reading is more than a matter of making 
decisions, the relevance of the information-processing perspective 
is limited. 
4. INFORMATION AND EXPERIENCE 
71 
Notes to chapter 4 begin on page 255 covering: 
Measuring information and uncertainty 
Measuring redundancy 
Limitations of information theory 
Computers and people 
Between Eye and Brain 
The eyes are given altogether too much credit for seeing. Their role 
in reading is frequently overemphasized. The eyes don’t see at all, in 
a strictly literal sense. The eyes look; they are devices for collecting 
information for the brain, largely under the direction of the brain, 
and it is the brain that determines what we see and how we see it. 
Our perceptual decisions are based only partly on information from 
the eyes, greatly augmented by knowledge we already possess. 
The present chapter is not intended to be a comprehensive phys- 
iology of the visual system, but it does outline a few characteristics 
of eye -brain function that make critical differences to reading. 
Three particular features of the visual system are considered: 
1. We don’t see everything that is in front of our eyes. 
2. We don’t see anything that is in front of our eyes immediately. 
3. We don’t receive information from our eyes continuously. 
Together these three considerations lead to three important im- 
plications for reading, and for learning to read: 
1 . Reading must be fast. 
2. Reading must be selective. 
3. Reading depends on what the reader already knows. 
72 
5. BETWEEN EYE AND BRAIN 
73 
The remainder of this chapter discusses the preceding six 
points in order, after considering the importance of what goes on 
behind the eyes in reading. 
TWO SIDES OF READING 
Obviously, reading is not an activity that can be conducted in the 
dark. To read you need illumination, some print in front of you, 
your eyes open, and possibly your spectacles on. In other words, 
reading depends on some information getting through the eyes to 
the brain. This can be called visual information. It’s easy to char- 
acterize the general nature of visual information — it goes away 
when the lights go out. 
Access to visual information is a necessary part of reading, but 
not sufficient. You could have a wealth of visual information in a 
text before your open eyes and still not be able to read. For exam- 
ple, the text might be written in a language you don’t understand. 
Knowledge of the relevant language is essential for reading, but you 
can’t expect to find it on the printed page. Rather it is information 
that you must have already, behind the eyeballs. It can be distin- 
guished from the visual information that comes through the eyes 
by being called nonvisual information or “prior knowledge." 
There are other kinds of nonvisual information apart from 
knowledge of language. Knowledge of subject matter is equally im- 
portant. Give many people an article on deconstructionism, sub- 
atomic physics, or the differential calculus, and they will not be 
able to read — not because of some inadequacy in the text, which 
specialists can read perfectly well, nor because there is anything 
wrong with their eyes, but because they lack appropriate nonvisual 
information. Experience in reading is another kind of nonvisual in- 
formation of evident importance in making reading possible, al- 
though it has nothing to do with the lighting, the print, or the state 
of one’s eyes. Nonvisual information is easily distinguished from 
visual information — it is carried around by the reader all the time; 
it doesn’t go away when the lights go out. 
The Trade-Off Between Visual and Nonvisual Information 
The distinction between visual and nonvisual information may 
seem obvious; nevertheless, it is so critical in reading and learning 
to read that I put it into diagram form (Fig. 5. 1 ). 
The reason that the distinction between visual and nonvisual in- 
formation is so important is simply stated — there is a reciprocal 
74 
UNDERSTANDING READING 
FIG. 5.1. Two sources of information in reading. 
relationship between the two. Within certain limits, one can be 
traded off for the other. The more nonvisual information a reader 
has, the less visual information the reader needs. The less 
nonvisual information that is available from behind the eyes, the 
more visual information is required. This reciprocal relationship 
is represented by the curved line between the two kinds of informa- 
tion in Fig. 5.1. 
Reading always involves a combination of visual and nonvisual 
information. Informal demonstrations of the trade-off between the 
two sources of information are not difficult to give. Popular novels 
and newspaper articles tend to be easy to read — they can be read 
relatively quickly, in poor light, despite small type and poor quality 
printing. They are easy to read because of what we know already: 
we have a minimal need for visual information. On the other hand, 
technical materials or difficult novels — or even the same material 
when read by someone not as familiar with the language or the con- 
ventions of the text — require more time and more effort, larger 
type, clearer print, and superior physical conditions. The names of 
familiar towns on traffic signs can be read from further away than 
the same size place names of unfamiliar localities. It is easier to 
read letters on a wall when they are arranged into meaningful 
words and phrases than the same size letters in the random order 
of an optometrist’s test chart. In each case the difference has noth- 
ing to do with the quality of the visual information available in the 
print but with the amount of nonvisual information that the reader 
can bring to bear. The less nonvisual information the reader can 
employ, the harder it is to read. 
Making Reading Difficult 
Now we can see one reason why reading can be so very much 
harder for children, quite independently of their actual reading 
5. BETWEEN EYE AND BRAIN 
75 
ability. They may have little relevant nonvisual information. Some 
beginning reading materials are perversely designed to prevent the 
use of prior knowledge. At other times, adults may unwittingly or 
even deliberately discourage its use, by prohibiting “guessing.” For 
whatever cause, insufficient nonvisual information makes reading 
more difficult. 
Insufficient nonvisual information can even make reading im- 
possible, because there is a limit to how much visual information 
the brain can handle at any one time. There is a bottleneck in the vi- 
sual system between the eye and the brain, as indicated in Fig. 5.2. 
Because of this bottleneck a reader can temporarily become func- 
tionally blind. It is possible to look but not to see, no matter how 
good the physical conditions. A line of print that is transparently 
obvious to a teacher (who knows what it says in the first place) may 
be almost completely illegible to a child whose dependence on vi- 
sual information can limit perception to just two or three letters in 
the middle of the line. 
Being unable to discern the words for the print is not a handicap 
that is restricted to children. Experienced readers may find them- 
selves in exactly the same situation for essentially the same rea- 
sons — by being given difficult material to read, by being required to 
pay a lot of attention to every word , or by being put into a condition 
of anxiety, all of which increase the demand for visual information 
and have the paradoxical consequence of making it harder to see 
the text. 
Later in this chapter, I show how the relative proportions of vi- 
sual and nonvisual information required in reading can be esti- 
mated and also indicate how narrow the bottleneck is, so narrow 
that at least three quarters of the visual information available in 
text must usually be ignored. In the words of psychologist Paul 
Kolers (1967), "Reading is only incidentally visual.” 
FIG. 5.2. The bottleneck in reading. 
76 
UNDERSTANDING READING 
LIMITATIONS OF VISION 
We Don't See Everything 
The fact that the eyes are open is not an indication that visual infor- 
mation from the world around is being received and interpreted by 
the brain. We don't see the world as its image falls on our eyes. How 
could we, when that image must often be a kaleidoscopic blur as the 
eyes flick from place to place in their fitful investigations of the 
world? But the argument is more complex than the relatively simple 
fact that the world we see is stable although the eyes are frequently 
in movement. The scene we perceive has very little in common with 
the information the eyes receive from the surrounding world. 
No single nerve fiber runs directly from the eye to the brain; in- 
stead, there are at least six interchanges where impulses along one 
nerve may start — or inhibit — propagation of a further pattern of 
impulses along the next section of the pathway. At each of these 
neural relay stations there are large numbers of interconnections, 
some of which determine that a single impulse arriving along one 
section may set off a complex pattern of impulses in the next, while 
others may relay the message only if a particular combination of 
signals arrives. Each interconnection point is, in fact, a place 
where a complex analysis and transformation take place. 
Three layers of interconnections are located in the retina of the 
eyes, which is, in terms of both function and embryonic develop- 
ment, an extension of the brain. A tremendous compression takes 
place within the retina itself. When the nerve fibers eventually leave 
the eye on their journey to the brain (the pencil-thick bundle of 
nerve fibers is collectively called the optic nerve), the impulses 
from about 120 million light-sensitive cells in the retina where the 
neural messages originate have been thinned out over a hundred- 
fold; the optic nerve consists of barely a million neural pathways. 
The actual nature of the impulses that pass along this complex ca- 
ble of nerves is also very different from our perception or belief of 
what the visual stimulus is like. Every nerve in our body is limited to 
conveying only one type of signal — either it fires, or it doesn't. The 
speed of the impulse may vary from nerve to nerve, but for any one 
nerve it is fixed; the response is “all or none." The nerve impulse is 
relatively slow; The fastest rate, for some of the long thick nerve fi- 
bers that travel severed feet along the body, is perhaps 300 feet per 
second (about 200 miles per hour). The smaller nerves, such as 
those in the visual system and brain, transmit at only a tenth of that 
speed (about 20 miles per hour). 
5. BETWEEN EYE AND BRAIN 
77 
Many examples of the way in which the brain imposes stability 
on the ever-changing perspective of the eyes are provided by what 
psychologists call the visual constancies. For example, we always 
see a known object as a constant size; we don’t think that a person 
or automobile moving away from us gets smaller as the distance in- 
creases, although the actual size of the image on the retina is 
halved as the distance doubles. We don’t think the world changes 
color just because the sun goes in, nor do we see a lawn as being 
different shades of green because parts of it are in the shade. We 
"see” plates and coins as circular, although from the angle at which 
such objects are usually viewed, the image hitting the eye is almost 
invariably ovoid. 
One might think that at least the perception of movement is de- 
termined by whether or not the image that falls on the retina is 
moving, but that is not the case. If our eyes are stationary and a 
moving image falls across them, we do indeed normally see move- 
ment. But if a similar movement across the eyes occurs because we 
move our eyes voluntarily — when we look around a room, for ex- 
ample — we don't see the world moving. Our perception of whether 
or not something is moving depends as much on the knowledge we 
have about what our eye muscles are doing as on the visual infor- 
mation being received by the eye. We can easily fool our own brain 
by sending it false information. If we “voluntarily" move an eye up 
or down the page of a book by the use of our eye muscles, we don’t 
see the book move, but if we move the eye in the same way by pok- 
ing it with a finger — moving the eye without moving the eye mus- 
cles — then we do see the book in movement. The brain “thinks” 
that if the eye muscles haven’t been actively involved, then the 
changing image on the eye must mean external movement, and 
constructs our perception accordingly. 
Tachistoscopes and Tunnel Vision 
Seeing is not a simple matter of an inner eye in the brain examining 
snapshots or video images of complete scenes from the outside 
world. The brain may generate a feeling that we are able to see most 
of what is in front of our eyes most of the time, but that is what it 
is — a feeling, generated by the brain. Upon analysis we may find 
that in fact we see very little. The eyes are not windows, and the 
brain doesn’t look through them. No pictures pass between the eye 
and brain, and no little person (no homunculus) sits inside the 
brain inspecting them. Not only what we see, but our conviction of 
seeing, is a fabrication of the brain. 
78 
UNDERSTANDING READING 
Take the case of reading. When we look at a page of print we may 
feel that we see entire lines at a time. In practice, we probably see 
very much less. And in extreme circumstances we may be almost 
blind. Paradoxically, the harder we try to look, the less we may ac- 
tually see. To understand the research that underlies these asser- 
tions, it is necessary to acquire some familiarity with a venerable 
piece of psychological instrumentation and with a rather precise 
way of talking about very small units of time. The small unit of time 
is the millisecond, usually abbreviated to msec. One millisecond is 
a thousandth part of a second; 10 milliseconds is a hundredth of a 
second; 100 msec is a tenth; 250 msec, a quarter; 500 msec, a half 
a second; and so forth. Ten milliseconds is about the amount of 
time the shutter of a camera requires to be open in normal condi- 
tions to get a reasonable image on a film. It can also be sufficient 
time for information to be available to the eye for a single percep- 
tual experience to result. Much more time is required for neural 
impulses to get from the eye to the brain or for the brain to make a 
perceptual decision. 
The venerable piece of psychological equipment is the 
tachistoscope, a device that presents information to the eyes for 
very brief periods of time. In other words, a tachistoscope dis- 
closes how much we can see at any one time. It doesn't allow the 
reader a second look. 
In its simplest form, a tachistoscope is a slide projector that 
throws a picture on a screen for a limited amount of time, usually 
only a fraction of a second. In experimental laboratories today, 
brief presentations are usually controlled with great precision by 
computers. One of the first discoveries made through the use of 
tachistoscopic devices during the 1890s was that the eye had to be 
exposed to visual information for very much less time than gener- 
ally thought. If there is sufficient intensity, an exposure of 50 msec 
is more than adequate for all the information the brain can manage 
on any one occasion. This doesn’t mean that 50 msec is adequate 
for identifying everything in a single glance; obviously it is not. You 
can’t inspect a page of a book for less than a second and expect to 
have seen every word. But 50 msec is a sufficient exposure for all 
the visual information that can be gained in a single fixation. It will 
make no difference if the source of the visual information is re- 
moved after 50 msec or left for 250 msec; nothing more will be 
seen. Eyes pick up usable information for only a fraction of the 
time that they are open. 
The second significant finding from the tachistoscopic and other 
studies was that what could be perceived in a single brief presenta- 
5 . BETWEEN EYE AND BRAIN 
79 
tion, in one glance, depended on what was presented and on the 
viewer’s prior knowledge. If random letters of the alphabet were 
presented — a sequence like KYBVOD — then only four or five letters 
might be reported. But if words were presented for the same 
amount of time, two or three might be reported, comprising a total 
of perhaps 12 letters. And if the words happened to be organized 
into a short sentence, then four or five words, a total of perhaps 25 
letters, might be perceived from the same exposure duration. 
The preceding paragraph reports a finding that is central to an 
understanding of reading. To underline its importance, the main 
points are reiterated in the form of a diagram (Fig. 5.3). 
In the Notes section at the end of this book, it is shown that the 
eye and brain are doing the same amount of work in each of the 
three situations depicted in Fig. 5.3. The eyes are sending the same 
amount of visual information to the brain and the brain is making 
sense of the same proportion of it. But the more sense the letters 
make — which means the more the brain is able to use nonvisual in- 
formation — the more can be seen. The difference lies in the num- 
ber of alternatives confronting the brain in making perceptual 
decisions. If the letters are random — or as good as random to the 
person trying to read them — they are basically unpredictable and 
demand a good deal of visual information for each identification 
decision. The reader consequently sees very little and is in a condi- 
tion known as “tunnel vision” (Mackworth, 1965), very similar to 
trying to examine the world through a narrow paper tube. Every- 
one can have tunnel vision; it has nothing to do with the health or 
efficiency of the eyes. Tunnel vision is a result of trying to handle 
too much visual information. Airline pilots can suffer from tunnel 
K B 0/6 \/ W/£ ( P J M S O^ T/X Oyd M/6 1/6 S/6 
Random letters: four or five 
(jump wheat) XX 
Unrelated words: about two (ten-twelve letters) 
(knights rode horses into war) 
A meaningful phrase: four or five words (about 25 letters) 
FIG. 5.3. What can be seen in one glance. 
80 
UNDERSTANDING READING 
vision, especially during takeoff and landing. That is why it takes 
more than one pilot to fly large planes. All readers can be afflicted 
with tunnel vision when the material they are trying to read is unfa- 
miliar, opaque, or otherwise difficult — or when through the partic- 
ular demands of the task or sheer anxiety they try to handle too 
much visual information. Beginning readers are prime candidates 
for having tunnel vision much of the time, especially if the books 
they are supposed to read make little sense to them. Tunnel vision, 
in other words, is caused by information overload. 
On the other hand, if the text is easily comprehended, entire 
lines can be seen at one time. So for a teacher who points to some 
words in a book and says to a child, “There, you can see that clearly 
enough, can't you?” the answer is probably “No.” The teacher who 
can see the entire line knows what the words are in the first place. 
The fact that the teacher is pointing can make the situation even 
worse and ensure that the child sees nothing very much beyond the 
tip of a finger. 
You can’t read if you see only a few letters at a time. Ttinnel vision 
makes reading impossible. And the situation can’t be retrieved by 
trying to look at the words more often. Seeing takes time, and there 
is a limit to the rate at which the brain can make its visual decisions. 
Seeing Takes Time 
We usually feel that we see what we are looking at immediately. But 
this is another illusion generated by the brain. It takes time to see 
anything because the brain requires time to make perceptual deci- 
sions. And the time that is required is again directly related to the 
number of alternatives confronting us. The more alternatives we 
have to consider and discard, the longer it takes the brain to make 
up its mind, so to speak, and for seeing to occur. 
The tachistoscope can again be used for an experimental dem- 
onstration. If a single letter of the alphabet is briefly but clearly 
displayed, say. A, the delay before the viewer succeeds in saying 
“A" will depend on the number of letters that could have occurred 
instead of A. Give the viewer no clue, so that the letter might be 
any one of 26 alternatives, and the delay — the “reaction time" — 
can be as long as 500 msec, half a second. Say in advance that the 
letter is a vowel, and the reaction time will be much briefer. Tell 
the viewer that the letter is either A or B, and reaction time may 
drop to as little as 200 msec. With fewer alternatives the brain of 
the viewer has much less work to do, and the decision comes very 
much faster. 
5. BETWEEN EYE AND BRAIN 
81 
In reading, it is imperative that the brain should make use of 
anything relevant that we already know in order to reduce the 
number of alternatives. The rather slow rate at which the brain 
can make decisions can be extremely disruptive. If the brain has 
to spend too long deciding among the alternatives, the visual in- 
formation that the eye makes available to the brain will be gone. 
That is the explanation of tunnel vision — the brain loses access to 
visual information before it has had time to make many decisions 
about it. 
Visual information doesn’t stay available to the brain for very 
long after being picked up by the eye. Obviously, visual information 
remains somewhere in the head for a short period of time, while 
the brain works on the information collected by the eye in the first 
few milliseconds of each look. Psychologists have even coined a 
name for the place where this information is supposed to reside 
between the time it has been sent back from the eye and the time 
the brain has made its decisions. This place is known as the sen- 
sory store , although it so far has remained a purely theoretical con- 
struct without any actual known location in the brain. But 
wherever and whatever the sensory store might be, it doesn’t last 
very long. Estimates of its persistence vary from half a second to — 
under optimum conditions — 2 seconds. But it is just as well that 
sensory store does persist briefly because a full second is required 
for the brain to decide even about the limited amount that it is usu- 
ally able to perceive in a single glance. The visual information that 
can be utilized in a single glance or tachistoscopic exposure — re- 
sulting in the identification of four or five random letters, a couple 
of unrelated words, or a meaningful sequence of four or five words 
— in fact requires a full second. The caption to Fig. 5.3 could be 
amended to read not just “What can be seen in one glance ” but 
also “What can be seen in one second .” The basic physiological 
limitation on the rate at which the brain can decide among alterna- 
tives seems to put the limit on the speed at which most people can 
read meaningful text aloud , which is usually not much more than 
250 words a minute (about 4 words a second). People who read 
very much faster than that rate are generally not reading aloud and 
certainly not delaying to identify every word. 
Information isn’t usually allowed to stay in sensory store for its 
full term of a second or so. Every time the eyes send another por- 
tion of visual information to the brain — which means every time we 
shift our gaze to a new focal point or at least blink to take a second 
look at the same place — then the arrival of new visual information 
erases the previous contents of sensory store. This phenomenon is 
82 
UNDERSTANDING READING 
referred to as masking. It is by the controlled use of masking in 
tachistoscopic experiments that psychologists have determined 
that the brain does indeed require a substantial amount of time to 
make perceptual decisions. The experiments illustrating how 
much can be seen in a single glance only work if a second exposure 
to visual information doesn't follow the first before the brain has 
had time to make sense of it. If a second exposure is presented less 
than half a second after the first, the viewer is unlikely to report the 
full four or five random letters or words that would otherwise be 
perceived. If the two events occur too close together — say within 50 
msec of each other — then the second can completely obliterate the 
first. Because masking occurs before the brain has time even to de- 
cide that something has taken place, the viewer will be completely 
unaware of a visual event that would otherwise be seen quite 
clearly. Seeing is a relatively slow process. 
On the other hand, because information in sensory store doesn’t 
persist more than about a second under most conditions, we can’t 
see more simply by looking longer at one spot. The eyes must be 
constantly active to replenish the fading stock of visual informa- 
tion in sensory store. A person who stares is not seeing more, but 
rather is having difficulty deciding what was looked at in the first 
place. Because the contents of sensory store decay rapidly and can- 
not be replenished from an eye that remains fixed in the same posi- 
tion, the eyes of anyone alert to the visual environment tend to be 
constantly on the move, even though the brain attends to only the 
first few moments of every new look. Visual information is con- 
stantly subject to interruption. 
Seeing Is Episodic 
Our eyes are continually in movement — with our knowledge and 
without it. If we pause to think about it, we know that our eyes are 
scanning a page of text, or glancing around a room, or following a 
moving object. These are the eye movements we observe if we watch 
another person’s face. These movements are rarely random — we 
would be quickly alarmed if our own eyes or someone else’s began 
rolling around uncontrollably — but instead the eyes move systemat- 
ically to where there is the most information or interest for us. The 
movements of the eye are controlled by the brain, and by examining 
how the brain directs the eye, we can get a basis for understanding 
what the brain is looking for, in general terms at least. 
But first, we must consider quite a different kind of eye move- 
ment, one not apparently under the direct control of the brain nor 
5. BETWEEN EYE AND BRAIN 
83 
one that is noticeable either in ourselves or others, but that none- 
theless can help to underline a point about the constructive nature 
of vision. Regardless of whether we are glancing around the envi- 
ronment, following a moving object, or maintaining a single fixa- 
tion, the eyeball is in a constant state of very fast movement. This 
movement, or tremor, occurs at the rate of 50 oscillations a sec- 
ond. We don’t notice the tremor in other people, partly because it is 
so fast, but also because the movement covers only a very short dis- 
tance; it is more a vibration than a movement from one place to an- 
other. But although the movement is normally unnoticeable, it 
does have a significant role in the visual process; the tremor en- 
sures that more than one group of retinal cells is involved in even a 
single glance. The tremor provides another illustration of the now 
familiar point that if the perceptual experience were a simple re- 
production of whatever fell on the retina, then all we should ever 
see would be a giddy blur. 
The constant tremor of the eye is essential for vision— cancel it 
and we are quickly blinded. The cancelation has been accom- 
plished by an ingenious experimental procedure called “stabilizing 
the image” (Heckenmueller, 1965; Pritchard, 1961). Information 
coming to the eye is made to oscillate at the same rate and over the 
same distance as the movement of the eye itself by being reflected 
through a small mirror mounted directly on the eyeball. The con- 
sequence of stabilizing the image is not that the viewer suddenly 
perceives a super- sharp picture of the world; on the contrary, per- 
ception disappears. 
The image doesn't seem to disappear instantaneously, nor does 
it fade slowly like a movie scene. Instead, entire parts drop away in 
a systematic fashion. If the outline of a face has been presented, 
meaningful parts will vanish, one by one. first perhaps the hair, 
then an ear, then perhaps the eyes, the nose, until the only thing re- 
maining may be, like the Cheshire cat, the smile. The word BEAT 
might disintegrate by the loss of its initial letter, leaving EAT, and 
then by dissolution to AT and A. By itself, the letter B might lose one 
loop to become P and then another to leave I. The phenomenon 
shows that the brain holds on to a disappearing image in the most 
meaningful way possible. Presumably the overworked retinal cells, 
deprived of the momentary respite the tremor can give them, be- 
come fatigued and send less and less information back to the 
brain, while the brain continues to construct as much of a percept 
as it can from the diminishing material that it receives. 
Other kinds of eye movement need not detain us. There is a kind 
of slow drift, a tendency of the eye to wander from the point of fo- 
84 
UNDERSTANDING READING 
cus, which is probably not very important because the eye has 
picked up all the useful information it is going to get during the first 
few milliseconds. There are "pursuit” movements that the eye 
makes when it follows a moving object. The only time that the eyes 
move smoothly and continuously from one position to another is in 
the course of a pursuit movement. Looking a person up and down 
with a single sweep of the eyes occurs only in fiction. 
The eye movement that is really of concern in reading is, in fact, 
a rapid, irregular, spasmodic, but surprisingly accurate jump from 
one focal position to another. It is perhaps a little inappropriate to 
call such an important movement a jump, so it is dignified by the 
far more elegant-sounding French word saccade (which literally 
translated into English, however, means “jerk”). 
Fixations and Regressions 
A saccade is by no means a special characteristic of reading, but 
rather the way we normally sample our visual environment for in- 
formation about the world. We are skilled in making saccadic 
movements of the eye. Guided by information received in its pe- 
riphery, the eye can move very rapidly and accurately from one side 
of the visual field to the other, from left to right, up and down, even 
though we may be unaware of the point or object upon which we 
will focus before the movement begins. Every time the eye pauses 
in this erratic progression, a fixation is said to occur. 
For reading English text, fixations are generally regarded as 
proceeding from left to right across the page, although, of course, 
our eye movements must also take us from the top of the page to- 
ward the bottom and from right to left as we proceed from one line 
to the next. Experienced readers often don't read "from left to 
right” at all — they may not make more than one fixation a line and 
may skip lines in reading down the page. We consider in a later 
chapter how such a method of reading can be possible. All read- 
ers make another kind of movement that is just another saccade 
but has got itself something of a bad name — a regression. A re- 
gression is simply a saccade that goes in the opposite direction 
from the line of type — from right to left along a line or from one 
line to an earlier one. Regressions can be just as productive as 
saccades in a forward, progressive direction. 
During the saccade, while the eye is moving from one position to 
another, very little is seen at all. The leaping eye is functionally 
blind. Information is picked up between saccades when the eye is 
relatively still — during fixations. The sole purpose of a saccade, in 
5. BETWEEN EYE AND BRAIN 
85 
whatever direction, is to move the eye from one position to another 
to pick up more visual information, like a bee foraging for pollen. 
The information collection occurs only once during a fixation — for 
the few hundredths of a second at the beginning, when information 
is being loaded into the sensory store. After that time, the back- 
room parts of the visual system are busy, perhaps for the next 
quarter of a second, making sense of the information. 
Saccades are fast as well as precise. The larger saccades are 
faster than the short ones. The movement of the eyes through 1 00 
degrees, say from the extreme left to the extreme right of the visual 
field, takes about 100 msec, a 10th of a second. A movement of 
only a 20th of that distance — about two or three words at a normal 
reading distance — might take 50 msec. But the fact that a saccade 
can be made in 50 msec doesn't mean that we can take in new infor- 
mation by moving the eye 20 times a second. The limit on the rate 
at which we can usefully move from one fixation to another is set by 
the time required by the brain to make sense of every new input. 
That is why there can be little “improvement” in the rate at which 
fixations are made during reading. You can’t accelerate reading by 
hurrying the eyes along. 
The number of fixations varies with both the skill of the reader 
and the difficulty of the passage being read, but not to any remark- 
able extent. In fact, fixation rate settles down by about Grade 4. 
There is a slight tendency for skilled readers to change fixations 
faster than unskilled readers, but the difference is only about one 
extra fixation a second; adults may average four and children just 
starting to read change fixation about three times a second. For any 
reader, experienced or novice, reading a difficult passage may cut 
about one fixation a second off the fastest reading rate. 
Children tend to make more regressions than fluent readers, 
but not so many more, perhaps one for every four progressive fixa- 
tions compared with one in six for adults. Once again, the rate of 
occurrence is determined as much by the difficulty of the passage 
as by the skill of the reader. Faced with a moderately difficult pas- 
sage, skilled readers will produce as many regressions as begin- 
ning readers with a passage that they find relatively easy. Readers 
who don’t make any regressions may be reading too slowly, too 
cautiously. When children make a lot of regressions, it is a signal 
that they are having difficulty, not a cause of difficulty. The number 
of regressions that readers make is an indication of the complexity 
to them of the passage they are trying to read. 
In short, the duration of fixations and the number of regres- 
sions are not reliable guides for distinguishing between good and 
86 
UNDERSTANDING READING 
poor readers. What does distinguish the fluent from the less- 
skilled reader is the number of letters or words — or the amount of 
meaning — that can be identified in a single fixation. As a result, a 
more meaningful way to evaluate the eye movements of a poor 
reader and a skilled one is to count the number of fixations re- 
quired to read a hundred words. Skilled readers need far fewer 
than the beginners because they are able to pick up more informa- 
tion on every fixation. A skilled college graduate reader might 
pick up enough information to identify words at an average rate of 
over one per fixation (including regressions) or about 90 fixations 
per 100 words. The beginner might have to look twice for every 
word, or 200 fixations per 100 words. The beginner tends to have 
tunnel vision. 
IMPLICATIONS FOR READING 
I said at the beginning of this chapter that discussion of the visual 
system would lead to three important implications for reading and 
for learning to read — that reading must be fast, that it must be se- 
lective, and that it depends on nonvisual information. By now the 
basic arguments underlying these implications may have become 
self-evident, but for emphasis they should be elaborated upon 
briefly. 
Reading Must Be Fast 
What is meant, of course, is that the brain must always move ahead 
quickly, to avoid becoming bogged down in the visual detail of the 
text to the extent that tunnel vision might result. This is not to sug- 
gest that the eyes should be speeded up. As I said earlier, reading 
can’t be improved by accelerating the eyeballs. There’s a limit to the 
rate at which the brain can make sense of visual information from 
the eyes, and simply increasing the rate at which fixations are 
made would have the consequence of further overwhelming the 
brain rather than facilitating its decisions. 
In fact, the customary reading rate of three or four fixations a 
second would appear to be an optimum. At a slower rate the con- 
tents of sensory store may begin to fade, putting the reader in the 
position of staring at nothing. At a faster rate than four fixations a 
second, masking can intrude so that the reader loses information 
before it is properly analyzed. 
The "slow reading” that must be avoided is the overattention to 
detail that keeps the reader on the brink of tunnel vision. Trying 
5. BETWEEN EYE AND BRAIN 
87 
to read text a few letters or a single word at a time keeps a reader 
functioning at the level of nonsense and precludes any hope of 
comprehension. Classroom advice to slow down in case of diffi- 
culty to be careful and examine every word closely can easily lead 
to bewilderment. 
Both word-perfect reading aloud and extensive deliberate mem- 
orization may require that a passage should be read more than 
once. A reader is unlikely to comprehend while reading more 
slowly than 200 words a minute, because a lesser rate would imply 
that words were being read as isolated units rather than as mean- 
ingful sentences. As we see in the next chapter, limitations of mem- 
ory also prevent sense being built up from isolated words. 
Comprehension demands relatively fast reading but memorization 
slows the reader down. Therefore, heavy memory burdens should 
be avoided when one is learning to read or unfamiliar with the lan- 
guage or subject matter. 
Reading Must Be Selective 
The brain just doesn’t have the time to attend to all the visual infor- 
mation available in most texts and can be easily inundated. Nor is 
memory able to cope with all the information that might be avail- 
able from the page. The secret of reading efficiently is not to read 
indiscriminately but to sample the text. The brain must be parsi- 
monious, making maximum use of what is already known and ana- 
lyzing the minimum of visual information required to verify or 
modify what can be predicted about the text. All this may sound 
very complicated, but in fact it is something that every experienced 
reader can do automatically, and almost certainly what you are do- 
ing now if you can make sense of what you read. It is no different 
from what you do when you look around a room or at a picture. 
But like many other aspects of fluent reading, selectivity in pick- 
ing up and analyzing samples of the available visual information in 
text comes with experience. Once again the initiative for how the 
eyes function rests with the brain. When the brain has got all the vi- 
sual information it requires from a fixation, it directs the eyes very 
precisely where to move next. The saccade will be either a progres- 
sive or regressive movement, depending on whether the next infor- 
mation that the brain requires is further ahead or further back in 
the page. The brain is able to direct the eyes appropriately, in read- 
ing as in other aspects of vision, provided it "understands” what it 
needs to find out. The brain must always be in charge. Trying to 
control eye movements in reading can be like trying to steer a horse 
88 
UNDERSTANDING READING 
by the tail. If the eyes don’t go to an appropriate place in reading, it 
is probably because the brain doesn’t know where to put them, not 
because the reader has insufficient visual ability to switch gaze to 
the right place at the right time. 
Reading Depends on Nonvisual Information 
Everything I have said so far should underline this final point. The 
brain — with its purposes, expectations, and prior knowledge — has 
to be in control of the eyes in reading. To assert that reading should 
be fast doesn’t mean recklessly so. A reader must be able to use 
nonvisual information to avoid being swamped by visual informa- 
tion from the eyes. To say that a reader should only sample the vi- 
sual information doesn’t imply that the eyes can go randomly from 
one part of the page to another. Rather, the reader should attend to 
just those parts of the text that contain the most important infor- 
mation for the reader’s purpose, whether every word, selected 
words, or selected portions of words. And this again is a matter of 
making maximum use of what is already known. 
The experienced reader employs no more visual information to 
comprehend four words in a single glance than the beginning 
reader who requires two fixations to identify a single word. All the 
additional information that skilled readers require is contributed 
by what they know already. When fluent readers encounter a pas- 
sage that is difficult to read — because it is poorly written or 
crammed with new information — the number of fixations (includ- 
ing regressions) they make increases, and reading speed goes 
down. Because of the additional uncertainty in the situation, they 
are forced to use more visual information to try to comprehend 
what they read. 
The relative ability to use prior knowledge has consequences in 
all aspects of vision. Experts — whether in reading, art, sports, or 
engineering — may be able to comprehend an entire situation at a 
single glance, but the greater uncertainty of novices handicaps 
them with tunnel vision. When readers in a tachistoscopic experi- 
ment are presented with words in a language they don't compre- 
hend, they are able to identify only a few letters. The fact that the 
words make sense to someone who knows the language is irrele- 
vant; to the uninformed reader the letters are essentially random, 
and inability to see very much will result. The implication for any- 
one involved in teaching reading should be obvious. Whenever 
readers cannot make sense of what they are expected to read — be- 
cause the material bears no relevance to any prior knowledge they 
5. BETWEEN EYE AND BRAIN 
89 
might have — then reading will become more difficult and learning 
to read impossible. 
ON SEEING BACKWARD 
One thing the eyes and the brain can’t do is see backward. I men- 
tion this fact because a belief exists that a visual handicap of this 
kind causes some children problems in learning to read. The basis 
of the myth is the indisputable evidence that many children at 
some point in their reading careers confuse reversible letters like b 
and d, p and q, and even words like was and saw, or much and 
chum. But seeing backward is both a physical and a logical impos- 
sibility, and a much simpler explanation is available. 
It is physically impossible to see part of the visual field in a dif- 
ferent orientation from the rest — to see one dog facing one way and 
one the other when they are both looking in the same direction. 
And it is logically impossible to see everything reversed because 
everything would still be seen in the same relationship to every- 
thing else and therefore nothing would be different; everything 
would still be seen the right way round. Of course, it is possible to 
make a mistake, to think a dog is facing east when it is facing west, 
especially if we are not familiar with the particular kind of dog, but 
that must be attributed to lack of adequate information or knowl- 
edge, not to a visual defect. 
And indeed, the simple explanation of why so many children 
confuse b and d is lack of appropriate experience. The discrimina- 
tion is not an easy one and can confuse adults whose information is 
limited, just as fluent readers of English become confused with the 
identification of similar letters in unfamiliar alphabets. The differ- 
ence between b and d is minimal — a matter of whether the upright 
stroke is on the left or the right of the circle — and is not a difference 
that is significant or even relevant in most aspects of children’s ex- 
perience. A dog is a dog whichever way it faces. Those more general 
discriminations that do require distinctions of actual or relative di- 
rection, such as “left” and “right” or telling the time from the hands 
of a clock, are notoriously difficult for most children to learn. 
Fluent readers don’t usually mistake b and d when they read, 
but that is primarily because they have so many other clues and 
need not be concerned with individual letters. But to distinguish b 
from d when they occur in isolation, one at a time, is much harder, 
and the fact that we can normally do so with facility must be attrib- 
uted to the years of experience we have had and the amount of time 
we are given, relatively speaking, to inspect the evidence. Being 
90 
UNDERSTANDING READING 
able to distinguish b from d doesn’t make a reader, but being a 
reader makes the discrimination easier. 
J3ecause the difference between b and d is both unusual and dif- 
ficult to perceive, it is relatively difficult for children to learn, espe- 
cially if they don't understand the significance of the difference in 
the first place. That is why the appropriate experience for such 
children is not more drill on isolated letters, which are meaning- 
less, but more meaningful reading. Children who have difficulty, 
perhaps confusing words like big and dig, must be reading words 
or sentences that are essentially meaningless (or as if they are 
meaningless). No one who is reading for sense could confuse 
words like big and dig, or was and saw, in a meaningful context. 
Unfortunately, children with a “reversals problem" are often given 
concentrated exercises on distinguishing word pairs like big and 
dig in isolation, increasing their apprehension and bewilderment. 
And if they show no progress with words in isolation they may be 
restricted to drills with b and d alone. But letters in isolation are 
considerably more difficult than letters in words because an im- 
portant relational clue has been removed. The difference between 
b and d at the beginning of a word is that the upright stroke is on 
the outside for b (as in big) but on the inside for d (as in dig). But 
“outside” and "inside” are meaningless for letters in isolation. 
There is only one possible way of making learning to distinguish b 
from d even more difficult, and that is to show them one at a time. 
This removes every relational clue, and puts the learner into situa- 
tions likely to confound even experienced readers. 
It is sometimes argued that children see letters backward be- 
cause they write them that way. But writing requires quite different 
kinds of skill. We all recognize faces and figures that we couldn’t 
possibly draw. If my drawing of a face looks like a potato, that does- 
n’t mean that I see a face as a potato, it means that I’m a poor artist. 
A child may draw a human figure as a circular head with match- 
stick arms and legs, but show them their own distorted efforts and 
an artist’s representation, and they will readily indicate which 
looks most like what they see. Children don’t and can't draw what 
they see, and the fact that they might write a few or many letters 
backward says nothing about their vision, simply that they haven't 
yet learned the difficult task of writing letters conventionally. 
A LITTLE MORE PHYSIOLOGY 
Anatomically, the brain is not all of a piece. In particular it is deeply 
split along a center line from the back of the head to behind the 
5. BETWEEN EYE AND BRAIN 
91 
nose into two roughly symmetrical hemispheres, the left and the 
right. These two hemispheres are relatively tenuously connected to 
each other, at least near the surface areas of the brain, including ar- 
eas that seem to be particularly involved with the organization of 
cognitive and motor functions. An old psychological and physiolog- 
ical puzzle, still not resolved, is how we piece together a coherent 
visual image of the scene in front of our eyes when the left half of the 
visual field (from both eyes) goes to one hemisphere and the right 
half goes to the other and there are no direct hemispheric connec- 
tions between the two. Physiologically the picture is split down the 
middle, but subjectively the seams don’t show. 
It has long been known that the left hemisphere of the brain is 
usually largely responsible for motor and sensory control of the 
right side of the body and the right hemisphere for the left. Because 
of this general cross-laterality , people who suffer strokes or other 
forms of injury to the left side of their brain tend to lose motor con- 
trol and possibly sensation in areas on the right side of their body, 
and damage to the right side of the brain affects the left side of the 
body. And it has also been known for over a century that for the ma- 
jority of people, especially the right-handed, areas of the left side of 
the brain tend to be particularly involved with language. For such 
people, strokes or other injuries to the right side of the brain may 
leave language abilities largely unimpaired, and accidents to the 
left side of the brain are more likely to be associated with language 
loss. However, this hemispheric specialization is by no means uni- 
versal or necessary. About 10 percent of the population has the 
right hemisphere primarily involved with language functions, and 
children who are born with or who early in life suffer damage to the 
left side of the brain can develop language relatively fluently with 
the right, although it becomes much harder to transfer language or 
to relearn it with the opposite hemisphere as they get older. 
Many ingenious studies have been made of the general modes of 
functioning of the two sides of the brain, particularly with the living 
brains of people unfortunate enough to have had the surface con- 
nections between their two hemispheres severed by accident or un- 
avoidable surgery. Such studies have shown that the two sides of 
the brain have quite different styles of operation. The left hemi- 
sphere (in most people) seems to be particularly involved in activi- 
ties that are analytic and sequential (like language), for intellectual 
calculations and planning. In such people, the right hemisphere s 
characteristic responsibilities and mode of operation are more ho- 
listic and spatial; it is concerned with global, subjective, and emo- 
tional matters. The left side maybe busier when we write a letter or 
92 
UNDERSTANDING READING 
plan an excursion, the right side when we listen to music or imag- 
ine a scene. 
All this is fascinating and indeed a significant step forward to- 
ward understanding the mechanisms of the brain. However, it can 
also be misleading and conducive to spurious and even damaging 
conclusions if interpreted too literally. For the great majority of 
people the brain functions as a whole. We draw on the resources of 
both hemispheres to produce and understand language, just as we 
use all of our brain in the rest of our experience. It is a mistake to 
regard the two hemispheres as separate entities that function inde- 
pendently and even in opposition. Unfortunately, some educators 
and psychologists who know little about physiology (and some 
physiologists and neuroanatomists who know little about language 
and learning) talk as if we have two brains rather than two sides of 
a single brain. The hemispheres are sometimes referred to as the 
left and right brains, although this is literally (and only approxi- 
mately) true for just a handful of people whose brains have been 
surgically or accidentally sectioned. 
One danger of such reasoning is that it confuses a structural ar- 
rangement relevant only to the internal working of the brain with 
the way in which a person functions as a whole. There are no peo- 
ple who think only with the left or the right side of their brains, even 
if personalities and behavior reflect proclivities for more analytic 
or reflective approaches to life and learning. It is appropriate to say 
that a particular kind of activity or preference is dominant in a per- 
son, but not that the hemisphere is dominant. It makes sense to 
say a person has a good spatial orientation and tends to the con- 
templative; this gives a way of understanding the person and per- 
haps adapting to idiosyncratic learning preferences. It adds 
nothing to say that the person is right-brain dominant, and may in- 
deed reduce understanding by switching attention from percepti- 
ble characteristics to an assumed and probably mythical cause. 
Anyone capable of learning to produce and understand the lan- 
guage of a familiar environment has the ability to use both sides of 
the brain and to do all those things the two hemispheres are sup- 
posed to specialize in. No child comes to school with only half a 
brain, and hemispheric specialization or ‘‘asymmetry" should not 
be proposed as an explanation of difficulty in learning to read, es- 
pecially when there are, as we shall see, so many alternative possi- 
bilities. In particular, it is wrong to work backward and to assume 
that because a child is slow or reluctant in learning to read there 
must be an imbalance or inadequacy of hemispheric function. 
There is no reading center in the brain. Many areas of the brain are 
5. BETWEEN EYE AND BRAIN 
93 
active when we read, but none is involved in reading to the exclu- 
sion of anything else. Illness, injury, or very occasionally an inher- 
ent defect may affect the working of the brain so that ability to read 
is disturbed, but there is nothing physiologically or intellectually 
unique about reading. Advances in mapping the architecture of the 
brain help us mostly to understand the brain (or to further respect 
its complexity), not to understand language or learning. 
It is important to know the possibilities and limitations of the 
brain, which is the reason we have already been so concerned 
with the mechanisms of perception and memory, but the actual 
manner in which the brain internally organizes its own affairs 
doesn’t have a great relevance to education. At least, that is my 
view. I don't believe it would or should make the slightest differ- 
ence to how reading should be taught if it were discovered tomor- 
row that we all have a critical neural center for reading in the foot 
(the left foot, for most people). Instruction should always be 
adapted to the circumstances in which an individual learns and 
understands best, but this is not promoted by speculation about 
hypothetical brain structures. 
I stress this because there continues to be in education what I 
and some other researchers believe is a readiness little short of 
tragic to attribute learning and teaching failures to pseudomed- 
ical or pseudoscientific causes. Failures are blamed on percep- 
tual or cognitive handicaps with evidence no more specific than 
the fact that the failure occurs and a convenient medical or scien- 
tific theory happens to be around. If a plausible explanation can’t 
be found in terms of visual, acoustic, memory, or intellectual in- 
adequacy, then an even vaguer “minimal learning disability” may 
be blamed. And the current excuse, the most popular explana- 
tion, always seems to follow the area of scientific research that is 
generating the most interest and receiving the most popular at- 
tention. Failure to learn is explained in terms of fad rather than 
fact. The association of subtle differences in learning, behavior, 
attitude, and personality with presumed differences in the archi- 
tecture of the brain should not become a new phrenology, as un- 
scientific as making judgments about people’s character from the 
bumps on their skulls. 
ISSUES 
The mountain of experimental data on eye movements grows with 
every technological development. Basic “facts” about how the eyes 
move are rarely in dispute, but their interpretations are, especially 
94 
UNDERSTANDING READING 
when researchers feel themselves compelled to pronounce on “the 
nature of reading” and “instructional implications.” At issue are 
such matters as whether the tendency of readers (in experimental 
situations) to direct fixations to every word in front of their eyes 
means that they are actually "reading” every word, identifying one 
word (or letter) at a time and comprehending one word at a time, or 
reading the way they normally would when they, not the experi- 
menters, control the purpose, selectivity, anticipation, and com- 
prehension of the reading. 
SUMMARY 
Reading is not just a visual activity. Both visual information and 
nonvisual information are essential for reading, and there can be a 
trade-off between the two. Reading is not instantaneous; the brain 
cannot immediately make sense of the visual information in a page 
of print. The eyes move in saccades , pausing at fixations to select 
visual information, usually progressing in a forward direction but, 
when needed, in regressions. Slow reading interferes with com- 
prehension. Reading is accelerated not by increasing the fixation 
rate but by reducing dependency on visual information, mainly 
through making use of meaning. 
Notes to chapter 5 begin on page 262 covering: 
Vision and information 
The rate of visual decision making 
Eye movements in reading 
On seeing backward 
Hemispheric specialization 
Bottlenecks of Memory 
In chapter 2 I wrote of the diverse and massive amounts of knowl- 
edge that together comprise every individual’s theory of the world. 
But I didn’t discuss how we manage to deposit and maintain all of 
this knowledge in the vaults of memory nor how we draw on it 
when we need to. 
Two chapters at the end of this book are concerned with the gen- 
eral topic of learning — with the circumstances in which our theory 
of the world develops and grows. But there are some specific issues 
related to how much can get into memory at any one time that are 
more appropriately considered at this point. These issues are re- 
lated to the bottlenecks in perception that I discussed in the previ- 
ous chapter — for example, the fact that beginning readers (or any 
reader confronted by an unfamiliar text) can see only a small 
amount at any one time, even as little as four or five letters. Now I 
must turn to some additional constraints that confront all readers, 
but especially those in difficulty. These are constraints on how 
quickly specific things can be taken into or out of memory. 
Why should tunnel vision, the temporary inability to see what is 
contained in more than a small area in front of the eyes, be such a 
crippling handicap for readers, whatever their experience and abil- 
ity? If a beginning reader can see only a few letters at a time — say, 
the first half of a word such as ELEP . . . — why can’t these letters be 
95 
96 
UNDERSTANDING READING 
remembered for the fraction of a second that the child requires to 
make a new fixation and see the rest of the word . . . HAN'T? Unfortu- 
nately, memory has its own limitations and can’t be called on to ex- 
ceed its capacity when the visual system is overworked. Fluent 
reading demands not only parsimony in the use of visual informa- 
tion but also restraint in the burdens placed on memory. In both 
cases there are limits to how much the brain can handle. Over- 
loading memory doesn’t make reading easier and can contribute to 
making reading impossible. 
There are a number of paradoxes about the role of memory in 
reading. The more we try to memorize, the less we are likely to re- 
call. The more we try to memorize, the less we are likely to compre- 
hend, which not only makes recall more difficult — it makes recall 
pointless. Who wants to remember nonsense? On the other hand, 
the more we comprehend, the more memory will take care of itself. 
An implication of these paradoxes is that the prior knowledge al- 
ready in memory is far more important in reading than efforts to 
memorize everything in a text. To repeat a theme that by now 
should be familiar, nonvisual information is critical. 
THREE ASPECTS OF MEMORY 
To begin, terms need to be clarified a little. We can use the word 
memory in a variety of ways, sometimes to refer to how well we can 
put something away for future use, sometimes to how long we can 
retain it, and sometimes to how well we can get to it. In this chapter, 
we consider four specific aspects or operating characteristics of 
memory: input (how material goes in), capacity (how much can be 
held), persistence (how long it can be held), and retrieval (getting it 
out again). We also consider what would appear to be several kinds 
of memory, because memory doesn’t always look the same when 
examined in different ways. 
Psychologists often distinguish three kinds or aspects of mem- 
ory, depending on the time that elapses between the original pre- 
sentation of something to be remembered and the test to see what 
can be retrieved. The first aspect, termed sensory store, we met in 
the previous chapter. It is related to information from its arrival at 
a receptor organ, such as the eye, until a perceptual decision is 
made, for example, the identification of letters or words. The sec- 
ond aspect, usually called short-term memory, involves the brief 
time we can maintain attention to something immediately after its 
identification, for example, remembering an unfamiliar tele- 
phone number as we dial it. Finally, there is long-term memory. 
6. BOTTLEN ECKS OF M EMORY 
97 
which is everything we know about the world, our total stock of 
nonvisual information. 
These three aspects of memory are often depicted in textbook 
“flow diagrams" as if they are separate locations in the brain or suc- 
cessive stages in the process of memorization, as indicated in Fig. 
6.1. But such a diagram should not be taken too literally. I’m not 
sure that it is most appropriate to refer to different “kinds” of mem- 
ory, so I use the more neutral term aspects. 
There is no evidence that different memories exist in different 
places in the brain, nor that one memory starts functioning when 
the other leaves off, as the diagram might suggest. It is definitely 
misleading to imply that there is movement in just one direction 
from short-term to long-term memory, and to ignore the fact that 
there is always selectivity about how much is remembered and the 
manner in which remembering takes place. 
However, a discussion must begin somewhere and proceed in 
some kind of sequence, so for convenience the three aspects of 
memory are dealt with in the left-to-right order. Later an alterna- 
tive representation of memory is offered. 
Sensory Store 
The first aspect of memory can be quickly disposed of, because it is 
a theoretical necessity rather than a known part of the brain. Sen- 
sory store is a metaphor, hypothesized to account for the persis- 
tence of visual information after it is received by the eye, at the 
beginning of each fixation, while the brain is working on it. The op- 
erating characteristics of sensory store are quickly stated — input 
is very fast (the first few milliseconds of a fixation), capacity is at 
least large enough to hold visual information equivalent to 25 let- 
ters (although the brain may not be fast enough to identify any- 
where near that number), persistence is very brief (about a second 
under optimal conditions, but normally erased before that time by 
another fixation), and retrieval depends on how fast sense can be 
made of the information. 
Visual information 
} 
FIG. 6.1 . A typical flow diagram for memory. 
98 
UNDERSTANDING READING 
Sensory store has little significance for reading instruction be- 
cause there is nothing that can or need be done about it. Sensory 
store cannot be overloaded, nor can its capacity be increased by ex- 
ercise. There is no evidence that children’s sensory stores are less 
adequate than adults’. What needs to be remembered is that sense 
must be made of the contents of sensory store and that the con- 
tents don’t persist very long. As a result, there is little point in 
speeding up fixations (which will simply erase sensory store faster) 
nor in slowing them down (which will result in blank stares). What 
makes the difference in reading is the effectiveness of the brain in 
using what is already known (nonvisual information) to make 
sense of incoming information (visual information) briefly held in 
sensory store. 
Short-Term Memory 
Can you repeat the sentence you are reading at this moment, with- 
out taking a second look at it? Whatever you can do to repeat what 
you have just read is a demonstration of the function of short-term 
memory. Short-term memory is — metaphorically speaking — a 
"working memory’’ or “buffer memory," where you retain in the 
forefront of your mind whatever you are attending to at a particular 
moment. As far as language is concerned, the contents of short- 
term memory are usually the last few words you have read or lis- 
tened to or whatever thoughts you had in your mind instead. Some- 
times short-term memory is occupied by what you are about to say 
or write, by an address you are looking for. or by a telephone num- 
ber you want to call. Short-term memory is whatever is holding 
your attention. And short-term memory is of central importance in 
reading. It is where you lodge the traces of what you have just read 
while you go on to make sense of the next few words. It is where you 
try to retain facts that you want to commit to rote memorization. 
Short-term memory would appear to have both strengths and 
weaknesses, just by virtue of the way it functions. On the credit 
side, there doesn’t appear to be any undue delay in getting some- 
thing into short-term memory. In fact if someone asks you to call a 
certain telephone number, your best strategy is to get on your way 
to dial the number, not to stand around trying to commit the num- 
ber to memory. Similarly, there doesn’t appear to be any particular 
problem about retrieving items from short-term memory. If some- 
thing is in short-term memory you can get it out again at once. In- 
deed, if you can’t immediately retrieve what you want — say, the 
telephone number — then you might just as well go back and ask for 
6. BOTTLENECKS OF MEMORY 
99 
it again. Either you have retained the number in short-term mem- 
ory, in which case it is accessible without delay, or it is gone for 
good. Short-term memory is what we happen to be attending to at 
the moment, and if our attention is diverted to something else, the 
original content is lost. 
But if short-term memory seems a reasonably efficient device as 
far as input and output operations are concerned, in other re- 
spects it has its limitations. Short-term memory can’t contain very 
much at any one time — little more than half a dozen items. A se- 
quence of seven unrelated digits is about as much as anyone can re- 
tain. It is as if a benevolent providence had provided humanity with 
sufficient short-term memory capacity to make telephone calls and 
then had failed to anticipate area codes. If we try to hold more than 
six or seven items in short-term memory, then something will be 
lost. If someone distracts us when we are on our way to make that 
telephone call, perhaps by asking us the time of day or the location 
of a room, then some or all of the telephone number will be forgot- 
ten and there will be absolutely no point in cudgeling our brains for 
the number to come back. We shall just have to ascertain the num- 
ber once again. For as much as we try to overload short-term mem- 
ory, that much of its contents will be lost. 
This is the reason why short-term memory can’t be used to over- 
come tunnel vision. The child who has seen only ELEP . . . just can’t 
hold those letters in short-term memory, read another four or five 
letters, and get them all organized in a way that makes sense. As 
the fragments of one fixation go into short-term memory, the frag- 
ments from the previous fixation will be pushed out. This is not the 
same as the masking or erasure of sensory store — it is possible to 
hold a few items in short-term memory over a number of fixations. 
But holding such items in the forefront of our attention simply pre- 
vents very much more going in and has the obvious result of mak- 
ing reading much more difficult. Not much reading can be done if 
half your attention is preoccupied with earlier bits of letters and 
words that you are still trying to make sense of. 
The second limitation of short-term memory involves its persis- 
tence. Nothing stays around very long in short-term memory. It is 
impossible to state an exact amount of time for the persistence of 
something in short-term memory, for the simple reason that its 
longevity depends on what you do with it. Ignore something in 
short-term memory for less than a second, and it will be gone. To 
retain it, you must keep giving it your attention. Rehearsal is the 
technical term often employed. To keep that telephone number in 
your head you must keep repeating it; it can’t be allowed to elude 
100 
UNDERSTANDING READING 
your attention. Theoretically, material can be kept in short-term 
memory indefinitely, but only if constantly rehearsed, a procedure 
that is generally impractical because it prevents you from thinking 
about anything else. Material in short-term memory must be dealt 
with as expeditiously as possible. Retaining something for longer 
than a fixation or two, for example, preempts attention that is re- 
quired for the task on hand in reading and promotes a further loss 
of comprehension. The more a reader fills short-term memory 
with unrelated letters, bits of words, and other meaningless items, 
the more the letters and bits of words that the reader is currently 
trying to understand are likely to be nonsense. 
Long-Term Memory 
Of course, memory is far more than whatever we happen to be 
thinking about at the moment. There is a vast amount that we 
know all the time, ranging from names and telephone numbers to 
all the complex interrelationships that we can perceive and predict 
among objects and events in the world around us, and only a small 
part of all of this knowledge can be the focus of our attention at any 
one time. Anything that persists in our minds quite independently 
of rehearsal or conscious knowledge is long-term memory, our 
continuous knowledge of the world. Long-term memory has some 
distinct advantages over short-term memory, especially with re- 
gard to its capacity. Nevertheless, long-term memory can't be used 
as a dump for any overflow of information from short-term mem- 
ory, for long-term memory also has limitations. 
Let’s begin with the positive side. Where short-term memory is 
restricted in capacity to barely half a dozen items, the capacity of 
long-term memory would appear to be infinite. No limit has been 
discovered to how much can be lodged in long-term memory. We 
never have to erase an old friend's name to make room for the 
name of a new acquaintance. 
Similarly, there is no apparent limit to the persistence of long- 
term memory. No question of rehearsal here. Memories we may not 
even be aware that we had, recollections of a childhood incident, 
for example, can quite unexpectedly revivify themselves, triggered 
perhaps by a few nostalgic bars of music, an old photograph, or 
even a certain taste or aroma. 
But as everyone knows, the fact that there seems to be no theo- 
retical limit to the capacity or persistence of long-term memory 
doesn’t mean that its contents are constantly accessible. It is here 
that some failings of long-term memory become apparent. Re- 
6 . BOTTLENECKS OF MEMORY 
101 
trieval from long-term memory Is by no means as immediate and 
effortless as retrieval from short-term memory. 
Indeed, retention and retrieval seem quite different in long- and 
short-term memory. Short-term memory is like a set of half a 
dozen small boxes, each of which can contain one separate item, by 
definition immediately accessible to attention because it is atten- 
tion that holds them in short-term memory in the first place. But 
long-term memory is more like a network of knowledge, an orga- 
nized system in which each item is related in some way or another 
to everything else. The organization and operation of long-term 
memory — our theory of the world — were discussed in chapter 2. 
Whether or not we can retrieve something from long-term memory 
depends on how it is organized. The secret of recall from long-term 
memory is to tap into the interrelationships. 
Sometimes the effort to get hold of something in long-term mem- 
ory can be frustrating. We know something is there but can’t find a 
way to get to it. An illustration is the “tip-of-the-tongue” phenome- 
non (Brown & McNeill, 1966). We know someone’s name begins 
with an S and has three syllables — and we are sure it is not Sander- 
son or Somerset or Sylvester. Suddenly the name appears in the 
set of alternatives that our mind is running through, or perhaps 
when someone else mentions the name, and then we recognize it at 
once. It was in long-term memory all the time, but not immediately 
accessible. 
Success at retrieving something from long-term memory de- 
pends on the clues we can find to gain access to it and on how well it 
was organized in long-term memory in the first place. Basically, ev- 
erything depends on the sense that we made of the material when 
we originally put it into memory. It is pointless to try to put an over- 
flow of unrelated fragments from short-term memory into long- 
term memory — that is why rote learning is so often unproductive. 
It is not just that nonsense that goes in will be nonsense when it co- 
mes out; it is extremely difficult to get nonsense out at all. 
There is another reason why it is not feasible to accommodate 
an overflow from short-term memory in long-term memory, con- 
cerned with the rate at which long-term memory can accept some- 
thing new. In contrast to the practically immediate input of half a 
dozen items into short-term memory, committing something to 
long-term memory is extremely and surprisingly slow. To put one 
item into long-term memory takes 5 seconds — and in that 5 sec- 
onds there is little attention left over for anything else. The tele- 
phone number that will tax short-term memory to its capacity is 
at least accepted as quickly as it is read or heard, but to hold the 
102 
UNDERSTANDING READING 
same number in long-term memory so that it can be dialed the 
next day will require a good half minute of concentration, 5 sec- 
onds for each digit. 
Committing fragments of text to long-term memory is not some- 
thing that can be resorted to in reading to overcome limitations of 
the visual system or of short-term memory. Quite the contrary. Ef- 
forts to cram long-term memory will have the effect of interfering 
with comprehension. Beginning readers with tunnel vision, who 
cannot hold in short-term memory more than the few letters they 
see in a single fixation, are even more confounded if they try to put 
isolated letters or bits of words into long-term memory. 
Fluent readers can find reading impossible if they overburden 
long-term memory, even if they are trying to read material that 
they would find completely comprehensible if they relaxed and 
were content to enjoy it. This problem can be acute for students 
trying to read a novel or Shakespearean play and at the same time 
trying to commit to memory the unfamiliar names of all the char- 
acters and every trivial detail or event. Memorization interferes 
with comprehension by monopolizing attention and reducing in- 
telligibility. Most readers have encountered the perverse textbook 
that is incomprehensible the day before the examination — when 
we are trying to retain every fact — yet transparently obvious the 
day after — when we are reading to discover what we missed. If you 
are having difficulty comprehending what you are reading right 
now, it may be because you are trying too hard to memorize. On 
the other hand — as I shall demonstrate in a number of ways — 
comprehension takes care of memorization. If you comprehend 
what you read or hear, then long-term memory will reorganize it- 
self so efficiently and effortlessly that you will not be aware that 
you are learning at all. 
Long-term memory is extremely efficient, but only if the acquisi- 
tion and organization of new material are directed by what we 
know already. Once again we find that what we know already tips 
the balance, making reading possible. It is time now to look at how 
prior knowledge helps to overcome the limitations of both 
short-term and long-term memory. 
OVERCOMING MEMORY LIMITATIONS 
There are some paradoxes to be resolved. The experimental evi- 
dence is that we can hold no more than half a dozen random letters 
in short-term memory, yet it is usually not difficult to repeat a sen- 
tence of a dozen words or more that we have just read or heard for 
6. BOTTLENECKS OF MEMORY 
103 
the first time. It appears that no more than one letter or digit goes 
into long-term memory every 5 seconds, yet we can commonly re- 
call many of the larger themes and significant details that we have 
read in a novel or seen in a film. 
To explain these discrepancies I must clarify some rather loose 
language that I’ve been using. I’ve been talking about retaining “ma- 
terial,” or half a dozen "things” or “items,” in short-term memory 
and putting just one “item” into long-term memory. What are these 
“things” or “items”? The answer depends on the sense you are 
making of what you are reading or listening to. These “things” or 
“items” are units that exist in long-term memory already. 
If you are looking for letters — or if you can only find letters in 
what you are looking at — then you can hold half a dozen letters in 
short-term memory. But if you are looking for words, then short- 
term memory will hold half a dozen words , the equivalent of four or 
five times as many letters. 
It is a question of what you already know. Short-term memory is 
filled by a seven-digit telephone number, which also requires half a 
minute to put into long-term memory. But not if the number hap- 
pens to be 123-4567 because that is a sequence that you already 
know. The number 1234567 will occupy just one part of short- 
term memory and will enter long-term memory within a few sec- 
onds because, in a sense, it is there already. Can you hold the let- 
ters THEELEPELTJE in short-term memory? Only if you recognize 
them as a word, which you will do if you can read Dutch. To put the 
same sequence of letters into long-term memory would require a 
good minute of concentration — and even then it is unlikely that you 
would be able to recall them all tomorrow — unless you already 
know the word, in which case you will commit it to memory as rap- 
idly as the English word teaspoon , which the Dutch word happens 
to mean. 
Psychologists refer to this process of storing the largest mean- 
ingful unit in short-term memory as chunking, which is a conve- 
niently picturesque term but also a bit misleading. The term 
suggests that at the beginning we first attend to the small fragments 
(individual letters or digits), which we subsequently organize into 
larger units for efficiency in memory. But we are looking for the 
larger units all the time. When we move on to consider reading 
more specifically, it will be seen that written words can be identi- 
fied without any reference to letters, and meaning without refer- 
ence to specific words. It is not that we perceive letters, which we 
then — if we can — chunk into words, but that we can perceive words 
or meaning in the first place and never bother the visual system or 
104 
UNDERSTANDING READING 
memory with letters. The “items” that we commit to memory are 
the largest meaningful units we can find. In other words, what we 
put into short-term memory is determined by the largest units that 
we have available in long-term memory. It is what we know and are 
looking for that determines the content of short-term memory, 
which is the reason I present an alternative diagram of memory to 
that of Fig. 6. 1. In Fig. 6.2, short-term memory is shown as part of 
long-term memory — the part concerned with what we happen to be 
attending to at the time. Short-term memory is not an antecham- 
ber of long-term memory but that part of long-term memory that 
we use to attend to, and make sense of, a current situation. 
The arrow between short-term memory and sensory store is 
double-ended to acknowledge that the brain is selective about the 
visual information that it attends to, and the arrows between 
short-term and long-term memory are double-ended to represent 
their continual interaction. 
One final qualification must be made. We can hold in short- 
term memory a few letters or a few words. But we can also put into 
short-term memory something far more mysterious — we can 
hold there large rich chunks of meaning. It is impossible to put a 
number to this — units of meaning can’t be counted the way we 
count letters or words. But just as we can hold the letters con- 
tained in words in memory far more efficiently than letters that 
are unrelated to each other, so we can hold meaningful sequences 
of words in memory far more efficiently than we can hold individ- 
ual unrelated words. The same applies to long-term memory: we 
can put an entire “meaning” away in just a few seconds — without 
any conscious awareness that we are doing so — even though that 
meaning might have been embedded in a dozen words or more. 
And by definition, any “meaning” that we put into long-term mem- 
ory is going to be far easier to retain and retrieve because mean- 
Vlsual information 
FIG. 6.2. An alternative representation of memory. 
6. BOTTLENECKS OF MEMORY 
105 
ingfulness implies that the input is related to what we already 
know and makes sense to us. 
We must get used to the notion that meaning is not dependent on 
specific words. This crucial point is elaborated many times in this 
book. When we retain a meaningful sequence of words in mem- 
ory — either short-term or long-term — we are not primarily storing 
the words at all but rather the meaning that we attribute to them. 
“Meaning" is the largest and most efficient unit of analysis that we 
can bring to bear from what we know already to what we are trying 
to read (or hear) and understand. For the moment, I offer just one 
illustration of the fact that not only do we look for meaning, rather 
than specific words, when we comprehend speech or print but also 
that this is the most natural thing to do. 
I have said that we can hold a dozen or more words in short-term 
memory if they are in a meaningful sequence, but that six or seven is 
the limit for words that make no collective sense, like the same se- 
quence of words in reverse. Try memorizing: memory term short in 
words more or dozen a hold can we. We don’t hold a sequence of a 
dozen or so words in short-term memory, but their meaning. If you 
ask a person to repeat a sentence, you will often get back the right 
meaning but not exacdy the same words. The person who remem- 
bers is not so much recalling words as reconstructing a sentence 
from the meaning that was remembered. We don't attend to words; 
we attend to meanings. So a substantial “mistake” might be made in 
repeating exact words — the word automobile might be recalled as 
the word car, for example — but there is rarely the substitution of a 
tiny word that makes a big difference to meaning, such as not. We 
shall see, incidentally, that “errors” of this kind, that preserve mean- 
ing, are committed by experienced readers and also by children 
learning to read who are in the process of becoming good readers. 
Reading involves looking for meaning, not specific words. 
Memory Without Bottlenecks 
Much of what we have considered so far might be called contrived 
memory — whether by researchers, teachers, or ourselves — when 
remembering is in effect put under external control. In memory re- 
search, subjects are usually told what they must remember and re- 
call. When we make a deliberate effort to commit a particular thing 
to memory or to get a particular thing out, we usually encounter the 
frustrating bottlenecks I have been describing. 
But most of the time no particular effort is involved in remem- 
bering — and memory seems to be much more efficient. I have 
106 
UNDERSTANDING READING 
talked about how easily something goes into long-term memory 
provided we can make sense of it, provided it is relevant to what we 
are doing at the time. These are the occasions when memory could 
be said to be controlled “from the inside,” by a brain naturally op- 
erating on the world and making sense of it. It is sometimes sur- 
prising to discover how much has gone into memory without our 
awareness. I have more to say about this in the chapters on learn- 
ing, in which I stress that efficient and effortless memorizing de- 
pends on how well it is integrated with our current knowledge, 
purposes, and predictions. 
I have also talked in the present chapter about how easily things 
can be retrieved from long-term memory, provided they are organized 
in relationships with relevant parts of our current theory of the world. 
The most relevant kind of relationship is when what we need to re- 
member is part of what we are actually involved in at the time. 
I am referring to an aspect of memory that we rely on just about 
every moment of the day, one that is incredibly efficient and rarely 
lets us down. It is memory that is not contrived in any way; it func- 
tions spontaneously, without effort or conscious management. 
Ironically, because of their predilection for studying aspects of 
memory that can be brought under control in the laboratory, exper- 
imental psychologists have largely ignored this aspect of remem- 
bering, which takes place without the conscious manipulation of 
the experimenters or their subjects. Because of their rather nar- 
row viewpoint, such researchers have even persuaded themselves 
and many educators that remembering is normally difficult, 
effortful, and frequently unsuccessful. 
But most of the time we remember automatically and without 
strain, even without awareness that we are making demands on 
memory. We don’t usually have difficulty remembering our own 
name, where we live, or our telephone number. We remember our 
birthday and that of a few other people as well, and we remember 
holidays. We remember our friends’ names, and how they look, and 
where they live, and even some of their telephone numbers. We re- 
member everything about the world that is familiar to us. We re- 
member that trees are called "trees" and birds “birds." even though 
we may not remember the names of particular kinds. We remember 
the meanings of just about every word we know, how these words 
are pronounced, and how many of them are spelled as well. We not 
only remember facts, we remember scenes, procedures, scenarios, 
scripts — all of the cognitive schemes discussed in chapter 2. We re- 
member innumerable things. We don’t remember these things all 
the time, of course. Our heads would be continually cluttered if we 
6. BOTTLENECKS OF MEMORY 
107 
did. They come to mind just when it is appropriate for us to remem- 
ber them, when they help us to make sense of the world we are in at 
the moment. 
Psychologist George Mandler (1985) termed this everyday as- 
pect of memory reminding, although he didn’t want to say that it is 
different from any other facet of memory. In fact, Mandler sug- 
gested that apparent differences in kinds of memory are really only 
differences in the forms of tests, in the way memory is examined. If 
we look at the recall of something soon after it has come to our at- 
tention, then we talk about “short-term memory.” If we consider 
something over a longer period, then we refer to “long-term mem- 
ory." And instead of looking at particular things that we (or re- 
searchers) happen quite arbitrarily to be concerned with, if we 
consider what we continually remember for our own purposes, 
then we have this phenomenon of “reminding." 
And as with comprehension and with putting things into mem- 
ory, the conditions that make reminding fluent and effortless are 
meaningfulness, relevance, and personal involvement. We remem- 
ber most easily when what we need to recall is most relevant to 
what we are engaged in doing and when we have no anxiety about 
not remembering. This is all part of the continually ongoing activity 
of "thinking.” Memory is not a special faculty of the brain that func- 
tions independently of everything else. Thinking and the “remind- 
ing" aspect of memory are inseparable. 
When we are able to read with comprehension, we are being re- 
minded all the time. The events of the story (or the steps in the ar- 
gument) carry us along as if we were experiencing them at first 
hand, and we rarely have to struggle to exercise our memory. We re- 
member the meaning of particular words when we are reading 
those words; we recall what we have already read (and other as- 
pects of nonvisual information) when it is appropriate to do so. 
And we are reminded of all manner of things that happen to be rele- 
vant to our understanding, and to our purposes in reading, at that 
particular time. 
As a simple example, we are reminded of the appropriate mean- 
ing of the words (or we predict those meanings) in a sentence like 
He wiped the tear from the child’s eye, but quite a different mean- 
ing in He repaired the tear in the child’s jacket. Normally we 
would not even notice the word tear could have a different pronun- 
ciation and meaning than those that are appropriate for the context 
the word is in. 
Of course, memory lets us down sometimes. We fail to remind 
ourselves to buy something at the store on the way home or to 
108 
UNDERSTANDING READING 
make a telephone call that we had planned. Often there is an expla- 
nation of why we forget — we are distracted, or confused, or possi- 
bly we may not even want to remember. If we fail to recall 
something, it is usually not because we have suffered a permanent 
loss of memory. We just cannot get access for the moment. And 
sometimes memory is frustratingly difficult. Remembering be- 
comes difficult when it is conscious, when we contrive deliberately 
to remember something that has not sprung to mind at once. 
Children don’t need to be taught to use memory efficiently — to 
avoid overloading short-term memory and to refrain from forcing 
pointless detail into long-term memory. They naturally do these 
things. But reading instruction may make these natural efficien- 
cies impossible. Anxiety while learning to read can force children 
into inefficient uses of memory. Reading, and therefore learning to 
read, depend on what you already know, on what you can make 
sense of. Reading teachers help to avoid overloading pupils' memo- 
ries when they ensure that the material the children are expected to 
read makes sense to them, so that they are not required — either by 
the material or by the instruction — to engage in extensive and 
pointless memorization. 
ISSUES 
There is little dispute about the characteristics of memory I have 
outlined (although there is much technical debate about physiolog- 
ical aspects). Problems arise when the limitations of memory are 
ignored, for example, in expecting large amounts of detailed mem- 
orization to take place during reading or learning, or on requiring 
the memorization of material that is not meaningful to the learner. 
SUMMARY 
Short-term memory and long-term memory both have their limi- 
tations, but these are handicaps only to the readers who can make 
little sense of what they are doing in the first place. The differing 
characteristics of memory are summarized in Fig. 6.3. When a 
reader can make sense of text and doesn’t strain to memorize, 
there is no awareness of the bottlenecks of memory. Fluent readers 
are immediately reminded of what is relevant for their current situ- 
ation and purposes. 
6. BOTTLENECKS OF MEMORY 
109 
Short-term memory 
(working memory ) 
Long-term memory 
( permanent memory) 
Capacity 
limited 
practically unlimited 
Persistence 
very brief 
practically unlimited 
Retrieval 
immediate 
depends on organization 
Input 
very fast 
relatively slow 
FIG. 6.3. Characteristics of short-term and long-term memory. 
Notes to chapter 6 begin on page 269 covering: 
Theories of memory 
Chunking 
Imagery 
Children’s memory 
Letter Identification 
After all the preliminaries, here is the first of four chapters specifi- 
cally on the topic of reading. The point I’m heading for is that fluent 
reading doesn’t normally require the identification of individual 
letters or words. But the most convenient route to that destination 
begins with a discussion of letter identification , focusing on an as- 
pect of reading where the issue can be concisely stated: How is it 
that anyone who knows the alphabet is able to distinguish and 
name any of the 26 alternatives on sight? I’m talking about sepa- 
rate letters, not letters in sequences of any kind. 
Unlike the reading of words, the question doesn’t arise of 
whether isolated letters are read a bit at a time or all at once. Indi- 
vidual letters can’t be “sounded out”: their appearance has a purely 
arbitrary relationship to the way they are pronounced. And there 
can be little question about their meaning; we “comprehend" a let- 
ter when we can say its name, and that is that. 
Yet despite this simplification, letter and word identification are 
alike in one important aspect — both involve the discrimination 
and categorization of visual information. Later we'll see that the 
manner in which letters are identified is relevant to an understand- 
ing of the identification of words. 
A brief digression may be in order to discuss the casual use of 
the terms identification and recognition as labels for how a letter 
110 
7. LETTER IDENTIFICATION 
111 
or word (or meaning) is distinguished and named. They are not, 
strictly speaking, synonymous. Identification involves a decision 
that an object should be put into a particular category. There is no 
implication that the object being identified should have been met 
before. Recognition, on the other hand, literally means that the ob- 
ject has been seen before, although identification may not be in- 
volved. We identify people when we put a name to them, whether or 
not we have met them before (identification cards aren’t recogni- 
tion cards). We recognize people when we know we have seen them 
before, whether or not we can put a name to them (police identifica- 
tion parades should be called recognition parades). 
Experimental psychologists and reading specialists usually talk 
about letter and word recognition, but the use of the term is doubly 
inappropriate. First, they would hardly consider a word to be rec- 
ognized unless its name could be given; they wouldn’t consider that 
a child recognized a word if all the child could say was “That’s the 
same squiggle I saw yesterday.” Second, skilled readers can very 
often attach a name to visual information that they have never met 
before. As a rather extreme case, do you “recognize” or “identify” 
the visual information rEaDiNg as the word reading ? You almost 
certainly have never seen the word written that way before. The 
weight of evidence would seem to favor identification, and the term 
is therefore used in this book for formal purposes, such as chapter 
headings. But having made the distinction, we need not be dog- 
matic about it; “identify,” “recognize,” “categorize,” “name,” and 
even “read” will, in general, continue to be used as conventional us- 
age and the context suggest. 
It is also not strictly correct to refer to the characters that we 
strive to identify on particular occasions as “letters”; this implies 
that the perceptual decision has already been made. Whether a 
particular mark on the page should be characterized as a letter de- 
pends on the perspective of the viewer (whether the reader or the 
writer). As we have seen, IO may be identified as two letters or two 
numerals. Prior to an identification decision, the visual informa- 
tion that is IO is merely a pattern of contrasting ink marks on pa- 
per, more precisely referred to as a visual configuration, a visual 
array, or even a visual stimulus. 
THEORIES OF PATTERN RECOGNITION 
Letter identification is a special problem within the broader theo- 
retical area of pattern recognition — the manner in which any two 
visual configurations are “cognized" to be the same. Recognition is 
112 
UNDERSTANDING READING 
of classical philosophical concern because it has been realized for 
over 2,000 years that no two events are ever exactly the same; the 
world is always in flux, and we never see an object twice in precisely 
the same form, from the same angle, in the same light, or with the 
same eye. A topic of general interest to psychology is what exactly 
determines whether two objects or events shall be considered to be 
equivalent. The equivalence decision clearly rests with the viewer 
and not in any property of the visual array. Are J and j the same? A 
printer would say no, although JOY and joy are the same word. Are 
two automobiles of the same year, model, and color identical? Pos- 
sibly to everyone except their owners. It is the viewer, not the ob- 
ject, that determines equivalence. 
We organize our lives and our knowledge by deciding that some 
things should be treated as equivalent — these are the things that 
we put into the same category — and some as different. Those dif- 
ferences between objects or events that help us to place them in 
category systems may be called by a variety of names, such as de- 
fining attributes or central attributes or distinctive features ; in 
essence, they are the differences that we choose to make signifi- 
cant. The differences that we choose to ignore, the ones that don't 
influence our decision, are often not noticed at all. Obviously it is 
more efficient to pay attention only to significant differences, par- 
ticularly in view of the limited information-processing capacity of 
the human brain. It is, therefore, hardly surprising that we may 
overlook differences that we are not looking for in the first place, 
like the sudden absence of our friend’s beard, or the pattern of his 
tie, or the misspelling in the newspaper headline. Human beings 
owe their preeminent position in the intellectual hierarchy of living 
organisms to their capacity to perceive things as the same accord- 
ing to criteria that they themselves establish, selectively ignoring 
what might be termed differences that don’t make a difference. 
The manner in which particular letters or words are treated as 
equivalent has become a focus of theoretical attention because of 
its particular application to computer technology. There is an obvi- 
ous economic as well as theoretical interest in designing comput- 
ers that might be able to read. The construction of a computer with 
any fluent degree of reading ability has proved difficult for a num- 
ber of reasons, one of which is that language can be understood 
only if there is an underlying understanding of the topic to which 
the language refers and the ability of computers to "understand" 
any topic is very limited indeed. It has not proved feasible to pro- 
vide a computer with rules for identifying all the printed and hand- 
written letters that people are able to identify, let alone words or 
7. LETTER IDENTIFICATION 
113 
meanings, with anything like the facility with which humans can 
identify them. But if we consider the problems of pattern recogni- 
tion from the computer point of view, we get some insights into 
what must be involved in the human skill (even though people 
aren’t computers). 
There are two basic ways in which a computer might be con- 
structed to recognize patterns, whether numbers, letters, words, 
texts, photographs, fingerprints, voiceprints, signatures, dia- 
grams, maps, or real objects, like faces. The two ways are essen- 
tially those that appear to be open theoretically to account for the 
recognition of patterns by humans. The alternatives may be called 
template matching and feature analysis , and the best way to de- 
scribe them is to imagine trying to construct a computer capable of 
identifying the 26 letters of the alphabet in their various forms. 
For both the template-matching and the feature-analytic de- 
vices, the ground rules are the same. At the input end is an optical 
scanner or "eye” to examine various patterns of visual information 
for each letter of the alphabet, and at the output end there is a set of 
26 alternative responses, the names or representations of each of 
the letters. The aim is to construct a system between the input and 
output mechanisms to ensure that given an input of &, the com- 
puter will say or print out “g.” 
Template Matching 
For a template-matching device, a series of internal representa- 
tions must be constructed to be, in effect, a reference library for 
the letters that the device is required to identify. We might start 
with one internal representation, or template, for each letter of 
the alphabet. Each template is directly connected with an appro- 
priate response, and between the optical detector (the “eye”) and 
the templates we shall put a device capable of comparing any in- 
put letter with all of the set of templates. Any letter that comes into 
the computer’s field of view will be internalized and compared 
with each of the templates, at least until a match is made. Upon 
matching the input with a template, the computer will perform 
the response associated with that particular template and the 
identification will be complete. 
There are obvious limitations to such a system. If the computer 
is given a template for the representation A, what will it do if con- 
fronted with A or A, not to mention .flor A? Of course, some flexibil- 
ity can be built into the system. Inputs can be normalized to iron 
out some of the variability; they can be scaled down to a standard 
114 
UNDERSTANDING READING 
size, adjusted into a particular orientation, have crooked lines 
straightened, small gaps filled, and minor excrescences removed; 
in short, a number of things can be done to increase the probability 
that the computer will not respond “I don’t know” but will instead 
match an input to a template. But, unfortunately, the greater the 
likelihood that the computer will make a match, the greater is the 
probability that it will make a mistake. This is the signal detection 
problem of chapter 4. A computer that can normalize A to make it 
look like the template A will be likely to do the same with 4 and H. 
The only remedy will be to keep adding templates to try to accom- 
modate all the different styles and types of lettering the device 
might meet. Even then, such a computer will be unable to make use 
of all the supporting knowledge that human beings have. 
Critical limitations of template-matching systems, for both com- 
puter and human, lie in their relative inefficiency and costliness. A 
single set of templates, one for each category, is highly restricted in 
the number of inputs that it can match, but every increase in the 
number of templates adds considerably to the size, expense, and 
complexity of the system, and also to the probability of false 
alarms. The template-matching model can work, but usually by 
cheating its way around the problem of the diversity of input repre- 
sentations. Instead of providing the computer with templates to 
meet many innumerable character styles, it makes sure the com- 
puter eye meets only a limited set of alternatives, like the num- 
bers 5 2101 * that are printed on our checks. 
Feature Analysis 
The alternative method of pattern recognition, feature analysis, 
dispenses with internal representations completely. There is no 
question of attempting to match the input with anything. Instead, a 
series of tests is made on the input. The results of each test elimi- 
nate a number of alternatives until, finally, all uncertainty is re- 
duced and identification is achieved. The features are properties of 
the visual array that are subjected to tests to determine which al- 
ternative responses should be eliminated. Decisions about which 
alternatives each test will eliminate are made by the viewers (or 
computer programmers) themselves. 
Let’s again imagine constructing a device capable of identifying 
letters, this time using feature analysis. Remember, the problem is 
essentially one of using rules to decide into which of a limited num- 
ber of categories might a very large number of alternative events be 
placed. In other words this is a matter of establishing equivalences. 
7. LETTER IDENTIFICATION 
115 
At the input end of the system, where the computer has its opti- 
cal scanner, we establish a set of “feature analyzers.” A feature ana- 
lyzer is a specialized kind of detector that looks for— is sensitive 
to — Just one kind of feature in the visual information and that 
passes just one kind of report back. We might imagine that each an- 
alyzer looks for a particular distinctive feature by asking a ques- 
tion: One asks, “Is the configuration curved?” (like C or O); another 
asks, "Is it closed?” (like O or P); a third asks, “Is it symmetrical?” 
(like A or W); and a fourth asks, “Is there an intersection?” (as in T 
or K). Every analyzer is in effect a test and the message it sends 
back is binary — either “yes” or “no.” Without looking too closely at 
the question of what constitutes a distinctive feature, we can say it 
is a property of visual information that can be used to differentiate 
some visual configurations from others — a “significant difference.” 
A distinctive feature must be common to more than one object or 
event; otherwise, it could not be used to put more than one into the 
same category, it would be all one needs to know. But on the other 
hand, if the feature were present in all objects or events, then we 
could not use it to segregate objects into different categories: it 
would not be “distinctive.” In other words, a feature permits the 
elimination of some of the alternative categories into which a stim- 
ulus might be allocated. 
For example, a “no” answer to the test "Is the configuration 
curved?” would eliminate rounded letters such as a, b, c, d but not 
other letters such as i, k, l, w, x, z. A “yes” answer to “Is it closed?” 
would eliminate open letters such as c,f, w but not b, d, o. A ques- 
tion about symmetry would distinguish letters like m, o, w, v from 
d,f, k, r. Different questions eliminate different alternatives, and 
relatively few tests would be required to distinguish among 26 al- 
ternatives in an alphabet. In fact, if all tests eliminated about half of 
the alternatives, and there was no test that overlapped with any 
other, only five questions would be needed to identify any letter. 
(The logic of the previous statement is set out in the notes.) No one 
would actually suggest that as few as five tests are employed to dis- 
tinguish among 26 letters, but it is reasonable to assume that there 
need be many fewer tests than categories — that is one of the great 
economic advantages of a feature-analytic system. 
With an input bank of feature analyzers built into the letter-iden- 
tification device, a link has to be provided to the 26 responses or 
output categories; “decision rules” must be devised so that the re- 
sults of the individual tests are integrated and associated with the 
appropriate letter names. The most convenient way to set up the 
rules is to establish a feature list for each category, that is, for each 
116 
UNDERSTANDING READING 
of the 26 letters. The construction of the feature lists is the same for 
every category, namely, a listing of the analyzers that were set up to 
examine the visual configuration. The feature list for every category 
also indicates whether each particular analyzer should send back 
a "yes” or "no” signal for that category. For the category c, for exam- 
ple, the feature list should specify a “yes” for the “curved?” ana- 
lyzer, a "no” for the “closed?” analyzer, a “no” for the “symmet- 
rical?” analyzer, and so forth. 
The actual wiring of the letter-identification device presents no 
problems — every feature analyzer is connected to every category 
that lists a “yes” signal from it, and we arrange that a categorizing 
decision (an "identification”) is made only when “yes” signals are 
received for all the analyzers listed positively on a category’s fea- 
ture list. In a sense, a feature list is a specification of what the char- 
acteristics of a particular letter should be. Descriptions of a letter 
that is being looked at — the input — are compared with specifica- 
tions of what the letter might be until a match is found. 
And that, in schematic form, is a feature-analytic letter-identifi- 
cation device. The system is powerful, in the sense that it will do a 
lot of work with a minimum of effort. Unlike the template model, 
which to be versatile requires many templates for every decision 
that it might make together with complex normalizing devices, the 
feature-analytic device demands only a very small number of ana- 
lyzers compared with the number of decisions it makes. Theo- 
retically, such a device could decide among over a million 
alternatives with only 20 questions. 
Functional Equivalence and Criterial Sets 
A considerable advantage of the feature-analytic model over tem- 
plate matching is that the former has much less trouble adjusting 
to inputs that ought to be allocated to the same category but which 
vary in size or orientation or detail, for example. A, A, and H The 
types of tests that feature analyzers apply are far better able to 
cope with distortion and noise than any device that requires an 
approximate match. But far more important, very little is added 
in the way of complexity or cost to provide one or more alternative 
feature lists for every category. With such a flexibility, the system 
can easily allocate not only the examples already given, but also 
forms as divergent as A\ a, a, and A to the category “A.” The only 
adjustment that need be made to the battery of analyzers is in wir- 
ing additional connections between them and the categories to 
which the analyzers are relevant, so that an identification will be 
7. LETTER IDENTIFICATION 
117 
made on any occasion when the specifications of any of the alter- 
native lists are satisfied. 
Call any set of features that meet the specifications of a particu- 
lar category a crlterial set. With the type of feature -analytic device 
being outlined, more than one criterial set of features may exist for 
any one category. Obviously, the more criterial sets that exist for a 
given device, the more efficient that device will be in making accu- 
rate identifications. 
It is also useful to give a special name to the alternative criterial 
sets of features that specify the same category — we shall say that 
they are functionally equivalent. A, CL, and a are functionally 
equivalent for our imagined device because they are all treated as 
being the same as far as the category “A” is concerned. Of course, 
the configurations are not functionally equivalent if they are to be 
distinguished on a basis other than their membership of the al- 
phabet; a printer, for example, might want them categorized into 
different fonts or type styles. But as I pointed out earlier, it is the 
prerogative of the viewer, not a characteristic of the visual configu- 
ration, to decide which differences shall be significant — which 
sets of features shall be criterial — in the establishment of equiva- 
lences. All that is required to establish functional equivalence for 
quite disparate visual configurations is alternative feature lists 
for the same category. 
Another powerful aspect of the feature -analytic model of pattern 
recognition is that it can work on a flexible and probabilistic basis. 
If a single feature list specifies the outcomes of ten analyzer tests 
for the categorical identification of a letter of the alphabet, a con- 
siderable amount of redundant information must be involved. Re- 
dundancy, as I noted in chapter 4, exists when more information is 
available than is required to reduce the actual amount of uncer- 
tainty. Ten analyzer tests could provide enough information to se- 
lect from over a thousand equally probable alternatives, and if 
there are only 26 alternatives, information from five of those tests 
could be dispensed with and there might still be enough data to 
make the appropriate identification. Even if analyzer information 
were insufficient to enable an absolutely certain selection, it might 
still be possible to decide which alternatives are more likely, given 
the particular pattern of features that is discriminated. By not de- 
manding that all the specifications of a particular feature list be 
satisfied before a category identification is made, the system can 
greatly increase its repertoire of functionally equivalent criterial 
sets of features. Such an increase significantly enhances the effi- 
ciency of the device at a cost of little extra complexity. 
118 
UNDERSTANDING READING 
The fact that different criterial sets can be established within a 
single feature list provides an advantage that was alluded to in the 
previous paragraph — a feature -analytic system can make use of re- 
dundancy. Let’s say that the system already “knows” from some 
other source of information that the configuration it is presented 
with is a vowel; it has perhaps already identified the letters THR . . . 
and it has been programmed with some knowledge of the spelling 
patterns of English. The device can then exclude from consider- 
ation for the fourth letter all those feature lists that specify conso- 
nant categories, leaving considerably reduced criterial sets for 
selection among the remaining alternatives. (Three tests might 
easily distinguish among five or six vowels.) 
A final powerful advantage of the feature -analytic model has also 
already been implied; it is a device that can easily learn. Every time 
a new feature list or criterial set is established there is an instance 
of learning. All that the device requires in order to learn is feed- 
back from the environment. It establishes, or rejects, a new feature 
list for a particular category (or category for a particular feature 
list) by "hypothesizing” a relationship between a feature list and a 
category and testing whether that relationship is appropriate. 
You may have noticed that the feature -analytic discussion devel- 
ops easily into such topics as “learning" and “thinking." It is evident 
that the more efficient and sophisticated we make our hypothetical 
letter-identifying device, the more we are likely to talk about its re- 
alization in human rather than computer terms. It is time to dis- 
card the computer analogy and to direct a more specific focus on 
the human pattern recognizer. 
The Human Letter Identifier 
I’ll use feature analysis as a model for the way in which letters are 
identified by readers. We learn to identify the letters of the alphabet 
by establishing feature lists for the required 26 categories, each of 
which is interrelated to a single name “A,” “B.” “C,” and so forth. 
The visual system is equipped with analyzers that respond to those 
features in the visual environment that are distinctive for alpha- 
betic discriminations (and many other visual discriminations as 
well). The results of the analyzer tests are integrated and directed 
to the appropriate feature lists so that letter identifications can oc- 
cur. The human visual perceptual system is biologically competent 
to demonstrate all the most powerful aspects of the feature-ana- 
lytic model outlined in the last section — to establish manifold 
criterial sets of features with functional equivalence, to function 
7. LETTER IDENTIFICATION 
119 
probabilistically, to make use of redundancy, and to learn by test- 
ing hypotheses and receiving feedback. Establishing feature lists, 
in other words, is natural. 
Two aspects of letter identification must be distinguished. The 
first aspect is the establishment of cognitive categories them- 
selves and especially the allocation of category names to them, 
such as “A,” “B,” “C.” This might be termed “learning to say the let- 
ters of the alphabet.” The second aspect of letter identification is 
the allocation of visual configurations to various cognitive catego- 
ries — the discrimination of various configurations as different, as 
not functionally equivalent. This might be termed “learning to 
recognize (or identify) the letters of the alphabet.” The greater 
part of perceptual learning involves finding out what exactly are 
the distinctive features by which various configurations should be 
categorized as different from each other and what are the sets of 
features that are criterial for particular categories. These are pre- 
cisely the two aspects of object or concept learning involved in dis- 
tinguishing one face from another, or cats from dogs. Categories 
must be established with unique names (like “cat” and “dog”), and 
rules must be devised for allocating particular instances to the 
appropriate categories. 
The association of a name with a category is neither necessary 
nor primary in visual discrimination. It is quite possible to segre- 
gate visual configurations into different categories without having a 
name for them. We can see that A and & are different and know they 
should be treated differently, even though we may not have a cate- 
gory name, or even a specific category, for &. In fact, we can’t allo- 
cate a name to & unless first we acquire some unconscious rules 
for discriminating it from A and from every other visual configura- 
tion with which it should not be given functional equivalence. The 
motivation for the establishment of a new category may come from 
either direction: Either a configuration such as & can’t be related to 
any existing category, or a new name such as “ampersand” can't be 
related to an existing category. The intermediate steps that tie the 
entire system together are the establishment of the first feature 
lists and criterial sets for the category so that the appropriate fea- 
ture tests and the category name can be related. 
Not only is relating a name to a category not primary: it is not dif- 
ficult. The complicated part of learning to make an identification is 
not in remembering the name of a particular category but in dis- 
covering the criterial sets of features for that category. Children at 
the age when they are often learning to read are also learning thou- 
sands of new names every year — names of friends and celebrities 
120 
UNDERSTANDING READING 
and automobiles and animals — as well as the names of letters and 
words. The person they learn from, the informal instructor, usu- 
ally points to or refers to an object and says “That’s an X,” leaving 
the child to work out what the significant differences must be. The 
complicated part of learning is the establishment of functional 
equivalences for the categories with which names are associated. 
The reason that “learning names" is frequently thought to be 
difficult is that all the intermediate steps are ignored and it is as- 
sumed that a name is applied directly to a particular visual con- 
figuration. Children may find it difficult to respond with the 
right name for the letters b or d (or for the words house and 
mouse, or for an actual dog or cat) but not because they can’t put 
the name to the configuration. Their basic problem is to find out 
how two alternatives are significantly different. Once learners 
can make the discrimination, so that the appropriate functional 
equivalences are established, the allocation of the correct verbal 
label is a relatively easy problem because the label is related di- 
rectly to the category. 
The Letter Identifier in Action 
Is there evidence to support the feature -analytic model of the hu- 
man visual system? Some of the physiological evidence has already 
been indicated. There is no one-to-one correspondence between 
the visual information impinging on the eye and anything that goes 
on behind the eyeball. The eye doesn’t send “images” back to the 
brain; the stammering pattern of neural impulses is a representa- 
tion of discrete features detected by the eye, not the transfer of a 
“picture.” In the brain itself, there is no possibility of storing tem- 
plates or even acquiring them in the first place. The brain doesn’t 
deal in veridical representations; it shunts abstractions through 
its complex neural networks. It is true that one aspect of the brain’s 
output, our subjective experience of the world, is generated in the 
form of “percepts” that might be regarded as pictures, but this ex- 
perience is a consequence of the brain's activity, not something the 
brain “stores” and compares with inputs. Our visual experience is 
the product of the perceptual system, not part of a visual process. 
Now we can examine evidence for the feature -analytic model 
from two kinds of letter-identification experiments. (Details are 
given in the notes. ) The basic assumption to be tested is that letters 
are actually conglomerates of features, of which there are perhaps 
a dozen different kinds. The only way in which letters differ visu- 
ally from each other is in the presence or absence of these features. 
7. LETTER IDENTIFICATION 
121 
Letters that have several features in common will be regarded as 
similar, and letters that are constructed of quite different feature 
combinations will be considered quite dissimilar in appearance. 
How does one assess similarity? Letters are similar — they are pre- 
sumed to share many features — if they are frequently confused 
with each other. Letters that are rarely confused with each other 
are assumed to have very few features in common. 
Of course, we don't confuse letters very often, and when we do 
the character of the error is usually influenced by nonvisual fac- 
tors. We might, for example, think that the fourth letter in the se- 
quence REQW ... is 17, not because W and U are visually similar but 
because we normally expect a U to follow Q. However, large num- 
bers of visual letter confusions can be generated by experimental 
techniques in which the letter to be identified is so “impoverished” 
that viewers can’t see it clearly, although they are forced to make a 
guess about what the letter probably is. In other words, the experi- 
mental viewers must make letter-identification decisions on mini- 
mal visual information. The research assumption is that viewers 
who can’t see the letter clearly must lack some vital information 
and thus be unable to make some feature tests. And if they are un- 
able to make certain feature tests, then the tests they are able to 
make will not reduce all uncertainty about the 26 alternative re- 
sponses. Viewers will still be left in doubt about a few possibilities 
that could be differentiated only by the tests that they have been un- 
able to perform. 
The actual method of visual degeneration is not important. The 
presentation may be a brief tachistoscopic presentation, or it may 
involve an image that is difficult to discriminate, projected at a low 
intensity or hidden behind a lot of visual noise. As soon as viewers 
start making errors, one can assume that they are not getting all 
the information that they need to make an identification. They are 
deciding on something less than a criterial set of features. 
There are only two possibilities if viewers are forced to identify a 
letter on insufficient visual information: Either their guesses will 
be completely random, or they will respond in some systematic 
way. If guesses are random they could be to respond with any of the 
26 letters of the alphabet. But if responses are systematic, the in- 
teresting possibility is that the viewers are selecting only from 
those alternative responses that remain after the features that can 
be discriminated have been taken into account. In such a circum- 
stance, it is to be expected that the confusions will “cluster”; in- 
stead of 25 types of confusion, one for each possible erroneous 
response, there will be only a few types. 
122 
UNDERSTANDING READING 
The evidence is that letter confusions fall into tightly packed 
clusters, and over two thirds of the confusions for most letters can 
be accounted for by three or four confusion types. If a mistake is 
made in the identification of a letter, the nature of the erroneous re- 
sponse is highly predictable. Typical confusion clusters can be very 
suggestive about the kind of information the eye must be looking 
for in discriminating letters. Some typical confusion clusters are 
(a. e, n, o, u), (t,J, i), and (h, m, n) (Dunn-Rankin, 1968). 
The specific conclusion to be drawn from the kind of experiment 
just described is that letters are indeed composed of a relatively 
small number of features. Letters that are easily confused, like a 
and e or t and f, must have a number of features in common, and 
those that are rarely confused, like o and w or d and y, must have 
few if any features in common. The general conclusion that may be 
drawn is that the visual system is indeed feature analytic. Letter 
identification is accomplished by examination of the visual envi- 
ronment for featural information that will eliminate all alternatives 
except one, thus permitting an accurate identification to be made. 
There is a second line of experimental evidence supporting the 
view that letters are arrangements of smaller elements, and this is 
related to the fact that recognition is faster or easier when there are 
fewer alternatives for what each letter might be. The classic exam- 
ple of such evidence has already been described in the tunnel vi- 
sion demonstration of chapter 5, where it was shown that 
nonvisual information can be employed to reduce the amount of vi- 
sual — or distinctive feature — information required to identify let- 
ters. There are other illustrations when the identification of words 
is considered in the next chapter. 
WHAT IS A FEATURE? 
The entire discussion of letter identification by feature analysis has 
been conducted without actually specifying what a feature is. The 
omission has been deliberate, because nobody knows what the dis- 
tinctive features of letters are. Not enough is known about the hu- 
man visual system to say exactly what is the featural information 
that the system looks for. 
Of course, general statements about features can be made. 
There have been a number of attempts to do this, with reasoning 
like “The only difference between c and o is that the circle of o is 
‘closed’; therefore, being closed must be a distinctive feature,” or 
“The only difference between h and n is the ’ascender' at the top of 
h; therefore, an ascender must be a distinctive feature." But we re- 
7. LETTER IDENTIFICATION 
123 
ally don't know whether, or how, the eyes might look for “closed- 
ness” or for “ascenders.” It can be argued that these hypothesized 
features are really properties of whole letters, and it is far from 
clear how a property of the whole could also be an element out of 
which the whole is constructed. It is obviously a reasonable pro- 
posal that the significant difference between h and n has something 
to do with the ascender, but it is an oversimplification to say that 
the ascender is the actual feature. 
Fortunately, it is not necessary to know exactly what the features 
are in order to assist children in discriminating letters. We can 
trust children to locate the information required provided the ap- 
propriate informational environment is available. The appropriate 
informational environment is the opportunity to make compari- 
sons and discover what the significant differences are. Remember, 
the primary problem of identification is to distinguish the pre- 
sented configuration from all those to which it might be equivalent 
but is not; the configuration has to be subjected to feature analysis 
and put in the appropriate category. Presenting h to children 50 
times and telling them it is “h” because it has an ascender will not 
help them to distinguish the letter. The presentation of h with other 
letters that are not functionally equivalent is the kind of informa- 
tion required for the brain to find out very quickly what the distinc- 
tive features are. 
ISSUES 
Whether physiological evidence — for example, from eye-movement 
or brain-function studies — should make any difference to how 
reading should be taught is a contentious question. Should learn- 
ers be taught “feature recognition,” for example, when the status of 
features is hypothetical and everyone who has learned to read in 
the past has coped without such specialized knowledge? This is 
part of a general issue that comes up constantly in this book, al- 
though frequently taken for granted elsewhere: whether descrip- 
tions of what readers do (or are hypothesized to do) should be the 
basis of what learners are taught. 
SUMMARY 
A feature identification model is proposed for letter identifica- 
tion. Feature lists permit the allocation of visual information to 
specific cognitive categories, in the present case for letters. Fea- 
ture lists are specifications of what visual information has to be 
124 
UNDERSTANDING READING 
like in order to be allocated to a particular category. The names of 
letters are part of the interrelations among categories. To permit 
the identification of the same letter when it has different configu- 
rations, for example.il A, a, and a., functionally equivalent fea- 
ture lists are established. For each feature list there will be a 
number of alternative criterial sets to permit identification deci- 
sions on a minimum of visual information, depending on the 
number and nature of the alternatives. 
Notes to chapter 7 begin on page 274 covering: 
Recognition versus identification 
Theories of pattern recognition 
Word Identification 
The preceding chapter was devoted to letter identification. In this 
chapter, I show that the prior identification ofletters is not required 
for the identification of entire words. This chapter is restricted to 
considering words in isolation, where there is no extrinsic clue to 
their identity. I am still not focusing on anything that might normally 
be regarded as reading, where a meaningful purpose and context 
are involved. But the chapter is another step toward a demonstra- 
tion that procedures permitting the visual identification of words 
without the prior identification of letters also permit comprehen- 
sion without the prior identification of words. 
THREE THEORIES OF WORD IDENTIFICATION 
There are three broad classes of theory about word identification: 
whole-word identification, letter-by-letter identification, and an 
intermediate position involving the identification of letter clusters, 
usually termed spelling patterns. In effect, these three views repre- 
sent three attempts to describe the manner in which a skilled 
reader is able to identify words on sight. They are accounts of what 
a reader needs to know and do in order to be able to say what a 
word is. One or another of the three views is apparent in practically 
every approach to reading instruction. 
125 
126 
UNDERSTANDING READING 
The whole-word view is based on the premise that readers don’t 
stop to identify individual letters (or groups of letters) in the identi- 
fication of a word. Instead the word is identified on the basis of its 
shape, or its "structure.” The view asserts that knowledge of the al- 
phabet and of the “sounds of letters” is irrelevant to reading. One 
source of support for the whole-word view has already been al- 
luded to — the fact that a viewer can report from a single tachisto- 
scopic presentation either four or five random letters or a similar 
number of words. Surely if a word can be identified as easily as a 
letter, then it must be just as much of a unit as a letter; a word must 
be recognizable as a whole, rather than as a sequence of letters. An- 
other piece of supporting evidence is that words may be identified 
when none of their component letters is clearly discriminable. For 
example, a name may be identifiable on a distant roadside sign, or 
in a dim light, under conditions that would make each individual 
letter of that name illegible if presented separately. If words can be 
read when letters are illegible, how can word recognition depend 
on letter identification? Finally, there is a good deal of evidence that 
words can be identified as quickly as letters. It has been shown that 
perception is far from instantaneous and that successively pre- 
sented random letters — or random words — can’t be identified 
faster than five or six per second (starting with Kolers & Katzman, 
1966; Newman, 1966). And if entire words can be identified as 
quickly as letters, how can their identification involve spelling 
them out letter by letter? 
A fundamental objection to the whole-word point of view is that 
it is not a theory at all; it merely rephrases the question it claims to 
answer. If words are recognized “as wholes," how are the wholes 
recognized? What exactly do readers know if they know what a 
word looks like? The qualification that words are identified “by 
their shapes” merely changes the name of the problem from “word 
identification” to “shape identification." Fluent readers are able to 
recognize at least 50,000 different words on sight (see Notes) — by 
what I call immediate word identification. Does that mean that 
readers have pictures of 50,000 different shapes stored in their 
minds and that for every word they encounter in reading they rum- 
mage through a pack of 50,000 templates in order to find a match? 
In what way would they sort through 50,000 alternatives? Surely 
not by starting at aarduark and examining each internal represen- 
tation until they find a match. If we are looking for a book in a li- 
brary, we don’t start at the entrance and examine every volume 
until we come across the one with a title that matches the title we 
are looking for. Instead, we make use of the fact that books are cate- 
8. WORD IDENTIFICATION 
127 
gorized and shelved in a predictable way; there are shortcuts for 
getting to the book we want. It would appear reasonable that we 
make use of shortcuts to make our word identification decisions 
quickly. We can usually find some explanation for any error that we 
make. We may misread said as sail or send (or even as reported in 
circumstances where the substitution would make sense), but 
never as elephant , plug , or predisposition. In other words, we ob- 
viously don’t select a word from 50,000 alternatives, but rather 
from a much smaller number. An unelaborated whole-word point 
of view can’t account for this prior elimination of alternatives. 
Besides, we have already discovered that 50,000 internal repre- 
sentations of shapes would be far from adequate to enable us to 
identify 50,000 different words. Even if we could identify HAT by 
looking up an internal representation, how could the same repre- 
sentation enable us to identify hat or Jaf or any of the many other 
ways in which the word may be written? 
The letter-by-letter theory, which the whole-word view is sup- 
posed to demolish, itself appears to have substantial supportive 
evidence. Readers are frequently sensitive to individual letters in 
the identification of words. The whole-word point of view would 
suggest that if viewers were presented with the stimulus fashixn 
tachistoscopically, they would either identify “the whole word" 
without noticing the x or else fail to recognize the word at all be- 
cause there would be no “match” with an internal representation. 
Instead, viewers typically identify the word but report that there is 
something wrong with it, not necessarily reporting that there is an 
x instead of an o, but offering such explanations as “There’s a hair 
lying over the end of it” — an observation first made over a century 
ago (Pillsbury, 1897). 
Furthermore, readers are very sensitive to the predictability of 
letter sequences. Letters don’t occur haphazardly. In English, for 
example, combinations like th, st, br, and almost any consonant 
and vowel pair are more likely to occur than combinations like tf, 
sr, bm, ae , or uo. The knowledge that readers acquire about these 
differing probabilities of letter combinations is demonstrated 
when words containing common letter sequences are more easily 
identified than those with uncommon sequences. Readers can 
identify sequences of letters that are not English words just as eas- 
ily as some English words, provided the sequences are “close ap- 
proximations” to English — which means that they are highly 
probable letter combinations (Miller, Bruner, & Postman, 1954). 
The average reader, for example, hardly falters when presented 
with sequences like vernalit or mossiant or ricaning — yet how 
128 
UNDERSTANDING READING 
could these be identified “as wholes” when they have never been 
seen before? 
A rather illogical argument is sometimes proposed to support 
the letter-by-letter theory. Because letters in some way spell out ap- 
proximations to the sound of a word (the “phonics” point of view), 
word identification must be accomplished by identification of the 
individual letters. It would be about as compelling to suggest that 
we must recognize models of cars by reading the manufacturer’s 
name on the back, simply because the name is always there to be 
read. Besides, the spelling of words is not a reliable guide to their 
sound. This question is so complex that phonics is given the next 
chapter to itself. For the moment we aren’t concerned with whether 
knowledge of letters can be used to identify words, but rather with 
whether skilled readers normally and necessarily identify words 
“that they know" by a time- and attention-consuming letter-by-let- 
ter analysis. 
The intermediate position — that words are identified through 
the recognition of clusters of letters — has the advantage of being 
able to account for the relatively easy identifiability of nonwords 
such as vernalit. It argues that readers become familiar with spell- 
ing patterns, such as ve and rn and even vern, which are recog- 
nized and put together to form words. The larger the spelling 
patterns we can recognize, the easier the word identification. The 
view is compatible with our normal experience that when a new 
word like zygotic or Helsincrfors halts our reading temporarily we 
don’t seem to break it down to individual letters before trying to put 
together what its meaning or sound must be. But many of the argu- 
ments that favor the whole-word position over letter analysis also 
work against the letter-cluster view. It may be useful, occasionally, 
to work out what a word is by analysis of letters or syllables, but 
normal reading doesn’t appear to proceed on this basis; in fact, it 
would seem impossible. There isn’t time to work out what words 
are by synthesizing possible letter or sound combinations. Be- 
sides, as the letter-cluster argument is pushed to its extreme it be- 
comes a whole -word approach because the largest and most 
reliable spelling patterns are words themselves. 
A Feature-Analytic Alternative 
Any serious attempt to understand reading must be able to explain 
why it might sometimes appear that words are identified as wholes 
and at other times through the apparent identification of compo- 
nent letters or groups of letters. In the previous chapter, two mod- 
8. WORD IDENTIFICATION 
129 
els for letter identification were examined: feature analysis and 
template matching. The traditional whole-word theory that words 
are identified because of the familiarity of their “shape” is essen- 
tially a template -matching model, and arguments for its inade- 
quacy have already been presented. The remainder of the present 
chapter considers the alternative, afeature-analytic model for the 
identification of individual words in isolation, or effectively so be- 
cause context is ignored. The identification of words in meaningful 
sequences, which is of course far more representative of most 
reading situations, is considered in chapter 10. 
Basically, the feature -analytic model proposes that the only dif- 
ference between the manner in which individual letters and indi- 
vidual words are identified lies in the complexity of the categories 
and feature lists that the viewer employs in the analysis of visual in- 
formation. The difference depends on whether the reader is look- 
ing for letters or for words; the process of looking and deciding is 
the same. If the reader’s objective is to identify letters, then the 
analysis of the visual configuration is carried out with respect to 
the feature lists associated with the 26 letter categories, one for 
each letter of the alphabet. If the objective is to identify words, then 
there is a similar analysis of features in the visual configuration 
with respect to the feature lists, or specifications, of a larger num- 
ber of word categories. 
What are the features of words? They obviously include the fea- 
tures of letters, because words are made up of letters. The arrays of 
marks on the printed page that can be read as words can also be 
distinguished as sequences of letters, so the “distinctive features” 
that constitute significant differences between one letter and an- 
other must also be distinctive features of words. For example, 
whatever visual information permits us to distinguish between h 
and n must also permit us to distinguish between hot and not. At 
first glance, many more discriminations and analyses of distinc- 
tive features would appear to be required to distinguish among 
tens of thousands of alternative words compared with only 26 al- 
ternative letters, but we shall see that the difference is not so great. 
In fact, no more information — no more featural tests — may be re- 
quired to identify a word in meaningful text than to identify a single 
letter in isolation. 
If the distinctive features of the visual configurations of letters 
are the same as those for the visual configurations of written 
words, it might be expected that feature lists for letter and word 
categories would be similar. However, feature lists for word catego- 
ries allow an additional dimension to those for letters — the analy- 
130 
UNDERSTANDING READING 
sis of word configurations involves the position of features within a 
sequence. The number oppositions” in a word feature list reflects 
the number of times a particular feature could occur in the se- 
quence of letters that constitute the word and obviously corre- 
sponds to the number of letters. A feature test that will be applied 
only once for the identification of a letter may be employed several 
times in the identification of a word, with the maximum number of 
tests equaling the number of letters in the word. This congruence 
between “position” and “letter” occurs because the distinctive fea- 
tures of individual letters become the distributed distinctive fea- 
tures of words. But it doesn’t follow that individual letters must be 
identified in order for words to be identified. I therefore use the 
term “position” rather than “letter” to avoid any implication that a 
word is identified from its letters, rather than by the distribution of 
features across its entire configuration. 
There could be a few distinctive features of words that are not 
features of letters, for example, the total width of the word configu- 
ration and — if the word is in lower case — the relative heights of dif- 
ferent parts. Many words in context can be identified by their 
contours (Haber & Haber, 1981; Haber, Haber. & Furl in, 1983). 
But as I have already noted, not enough is known of the visual sys- 
tem to assert what distinctive features might actually be. 
The feature-analytic view of letter identification proposes that be- 
cause there is redundancy in the structure of letters — because there 
is more than enough featural information to distinguish among 26 
alternatives — not all features need be discriminated for a letter to be 
identified. Therefore, a number of alternative criteria! sets of fea- 
tures may exist within each feature list, information about the fea- 
tures within any set being sufficient for an identification to be made. 
It would be expected that criteria! sets could also exist for words to 
be identified, except that now they would extend over a second di- 
mension and take into account feature combinations across the en- 
tire word. As I explained in chapter 7 and its notes, five or six 
features would be adequate to identify any letter, although there are 
doubtless more. That means that words, with an average length of 
about five letters, must consist of at least 25 distinctive features. The 
actual uncertainty of words in isolation (see page 264) requires only 
half that number (and only about a quarter for words in context). 
Redundancy Among Distinctive Features of Words 
Experienced readers acquire wide knowledge of the way in which 
letters are grouped into words, for example that th and sp are fre- 
8. WORD IDENTIFICATION 
131 
quent combinations and that tf and sr are not. This knowledge of 
the way words are spelled, or orthographic information, is an al- 
ternative nonvisual source of information to th e featural or visual 
information available to the eyes from the page. To the extent that 
both of these sources of information reduce the number of alterna- 
tive ways in which a written or printed word might be constructed, 
there is redundancy. The duplication of information resulting from 
orthographic predictability is a form of sequential redundancy , 
because its source lies in the fact that the different parts of a word 
are not independent; the occurrence of particular alternatives in 
one part of the sequence of letters limits the range of alternatives 
that can occur anywhere else in the sequence. 
The orthographic redundancy of English is enormous. If all 26 
letters of the alphabet could occur without restriction in each posi- 
tion of a five -letter word, there could be nearly 12 million different 
five-letter words, compared with perhaps 1 0,000 that actually exist. 
I have been referring to the sequential constraints that one letter 
places on the occurrence of other letters in a word. But precisely 
the same argument can apply to features. Obviously, if we can say 
that the occurrence of the letter T in the first position of a word re- 
stricts the possibilities for the second position to H, R, A, E, I, O, U, 
and Y, then we can also say that the occurrence of the features of the 
letter T restricts the possible features that can occur in the second 
position. In fact, we can avoid mentioning letters and specific posi- 
tions altogether and say that when certain features occur in one 
part of a word, there are limits to the kinds of feature combinations 
in other parts of the word and to what the word as a whole might 
be. A reader implicitly knowledgeable about such limitations is 
able to make use of sequential redundancy among features , and 
as a result will be able to identify words with so little visual infor- 
mation that the identification of letters is completely bypassed. 
Similar arguments concerning featural redundancy are made by 
linguists to explain the pronunciation and recognition of spoken 
words (Pinker, 1999, pp. 93-94). 
The visual identification of words becomes very fast and effi- 
cient. Fraisse (1984) observed that experienced readers could of- 
ten name printed words like house and tree flashed on a screen 
quicker than they could name drawings of the referents of the same 
words, even though they had previously seen the full set of words 
and drawings they would be asked to identify. 
To summarize, the difference between letter and word identifica- 
tion is simply the category system that is involved — the manner in 
which featural information is allocated. If the reader is examining 
132 
UNDERSTANDING READING 
an array of visual information in order to identify letters, the visual 
information will be tested and identifications made on the basis of 
the feature lists for the 26 categories. If the purpose is to identify 
words, the visual information will be tested with respect to the fea- 
ture lists for words, and there will be no question of letter identifi- 
cation. It follows from this argument that it should be impossible 
to identify a word and its component letters simultaneously, be- 
cause one can’t use the same information to make two different 
kinds of decision. 
Because letter identification and word identification involve the 
same featural information, it is not possible to identify a configura- 
tion both as a word and as a sequence of letters at the same time. 
We can see the configuration cat either as the letters c, a, t or as the 
word cat, but not as both simultaneously. Similarly, we can see the 
configuration read either as the word pronounced “reed" or as the 
word pronounced “red” but not as both at once; and IO can be seen 
either as numerals or as letters but not as both. We can't apply the 
same information to two categories simultaneously, just as we can’t 
use the same contour as part of two figures simultaneously — we 
can’t see the vase and the faces of Fig. 2.1 (page 18) simultaneously. 
LEARNING TO IDENTIFY WORDS 
Two aspects of learning to identify words are analogous to the two 
aspects of learning to identify letters outlined at the end of the last 
chapter. One aspect is establishing criterial sets of functionally 
equivalent distinctive features for each category, the specifications 
for qualifying configurations, and the other aspect is associating a 
name with a category. For letter identification, it was asserted that 
relating the name to the category was not a problem; children learn 
names for visual configurations all the time. In word identification, 
there may indeed be a problem in relating names to categories, be- 
cause children may have difficulty not in remembering the name for 
a category once they have found out what it is, but in ascertaining the 
category name in the first place. When children are beginning to dis- 
cover written language, helpful adults often act as mediators by say- 
ing what the printed words are, leaving to the child the more 
complex task of discovering how to distinguish one word from an- 
other. Someone says to them “That word is cat" or “This is how your 
name is written.” Or they read a simple story with them. 
Finding out the name of a category in the absence of outside help 
may be termed mediated word identification and is the topic of 
the next chapter. Word identification must be mediated when a 
8. WORD IDENTIFICATION 
133 
word can't be identified on sight by allocation to a category through 
an existing feature list. By contrast, I refer to word identification as 
discussed in this chapter as immediate word identification. The 
term immediate is used not in the sense of instantaneous, which 
we know is not the case, but to mean not mediated, indicating that 
a word is identified directly from its features. The aspect of learn- 
ing with which the remainder of this chapter is concerned is the es- 
tablishment of appropriate visual feature lists for immediate word 
identification. 
It will help if we imagine a specific instance. A child is about to 
learn to recognize a particular written word, say, John. The task 
confronting the child is to discover the rules for recognizing this 
event when it occurs again, which means finding out something 
about the configuration that will distinguish it from other configu- 
rations that should not be called “John.” Assume that the child has 
already discovered that a reliable distinguishing characteristic for 
the configuration is not the color of the paper that it is printed on or 
the color of the ink, both of which may be reasonable cues for other 
types of identification but which will sooner or later prove to be in- 
adequate for the allocation of visual information to word catego- 
ries. Also assume that the child at this time is not confronted by 
John in a number of different type styles. The ability to name any 
or all of the letters of the alphabet has no direct relevance in imme- 
diate word identification, although there will be an obvious (al- 
though by no means essential) advantage for children if they have 
learned to distinguish even a few letters, without necessarily being 
able to name them, because they will have begun to acquire cues 
about the features that distinguish words. 
The problem for the child is to discover cues that will distin- 
guish John from other configurations. The child may decide that a 
good cue lies in the length of the word, or the two upright strokes, 
or the shape of the “fish hook” at the beginning. In selecting a cue 
that will be the basis for recognition of the word, a child will estab- 
lish the first tentative distinctive features to be looked for in the fu- 
ture when testing whether to allocate a configuration to the 
category “John.” 
Exactly what the first distinctive feature will be depends on the 
other words from which the child tries to distinguish the configura- 
tion John. Until the child comes across another word that is not 
John, there is no problem; the child applies the single test and 
calls every configuration that passes the test John. But until the 
child comes across another word that is not John, there can be no 
learning. What brings a child to the development of feature lists 
134 
UNDERSTANDING READING 
that will serve for reading is having to distinguish John from all the 
other configurations with which it is not functionally equivalent. 
The child will only really be able to identify John after learning not 
to apply that name to every other word configuration that is met. It 
is when the child is confronted by a configuration that should go 
into a different category that the soundness of the tentative dis- 
crimination is tested, and, of course, it is soon found to be wanting. 
If the hypothesized distinctive features were related to the length of 
the word, then the child would respond John to the configuration 
Fred. If the hypothesis involved the initial fishhook, the child 
would say “John” to Jack, or June, or Jeremiah. The more non- 
equivalent configurations — the more different words — children 
have to discriminate among, the more they will come to select as 
distinctive features those that will be appropriate to the eventual 
task of fluent reading. But until children can understand what they 
have to distinguish John from, they will never acquire an appropri- 
ate set of distinctive features for identifying that word. 
The preceding statement doesn't mean that children must be able 
to name every other word they meet; not at all. All they have to do is 
see a representative sample of words that are not John, so that they 
can find out in what respects John is different. It doesn’t matter if 
they can’t discriminate among all the other words (although in 
learning to identify John they will learn something about all other 
words); the beginning can be the establishment of only two catego- 
ries: configurations that are John and configurations that aren’t 
John. Attempting to teach “one word at a time.” repeatedly insisting 
“This is John -, this is John," won’t help children to learn the word be- 
cause they will never learn how John may be distinguished from any 
other word. The notion that a child can learn to identify a word by 
repetition (or “practice") is a template theory. But there is no way for 
a child to transfer a picture of what is presented to the eyes into a 
storehouse in the brain. Children don’t need to be told interminably 
what a word is; they have to be able to see what it is not. 
Acquaintance with a wide variety of nonequivalent alternatives is 
everything. Through growing familiarity with the written form of 
language, children learn not only to discriminate distinctive fea- 
tures, to establish feature lists, and to recognize functional equiva- 
lences, but they also learn about redundancy. And by acquiring a 
pool of knowledge about the redundancy of words, they learn to 
identify words economically, on minimal quantities of visual infor- 
mation; they establish large numbers of alternative criterial sets. 
At no stage is there any need to belabor the presence of particular 
letters, or to make reference to their putative “sounds.” 
8. WORD IDENTIFICATION 
135 
It is perhaps a sobering thought that just about everything that a 
child must learn, as described in the preceding paragraph, is never 
explicitly taught. Among the many positive things reading teachers 
can do — providing relevant demonstrations, collaboration, and 
encouragement — they can’t include the provision of rules by which 
words are to be differentiated and recognized. That part of learning 
must be left to children themselves. They must be given the oppor- 
tunity, and if necessary the assistance, to gain experience in read- 
ing so that they can achieve all the learning that is necessary. 
A POSTSCRIPT ABOUT WORDS 
One of the inevitable consequences of examining closely a subject 
like reading, about which so much is taken for granted, is that it 
turns out to be more complicated and less well understood than we 
thought it to be. An obvious first step in my discussion of word iden- 
tification might have been to state clearly and precisely how many 
words the average fluent reader knows. This would give some useful 
knowledge about the dimensions of the problem. But the trouble 
with a simple request for a count of the words that a person knows is 
that the answer depends on what is meant by “word,” and that in any 
case there is no way to compute a reliable answer. 
Consider first the matter of deciding what we want to call a word. 
Should cat and cats (or walk and walked ) be regarded as two dif- 
ferent words or as two forms of the same word? Dictionaries usu- 
ally provide entries only for the base or root form of words, 
refusing to count as different words such variations as plurals, 
comparatives, adjectival forms, and various verb tenses. If we want 
to call cat and cats (or walk and walked ) different words (and cer- 
tainly we would not regard them as functionally equivalent visu- 
ally), the number of words we know on sight might turn out to be 
three or four times greater than the number of words the dictio- 
nary maker would credit us with. Furthermore, common words 
have many meanings as in “You can bank on the bank by the river 
bank." But if the same spelling is to be regarded as (at least) three 
different words because bank has several meanings, should a 
preposition like by, which has so many different senses, be 
counted as 40 words or more? 
The next problem is to count. Obviously, it is not good enough 
simply to count the number of words that a person reads or hears 
or produces during the course of a day, for many words will be 
used more than once and others will occur not at all. To count the 
number of different words a person produces, we have to examine 
136 
UNDERSTANDING READING 
a torrent of very familiar words. But in how big a torrent shall we 
look? How can we ever be certain that we have given sufficient op- 
portunity for all the words a person knows? Without a doubt, we 
shall find some new words in every additional sample of a thou- 
sand that we record, but surely a law of decreasing returns would 
apply. After analyzing, say, 100,000 words from one person, it 
would seem unlikely that many new ones would be produced. But 
such is not the case. Very many words with which we are quite fa- 
miliar occur less than once in every million, and it may take any- 
where from 2 months to 2 years for a person to produce that 
number of words. One very extensive analysis of nearly 5 million 
word occurrences in popular magazines (Thorndike & Lorge, 
1944) found over 3,000 words that occurred an average of less 
than once in every million, and almost all of these words would fall 
under our category of known. Here are some of the words that oc- 
curred only once in every 5 million words: earthiness, echelon, 
eclair, ejfluence, egotistic. One or two may be a little unusual, but 
by and large they are words that we can recognize. 
Obviously, it is not possible to count how many different words 
an individual might know, so an estimate is necessary. And many 
estimates have been offered, varying from 50,000 to over 800,000, 
depending on the definitions used and assumptions made. This 
gives one good answer to the question of how many words a person 
might know — it is impossible to say. 
When Is a Word Not a Word? 
Researchers who use experimental procedures in what they call 
reading studies frequently claim that their test materials are mean- 
ingful because they consist of words that appear in dictionaries, in 
contrast to nonword sequences of letters like ricaning, vernalit, 
msk, or wbc. It may be difficult to see why strings of letters that 
have no relevance to readers should be regarded as meaningful, 
even if the letters do happen to be in sequences that appear in dic- 
tionaries. A spoken word produced without expressive or commu- 
nicative intent is simply noise, and there is little reason to regard a 
similarly produced written word as anything different. 
ISSUES 
There is lively technical speculation about the neurological pro- 
cesses assumed to underlie the identification of written words, 
presumably no different from the processes underlying our re- 
8. WORD IDENTIFICATION 
137 
markably efficient facility for recognizing faces, places, dogs, cats, 
and innumerable other objects, and no better understood. The 
study of word identification is often confounded by the question of 
whether it is necessary to say a written word to oneself — to “de- 
code" (or “recode") it — in order to understand it. Obviously we 
don’t need to say to ourselves that an animal that suddenly con- 
fronts us is a dog in order to understand that it is a dog; rather, the 
wordless identification of “dogness” must precede the naming. 
Why then should we have to say that the written word dog is pro- 
nounced “dog” in order to understand it in a story? The relation- 
ship between the sounds of speech and the visual representations 
of written words is one of the most contentious issues in reading 
theory and research. 
SUMMARY 
Words, like letters, can be identified directly from the distinctive 
features that are the visual information of print. Immediate word 
identification takes place when feature analysis allocates a visual 
configuration to the feature list of a word category in cognitive 
structure, without the intermediate step of letter identification. 
Criterial sets of features within functionally equivalent feature lists 
permit the identification of words on minimal information, for ex- 
ample, when the reader can employ prior knowledge of the ortho- 
graphic redundancy within words. 
Notes to chapter 8 begin on page 277 covering: 
Template and feature -analytic theories 
Letter identification in words 
Use of redundancy by children 
Distributional redundancy among words 
9 
^ / Phonics and Mediated 
Word Identification 
The preceding chapter was concerned with immediate word iden- 
tification, and the manner in which visual feature lists may be es- 
tablished and used so that words can be recognized on sight, 
without “decoding to sound" or any other means of mediated word 
identification. In fact, the chapter argued that letter-by-letter iden- 
tification is unnecessary and even impossible for word identifica- 
tion in normal reading, thus leaving no room for decoding to 
sound. Immediate word identification is illustrated in Fig. 9.1. 
In the preceding chapter, I was talking about the identification of 
words where the "name” of the word — its pronunciation when read 
aloud — is either known to the reader or otherwise available to the 
learner. The learner doesn’t need to figure out what the visual con- 
figuration “says,” but only how it should be recognized on future 
occasions. The situation is identical to that of a child who is told 
that a particular animal is a cat and then left to discover how to rec- 
ognize cats on other occasions. 
Suppose, however, that the name of a word is not immediately 
available to the learner — that there is no one to identify an unfamil- 
iar word, and there are no context cues, perhaps because the word 
is seen in isolation or as part of a list of unrelated words. Now the 
138 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
139 
Name 
FIG. 9.1. Immediate word identification. 
learner has a double problem, not only to discover how to recog- 
nize the word in the future but to find out what the word is in the 
first place. This is like trying to discover how to distinguish cats 
and dogs without being told whether particular animals are cats or 
dogs. In such a situation in reading, a word obviously cannot be 
identified immediately ; its identification must be mediated by 
some other means of discovering what it is. The present chapter is 
about the use of phonics — a set of relationships between letters 
and sounds — and other methods of mediated word identification. 
The use of phonic rules to mediate word identification is illus- 
trated in Fig. 9.2. 
In particular, this chapter examines the extent to which knowl- 
edge of the sounds associated with letters of the alphabet helps in 
the identification of words. For many, this process of decoding the 
spelling of words to their sounds is the basis of reading, a view that 
I don't think is tenable. It is not necessary and sometimes it is im- 
possible to “say” what a written word is before we can comprehend 
its meaning; the naming of a word normally occurs after the identi- 
fication of its meaning. 
This chapter is still not the whole story of reading, even as far 
as words are concerned. In both the preceding and the present 
chapters, the assumption is made that the word a reader is trying 
to identify already exists in the reader’s spoken language vocabu- 
lary; its meaning is known. The reader’s problem is to identify the 
word, to discover or recognize its “name,” not to learn its mean- 
FIG. 9.2. Mediated word identification: the phonic model. 
140 
UNDERSTANDING READING 
ing. The next chapter deals with the situation of words that are 
truly new, where the meaning must be discovered as well as the 
name or pronunciation. 
THE COMPLEXITY AND LIMITATIONS OF PHONICS 
Mediated word identification is not the most critical part of read- 
ing, and phonics is not the only strategy available for mediated 
word identification. Nor is phonics the best strategy. Nevertheless, 
phonics frequently plays a central role in reading instruction, and 
it will clear the air if we examine the efficacy of phonics first. 
Rules and Exceptions 
The aim of phonics instruction is to provide readers with rules that 
will enable them to predict how a written word will sound from the 
way it is spelled. The value of teaching phonics depends on how 
many correspondences there are between the letters and sounds of 
English. A correspondence exists whenever a particular letter (or 
sometimes a group of letters) represents a particular sound (or ab- 
sence of sound). Thus, c is involved in at least four correspon- 
dences — with the sound /s/ as in city , with /k/ as in medical, as part 
of /ch/ as in much, and with no sound at all as in scientist. Alterna- 
tively, a correspondence exists whenever a particular sound is rep- 
resented by a particular letter or letters, as /f/ can be represented 
by f, ph, and gh. Thus, the total number of “spelling-sound” corre- 
spondences must be the same as the total number of "sound-spell- 
ing” correspondences. But by now it is probably no surprise that 
any question related to language involving a simple “how many” 
leads to a very complicated and unsatisfactory approximation to 
an answer. Phonics is no exception. 
The first problem concerns our expectations about rules. If we 
expect a rule to mean a correspondence that has no exceptions, 
then we will have a difficult task finding any rules in phonics at all. 
Here is a phonic rule that would appear to have impeccable creden- 
tials: Final e following a single consonant indicates that the preced- 
ing vowel should be long, as in hat and hate, or hop and hope. And 
here are two instant exceptions: axe has a single consonant but a 
short /a/, while ache has a double consonant but a long /a/. We have 
the choice of admitting that a familiar rule is not impervious to ex- 
ceptions, or else we have to make a rule for the exceptions. One ex- 
planation that might be offered is that x is really a double 
consonant, ks, and that ch is really a single consonant, k. But then 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
141 
we are in the rather peculiar position of changing the notion of 
what constitutes a single letter simply because we have a rule that 
doesn’t fit all cases. And if we have to say that the definition of what 
constitutes a letter depends on the pronunciation of a word, how 
can we say the pronunciation of a word can be predicted from its 
letters? Besides, what can we say about the silent e at the end of 
have or love, which is put there only because there is a convention 
that English words may not end with a v? Or the e at the end of 
house, which is to indicate that the word is not a plural? Or the o in 
money and women, which is there because early printers felt that 
a succession of up-and-down strokes, like mun and wim, would be 
too difficult to decipher? 
Having made the point that phonic rules will have exceptions, 
the next problem is to decide what constitutes an exception. Some 
exceptions occur so frequently and regularly that they would ap- 
pear to be rules in their own right. It is quite arbitrary how anyone 
decides to draw the line between rules and exceptions. We have a 
choice of saying that the sounds of written English can be predicted 
by relatively few rules, although there will be quite a lot of excep- 
tions, or by a large number of rules with relatively few exceptions. 
Indeed, if we care to say that some rules have only one application, 
for example, that acht is pronounced /ot/ as in “yacht,” then we can 
describe English completely in terms of rules simply because we 
have legislated exceptions out of existence. 
If the concept of a rule seems arbitrary, the notion of what con- 
stitutes a letter is even more idiosyncratic. It is true that in one 
sense there can be no doubt about what a letter is — it is one of the 
26 characters in the alphabet — but any attempt to construct rules 
of spelling- sound correspondence is doomed if we restrict our 
terms of reference to individual letters. To start with, there are 
only 26 letters, compared with about 40 or more different sounds 
of speech, so many letters at least must do double duty. We find, of 
course, that many letters stand for more than one sound, while 
many sounds are represented by more than one letter. However, 
many sounds are not represented by single letters at all — th, ch , 
ou and ue, for example — so that we have to consider some combi- 
nations of letters as quite distinct spelling units — rather as if th 
were a letter in its own right (as it is in Greek, in two different 
forms for two different pronunciations). It has been asserted, 
with the help of a computer analysis of over 20,000 words 
(Venezky, 1967, 1970), that there are 52 “major spelling units” in 
English, 32 for consonants and 20 for vowels, effectively doubling 
the size of the alphabet. 
142 
UNDERSTANDING READING 
The addition of all these extra spelling units, however, doesn’t 
seem to make the structure of the English writing system very 
much more orderly. Some of the original letters of the alphabet are 
superfluous. There is nothing that c or q or x can do that couldn’t 
be done by the other consonants. And many of the additional spell- 
ing units that are recognized simply duplicate the work of single 
letters, such as ph for /and dg for j. There are also compound vow- 
els whose effect duplicates the silent final e, like ea in meat com- 
pared with mete. Some combinations of letters have a special 
value only when they occur in particular parts of a word — gh may 
be pronounced as / (or as nothing) at the end of a word (rough, 
through) but is pronounced just like a single g at the beginning 
(ghost, ghastly). Often letters have only a relational function, sac- 
rificing any sound of their own in order to indicate how another let- 
ter should be sounded. An obvious example is the silent e; another 
is the u that distinguishes the g in guest from the g in gem. 
So for our basic question of phonics, what we are really asking 
is, how many arbitrarily defined rules can account for an indeter- 
minate number of correspondences between an indefinite set of 
spelling units and an uncertain number of sounds (the total and 
quality of which may vary from dialect to dialect)? 
Some aspects of spelling are simply unpredictable, certainly to 
a reader with a limited knowledge of word derivations, no matter 
how one tries to define a spelling unit. An example of a completely 
unpredictable spelling-sound correspondence is th, which is 
pronounced in one way at the beginning of words like this, than, 
those, them, then, and these, but in another way at the beginning 
of think, thank, thatch, thong, theme, and so on. There is only 
one way for a learner to tell whether th should be pronounced as 
in /this/ or as in /think/, and that is to identify the word first. On 
the other hand, in many dialects there is no difference between 
the sounds represented by w and wh, as in witch and which, so 
that in some cases it can be the spelling that is not predictable, not 
the sound. Almost all common words are exceptions — c/requires 
a rule of its own for the pronunciation of f, and was for the pro- 
nunciation of as. 
The game of finding exceptions is too easy to play. I give only one 
more example to illustrate the kind of difficulty one must encoun- 
ter in trying to construct — or to teach — reliable rules of phonic cor- 
respondence. How are the letters ho pronounced, when they occur 
at the beginning of a word? Here are 1 1 possible answers (all, you 
will notice, quite common words): hope, hot, hoot, hook, hour, 
honest, house, honey, hoist, horse, horizon. 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
143 
Of course, there are rules (or are some of them exceptions?) that 
can account for many of the pronunciations of ho. But there is one 
very significant implication in all these examples that applies to al- 
most all English words — in order to apply phonic rules, words must 
be read from right to left. The way in which the reader pronounces 
ho depends on what comes after it, and the same applies to the p in 
ph, the a in ate. the k in knot, the t in tion. The exceptions are very 
few, like asp and ash, which are pronounced differently if preceded 
by a w. The fact that sound “dependencies” in words run from right 
to left is an obvious difficulty for a beginning reader trying to sound 
out a word from left to right, or for a theorist who wants to maintain 
that words are identified on a left-to-right basis. 
In summary, English is far from predictable as far as its spell- 
ing-sound relationships are concerned. Just how much can be 
done to predict the pronunciation of a relatively small number of 
common words with a finite number of rules we see later. But be- 
fore this catalog of complications and exceptions is concluded, two 
points should be reiterated. The first point is that phonic rules at 
best can only be considered as probabilistic, as guides to the way 
words might be pronounced, and that there is rarely any indication 
of when a rule does or does not apply. The rule that specifies how to 
pronounce ph in telephone falls down in the face of haphazard or 
shepherd. The rule for oe in doe and woe will not work for shoe. 
The only way to distinguish the pronunciation of sh in bishop and 
mishap, or th in father and fathead, is to know the entire word in 
advance. The probability ofbeing wrong ifyou don't knowa word at 
all is very high. Even if individual rules were likely to be right three 
times out of four, there would still be only one chance in three of 
avoiding error in a four-letter word. 
The second point is that phonic rules look simple if you already 
know what a word is. I don’t intend to be facetious. Teachers often 
feel convinced that phonic rules work because letter-sound corre- 
spondences appear obvious when a word is known in advance; the 
alternatives are not considered. And children may appear to apply 
phonic rules when they can recognize a word in any case — or be- 
cause the teacher also suggests what the word is — thereby enabling 
them to identify or recite the phonic correspondences that happen 
to be appropriate. 
The Efficiency of Phonics 
A classic attempt to construct a workable set of phonic rules for 
English was made by Berdiansky, Cronnell, and Koehler (1969). 
144 
UNDERSTANDING READING 
The effort had modest aims — to see how far one could go in estab- 
lishing a set of correspondence rules for the 6,092 one- and two- 
syllable words among 9,000 different words in the comprehension 
vocabularies of 6- to 9-year-old children. (The remaining words, 
nearly one third of the children’s vocabularies, were all three or 
more syllables, adding too much complexity to the phonic analy- 
sis.) The words were all taken from books to which the children 
were normally exposed — they were the words that the children 
knew and ought to be able to identify if they were to be able to read 
the material with which they were confronted at school. 
The researchers who analyzed the 6,092 words found rather 
more than the 52 "major spelling units” to which I have already re- 
ferred — they identified 69 “grapheme units” that had to be sepa- 
rately distinguished in their rules. A group of letters was called a 
grapheme unit, just like a single letter, whenever its relationship to 
a sound could not be accounted for by any rules for single letters. 
Grapheme units included pairs of consonants such as ch, th ; pairs 
of vowels such as ea, oy; and letters that commonly function to- 
gether, such as ck and q u, as well as double consonants like hh and 
tt, all of which require some separate phonic explanation. The 
number of grapheme units should not surprise us. The previously 
mentioned 52 major units were not intended to represent the only 
spelling units that could occur but only the most frequent ones. 
An arbitrary decision was made about what would constitute a 
rule: It would have to account for a spelling-sound correspon- 
dence occurring in at least 10 different words. Any distinctive 
spelling-sound correspondence and any grapheme unit that did 
not occur in at least 10 words was considered an "exception." 
Actually, the researchers made several exceptions among the ex- 
ceptions. They wanted their rules to account for as many of their 
words as possible, so they let several cases through the net when it 
seemed to them more appropriate to account for a grapheme unit 
with a rule rather than to stigmatize it as an exception. 
The researchers discovered that their 6.092 words involved 211 
distinct spelling-sound correspondences. This doesn't mean that 
211 different sounds were represented, any more than there were 
211 different grapheme units, but rather that the 69 grapheme 
units were related to 38 sounds in a total of 2 1 1 different ways. The 
results are summarized in Table 9.1. 
Eighty-three of the correspondences involved consonant 
grapheme units, and 128 involved vowel grapheme units, includ- 
ing no fewer than 79 that were associated with the six “primary” 
single-letter vowels, a, e, i, o, u, y. In other words, there was a total 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
145 
TABLE 9.1 
Spelling-Sound Correspondences Among 6,092 One- and Two-Syllable 
Words in the Vocabularies of 9-Year-Old Children 
Consonants 
Spelling-sound 
83 
correspondences 
‘Rules'’ 
60 
“Exceptions" 
23 
Grapheme units 
41 
in rules 
Primary 
Vowels 
Secondary 
Vowels 
Total 
79 
49 
211 
73 
33 
166 
6 
16 
45 
6 
19 
69 
of 79 different ways in which single vowels could be pronounced. 
Of the 211 correspondences, 45 were classified as exceptions, 
about half involving vowels and half consonants. The exclusion of 
45 correspondences meant that about 10% of the 6,092 words had 
to be set aside as “exceptions.” 
The pronunciation of the remaining words was accounted for by a 
total of 166 rules. Sixty of these rules were concerned with the pro- 
nunciation of consonants (which are generally thought to have fairly 
“regular” pronunciations! and 106 with single or complex vowels. 
Conclusions that can be drawn from this research are far- 
reaching in their implications. The first is that phonics is compli- 
cated. Without saying anything at all about whether it is desirable 
to teach children a knowledge of phonics, we have an idea of the 
magnitude of the endeavor. We know that if we really expect to give 
children a mastery of phonics, then we are not talking about a 
dozen or so rules. We are talking about 166 rules, which will still 
not account for hundreds of words children might expect to meet 
in their early reading. 
It is obvious that the most that can be expected from a knowledge 
of phonic rules is that they may provide a clue to the sound (or 
“name”) of a configuration being examined. Phonics can provide 
only approximations. Even if readers do happen to know the 73 
rules for the pronunciation of the six vowels, they would still have no 
sure way of telling which rule applies — or even that they are not deal- 
ing with an exception. And in any case, the sounds that letters are 
supposed to represent don’t actually exist as separate units in 
speech — they are part of a continuous flow of sounds that overlap 
146 
UNDERSTANDING READING 
and constantly change. The sophisticated trick of being able to iso- 
late the supposed individual sounds of speech is called phonologi- 
cal (or phonemic) awareness, which has nothing to do with reading 
(except that it is easier for readers) but is regarded as being of su- 
preme importance in some quarters. There is more about phone- 
mic awareness in the notes to this chapter. 
There is still one issue to be considered concerning the effec- 
tiveness of phonics. Is the limited degree of efficiency that might be 
attained worth acquiring? Other factors have to be taken into ac- 
count related to the cost of trying to learn and use phonic rules. 
There is the possibility that reliance on phonics will involve read- 
ers in so much delay and confusion that short-term memory will be 
overloaded, and they lose the sense of what they are reading. A ten- 
dency — or requirement — to rely on phonic rules may create a 
handicap for beginning readers whose biggest problem is to find 
meaning and develop speed in reading. Working memories don't 
have an infinite capacity, and reading is not a task that can be ac- 
complished at too leisurely a pace. Other sources of information 
exist for finding out what a word in context might be. 
The Cost of "Reform" 
The convoluted relation between the spelling of words and their 
sound has led to frequent suggestions for modifying the alphabet 
or for rationalizing the spelling system. These intentions share 
misconceptions and difficulties. A number of contemporary lin- 
guists would deny that there is anything wrong with the way most 
words are spelled; they argue that a good deal of information 
would be lost if spelling were changed (Chomsky & Halle. 1968; 
Pinker, 1999). Most of the apparent inconsistencies in spelling 
have some historical basis; spellings are not arbitrary — they have 
become what they are for quite systematic reasons. And because 
spelling is systematic and reflects something of the history of 
words, much more information is available to the reader than we 
normally realize. (The fact that we are not aware that this informa- 
tion is available doesn’t mean that we don’t use it; we have already 
seen a number of examples of the way in which we have and use a 
knowledge of the structure and redundancy of our language that 
we can’t put into words.) 
Spelling reform might seem to make words easier to pronounce, 
but only at the cost of information about the way words share 
meanings, so that rationalizing words at the phonemic level might 
make reading more difficult at syntactic and semantic levels. As 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
147 
just one example, consider the “silent b” in bomb, bombing, and 
bombed, which would be an almost certain candidate for extinc- 
tion if spelling reformers had their way. But the b is something 
more than a pointless appendage; it relates the previous words to 
others like bombard, bombardier, and bombardment where the b 
is pronounced . And if you save yourself the trouble of a special rule 
about why b is silent in words like bomb, at another level there 
would be a new problem of explaining why b suddenly appears in 
words like bombard. Remove the g from sign and you must ex- 
plain where it comes from in signature. 
Another argument in favor of the present spelling system is that it 
is the most competent one to handle different dialects. Although 
there is almost universal acceptance of the idea that words should 
be spelled in the same way by everyone, we don't all pronounce 
words in the same way. If the spelling of words is to be changed so 
that they reflect the way they are pronounced, whose dialect will pro- 
vide the standard? Is a different letter required for every different 
sound produced in any dialect of English we might encounter? Or 
should we have different spelling systems for different dialects? 
Phonics instruction becomes even more complicated when it is real- 
ized that in many classrooms teacher and students don’t speak the 
same dialect and that both may speak a different dialect from the au- 
thorities who suggested the particular phonic rules they are trying to 
follow. The teacher who tries to make children understand a phonic 
difference between the pronunciation of caught and cot will have a 
communication problem if this distinction is not one that the chil- 
dren observe in their own speech. The teacher may not even pro- 
nounce the two words differently, so that although the teacher 
thinks the message to the child is “That word isn’t cot; it’s caught," 
the message coming across is “That word isn’t cot; it’s cot.” 
Spelling and Meaning 
The manner in which words are spelled in English becomes a 
problem primarily if reading is regarded as decoding written 
words to sound, and if the primary function of spelling is seen as 
representing sounds of spoken words. But spelling also reflects 
meaning, and where there is a conflict between pronunciation and 
meaning it is usually meaning that prevails, as if even the spelling 
system of written language recognized the priority of meaning. For 
example, the plural represented by a simple s in written language 
may be pronounced in three different ways in speech — as the /s/ 
sound on the end of cats, the /z/ sound on the end of dogs, and the 
148 
UNDERSTANDING READING 
/iz/ sound on the end of judges. Would print be easier to read if the 
past tense of verbs were not indicated by a consistent -ed but 
rather reflected the pronunciation, so that we had such variations 
as walkt and landid? The reason that medicine and medical are 
spelled as they are is not because c is sometimes arbitrarily pro- 
nounced /s/ and sometimes /k/ but because the two words have the 
same root meaning, represented by medic. This shared meaning 
would be lost if the two words were spelled medisin and medikal. 
It should be noted, incidentally, that the consistent representation 
of the various pronunciations of the plural meaning by s or the 
past-tense meaning by -ed rarely causes difficulty to readers, even 
beginners, provided they are reading for sense. It is not necessary 
for teachers to instruct children that -ed is often pronounced /t I 
(among other things). If a child understands a word, the pronunci- 
ation will take care of itself, but the effort to produce pronunciation 
as a prerequisite for meaning is likely to result in neither being 
achieved. 
Of course, spelling can be a problem, both in school and out, but 
it is a problem of writing , not of reading. Knowing how to spell 
doesn’t make a good reader because reading is not accomplished 
by decoding spelling. And good readers aren't necessarily good 
spellers; we can all read words that we can’t spell. I’m not saying 
that knowledge of spelling is unimportant, only that it has a mini- 
mal role in reading, and that undue concern with the way in which 
words are spelled can only interfere with a child's learning to read. 
There is a frequent argument that if spelling and decoding to 
sound are as irrelevant to reading as the preceding analyses indi- 
cate, why should we have an alphabetic written language at all? 
My view, set out in chapter 1 (p. 5), is that the alphabetic system is 
more of a help to the writer than to the reader, and more of a con- 
venience to the printer (or scribe) than to the author. For a variety 
of reasons, writing may be harder to learn and to practice than 
reading, at least if the writer aspires to be conventionally "correct" 
with respect to such matters as spelling, grammar, punctuation, 
neatness, layout, and so forth. But as we shall see in chapter 10, 
although writing may not do much to facilitate one's ability as a 
reader, extensive reading can take care of most of the learning 
problems of writers. 
The alphabet exacts its own price from readers. The Chinese 
ideographic system can be read by people from all over China, al- 
though they might speak languages that are mutually unintelligi- 
ble. If a Cantonese speaker cannot understand what a Mandarin 
speaker says, they can write their conversation in the nonalpha- 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
149 
betic writing system they share and be mutually comprehensible. 
This is something English speakers can’t do with speakers of other 
languages unless they employ the small part of their own writing 
systems that is not alphabetic, such as arithmetic symbols like 2 + 
3 = 5. Imagine the immense savings if all the spoken languages 
used in the United Nations shared the same meaning-based writing 
system. 
When we can't remember or don’t know how a word should be 
written, we have little recourse to anything but what we know 
about spelling. It doesn’t help much in writing to look at the words 
before and after the one that is giving difficulty. But in reading we 
have more effective alternatives before we need call on phonics for 
clues to identifying a word, and it is to these alternatives that we 
now turn. 
STRATEGIES OF MEDIATED WORD IDENTIFICATION 
To repeat, the problem we are concerned with is that of a reader 
who encounters a word that cannot be recognized on sight, for 
which a visual feature list must be established. The reader will 
know the meaning of the word once it is identified: the problem is 
to identify the word by some kind of mediation. 
Phonics, as we have seen, is one such strategy, but a most inade- 
quate and time-consuming one. Fortunately other strategies are of- 
ten available. An obvious alternative is simply being told what the 
word is. Before most children come to school, well-intentioned 
adults say to them “That word is John," or “That word is cereal," 
just as on other occasions they say, “That animal is a cat,” in all 
cases leaving it to the child to solve the more complex problem of 
working out exactly how to recognize the word or animal on future 
occasions. But when the children get to school this support is fre- 
quently taken away from them, at least as far as reading is con- 
cerned. Another well-intentioned adult is likely to say to them, 
“Good news and bad news today, children. The bad news is that no 
one is ever likely to tell you what a word is again. The good news is 
that we are going to give you 1 66 rules and 45 exceptions so that 
you can work it out for yourselves." 
Alternative Identification Strategies 
Ask experienced readers what they do when they come across a 
word they don’t recognize, and the most probable answer will be 
that they skip it, coming back to the word later if necessary. Passing 
150 
UNDERSTANDING READING 
over a word is a reasonable first strategy because it is unnecessary 
to understand every word to understand a passage of text, and lin- 
gering to try to decipher a word may be more disruptive to compre- 
hension than missing the word completely. The second preferred 
strategy is to predict, which doesn’t mean a reckless stab in the dark 
but making use of context to eliminate unlikely alternatives for what 
the unfamiliar word might be. The final strategy may be trying to 
work out what the word is from its spelling, not so much by decod- 
ing the word to sound with phonics as by making use of what is al- 
ready known about other words. The final strategy might be called 
identification by analogy , because all or part of the unknown word 
is compared with all or part of words that are already known. 
Examine the same question with a child who is gaining experi- 
ence in reading and again you are likely to get the same sequence of 
strategies. The best learners tend to skip occasional unknown 
words (unless constrained to “read carefully” and figure out every 
word). The second preference, especially if there is no helpful adult 
around to provide assistance, is to hypothesize what a word might 
be based on the meaning of the text, and the final choice is to use 
what is known about similar-looking words. Trying to sound out 
words without reference to meaning is a characteristic strategy of 
poor readers; it is not one that leads to fluency in reading. The act 
of reading, on the other hand, usually helps with learning unfamil- 
iar words, as we shall see. 
What is the best method of mediated word identification? The 
answer depends on the situation a reader is in. Sometimes the best 
strategy will indeed be to ignore the unfamiliar word, because suffi- 
cient meaning may be carried by the surrounding text, not only to 
compensate for lack of understanding of the unknown word but 
also to provide critical cues to what the unknown word might be on 
future occasions. For a child beginning to learn to read, or con- 
fronted by text where many words are unfamiliar, the best situation 
is probably to have a more competent reader to turn to, if neces- 
sary by reading the entire passage to the child. But if the reader can 
understand enough of the passage to follow its sense, then a most 
effective strategy may be identification by analogy, making use of 
what is already known about reading. 
Phonics in itself is almost useless for sounding out words letter 
by letter, because every letter can represent too many sounds. But 
uncertainty about the sound of a particular letter diminishes as let- 
ters are considered not in isolation but as part of letter clusters or 
“spelling patterns." This has led a number of theorists to argue that 
the basic unit for word recognition should be regarded as the sylla- 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
151 
ble rather than the individual letter, particularly syllables that 
rhyme with known words. And indeed it is true; the pronunciation 
of syllables is far less variable than the pronunciation of the indi- 
vidual letters that make up syllables. The trouble is that it would be 
impossible for a beginner to learn to read by memorizing the pro- 
nunciation of hundreds of syllables, because syllables in them- 
selves tend to be meaningless — there are relatively few one-syllable 
words — and the human brain has great difficulty in memorizing, 
and particularly in recalling, nonsense. What a reader can turn to 
for a ready-made store of syllabic information is words that have 
already been learned. It is far easier for a reader to remember the 
unique appearance and pronunciation of a whole word like photo- 
graph , for example, than to remember the alternative pronuncia- 
tions of meaningless syllables or spelling units like ph, to, gr or 
gra, and ph. A single word, in other words, can provide the basis 
for remembering different rules of phonics, as well as the excep- 
tions, because not only do words provide a meaningful way to orga- 
nize different phonic rules in memory, they also illustrate the 
phonic rules at work. 
Identification by Analogy 
The mediated word identification strategy of identification by anal- 
ogy means looking for cues to the pronunciation and meaning of an 
unfamiliar word from words that look similar. We don’t learn to 
sound out words on the basis of individual letters or letter clusters 
whose sounds have been learned in isolation but rather by recogniz- 
ing sequences of letters that occur in words that are already known. 
Such a strategy offers an additional advantage to the reader because 
it does more than indicate possible pronunciations for all or part of 
unknown words; it can offer suggestions about meaning. As I 
pointed out earlier, English spelling in general respects meaning 
more than sound — words that look alike tend to share the same 
sense. And as I have reiterated throughout this book, the basis of 
reading and of learning to read is meaning. The advantage of trying 
to identify unknown words by analogy with words that are known al- 
ready is not simply that known words would provide an immedi- 
ately accessible stock of pronunciations for relatively long 
sequences of letters but that all or part of known words can provide 
clues to meaning, which is always a far better clue to pronunciation 
than just the way an unknown word is spelled. 
This is not to suggest that the existence of spelling-sound corre- 
spondences should be concealed from children learning to read, but 
152 
UNDERSTANDING READING 
the correspondences will be learned as they are encountered in 
meaningful reading, not as a result of being drilled into beginners. 
There remain the three fundamental problems with spelling-sound 
correspondences: the total number of rules and exceptions, the time 
it takes to apply them in practice, and their general unreliability. The 
problem of the number of rules is solved if they are not taught in the 
abstract, outside of meaningful reading. The easiest way to learn a 
phonic generalization is to learn a few words that exemplify it, which 
means — another point that will bear some reiteration — that chil- 
dren master phonics as a result of reading rather than as a pre- 
requisite for reading. The time problem is overcome by resorting to 
phonics in actual reading as infrequently as possible. 
The problem of the unreliability of phonic generalizations is an- 
other matter. Phonic generalizations alone will not permit a reader 
to decode the majority of words likely to be met in normal reading, 
simply because there are always too many alternatives. This is the 
reason that producers of phonic workbooks prefer to work with 
strictly controlled vocabularies. There’s not as much uncertainty 
of pronunciation with The fat cat sat on the mat as there is with 
Two hungry pigeons flew behind the weary ploughman, a sen- 
tence that makes more sense but defies phonic analysis. Text that 
conforms to phonic regularities is referred to as "decodable text," 
which is usually synonymous with “unnatural English." 
Meaning Plus Phonics 
Phonic generalizations may be of limited utility if all they are re- 
quired to do is reduce alternatives, without being expected to iden- 
tify words completely or to decode them to sounds. To give an 
example, the use of phonics will never succeed in decoding horse if 
the word appears in isolation or is one of thousands of possible al- 
ternatives. But if the reader knows that the word is either horse, 
mule , or donkey, then the strategy will work effectively. Not only is 
a minimal amount of phonic analysis required to know that mule 
and donkey could not begin with h, but not much more can be ex- 
pected of phonic rules in any case. This is why I have asserted that 
phonics is easy — for the teacher and the child — if they know what a 
word is in the first place. And because of the allure of the alphabet, 
it can look as if phonics is doing all the work. 
Phonic strategies can’t be relied on to eliminate all uncertainty if 
the reader has no idea what the word might be. One way to reduce 
uncertainty in advance is to employ the mediating technique of 
making use of context. Understanding of the text, in general, will 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
153 
reduce the number of alternatives that an unknown word might be. 
The other way to reduce uncertainty in advance is to employ the al- 
ternative mediating technique of identification by analogy, compar- 
ing the unknown word with known words that provide hypotheses 
about possible meanings and pronunciations. The reason that we 
can so easily read nonwords like vernalit, mossiant, and ricaning 
is not because we have in our heads a store of pronunciations for 
meaningless letter sequences like uern, iant, or ric but because 
these close approximations to English are made up of parts of 
words or even entire words that are immediately recognizable, 
such as govern, vernal, lit, moss, and so forth. 
To ignore alternative means of reducing uncertainty is to ignore 
the redundancy which is a central part of all aspects of language. 
The prior elimination of unlikely alternatives is, after all, the foun- 
dation on which reading takes place, according to the analysis of 
comprehension that was offered in chapter 4. By this analysis, 
readers are usually unaware of the strategies employed as words 
are tentatively identified and feature lists established. Phonic gen- 
eralizations function almost as sentinels; they can’t decipher un- 
known words on their own, but they will protect readers against 
making reckless hypotheses. 
LEARNING MEDIATED WORD IDENTIFICATION STRATEGIES 
The basis of all learning — and especially of language learning — is 
sense. It is pointless to expect a child to memorize lists of rules, 
definitions, examples, even names, if these have no apparent pur- 
pose or utility to the child. Not only will learning be difficult, but re- 
call will be almost impossible. The mediated word identification 
strategies that have been discussed can fall into the category of 
meaningless learning for a child expected to acquire them outside 
a context of meaningful reading. A child should not be expected to 
memorize phonic rules, or the pronunciation of isolated syllables 
and letter clusters, prior to learning how to read. Identification by 
analogy can also only be fostered after a child has begun to read. 
Learning is self- directing and self-reinforcing when children are 
in a situation that makes sense to them, that can be related to what 
they know already. Rules that can’t be verbalized about many as- 
pects of the physical world and of language are hypothesized and 
tested with little conscious awareness. Where children can under- 
stand a relationship they will learn the relationship, whether it is 
the relationship of a name to a word, of a meaning to a word, or of a 
spelling-sound correspondence. 
154 
UNDERSTANDING READING 
What all this means is that reading guarantees increasing re- 
turns. The more experience that children have in reading, the more 
easily they will learn to read. The more they can recognize words, the 
more easily they will be able to understand phonic correspondences, 
to employ context cues, and to identity new words by analogy. The 
more that children are able to read — or are helped to read — the more 
they are likely to discover and extend these strategies for themselves. 
By acquiring an extensive “sight vocabulary" of immediately 
identifiable words, children are able to understand, remember, 
and utilize phonic rules and other mediated identification strate- 
gies. But such a summary statement doesn't imply that the way to 
help children read is to give them plenty of experience with isolated 
words and word lists. The easy way to learn words is to experience 
lots of them in meaningful text. We are considering only a limited 
and secondary aspect of reading when we restrict our attention to 
individual words. As we turn our attention to reading sequences of 
words that are grammatical and make sense, we find that word 
identification and learning are more easily explained theoretically 
and more easily accomplished by the child. 
The time has come to complete the picture of reading by ac- 
knowledging that words are rarely read or learned in meaningless 
isolation. Reading is easiest when it makes sense, and learning to 
read is also easiest when it makes sense. We are ready to view read- 
ing from a broader and more meaningful perspective. 
ISSUES 
The difficulty many children experience in learning phonics rules, 
or indeed in making sense of them, has led to the notion that such 
children lack “phonemic awareness,” that is, the ability to decons- 
truct the sounds of spoken words. It is taken for granted that the 
tenuous relationship between letters and sounds must be of cen- 
tral importance to readers of alphabetic writing systems (com- 
pared with readers of nonalphabetic systems like Chinese) — why 
else have an alphabet? An alternative view is that phonics appeals 
particularly to researchers and program developers who like to 
break reading down into parts that are easily controlled in instruc- 
tional systems and tests. 
SUMMARY 
Mediated word identification is a temporary expedient for identify- 
ing unfamiliar words while establishing feature lists to permit im- 
9. PHONICS AND MEDIATED WORD IDENTIFICATION 
155 
mediate identification. Alternative strategies for mediated word 
identification include asking someone, using contextual cues, 
analogy with known words, and a limited and controlled use of 
phonics (spelling-sound correspondences). Attempting to decode 
isolated words to sound is unlikely to succeed because of the num- 
ber, complexity, and unreliability of phonic generalizations. Phonic 
rules will help to eliminate alternative possibilities only if uncer- 
tainty can first be reduced by other means, for example if the unfa- 
miliar words occur in meaningful contexts. Spelling-sound 
correspondences are not easily or usefully learned before children 
acquire some familiarity with reading. 
Notes to chapter 9 begin on page 280 covering: 
A defining moment 
The relevance of phonics 
Phonological and phonemic awareness 
Word identification by analogy 
Spelling 
The Identification 
of Meaning 
Previous chapters showed how a system of feature analysis could 
be employed both for the identification of letters and for the direct 
identification of words. Word identification doesn't require the 
prior identification of letters, at least not when the word that read- 
ers are concerned with is familiar to them, a part of their “sight vo- 
cabulary.” It is only when words can’t be identified immediately 
that the prior identification of letters may become relevant at all, 
and then only to a limited extent depending on the amount of con- 
textual and other information that the reader might have available. 
The alternatives are summed up in Fig. 10.1. 
Now I want to show that comprehension, which in this chapter is 
referred to as the identification (or apprehension) oj meaning, 
doesn't require the prior identification of words. The same fea- 
ture-analytic procedure that underlies the identification of letters 
and words is also available for the immediate apprehension of 
meaning from print. Although mediated meaning identification 
may sometimes be necessary, if for some reason meaning can t im- 
mediately be assigned to text, attempting to make decisions about 
possible meaning by the prior identification of individual words is 
highly inefficient and unlikely to succeed. 
156 
10. THE IDENTIFICATION OF MEANING 
157 
FIG. 10.1. Immediate and mediated word identification. 
In other words, immediate meaning identification is as inde- 
pendent of the identification of individual words as immediate 
word identification is independent of the identification of individ- 
ual letters. The alternatives are represented in Fig. 10.2. The argu- 
ment is presented in three steps: 
1 . Showing that immediate meaning identification is accom- 
plished: that readers normally identify meaning without or 
before the identification of individual words. 
2. Proposing how immediate meaning identification is accom- 
plished. 
3. Discussing how immediate meaning identification is learned. 
In a final section, I briefly discuss the mediated identification of 
meaning, or what a reader might do when the direct apprehension 
of meaning is not possible. 
USING MEANING IN READING 
One demonstration has already been given that readers employ 
meaning (or sense) to assist in the identification of individual 
words rather than laboring to identify words in order to obtain 
FIG. 10.2. Immediate and mediated meaning identification. 
158 
UNDERSTANDING READING 
meaning. I am referring to the research showing that from a single 
glance at a line of print — the equivalent of about 1 second of read- 
ing — a reader can identify four or five words if they are in a mean- 
ingful sequence but scarcely half that amount if the words are 
unrelated to each other. The explanation in chapter 5 was that with 
meaningful text a reader could recruit nonvisual information to re- 
duce alternatives so that the amount of visual information that the 
brain can handle in a second would go twice as far, to identify four 
or five words instead of a couple. The nonvisual information that 
the reader already possesses can only be meaning, or prior knowl- 
edge of the way in which words go together in language that is not 
only grammatical but makes sense. 
The Constant Search for Sense 
It is important to understand that the reader in the situation just 
discussed is constantly making use of meaning; meaningfulness 
facilitates the identification of every word in the line. The reader 
doesn’t first identify one or two words by a word identification 
strategy, as if they had nothing to do with the other words in the 
line, and then make an educated guess about the rest. Indeed, that 
same research shows that if two words had to be identified to give a 
clue about the others, then there would be no time left for the oth- 
ers to be identified. TWo is the limit for words that have no mean- 
ingful relationships. Where a sequence of words does make sense, 
the identification of every word is facilitated, the first as well as the 
last, just as individual words can be identified in conditions in 
which none of their component letters would be individually 
discriminable. It is all a matter of the prior elimination of unlikely 
(and impossible) alternatives. 
The research just discussed is historic; it was first conducted 
and reported a century ago. But in a sense, the fact that meaning fa- 
cilitates the identification of individual words is demonstrated ev- 
ery time we read, because reading would be impossible if we 
labored along, blindly striving to identify one word after another 
with no prior insight into what the meaning of those words might 
be. Eye-movement studies may show that in some circumstances 
the eyes briefly come to rest on all or most words, but reading is not 
achieved saccadically. The eyes may “look at" words one at a time, 
but the brain deals with words in meaningful clusters. Slow read- 
ing is not efficient reading because it tends to create tunnel vision, 
overload short-term memory, and leave the reader floundering in 
the ambiguity of language. It’s impossible to read normally text that 
10. THE IDENTIFICATION OF MEANING 
159 
doesn't make sense, as you can experience for yourself if you try to 
read the following passage: 
Be might words those of meaning the what into insight prior no with, 
another after word one identify to striving blindly, along labored we if 
impossible be would reading because, read we time every demon- 
strated is words individual of identification the facilitates meaning 
that fact the, sense a in but. 
The words you have just tried to read are what I hope is a mean- 
ingful English sentence— because I used it myself in the previous 
paragraph — written backward. Any difference between the rate and 
ease with which you could read the backward and forward versions 
of that sentence can only be attributed to whether you were able to 
make sense of it. If you had read the backward passage aloud, inci- 
dentally, you probably would have sounded very much like many 
“problem readers” at school, who struggle to identify words one at a 
time in a dreary monotone, as if each word had nothing to do with 
any other. Such children seem to believe — and may well have been 
taught — that meaning should be their last concern; that sense will 
take care of itself provided they get the words right. 
Professional readers, for example broadcasters, know the im- 
portance of prior understanding of what they are about to read, 
which is why they like to scan through a script in advance. Looking 
ahead also helps the silent reading of novels and technical books — 
we can get a general idea of what will transpire and then go back 
where necessary to study particular points. General comprehen- 
sion comes out of fast reading, and the slow reading that might be 
necessary for memorization or for reflection on detail can only be 
accomplished if comprehension has already been taken care of. 
Conversely, meaning can interfere with some reading tasks. Proof- 
readers tend to overlook misprints if they attend to the sense of 
what they read; they see the spellings and words that should be on 
the page rather than those that actually are there. Sometimes 
proofreaders will deliberately read backward in order to give all 
their attention to spelling and individual words, but then of course 
they will overlook anomalies of meaning. Their dilemma highlights 
the fact that attention to individual words and attention to meaning 
are alternative and not concurrent aspects of reading. 
The prior use of meaning ensures that when individual words 
must be identified, for example in order to read aloud, a minimum 
of visual information will be used. And as a consequence, mistakes 
will occasionally occur. If a reader already has a good idea of what a 
160 
UNDERSTANDING READING 
word might be, there is not much point in delaying to make extra 
certain what the word actually is. As a result, it is not unusual for 
even highly experienced readers to make misreadings that are rad- 
ically different visually — like reading said when the word is an- 
nounced or reported — but that make no significant difference to 
the meaning. Beginning readers often show exactly the same ten- 
dency, demonstrating that children will strive for sense even as 
they learn to read (provided the material they are expected to learn 
from has some possibility of making sense to them). The mistakes 
that are made are sometimes called miscues rather than errors to 
avoid the connotation that they are something bad (Goodman, 
1969). The miscues show that these beginning readers are at- 
tempting to read in the way fluent readers do, with sense taking 
priority over individual word identification. Of course, reading 
with minimal attention to individual words will sometimes result 
in misreadings that do make a difference to meaning, but one of the 
great advantages of reading for meaning is that one becomes aware 
of mistakes that make a difference to meaning. An important dif- 
ference between children who are doing well in reading and those 
who are not is not that good readers make fewer mistakes, but that 
they go back and correct the mistakes that make a difference. 
Children who are not reading for sense have no chance of becom- 
ing aware of even important errors. 
The Priority of Meaning 
A unique illustration of the manner in which meaning takes prior- 
ity over the identification of individual words was provided by 
Kolers (1966), who asked bilinguals fluent in English and French 
to read aloud from passages of text that made sense but where the 
actual language changed from English to FYench every two or three 
words. For example: 
His horse, followed de deux bassets, faisait la terre resonner under 
its even tread. Des gouttes de verglas stuck to his manteau. Une 
violente brise was blowing. One side de l'horizon lighted up. and 
dans la blancheur of the early morning light, il apergut rabbits hop- 
ping at the bord de leur terriers. 
The subjects in this experiment could read and understand 
such passages perfectly well, but when they had finished they often 
could not remember whether particular sentences or words were 
in English or French. Most significantly, they frequently substi- 
1 0. THE IDENTIFICATION OF MEANING 
161 
tuted for a word in one language an appropriate word in the other. 
They might read porte when the word was door, or hand when 
main was given, getting the meaning right but the language wrong. 
This doesn’t mean that they weren’t looking at individual words at 
all — the passage was not completely predictable — but they were 
looking at words and finding meanings, just as an English speaker 
might look at 2,000 and understand “two thousand” while a 
French speaker would look at the same print and understand 
“deux milles.” 
An important and perhaps difficult point to understand from 
the preceding discussion is that it is possible to make meaning- 
ful decisions about words without saying exactly what the words 
are. In other words, we can see that the written word door means 
door without having to say aloud or to ourselves that the word is 
“door.” Written words convey meaning directly; they are not in- 
termediaries for spoken language. An obvious example is pro- 
vided for English by words that have different spellings for the 
same sound, like their and there. It is easy to detect the spelling 
error in The children left there books behind because there rep- 
resents the wrong meaning. The difference between their and 
there, read and reed, and so, sew, and sow is evidently not that 
the different spellings represent different sounds, because they 
don't, but that the different appearances of the words indicate 
different meanings. The visual appearance of each word indi- 
cates meaning directly. 
The fact that readers can, do, and must read directly for mean- 
ing is similarly apparent with a written language like Chinese, 
which doesn’t correspond to any particular sound system. It would 
be pointless to argue whether a particular Chinese symbol repre- 
sents the Mandarin or the Cantonese word for house because it 
simply represents a meaning. The fact that some written lan- 
guages are based on letters that are more or less related to the 
sounds of a spoken language is quite coincidental as far as readers 
are concerned. There is no evidence that fluent readers need to 
identify letters in order to identify familiar words, and English 
spelling is an inadequate guide to the identification of words that 
are unfamiliar. 
My final example that written language indicates meaning di- 
rectly comes from studies of brain-injured patients. People unable 
to find the exact word have been reported, for example, as reading 
the isolated word ill as “sick," city as “town,” and ancient as “his- 
toric” (Marshall & Newcombe, 1966) or injure as “hurt,” quiet as 
“listen.” and fly as “air” (Shallice & Warrington. 1975). 
162 
UNDERSTANDING READING 
COMPREHENSION AND THE REDUCTION OF UNCERTAINTY 
I have tried to show that meaning can take priority over the identi- 
fication of individual words in two ways, both for experienced 
readers and for beginners. In the first case, the meaning of a se- 
quence of words facilitates the identification of individual words 
with relatively less visual information. In the second case, written 
words can be understood without being identified precisely. 
Usually both aspects of meaning identification occur simulta- 
neously: we comprehend text using far less visual information 
than would be required to identify the individual words and with- 
out the necessity of identifying individual words. Both aspects of 
meaning identification are, in fact, reflections of the same under- 
lying procedure — the use of minimal visual information to make 
decisions specific to implicit questions (or predictions) about 
meaning on the part of the reader. 
I’m using a rather awkward expression, "meaning identifica- 
tion,” as a synonym for comprehension in this chapter to underline 
the fact that the way in which a reader makes sense of text is no dif- 
ferent from that by which individual letters or words may be identi- 
fied in the same text. I could also use the rather old-fashioned 
psychological term apprehension to refer to the way meaning must 
be captured, but that would cloud the similarities between com- 
prehension and letter and word identification. What is different 
about comprehension is that readers bring to the text implicit 
questions about meaning rather than about letters or words. The 
term meaning identification also helps to emphasize that compre- 
hension is active. Meaning doesn't reside in surface structure, 
waiting to be picked up. The meaning that readers comprehend 
from text is always relative to what they already know and to what 
they want to know. Put in another way, comprehension involves the 
reduction of a reader’s uncertainty, asking questions and getting 
them answered, which is a point of view already employed in the 
discussion of letter and word identification. Readers must have 
specifications about meanings, specifications that constantly 
change as meanings develop. 
A passage of text may be perceived in at least three ways: as a se- 
quence of letters from an alphabet, as a sequence of words of a par- 
ticular language, or as an expression of meaning in a certain 
domain of knowledge or understanding. But a passage of text is 
none of these things, or at least it is only these things potentially. 
Basically, written text is a conglomeration of marks on a page or 
screen, variously characterized as visual information, distinctive 
10. THE IDENTIFICATION OF MEANING 
163 
features, or surface structure. Whatever readers perceive in text — 
letters, words, or meanings — depends on the prior knowledge 
(nonvisual information) that they happen to bring and the implicit 
questions they happen to be asking. The actual information that 
readers find (or at least seek) in the text depends on their original 
uncertainty. 
Consider the sentence that you are reading at this moment. The 
visual information in the sentence can be used to make decisions 
about letters, for example, to say that the first letter is c, the second 
o, the third n, and so forth. Alternatively, exactly the same visual in- 
formation can be used to decide that the first word is consider, the 
second word the, the third word sentence, and so forth. The 
reader employs the same visual information, selects from among 
the same distinctive features, but this time sees words, not letters. 
What readers see depends on what they are looking for, on their im- 
plicit questions or uncertainty. Finally, exactly the same visual in- 
formation can be employed to make decisions about meaning in 
the sentence, in which case neither letters nor words would be seen 
individually. It is not easy to say precisely what is being identified in 
the case of meaning, but that is due to the conceptual difficulty of 
saying what meaning is — using words to describe something be- 
yond words — not because the reader is doing anything intrinsically 
different or difficult. The reader is using the same source of visual 
information to reduce uncertainty about meaning rather than 
about letters or words. 
As we have already seen, the amount of visual information re- 
quired to make a letter or word identification depends on the ex- 
tent of the reader’s prior uncertainty, on the number of alternatives 
specified in the reader’s mind (and also the degree to which the 
reader wants to be confident in the decisions to be made). With let- 
ters it is easy to say what the maximum number of alternatives 
is — 26 if we are considering just one particular letter in upper or 
lower case typeface in the English alphabet. It is similarly easy to 
show that the amount of visual information required to identify 
each letter goes down as the number of alternatives that the letter 
might be (the reader’s uncertainty) is reduced. The fewer the alter- 
natives, the more rapidly or easily a letter is identified, because 
fewer distinctive features need to be discriminated for a decision to 
be made. 
It is not so easy to say what the maximum number of alternatives 
is for words, because that depends on the range of alternatives that 
the reader is considering, but again it is not difficult to show that 
the amount of visual information required to identify a word goes 
164 
UNDERSTANDING READING 
down as the reader’s uncertainty is reduced. A word can be identi- 
fied with fewer distinctive features when it comes from a couple of 
hundred alternatives than from many thousands. 
Finally, it is quite impossible to say how many alternative mean- 
ings there might be for a passage of text , because that depends en- 
tirely on what an individual reader is looking for, but it is obvious 
that reading is easier and faster when the reader finds the material 
meaningful than when comprehension is a struggle. The less un- 
certainty readers have about the meaning of a passage, the less vi- 
sual information is required to find what they are looking for in the 
passage. 
The Use of Nonvisual Information 
In each of the preceding cases, nonvisual information can be em- 
ployed to reduce the reader’s uncertainty in advance and to limit 
the amount of visual information that a reader must attend to. 
The more prior knowledge a reader can bring to bear about the 
way letters go together in words, the less visual information is re- 
quired to identify individual letters. Prediction, based on prior 
knowledge, eliminates unlikely alternatives in advance. Similarly, 
the more a reader knows about the way words go together in 
grammatical and meaningful phrases — because of the reader's 
prior knowledge of the particular language and of the topic being 
discussed — the less visual information is required to identify in- 
dividual words. In the latter case, meaning is used as part of 
nonvisual information to reduce the amount of visual information 
required to identify words. 
Many people can follow the meaning of a novel or newspaper ar- 
ticle at the rate of a thousand words a minute, which is four times 
faster than their probable speed if they were identifying every 
word, even with meaning to help them. There is a prevalent mis- 
conception that for this kind of fast reading the reader must be 
identifying only one word in every four and that this gives sufficient 
information at least for the gist of what is being read. But it is easy 
to demonstrate that identifying one word in four won't contribute 
very much toward the intelligibility of a passage. Here is every 
fourth word from a movie review: “Many * * * been * * * face * * * 
business * * * sour ***If***to***." The passage is even less 
easy to comprehend if the selected words are in groups, with corre- 
spondingly larger gaps between them. It is somewhat easier to 
comprehend what a passage is about if every fourth letter is pro- 
vided rather than every fourth word, and. of course, my argument 
1 0. THE IDENTIFICATION OF MEANING 
165 
is that reading at a thousand words a minute is possible only if the 
omissions occur at the featural level. 
It should not be thought that there is a special kind of distinctive 
feature for meaning in print, different from the distinctive features 
of letters and words. There is no “semantic feature,” for example, 
that house and residence physically have in common that we 
should expect to find in print. What makes visual features distinc- 
tive as far as meaning is concerned is precisely what makes them 
distinctive for individual letters and words — the particular alter- 
natives that already exist in the reader’s uncertainty. The same fea- 
tures that can be used to distinguish the letter m from the letter h 
will also distinguish the word mouse from the word house and the 
meaning of we went to the mill from the meaning of we went to the 
hill. It is not possible to say what particular features a reader might 
employ to distinguish meanings; this would depend on what the 
reader is looking for, and, in any case it is not possible to describe 
the features of letters or words either. The situation is only addi- 
tionally complicated by the fact that we can’t say precisely what a 
meaning is. 
Capturing "Meaning" 
As I pointed out in chapter 3, in the discussion of the chasm between 
the surface structures and the deep structure of language, meaning 
lies beyond words. One can’t say what meaning is in general, any 
more than one can say what the meaning of a particular word or 
group of words is, except by saying other words that are themselves 
surface structure. The meaning itself can never be exposed. This in- 
ability to pin down meaning is not a theoretical defect or scientific 
oversight. We should not expect that researchers will soon make a 
wondrous discovery that will enable us to say what meanings are. 
Meaning, to repeat myself, can’t be captured in words. 
A reader doesn’t comprehend the written word table by saying 
the spoken word table either aloud or silently, any more than the 
spoken word table can itself be understood simply by repeating it 
to oneself. And neither the written nor the spoken word table is 
understood by saying silently to oneself “a four-legged flat-topped 
piece of furniture,” or whatever other definition might come to 
mind, because the understanding of the definition would itself 
still have to be accounted for. There can be no understanding and 
no explanation unless the web of language is escaped. The actual 
words, written or spoken, are always secondary to meaning, to 
understanding. 
166 
UNDERSTANDING READING 
We are normally unaware of not identifying individual words 
when we read because we are not thinking about words in any case . 
Written language (like speech) is transparent — we look through 
the actual words for the meaning beyond, and unless there are no- 
ticeable anomalies of meaning, or unless we have trouble compre- 
hending, we are not aware of the words themselves. (When we 
deliberately attend to specific words, for example, in the subtle 
matter of reading poetry, this is a consequence of asking a different 
kind of question in the first place. The sounds that we can give to 
the words don’t so much contribute to a literal interpretation as es- 
tablish a different — a complementary or alternative — kind of mood 
or meaning. ) 
Reading Aloud and Silently 
Of course, word identification is necessary for reading aloud, but 
as I have tried to show, the identification of words in this way de- 
pends on the prior identification of meaning. The voice lags be- 
hind the understanding and is always susceptible, to some extent, 
to diverging from the actual text. The substitution of words and 
even phrases with appropriate meanings is again not something a 
reader will be aware of; the reader’s main concern, even in read- 
ing aloud, must be with the sense of the passage. Misreadings 
(miscues) would have to be pointed out by a listener following 
both the text and the reading. Misreadings of this kind are not 
normally made if the words to be read are isolated or in arbitrary 
lists, but then there is no way that meaning could be sought in 
such words. Besides, in such circumstances readers usually have 
the leisure to scrutinize sufficient visual information to identify 
individual words precisely (giving the words a label rather than a 
meaning) because nothing is lost by reading slowly. According to 
Huey (1908), instruction at the beginning of the 20th century 
placed oral reading long after silent. Currently the trend is the re- 
verse. Huey was critical of any emphasis on reading aloud, which 
he considered much more difficult and unnatural than reading si- 
lently (p. 359). He considered “reading aloud" the opposite of 
“reading for thought.” 
Subvocalization (or reading silently to oneself) can’t in itself con- 
tribute to meaning or understanding any more than reading aloud 
can. Indeed, like reading aloud, subvocalization can only be accom- 
plished with anything like normal speed and intonation if it is pre- 
ceded by comprehension. We don’t listen to ourselves mumbling 
parts of words or fragments of phrases and then comprehend. If 
10. THE IDENTIFICATION OF MEANING 
167 
anything, subvocalization slows readers down and interferes with 
comprehension. The habit of subvocalization can be broken without 
loss of comprehension (Hardyck & Petrinovich, 1970). 
Most people don’t subvocalize as much as they think. If we “lis- 
ten” to ascertain whether we are subvocalizing, subvocalization is 
bound to occur. We can never hear ourselves not subvocalizing, but 
that doesn’t mean that we subvocalize all the time. Why do we sub- 
vocalize at all? The habit may simply be a holdover from our youn- 
ger days, when we were expected to read aloud. A teacher knows 
that children are working if their fingers are moving steadily along 
the lines and their lips are moving in unison. Subvocalization may 
also have a useful function in providing “rehearsal” to help hold in 
short-term memory words that can’t be immediately understood 
or otherwise dealt with. But in such cases subvocalization indi- 
cates lack of comprehension rather than its occurrence. There is a 
general tendency to subvocalize when reading becomes difficult, 
when we can predict less. 
Prediction and Meaning 
We don’t normally read with our minds blank, with no prior pur- 
pose and no expectation of what we might find in the text. We don’t 
look for meaning by considering all possibilities, nor do we make 
reckless guesses about just one; instead, we predict within the 
most likely range of alternatives. In this way we can overcome the 
information-processing limitations of the brain and also the inher- 
ent ambiguity of language. We can derive meaning directly from 
text because we bring expectations about meaning to text. The pro- 
cess is normally as natural, continuous, and effortless as the way 
we bring meaning to every other kind of experience in our life. 
Comprehension is not a matter of putting names to nonsense and 
struggling to make sense of the result, but of operating in the realm 
of meaningfulness all the time. 
LEARNING TO IDENTIFY MEANING 
There’s no need for a special explanation about how children learn 
to apprehend meaning from print because no special process is in- 
volved. Children naturally try to bring sense to print, as they try to 
bring sense to all their encounters with the world around them. For 
them there is no point in language that is not meaningful, whether 
spoken or written. They perceive spoken language by looking for 
meaning, not by focusing on the sounds of words. 
168 
UNDERSTANDING READING 
The Expectation of Sense 
There is a classic illustration of the priority that meaning takes as 
children learn to talk. Even when children try to “imitate,” it is 
meaning that they imitate, not meaningless sounds. McNeill 
(1967) reported an exchange between mother and child which 
went like this: 
Child: Nobody don't like me. 
Mother: No, say "Nobody likes me.” 
Child: Nobody don’t like me. 
(eight repetitions of this exchange) 
Mother: No, now listen carefully, say ‘‘Nobody likes me." 
Child: Oh! Nobody don’t likes me. 
Even when children are asked to perform a language exercise, 
they expect it to make sense. Like the child in the previous exam- 
ple, it takes them a long time to understand the task if they are re- 
quired to attend to surface structure, not to meaning. Children 
don’t need to be told the converse, to look for sense: that is their 
natural way of learning about language. Indeed, they won’t willingly 
attend to any noise that doesn’t make sense to them. 
Just as children don’t need to be told to look for meaning in ei- 
ther spoken or written language, so they also don’t need to learn 
special procedures for finding meaning. Prediction is the basis of 
comprehension, and all children who can understand the spoken 
language of their own environment must be experts at prediction. 
Besides, the very constraints of reading — the constant possibility 
of ambiguity, tunnel vision, and memory overload — serve as re- 
minders to learners that the basis of reading must be prediction. 
Certainly there is no need for a special explanation of how com- 
prehension should be taught. Comprehension is not a new kind of 
skill that has to be learned for reading but the basis of all learning. 
However, it may happen that children at school are taught the re- 
verse of comprehension, being instructed instead to take care to 
“decode” correctly and not “guess” if they are uncertain. They may 
even be expected to learn to read with materials and exercises spe- 
cifically designed to discourage or prevent the use of nonvisual in- 
formation. 
Of course, there are differences between the comprehension of 
written language and the comprehension of speech or of other 
10. THE IDENTIFICATION OF MEANING 
169 
kinds of events in the world, but these are not differences of pro- 
cess. The differences are simply that readers must use distinctive 
features of print to test predictions and reduce uncertainty. 
Children need to become familiar with these distinctive features of 
print and with how they are related to meaning. This familiarity 
and understanding can’t be taught, any more than rules of spoken 
language can be taught, but formal instruction is similarly unnec- 
essary and in fact impossible. The experience that children require 
to find meaning in print can only be acquired through meaningful 
reading, just as children develop their speech competence through 
using and hearing meaningful speech. And until children are able 
to do meaningful reading on their own account, they are clearly de- 
pendent on being read to, or at least on being assisted to read. Let 
me summarize everything I have just said: Children learn to read 
by reading. 
The Right to Ignore 
A final point. It is not necessary for any readers, and especially not 
for beginners, to understand the meaning of everything they at- 
tempt to read. Whether adults are reading novels, menus, or adver- 
tisements, they always have the liberty to skip passages and to 
ignore many small details, either because they are not comprehen- 
sible or because they are not relevant to their interest or needs. 
Children, when they are learning spoken language, seem able and 
willing to follow adult conversations and television programs with- 
out comprehending every word. A grasp of the theme, a general in- 
terest, and the ability to make sense on the basis of a few 
comprehensible parts can be more than sufficient to hold a child’s 
attention. Such partially understood material is indeed the basis 
for learning; no one will pay attention to any aspect of language, 
spoken or written, unless it contains something that is new. For 
children, a good deal that is not comprehensible will be tolerated 
for the opportunity to explore something that is new and interest- 
ing. But children are rarely given credit for their ability to ignore 
what they can’t understand and to attend only to that from which 
they will learn. 
Unfortunately, the right of children to ignore what they can’t un- 
derstand may be the first of their freedoms to be taken away when 
they enter school. Instead, attention may be focused on what each 
child finds incomprehensible in order to “challenge” them to fur- 
ther learning. Anything a child understands may be set aside as 
“too easy.” Paradoxically, many reading materials are made inten- 
170 
UNDERSTANDING READING 
tionally meaningless. In such cases there is no way in which chil- 
dren will be able to develop and profit from their ability to seek and 
identify meaning in text. 
MEDIATED MEANING IDENTIFICATION 
Reading normally involves bringing meaning immediately or di- 
rectly to printed text, without awareness of individual words or 
their possible alternative meanings. There are occasions, however, 
when the meaning of the text or of particular words can’t be imme- 
diately comprehended. On these occasions, mediated meaning 
identification may be attempted, involving the identification of in- 
dividual words before comprehension of a meaningful sequence of 
words as a whole. There are two cases to be considered, the first 
concerned with the mediated meaning identification of entire se- 
quences of words, such as phrases and sentences, and the second 
concerned with the mediated identification of the meaning of occa- 
sional individual words. 
I have already argued that the first is rarely possible. The meaning 
of a sentence as a whole is not understood by putting together the 
meanings of individual words (chap. 3). Individual words have so 
much ambiguity — and usually alternative grammatical functions as 
well — that without some prior expectation of meaning there is little 
chance for comprehension even to begin. In addition, constraints of 
visual information processing and memory arc difficult to overcome 
if the reader attempts to identify and understand every word as if it 
had nothing to do with its neighbors and came from many thou- 
sands of alternatives. So although some theories of reading and 
methods of reading instruction would appear to be based on the as- 
sumption that comprehension of written text is achieved one word 
at a time, the present analysis leaves little on this topic to be dis- 
cussed. To attempt to build up comprehension in such a way must 
be regarded as highly inefficient and unlikely to succeed. 
But the second sense of mediated meaning identification — where 
the passage as a whole is comprehensible and perhaps just one 
word is unfamiliar and not understood — is a more general charac- 
teristic of reading. In this case, the question is not one of trying to 
use the meanings of individual words to construct the meaning of 
the whole, but rather of using the meaning of the whole to provide a 
possible meaning for an individual word. And not only is this possi- 
ble, it is the basis of much of the language learning that we do. The 
bulk of the vocabulary of most literate adults must come from read- 
ing (Nagy, Herman, & Anderson, 1985). It’s not necessary to under- 
10. THE IDENTIFICATION OF MEANING 
171 
stand thousands upon thousands of words to begin to learn to 
read — basically, all that is required is a general familiarity with the 
words and constructions in the written material from which one is 
expected to learn, and then not all of those. And it seems highly un- 
likely that our understanding of many of the words that we have 
learned as a result of reading should be attributed to thousands of 
trips to the dictionary or to asking someone else what the word 
might be. We learn the meaning from the text itself. 
Informal evidence that we quite coincidentally learn new words 
while reading comes from those words whose meaning or refer- 
ence we know well but which we are not sure of pronouncing cor- 
rectly, so there is no way we could have learned about them from 
speech. I am referring to words like Penelope (Penny-loap?), 
Hermione (Hermi-own?), misled (mizzled? myzeled?), and gist 
(like guest or jest?), and perhaps slough and orgy, not to mention 
innumerable foreign words, names, and places. There is what I 
like to call the facky-tious phenomenon (tious to rhyme with pi- 
ous) after the occasion when the mother of a friend commented 
that one didn’t often hear the word facky-tious these days. My 
friend confessed that he couldn’t remember the last time he heard 
the word and asked what it meant. “A little sarcastic or supercil- 
ious,” he was told. Something clicked. “You mean facetious ,” he 
said. “No,” replied his mother thoughtfully, “though the two words 
do have a similar meaning. Come to think of it,” she added, “I don’t 
think I’ve ever seen the word facetious in print.” 
The question, of course, is where she got the correct meaning of 
facky-tious, which she had never heard in speech and had obvi- 
ously never asked anyone about. And the answer must be that she 
learned it the way most of us learn the meaning of most of the 
words we know — by making sense of words from their context, us- 
ing what is known to comprehend and learn the unfamiliar. Medi- 
ated meaning identification from context is something that 
experienced readers do frequently, without awareness, and is the 
basis not just of comprehension but of learning. 
Learning new words without interference with the general com- 
prehension of text is another example of the way that chil- 
dren — and all readers — continually learn to read by reading. The 
vocabulary that develops as a consequence of reading provides a 
permanent source of knowledge for determining the probable 
meaning and pronunciation of new words. If you know both the 
meaning and the pronunciation of auditor and visual, you will 
have little difficulty in comprehending and saying a new word like 
audiovisual. The larger your capital, the faster you can add to it — 
172 
UNDERSTANDING READING 
whether with words or material wealth. The best way to acquire a 
large and useful sight vocabulary for reading is by meaningful 
reading. If the text makes sense, the mediation and the learning 
take care of themselves. 
SO HOW DO YOU RECOGNIZE NEW WORDS? (A REPRISE) 
This section is repetitious, but I think it has to be. My contention 
in this book is that reading is a direct relationship between print 
and meaning. When you look at a line of type, your immediate ex- 
perience is a meaningful event. There is no “decoding" of written 
words to the sounds of words, or to speech. Reading is normally a 
silent affair, just as we normally recognize houses, cars, people, 
and other aspects of a street scene without having to say a name 
for everything we encounter. Our understanding of the situation is 
immediate. Actual names are usually irrelevant for our under- 
standing (unless we are telling someone else about the experi- 
ence, in which case we put words to our understanding). We don’t 
have to say the words first in order to get the meanings, the under- 
standing. 
The same applies to reading. We don’t have to say to ourselves, 
or to anyone else, what a word is in order to understand that word. 
The understanding has to come first. Saying the word is something 
extra you do in order to recite what you read to yourself, in an “in- 
ner voice,” or aloud to someone else. 
But the question inevitably arises, “What about words you’ve 
never met before? If you don’t use phonics, how do you learn to rec- 
ognize those words, and how do you learn to say them?” The ques- 
tion usually means, “How do I teach children to identify and say 
words they have never met before?" Teachers often feel they have to 
find things to do, to instruct children, rather than arrange situa- 
tions where the desired learning will take care of itself. 
I’ll discuss the issue in three steps: ( 1 ) summarizing how text is 
understood when the words are familiar, in silent reading and in 
reading aloud, (2) explaining how unfamiliar words are under- 
stood, and (3) explaining how unfamiliar words are read aloud. 
1 . How Familiar Words Are Understood 
In silent reading, which is the normal and natural way to read, the 
words in their sequence and setting are interpreted immediately. 
We move directly from the words to their combined meaning, with 
no analysis or transformation into any aspect of spoken language. 
10. THE IDENTIFICATION OF MEANING 
173 
We understand the words The dog jumped over the fence the way 
we would understand a picture of a dog jumping over a fence, with- 
out having to say to ourselves “The dog jumped over the fence.” The 
sounds of the words are irrelevant. 
In reading aloud, for our own purposes or to other people, an ex- 
tra step is required. First we have to understand what we are read- 
ing, then we have to say what we understand. We don’t transform 
the uninterpreted words (or their component letters) into sound; 
we put sound to the words that we have interpreted. We do this in 
exactly the same way that we identity the dog that we see jumping 
the fence. We don’t say "There’s a dog,” and then understand that it 
is a dog that we have seen. We recognize a dog, and then say the 
word that we have for animals that we recognize as dogs. 
The problem in all of this, as I explained in chapter 1 , is that it is 
so difficult to escape the lure of letters, to overcome the apparently 
self-evident fact that words are comprised of letters — just as build- 
ings may be made of bricks — and therefore the sounds the letters 
represent must have something to do with reading (although indi- 
vidual bricks have nothing to do with our recognition of buildings. ) 
Where possible I prefer to talk about the structure of written 
words, which happens to consist of letters, rather than talk about 
letters. No one expects structure to decode to sound. 
2. How Unfamiliar Words Are Understood 
I am talking now about understanding , not about reading aloud or 
about transposing written language into silent speech to ourselves. 
How do we get the meaning of an unfamiliar word before we even 
try to say it? The answer is that the meaning comes from the con- 
text in which the word occurs. Words we do know indicate the 
meaning of the word we don’t know. Or rather, the entire grammati- 
cal and semantic structure of the meaningful sequence in which 
the unknown word is embedded, together often with cues within 
the structure of the unknown word itself (words that look alike 
tend to share similar meanings), enmeshes the unknown word in a 
network of understanding, so that the probable meaning is imme- 
diately apparent. And if we make a mistake, the later meaningful 
context usually tells us that our assumption was wrong, and prob- 
ably suggests a more appropriate interpretation. In brief, we get 
the meaning of one small part from the meaning of the whole (just 
as you would have little difficulty inferring the meaning of the word 
glerp if I told you I left my glerp at home this morning and got 
soaked by rain later in the day). 
174 
UNDERSTANDING READING 
There is substantial evidence that readers quickly become ex- 
tremely proficient at attributing the correct meaning to unfamiliar 
words in the normal course of reading, not just experienced adult 
readers, but high school and even younger students. One encounter 
with an unfamiliar word in a meaningful context is enough to give an 
approximate meaning; half a dozen encounters are sufficient to 
draw an accurate conclusion. In this way teenaged readers can learn 
thousands of new words every year (see notes to chap. 12). 
It is a perfectly natural thing to do. Children from the second 
year of their lives accurately infer the meaning of new spoken 
words about 20 times a day, with no allowance for forgetting — both 
from the language itself and the situation in which it occurs. Some- 
one says “Like a drink of blunk?,” holds out a beaker of orange 
juice, and the child knows what blunk is. Children do this without 
instruction from adults or frequent visits to a dictionary. This is 
the way vocabularies grow, in spoken and written language. But it 
applies only to words that occur in meaningful contexts, either 
spoken or written; it doesn’t apply to words that are presented in 
lists or in any other contrived instructional context. As I frequently 
reiterate, we learn when we understand; learning is a by-product of 
understanding. In effect, children learn about language the way ar- 
chaeologists decipher ancient texts, by bringing sense to them. It’s 
all very natural. 
If the meaning of the unfamiliar word that we have identified is 
already in our spoken language vocabulary, then we can associate 
the written word with the spoken word, the sight with the sound. 
The association is mediated by meaning. Our inner "lexicon" is not 
a list of the sounds of words with meanings attached; it is a list of 
meanings with sounds and written forms attached. When we be- 
come familiar with the meaning of written words, we employ our 
lexicon of meanings to establish a relationship between the written 
word and its spoken counterpart. The sequence is 
written word 4 meaning 4 spoken word 
not 
written word 4 spoken word -> meaning 
If the written word has a meaning that we can't associate with a 
spoken word, then it remains one of those written words that we 
can recognize and understand without being able to put it into 
speech (or only with a very rough approximation of what the pro- 
nunciation of the word might be). It is precisely the situation we are 
often in when we encounter recognizable and comprehensible 
10. THE IDENTIFICATION OF MEANING 
175 
symbols that we haven’t put a name to, such as perhaps Q &£, 
0 4 and •. 
3. How Unfamiliar Words Are Read Aloud 
How can we put a sound (or a name) to an unfamiliar word? It de- 
pends on how many words we are familiar with. Words have many 
structural similarities — they begin with the same letters (one of the 
26 alternative shapes), they end with certain patterns like ed, ing, 
ance, and so forth. From our experience with other words, we know 
how they are pronounced, and also how parts of them are pro- 
nounced. We can pronounce familiar structures in unfamiliar 
words the same way we pronounce the same structures in familiar 
words. If we know the pronunciation of contain and persist, we can 
make a good attempt at the pronunciation of consist, especially if we 
can take meaning and grammatical structure into account as well. 
This is not the use of phonics, which we have seen in chapter 9 could 
not possibly work, nor does it demand prior learning of the parts of 
words in isolated exercises. All it demands is experience in reading. 
The use of analogies to indicate both meaning and sound is natural 
and automatic (provided “guessing” has not been inhibited). 
The system is not perfect, but no system for putting sounds to un- 
familiar words could be perfect. Linguists recognize that the spell- 
ing of words doesn’t even attempt to indicate aspects of speech that 
the reader/speaker might be expected to know already. For example, 
everyone knows — unconsciously — from their patterns of speech 
that the past-tense ending ed is pronounced /d/ at the end of words 
like pulled, but that it is pronounced /t/ at the end of words like 
walked, while in words like handed it is pronounced like /id/. Final 
s is pronounced /s/ at the end of words like cups but like /z/ at the 
end of words like beds. None of this has anything to do with phonics, 
which doesn’t acknowledge these distinctions. The difference comes 
from our knowledge (unconsciously acquired) of subtle and special- 
ized rules of spoken language. The component letters of written 
words also tell us nothing about the intonation, whether for example 
that stress falls at the beginning, like acorn, or at the end, like about. 
We recognize new and unfamiliar words because of what we al- 
ready know of words that are familiar. We can put meanings and 
pronunciations to them because of what we know about spoken 
words that have a similar appearance. 
All this is done very rapidly. Researchers have found that “fast 
mapping” of a tentative meaning takes place on the first encounter. 
176 
UNDERSTANDING READING 
and half a dozen more encounters suffice to fully round out the con- 
ventional meaning — with no “feedback” beyond the context in which 
the word occurs (see chap. 12 notes). For the sounds of written 
words the mapping is just as fast, if bothered with at all. The sounds 
are not always correct — we get the meaning but mispronounce the 
conventional sounds for words we haven't encountered in speech. 
Such words may then go into our spoken-language vocabulary with 
an inappropriate pronunciation, to be corrected fairly rapidly if we 
hear the word spoken, but otherwise to remain idiosyncratic. 
How do we get the correct pronunciation? Phonics can’t be de- 
pended on. Either someone tells us how the word is pronounced at 
a helpful time, or we subsequently hear the word spoken and make 
the connection to the word we have encountered in our reading. 
Nothing about finding meanings and pronunciations for new 
words is normally a problem for learners, only for people who 
think that writing is a visible form of speech, rather than that writ- 
ing and speech are related but independent forms of language. 
Neither the sounds of letters nor words themselves are repre- 
sented in speech the way they are in writing. If we didn't have a writ- 
ing system we wouldn’t know what words are, and no one would try 
to break down the flowing intermingling sounds of speech into in- 
dividual segments like letters. Only meaning can be common to 
spoken and written language, and meaning is not something that 
can be decomposed into segments in any form of language. 
ISSUES 
A persisting issue is whether individual words have to be specifi- 
cally identified in normal reading, one at a time, before the mean- 
ing of a phrase, sentence, or passage can be comprehended. Such 
issues are difficult to resolve experimentally — evidence of where 
the eyes are focused doesn’t necessarily indicate what the reader is 
thinking about. Meaning as a concept is difficult enough to contem- 
plate in any case. To try to pinpoint meaning by studying where the 
eyes fixate can be like trying to study digestion by analyzing knife 
and fork movements. As we shall continue to see, driving many of 
the controversies about the nature of reading are deep-rooted be- 
liefs about how children should be taught. 
SUMMARY 
Comprehension, the basic objective of reading, also facilitates the 
process of reading in two ways. Immediate meaning identifica- 
10. THE IDENTIFICATION OF MEANING 
177 
tion makes unnecessary the prior identification of individual 
words, and comprehension of a passage as a whole facilitates the 
comprehension and, if necessary, the identification of individual 
words. Mediated meaning identification increases the probability 
of tunnel vision, memory overload, and ambiguity caused by over- 
dependence on visual information. 
Notes to chapter 10 begin on page 289 covering: 
Effects of meaningful context 
Context and prediction 
"Dual process” 
Reading, Writing, 
and Thinking 
So far we have been primarily concerned with topics much broader 
than reading — like language, comprehension, and memory — or 
with narrow aspects of reading — like letter or word identification. 
In this chapter, the spotlight can finally be directed on reading it- 
self, on the specific act, when something meaningful is in front of a 
reader’s eyes, and the reader is looking at it for a purpose. What 
does it mean to read? What can be said to be happening? And what 
do readers need to know? 
Reading is never an abstract, meaningless activity, although it is 
frequently studied in that way by researchers and theorists and 
still taught in that way to many learners. Readers always read 
something , they read for a purpose, and reading and its recollec- 
tion always involve feelings as well as knowledge and experience. 
Reading can never be separated from the intentions and inter- 
ests of readers, or from the consequences that it has on them. This 
chapter is mainly concerned with what reading means to readers. 
Reading also can never be separated from writing or thinking. Al- 
though this book is not specifically directed to either of these large 
topics, their relevance can’t be ignored, and the chapter ends with 
brief comments on writing and thinking. Learning to read (which 
178 
I 1 . READING, WRITING, AND THINKING 
179 
also will be found inseparable from the act of reading itself) is the 
specific concern of chapter 13. 
On Definitions of Reading 
Books on reading often attempt to define their terms with formal 
statements like “reading is extracting information from print.” 
But such imposing declarations provide no insight into reading, 
and can lead to fruitless debates. A definition doesn’t justify its 
author using a common word differently from anyone else. For- 
mal definitions are useful only if there is a reason for using words 
in a specialized, narrow, or otherwise unpredictable way, and 
even then they can cause more trouble than they are worth be- 
cause readers prefer to interpret familiar words in familiar ways. 
Philosopher Karl Popper (1976) pointed out that precision in lan- 
guage can only be increased at the cost of clarity. As I have already 
discussed, common, easily understood words tend to have a mul- 
tiplicity of meanings, and what usually gives a word an unambigu- 
ous interpretation is neither prior agreement nor fiat but the 
particular context in which it happens to be used. As Popper also 
said, it is better to describe how a word is used than to define it. 
Take the question of whether reading necessarily involves com- 
prehension, an issue sometimes discussed at great length. Such a 
question asks nothing about the nature of reading, only about the 
way the word is used on particular occasions. And the only possi- 
ble answer is that sometimes the word reading implies compre- 
hension, and sometimes it doesn’t. When we suggest that someone 
should read a particular book, we obviously include comprehen- 
sion in our recommendation — it would be redundant if not rude to 
say “I think you ought to read and comprehend this book.” But on 
the other hand, our friend might reply, “I’ve already read it, but I 
didn’t understand it,” now obviously excluding comprehension 
from the meaning of the word reading. Everything depends on the 
general sense in which words are used, even in the same conversa- 
tion, in two successive sentences. If there is doubt, it is better to 
provide a more complete description of how the word is being used 
than to attempt a general definition. 
Consider, for example, the differences between reading a novel, 
a poem, a social studies text, a mathematical formula, a telephone 
directory, a recipe, the formalized description of some opening 
moves in chess, or an advertisement in a newspaper. Novels are 
usually read for the experience, for involvement in a situation, not 
unlike watching a play or movie or participating in actual events, 
180 
UNDERSTANDING READING 
where we are caught up with the characters and motivations of in- 
dividual people and with how circumstances will deal with them. 
To read a novel is to participate in life. A poem may evoke a much 
more intense experience, especially emotionally, involving a partic- 
ular mental attitude and a sensitivity to the sounds as well as to the 
meanings of words, akin in many ways to listening to music. The 
social studies text may lack the direct emotional and aesthetic con- 
nection of a novel or poem, but generate more detailed analytic 
thought — thinking that is more “off the page” and general than the 
details directly presented in the print. The mathematical formula 
is a tool, to be lifted (with understanding) from its position in the 
text and used elsewhere, and the telephone directory is like a col- 
lection of keys, each of which will unlock a particular connection. A 
recipe is a description of actions for the reader to follow, chess no- 
tation involves participation in a game, and a newspaper advertise- 
ment is a device for persuading readers to act in particular ways. 
These descriptions are clearly inadequate for the richness that 
is reading. My aim in attempting a list was to illustrate the richness 
by demonstrating the inadequacy. And even then, I oversimplified. 
There isn’t one kind of novel or one kind of advertisement, and the 
same texts can be read in different ways. A novel can be read like a 
social studies text, and a social studies text like a novel. A newspa- 
per advertisement may be read like a poem. Moreover, each of 
these different ways of reading texts is more like other forms of be- 
havior or experience that don't involve reading than they are like 
other forms of reading. I equated reading a novel with watching a 
play, not with reading a play, and reading a recipe is obviously more 
like cooking than like reading about any other kind of activity. 
There is no one activity that can be summed up as reading-, no de- 
scription that can be summarized as the "process” that is involved. 
The meaning of the word reading in all these senses depends on 
everything that is going on — not just on what is being read, but on 
why a particular reader is reading. It might be said that in all of the 
examples I have given, answers are sought to questions that vary 
with the person asking them. And the only thing that makes all of 
these different activities reading is that the answers are being 
sought in print. 
Asking Questions, Finding Answers 
Because of the limitations on the amount of visual information 
from a text that the brain can deal with, the location and nature of 
the answers must to some extent be predictable. Thus the reader 
11. READING, WRITING, AND THINKING 
181 
must have relevant expectations about the text. All questions must 
be couched within a prediction, a range of possible alternatives. 
This leads to a very broad description that I have already of- 
fered — that comprehension of text is a matter of having relevant 
questions to ask (that the text can answer) and of being able to find 
answers to at least some of those questions. To use a term I intro- 
duced earlier and will shortly elaborate more fully — reading de- 
pends on the reader’s specification of the text. 
The particular questions a reader might ask can range from the 
implications of a single word to matters related to the style, sym- 
bolism, and worldview of the author. I have avoided any attempt to 
list and characterize all these different questions because of their 
very specific — and sometimes specialized — nature. Instead, I have 
focused on three kinds of question that all fluent readers seem able 
to ask and answer in most reading situations, related to the identi- 
fication of letters, words, and meanings. These three kinds of ques- 
tion are alternatives, all three can't be asked simultaneously, and it 
is unnecessary for the reader to attempt to ask them in sequence. 
From this perspective, it doesn’t make sense to ask whether 
print basically consists of letters, words, or meanings. Print is dis- 
criminable visual contrasts, marks on paper or a monitor screen, 
that have the potential of answering certain questions — usually im- 
plicit— that readers might ask. Readers find letters in print when 
they ask one kind of question and select relevant visual informa- 
tion; they find words in print when they ask another kind of ques- 
tion and use the same visual information in a different way; and 
they find meaning in print, in the same visual information, when 
they ask a different kind of question again. It should be rare for a 
reader to ask questions about specific letters (except when letters 
themselves have a particular relevance, for example, as a person's 
initials or as a compass direction N. S, E, or VF). It should also be 
rare for a reader to attend specifically to words, unless again there 
is a particular reason to identify a word, for example, a name. 
Comprehension, as I have said, is relative; it depends on getting 
answers to the questions being asked. A particular meaning is the 
answer a reader gets to a particular question. Meaning therefore 
also depends on the questions that are asked. A reader “gets the 
meaning” of a book or poem from the writer’s (or a teacher's) point 
of view only when the reader asks questions that the writer (or 
teacher) implicitly expects to be asked. Disputes over the meaning 
of text, or the “correct” way to comprehend text, are usually dis- 
putes over the questions that should be asked. A particular skill of 
accomplished writers (and of accomplished teachers) is to lead 
182 
UNDERSTANDING READING 
readers to ask the questions that they consider appropriate. 
Thus, the basis of fluent reading is the ability to find answers in 
the visual information of written language to the particular ques- 
tions that are being asked. Written language makes sense when 
readers can relate it to what they know already ( including those 
occasions when learning takes place, when there is a comprehen- 
sible modification of what readers know already). And reading is 
interesting and relevant when it can be related to what the reader 
wants to know. 
READERS, WRITERS, AND TEXT 
Readers must bring meaning to texts; they must have a developing 
and constantly modifiable set of expectations about what they will 
find. This is their specification of the text. But obviously writers 
make a contribution too. They must have their own specifications. 
And there must be a point at which readers and writers intersect. 
That point is the text, and the next section is about the interconnec- 
tions of readers, writers, and texts. 
Global and Focal Predictions 
So far throughout this book I have talked as if predictions are 
made and dealt with one at a time. But predictions are usually mul- 
tiplex, varying widely in range and significance. Some predictions 
Eire overriding; they carry us across large expanses of time and 
space. Other predictions occurring concurrently are far more tran- 
sient, arising and being disposed of relatively rapidly. Our predic- 
tions are layered and interleaved. 
Consider the analogy of driving a car. We have a general expecta- 
tion that we will reach a certain destination at a certain time, lead- 
ing to a number of relatively long-range predictions about 
landmarks that will be met along the route. Call these predictions 
global, because they tend to influence large parts of the journey. No 
matter how much our exact path might have to be varied because of 
exigencies that arise on the way, swerving to avoid a pedestrian or 
diverting down a side street because of a traffic holdup, these over- 
riding global predictions tend to bring us always toward our in- 
tended goal. 
But although global predictions influence every decision until 
our intended goal is reached, we simultaneously make more de- 
tailed predictions related to specific events during the course of 
the journey. Call predictions of this nature focal , because they 
11. READING, WRITING, AND THINKING 
183 
concern us for short periods of time only and have no lasting con- 
sequence for the journey as a whole. Focal predictions must be 
made, often quite suddenly, with respect to the oncoming truck or 
the pedestrian or as a consequence of a minor diversion. In con- 
trast to global predictions, we can’t be specific about focal predic- 
tions before the journey begins. It would be useless to try to 
predict before starting the specific location of incidents that are 
likely to occur on the way. The occasion for a focal prediction may 
arise out of particular sets of local circumstances, but the predic- 
tion itself will still be influenced by our global expectations about 
the journey as a whole. For example, the modified focal predic- 
tions that will result if we have to make an unexpected detour will 
still be influenced by our overriding intention of eventually reach- 
ing a particular destination. 
We make similar global and focal predictions when we read. 
While reading a novel, for example, we may be concerned with a 
number of quite different predictions simultaneously, some global 
that can persist through the entire length of the book, others more 
focal that can rise and be disposed of in a single fixation. 
We begin a book with extremely global predictions about its 
content from its title and from what perhaps we have heard about 
it in advance. Sometimes even global predictions may fail — we 
discover that a book is not on the topic we anticipated. But usu- 
ally global predictions about content, theme, and treatment per- 
sist throughout the book. At a slightly more detailed level, there 
are likely to be still quite global expectations that arise and are 
elaborated within every chapter. At the beginning of the book we 
may have such predictions about the first chapter only, but in the 
course of reading the first chapter expectations about the second 
arise, the second leads to expectations about the third, and so on 
to the end. Within each chapter there will be rather more focal 
predictions about paragraphs, with each paragraph being a ma- 
jor source of predictions about the next. Within each paragraph 
there will be predictions about sentences and within each sen- 
tence predictions about words. 
Lower level predictions arise more suddenly; we will rarely 
make focal predictions about words more than a sentence ahead of 
where we are reading, nor predictions about sentences more than a 
paragraph ahead, nor predictions about paragraphs more than a 
chapter ahead. The more focal the prediction, the sooner it arises 
(because it is based on more immediate antecedents) and the 
sooner it is disposed of (because it has fewer long-range conse- 
quences). In general, the more focal a prediction, the less it can be 
184 
UNDERSTANDING READING 
specifically formulated in advance. You would be unlikely to pre- 
dict the content of the present sentence before you had read the 
previous sentence, although the content of the paragraph as a 
whole was probably predictable from the previous paragraph. On 
the other hand, predictions at the various levels inform each other. 
The entire process is at once extremely complex and highly dy- 
namic; Fig. 11.1 is an attempt to illustrate it in a simplified and 
static diagram. 
In general, the expectations of Fig. 11.1 should be regarded as 
developing from left to right; the past influences our expectations 
for the future. But it can occasionally help at all levels of prediction 
in reading to glance ahead. The sequence of reading doesn't have to 
follow the page numbering of the book. Similarly, there should per- 
haps be diagonal lines all over the diagram as the outcomes of local 
predictions have their effect on global predictions and the global 
expectations exert their constant influence on specific focal predic- 
tions. At any moment, the character of our existing expectations 
about the book, chapter, paragraph, sentence, and word is our 
ever-changing specification of the text. 
Don’t interpret the diagram too rigidly. It isn’t necessary to pre- 
dict at every level all of the time. We may become unsure of what a 
book as a whole is about and, for a while, hold our most global 
predictions to the chapter or even to a lower level while we try to 
grasp where the book might be going. Sometimes we may have so 
much trouble with a paragraph that we find it impossible to main- 
tain predictions at the chapter level. At the other extreme, we may 
find a chapter or paragraph so predictable, or so irrelevant, that 
Book expectations 
First Page 
Time 
Last Page 
*■ 
FIG. 11.1. Layers of prediction in reading a book. 
1 1 . READING, WRITING, AND THINKING 
185 
we omit predictions at lower levels altogether. In plain language, 
we skip. It is only when we can make no predictions at all that a 
book will be completely incomprehensible. It should also not be 
thought that there are clearly defined boundaries between the dif- 
ferent levels of prediction; the global-focal distinction doesn’t de- 
scribe alternatives but rather the extreme ends of a continuous 
range of possibilities. 
The Writer's Point of View 
We may consider now the intentions of writers, using the frame- 
work of Fig. 11.1 that represented the predictions of readers. To 
some extent, the patterns of predictions and intentions can be seen 
as reflections of each other. 
Writers of books often begin with only global intentions of what 
the book as a whole will be about and of the way the subject will be 
treated. These global intentions, in due course, determine lower 
level intentions for every chapter. Within each chapter more focal 
intentions arise about every paragraph, and within each paragraph 
quite detailed focal intentions arise regarding sentences and 
words. And just as the more focal predictions of the reader tend to 
arise at shorter notice and to be dispensed with more quickly, so 
the more focal intentions of the writer extend over a shorter range 
in both directions. What I want to say in the present sentence is 
most specifically determined by what I wrote in the previous one 
and will, in turn, place a considerable constraint on how I compose 
the following sentence. But these focal constraints are at the de- 
tailed level. My intention in every sentence that I write is also influ- 
enced by the more global intentions for the paragraph as a whole, 
and of course my intention in every paragraph reflects the topic I 
have selected for the chapter and more generally for the book. 
The intentions of writers can be represented by exactly the same 
framework that I have used to represent the predictions of readers 
in Fig. 11.1. The only difference would be that now the diagram 
should be captioned “Layers of intention in writing a book,” with 
the word intentions replacing expectations at every level from 
global to focal. The same qualification would also apply about not 
taking the diagram too literally. Authors may at times be fairly sure 
about their global intentions at book, chapter, and even paragraph 
levels but be lost for focal intentions concerning particular sen- 
tences and individual words. At other times the words may flow 
without any clear indication of where they are going, with the para- 
graph and other more global intentions remaining obscure. 
186 
UNDERSTANDING READING 
Global and Focal Conventions 
The cascading diagram of Fig. 11.1 can be used for a third time, to 
represent the basic relationship between writers and readers. First 
I used Fig. 11.1 to represent the reader’s point of view, the texture 
of predictions. Then with a slight modification of labeling it was 
used from the writer’s point of view, as a network of intentions. 
Finally it can be employed as a representation of the text itself, the 
meeting ground of writer intentions and reader expectations. 
In what way do writers manifest their various intentions, and 
what is it that readers predict at the various global and focal levels? 
As I outlined in chapter 3, the answer is conventions. Conventions 
exist in every aspect of language; they correspond to every kind and 
level of intention and expectation. In considering the written lan- 
guage of books. Fig. 11.1 needs simply to be relabeled “Layers of 
convention in a book," with the word convention replacing expec- 
tations (or intentions) at every level. There are global conventions 
for books as a whole — these are genre schemes, story grammars, 
and the conventions of register. There are conventions for the way 
paragraphs are arranged into chapters and chapters into books — 
these are the discourse structures. There are conventions for the 
way sentences are organized into paragraphs — these are the con- 
ventions of cohesion. There are the conventions for the organiza- 
tion of words in sentences, the conventions of grammar and of 
idiom. And there are conventions for the words themselves, the 
conventions of semantics, and for the physical representation of 
those words, the conventions of spelling. Complete sets of conven- 
tions exist for traditional texts — and for the hypertexts of the 
Internet. 
When labeled for conventions. Fig. 11.1 is. I think, a reasonably 
appropriate way to characterize an entire text. Texts are static — 
they don’t change their structure from moment to moment (unless 
someone is working on them). But the figure offers only a way of 
thinking about readers and writers: I wouldn’t want to suggest that 
such a structure ever exists in its entirety or in a stable form in any- 
one’s head. We can inquire into particular global and focal inten- 
tions or predictions in writers’ and readers' minds at particular 
times, but we should never expect to find a complete or unchanging 
set of them the way the diagram might suggest. Instead we would 
find that writers and readers, each in their own way, have in their 
minds a specification of a text, a specification of global and focal el- 
ements far less complete and detailed than Fig. 11.1, but far more 
dynamic and flexible. 
1 1 . READING, WRITING, AND THINKING 
187 
The Specification of a Text 
Consider the matter first from the writer’s point of view. What does 
a writer have in mind (a) before a text is begun to direct the writing 
that will be done, (b) while the text is being written to ensure that it 
follows the writer’s developing intentions, and (c) when the text is 
done, when the writer can say “That's what I intended to write”? My 
answer each time is a specification. 
The specification of a text is similar in many ways to the specifi- 
cation of a house. Such a specification is not the house itself, nor 
is it the plans for a house. It is a cluster of intentions and expecta- 
tions, of constraints and guidelines, which determine what the 
plans and ultimately the house will be like. Specifications are 
never complete — we wouldn’t say to the architect, “This is exactly 
how we want the house,” because in that case we wouldn’t need 
the architect. Specifications will have gaps, they may even be in- 
ternally inconsistent, and during the designing of the plans we or 
the architect may find a need for the specifications to be changed. 
Indeed, specifications should be expected to change as the execu- 
tion of the plans develops, so that eventually there is a match be- 
tween the plans (and house) and the specifications, between the 
aim and its fulfilment, partly because the house was designed 
around the constraints of the specifications but also because the 
specifications were changed and developed to meet the contingen- 
cies of actually designing and building the house. A different ar- 
chitect might have designed a different house, but we would still 
say “That is what we wanted" if the design is in accordance with 
our final specifications. 
So it is with the writer. The book (or any other kind of text) that 
the author plans will initially develop in conformity with certain 
specifications that don’t contain all the details of the text. And as 
the text develops the specifications will change, partly as the de- 
mands of the text change but also as a consequence of what has al- 
ready been written. And at the end, if the final text is compatible 
with the final specification, the author will say, “That’s what I 
wanted to write,” even though the constantly changing specifica- 
tion at no time spelled out exactly what the book would contain at 
all of its global and focal levels, and even though a different book 
might have been written to the same initial specifications on a dif- 
ferent occasion. 
So it is too with readers. We begin with a sketchy specification 
of the text (“This is a book about reading”), which develops in the 
course of our reading, consolidating in terms of what we have 
188 
UNDERSTANDING READING 
read so far and elaborating when necessary for the prediction of 
what is to come. Focal aspects of the specification are developed 
to make sense of detail as we come to it but then discarded as we 
move on to the next detail. Apart from the occasional quotation or 
specific idea that might lodge in our mind, we shall in general be 
far more concerned with the persisting global aspects of our spec- 
ification than with the transient focal ones. And at the end we will 
have a specification that is still not the book itself but that is our 
ultimate comprehension of the book (just as the specification we 
can put together a week or a month later is our memory of the 
book at that time). 
How we comprehend when we read is a matter of the richness 
and congruence of the specification that we bring to the text and of 
the extent to which we can modify the specification in the course of 
reading the text. What we comprehend and what we are left with in 
memory as a consequence of the reading depend on how our expe- 
rience with the text modifies our specification. Subsequent reflec- 
tion may change the specification even more, of course, which is 
the reason that we often can’t distinguish in memory what we read 
in a text from what we read into it. 
FLUENT READING AND DIFFICULT READING 
A distinction is often drawn between fluent reading and beginning 
reading to contrast the virtuoso manner in which experienced 
readers are supposed to read with the stumbling, less proficient 
behavior of learners. But the distinction isn’t valid. It’s usually pos- 
sible to find something that any beginning reader can read easily, 
even if only one word. And it’s always possible to find something an 
experienced reader can't read without difficulty. The advantage of 
an experienced reader over a neophyte lies in familiarity with a 
range of different kinds of text, not in the possession of skills that 
facilitate every kind of reading. 
For beginners and experienced readers alike, there is always the 
possibility of fluent reading and the possibility of difficult read- 
ing. There is no sudden transition from beginning reading, when 
nothing can be read without difficulty, to fluent reading, when all 
reading is easy. The more we read, the more we are able to read. 
Learning to read begins with one word and one kind of text, contin- 
ues a word and a text at a time, and the learning never stops. Every 
time a reader meets a new word, something new is likely to be 
learned about the identification and meaning of words. Every time 
a new text is read, something new is likely to be learned about read- 
11. READING, WRITING, AND THINKING 
189 
ing different kinds of text. Learning to read is not a process of 
building up a repertoire of specific skills, which make all kinds of 
reading possible. Instead, experience increases the ability to read 
different kinds of text. 
Even experienced readers have difficulty in reading some texts — 
because of the way the texts are written, or because of inadequate 
nonvisual information on the reader’s part, and sometimes be- 
cause of pressures or anxieties involved in the particular act of 
reading. And when otherwise competent readers experience diffi- 
culty in reading, they tend to read like beginners. By the same to- 
ken, when beginners find easy material to read, they tend to read 
like experienced readers. 
In other words, the critical difference is not between experienced 
and beginning reading, or even between “good reading” and “poor 
reading,” but between fluent reading, which even beginners can do 
in the right circumstances, and difficult reading, a situation in 
which even experienced readers can sometimes find themselves. 
The problem for children learning to read is that everything they 
might attempt to read is likely to be difficult. 
Fluent reading involves pursuing a complex and ever-changing 
set of objectives in order to make sense out of print in ways that 
are relevant to the purposes of the reader. Neither individual let- 
ter identification nor individual word identification are involved 
unless they are relevant to the particular requirements of the 
reader. Nor is every potential “meaning” on a page examined un- 
less it has some bearing on the reader's purposes. Fluent reading 
is based on a flexible specification of intentions and expectations, 
which change and develop as a consequence of the reader’s pro- 
gression through a text. Thus, fluent reading demands knowledge 
of the conventions of the text, from vocabulary and grammar to 
the narrative devices employed. How much conventional knowl- 
edge is required depends on the purposes of the reader and the 
demands of the situation. Knowledge need not be complete; in 
fact, provided there is sufficient comprehension to maintain the 
reader’s attention, learning is likely to take place wherever spe- 
cific knowledge is lacking. 
The Consequences of Reading 
Reading is more than just a pleasant, interesting, and informative 
experience. It has consequences, some of which are typical of any 
kind of experience we might have. Other consequences are 
uniquely particular to reading. 
190 
UNDERSTANDING READING 
General consequences of experience are an increase in specific 
memories and knowledge. I haven’t found any studies of how much 
individuals normally remember from what they read (outside of 
artificial experimental situations looking at how much can be re- 
called of specific items determined by the researcher). But com- 
mon observation would suggest that individuals remember as 
much about books that they find interesting and readable as they 
do about “real life" experiences in which they are involved. Many 
anecdotal reports indicate remarkable memories on the part of 
readers for the appearance, titles, authors, characters, settings, 
plots, and illustrations of books that were important to them, often 
extending back to childhood. With books, as with every other kind 
of experience, we remember what we understand and what is sig- 
nificant to us. 
There are also specific consequences. Experience always results 
in learning. Experience in reading leads to more knowledge about 
reading itself. Not surprisingly, students who read a lot tend to 
read better (Anderson, Hiebert, Scott, & Wilkinson, 1985). They 
don’t need to read better in order to read a lot, but the more they 
read, the more they learn about reading. The same researchers re- 
ported that students who read more also tended to have larger vo- 
cabularies, better comprehension, and generally did better on a 
range of academic subjects. In other words, reading makes people 
smarter. 
Other things are learned through reading. I've argued at length 
(Smith, 1983b, 1994) that it is only through reading that anyone 
can learn to write. The only way to learn all the conventions of 
spelling, punctuation, capitalization, paragraphing, grammar, and 
style is through reading. Authors teach readers about writing. 
In the next two chapters, I describe learning in metaphorical 
terms as the membership of clubs. By joining the club of readers, 
even as beginners, individuals can learn to become readers and 
writers. But reading also opens the doors to any club that can be 
the topic of a book, which probably means most of the clubs in the 
world and certainly many clubs that could not exist in the world as 
we know it. Reading is the club of clubs, the only possibility for 
many experiences of learning. 
And finally, there are emotional concomitants and conse- 
quences of reading. Reading, like everything else, inevitably in- 
volves feelings. On the positive side, reading can provide interest 
and excitement, stimulate and alleviate curiosity, console, encour- 
age, rouse passions, relieve loneliness, assuage tedium or anxiety, 
palliate sadness, and on occasion induce sleep. On the negative 
1 1 . READING, WRITING, AND THINKING 
191 
side, reading can bore, confuse, and generate resentment. The 
emotional response to reading is treated insufficiently in most 
books about literacy (not excluding the present volume), although 
it is the primary reason most readers read, and probably the pri- 
mary reason most nonreaders don’t read. 
Because of the range and depth of feelings involved, attitudes to- 
ward reading become habitual. Reading can become a desired ac- 
tivity or an undesirable one. People can become inveterate readers. 
They can also become inveterate nonreaders, even when they are 
capable of reading. One of the great tragedies of contemporary edu- 
cation is not so much that many students leave school unable to 
read and to write, but that many graduate with an antipathy to 
reading and writing, despite the abilities they might have. Nothing 
about reading or its instruction is inconsequential. 
READING AND THINKING 
The heading may be a trifle misleading. Reading is thinking, as I 
hope I have demonstrated throughout this chapter. And the think- 
ing we do when we read, in order to read, is no different from the 
thinking we do on other occasions. Just as we can’t talk without 
thinking, or understand what someone is saying without think- 
ing, or make any sense of the world without thinking, so it is im- 
possible to read and not think. (If we sometimes say that we have 
spoken without thinking, we mean that we didn’t consider all the 
implications of what we said. ) Reading is thinking that is partly fo- 
cused on the visual information of print; it’s thinking that is stim- 
ulated and directed by written language. The only time we might 
attempt to read without thinking is when the text we are trying to 
read is meaningless to us, a situation unlikely to persist in normal 
circumstances. 
It is true that we may read a story or magazine to relax, in order 
not to think about particular things — but we obviously have to 
think enough about whatever we are reading in order to be dis- 
tracted from other thoughts. If we fail to read every story with the 
intensity and acumen of a literary critic, it is probably not because 
we can’t think, but because we aren’t interested in reading like a lit- 
erary critic. 
The thought in which we engage while reading is like the 
thought we engage in while involved in any kind of experience. 
Fulfilling intentions, making choices, anticipating outcomes, and 
making sense of situations are not aspects of thinking exclusive to 
fluent reading. We must draw inferences, make decisions, and 
192 
UNDERSTANDING READING 
solve problems in order to understand what is going on in situa- 
tions that involve reading and situations that don’t. Reading de- 
mands no unique forms or “skills" of thought. 
An enormous advantage of reading over thinking in other cir- 
cumstances is the control that it offers over events. Readers can 
stop the action, and pause in the middle of an experience for reflec- 
tion. Readers can relive reading experiences, as often as they wish, 
and examine them from many points of view. Readers can even 
skip over experiences they are not interested in having or that 
would disrupt their flow of thought. Readers have power. 
Reading is no different in essence from any other manifestation 
of thoughtful activity — but it may be the most natural and satisfy- 
ing form of thinking available to us. The human brain runs on sto- 
ries. Our theory of the world is largely in the form of stories. Stories 
are far more easily remembered and recalled than sequences of 
unrelated facts. The most trivial small episodes and vignettes are 
intrinsically more interesting than data. We can’t see random pat- 
terns or dots (or clouds or stars) without putting faces or figures to 
them. We can’t even observe small points of light moving randomly 
against a dark background without seeing them “interact” with 
each other in a narrative fashion (Michotte, 1946). 
Thinking thrives on stories, on the construction and exploration 
of patterns of events and ideas, and reading often offers greater 
scope for engaging in stories than any other kind of activity. 
ISSUES 
The rift between the “experience” and “information” approaches to 
teaching reading is less an unresolved problem than a gulf between 
totally antithetical points of view, making a tremendous difference 
to how research is interpreted, theories developed, and teaching 
and learning perceived. The nature of reading itself is at issue: 
whether it is a process of acquiring information from print that 
may be turned on in any circumstances, or a creative experiential 
interaction in an environment of print. When teachers and learners 
are evaluated on “performance indicators” or on the “product" or 
“output” of reading instruction, it is almost invariably acquisition 
of information rather than quality of experience that is assessed. 
SUMMARY 
Reading — like writing and all other forms of thinking — can never 
be separated from the purposes, prior knowledge, and feelings of 
11. READING, WRITING, AND THINKING 
193 
the person engaged in the activity nor from the nature of the text be- 
ing read. The conventions of texts permit the expectations of read- 
ers and the intentions of writers to intersect. Global and focal 
expectations and intentions form a personal specification that 
readers and writers develop and modify as they proceed through a 
text. The fluency of reading depends as much on characteristics of 
the text and reader as on reading ability. Experienced readers who 
find a text difficult may read like beginners. 
Notes to chapter 1 1 begin on page 294 covering: 
Comprehension and thinking 
Reading speed 
Comprehension and context 
Benefits of reading 
Learning About 
the World 
This chapter introduces the topic of learning. It is not specifically 
concerned with learning to read, a matter postponed to the next 
and final chapter. But this chapter is relevant to the manner in 
which children learn to read, because this is the same as the man- 
ner in which children achieve mastery of spoken language and, 
even earlier, begin learning about the world in general through 
their first elaborations of a theory of the world. 
The chapter is linked with many of the preceding chapters, with 
their emphasis on meaning and comprehension in reading, be- 
cause it shows that the basis of all learning, including learning to 
read, is comprehension. Children learn by relating their under- 
standing of the new to what they know already, modifying or elabo- 
rating their prior knowledge. Learning is continuous and 
completely natural, and it is not necessary to propose separate 
“processes” of motivation and reinforcement to sustain and con- 
solidate learning (nor should it be necessary for teachers to regard 
incentives and rewards as separate concerns that can be grafted 
onto reading instruction). Children may not always find it easy or 
even necessary to learn what we try to teach them, but they find the 
state of not learning anything intolerable. 
194 
12. LEARNING ABOUT THE WORLD 
195 
CONSTRUCTING A THEORY OF THE WORLD 
Chapter 2 discussed the complex yet precise and accurate theory 
of the world that we all possess. Obviously, we were not born with 
such a theory. The ability to construct a theory of the world and to 
predict from it may be innate, but the actual content of the theory, 
the specific detail underlying the order and structure that we come 
to perceive in the world, is not part of our birthright. But equally 
obviously, very little of our theory can be attributed to instruction. 
Only a small part of what we know is actually taught to us. 
The Cat and Dog Problem 
Consider again what we know that enables us to tell the difference 
between cats and dogs. What were we taught that has given us this 
ability? It is impossible to say. Just try to write a description of 
cats and dogs that would enable a being from outer space — or a 
child who has never seen cats and dogs before — to tell the differ- 
ence. Anything you might want to say about the appearance of 
some dogs, that they have long tails or pointed ears or furry coats, 
will apply to some cats and not to other dogs. The difference be- 
tween cats and dogs is implicit knowledge that we can’t put into 
words. Nor can we communicate this knowledge by pointing to a 
particular part of cats and dogs and saying “That’s where the dif- 
ference lies.” 
Differences obviously exist between cats and dogs, but you can’t 
find and don’t need language to distinguish them. Children without 
language can tell the difference between cats and dogs. Cats and 
dogs can tell the difference between cats and dogs. But if we can’t 
say what this difference is, how can we teach it to children? What 
we do. of course, is point out to children examples of the two kinds 
of animal. We say “That’s a cat” or “There goes a dog.” But pointing 
out examples doesn’t teach children anything: it merely confronts 
them with the problem. In effect, we say, “There is something I call 
a cat. Now you find out why.” The “teacher” poses the problem and 
leaves the child to discover the solution. 
The same argument applies to just about everything we can dis- 
tinguish in the world, to the letters of the alphabet, to numbers, 
chairs and tables, houses, foodstuffs, flowers, trees, utensils, and 
toys, to every kind of animal, bird, and fish, to every face, every car 
and plane and ship, thousands upon thousands of objects that we 
can recognize not only by sight but by other senses as well. And 
when did anyone tell us the rules? No one has ever told us, “Chairs 
196 
UNDERSTANDING READING 
can be recognized because they have four legs and a seat and possi- 
bly a back and arms.” (You can see how inadequate a description 
would be.) Instead, somebody once said in passing, “There’s a 
chair,” and left us to decide not only how to recognize chairs on 
other occasions but also to discover what exactly the word chair 
means, how chairs are related to everything else in the world. 
With reading we don’t even need someone to pose the problem 
in the first place. Reading at the same time presents both the 
problem and the possibility of its solution. Just by virtue of being 
a reader, every one of us has acquired a sight vocabulary of at least 
50,000 words, words that we can identify on sight the way we rec- 
ognize familiar faces and houses and trees. How did we acquire 
this enormous talent? Fifty thousand flashcards? Fifty thousand 
times a teacher wrote a word on a board and told us what it was? 
Fifty thousand times we blended together the sound of a word 
through phonics? We have learned to recognize words by reading. 
Not only can we recognize 50,000 words on sight — and also, of 
course, by sound — we can usually make sense of all these words. 
Where have all the meanings come from? Fifty thousand trips to 
the dictionary? Fifty thousand vocabulary lessons? We have 
learned all the conventions of language by making sense of it. What 
we know about language is largely implicit, just like our knowledge 
of cats and dogs. So little of our knowledge of language is actually 
taught; we underestimate how much of language we have learned. 
Most of our theory of the world, including most of our knowledge 
of language, whether spoken or written, is not the kind of knowl- 
edge that can be put into words; it is more like the implicit cat- 
and-dog kind of knowledge. Knowledge that no one can put into 
words is not knowledge that can be communicated by direct in- 
struction. 
How, then, do we acquire and develop the theory of the world 
we have in our heads? How does it become so complex and pre- 
cise and efficient? There seems to be only one answer: by testing 
hypotheses. 
Learning by Hypothesis Testing 
Children learn by testing hypotheses. For example, a child might 
hypothesize that the difference between cats and dogs is that cats 
have pointed ears. The child can then test this hypothesis by saying 
"There’s a cat” or "What a nice cat” when any animal with pointed 
ears passes by, and “There’s a dog” (or “That's not a cat ”) for any an- 
imal without pointed ears. Any reaction tells the child whether the 
12. LEARNING ABOUT THE WORLD 
197 
hypothesis is justified or not. If someone says, “Yes, there’s a pretty 
cat,” or accepts the child’s statement by making no overt response, 
then the child has learned that the hypothesis has worked, on this 
occasion at least. The child’s theory can be tentatively modified to 
include a rule that cats are animals with pointed ears. But if some- 
one says to the child. “No, that’s a dog,” then the child knows that 
the hypothesis has failed. Another hypothesis must be selected 
and tested. Clearly, more than one test will be required; it will take 
experience with cats and dogs before a child can be reasonably cer- 
tain of having uncovered dependable differences between them 
(whatever the differences may be). But the principle is always the 
same: Stay with your theory for as long as it works; modify your 
theory — look for another hypothesis — whenever it fails. 
Note that it is essential for the child to understand the problem 
in the first place. Children won’t learn to recognize cats simply by 
being shown cats; they will not know what to look for. Both cats and 
dogs must be seen in order for the hypothesis about their relevant 
differences to arise. Children learn each letter of the alphabet by 
seeing them all; they must see what the alternatives are. 
There is an intimate connection between comprehension and 
learning. Children’s tests never go beyond their theories; they must 
comprehend what they are doing all the time they are learning. 
Anything that bewilders a child will be ignored; there is nothing to 
be learned there. It isn’t nonsense that stimulates children to learn, 
but the possibility of making sense; that’s why children grow up 
speaking language and not imitating the noise of the air condi- 
tioner. Children don’t learn by being denied access to problems. A 
child learning to talk must be immersed in spoken language, and it 
is far better that a beginning reader having difficulties should be 
helped to read than be deprived of reading. 
This process of hypothesis testing goes on instinctively, below 
the level of awareness. If we were aware of the hypotheses we test, 
then we could say what it is that enables us to tell the difference be- 
tween cats and dogs. We are no more conscious of the hypotheses 
that underlie learning than we are of the predictions that underlie 
comprehension, or of the theory of the world itself. Indeed, there is 
basically no difference between comprehension and learning; hy- 
potheses are simply tentative predictions. 
LEARNING ABOUT LANGUAGE 
When does all this testing take place? I think that for young chil- 
dren there is only one answer: They are testing hypotheses all the 
198 
UNDERSTANDING READING 
time. Their predictions are always tentative. This assertion is best 
illustrated with respect to the topic with which we are most con- 
cerned, namely, language. 
Bringing Meaning to Speech 
Children who have just begun to talk frequently make statements 
that are completely obvious. A child looking out of a window with 
you will say something like “See big plane” although you may even 
have pointed out the plane in the first place. Why then should the 
child bother to make the statement? The answer is, because the 
child is testing hypotheses. In fact, a child could be conducting no 
fewer than three different tests at the same time in that one simple 
situation. 
The child could be testing the hypothesis that the object you can 
both clearly see in the sky is a plane, that it is not a bird or some 
other unidentified flying object. When you say “Yes. I see it,” you are 
confirming that the object is a plane. Even silence is helpful, be- 
cause the child would expect you to make a correction if the hy- 
pothesis were in error. The second hypothesis that the child might 
be testing concerns the sounds of the language, that “plane” is the 
right name for the object, rather than "pwane," “prane.” or what- 
ever else the child might say. Once again the child can assume that 
if you don’t take the opportunity to make a correction, then there is 
nothing to be corrected. A test has been successfully conducted. 
The third hypothesis that the child may be testing is linguistic, 
whether “See big plane” is a grammatically acceptable and mean- 
ingful sentence in adult language. The feedback comes when the 
adult says “Yes, I can see the big plane.” The child learns to produce 
sentences in your language by using tentative sentences for which 
you both already know the meaning, in a situation that you both 
comprehend. 
The same principle of making sense of language by understand- 
ing the situation in which it is used applies in the other direction as 
children learn to comprehend adult speech. At the beginning of 
language learning, infants must be able to understand what adults 
say before they can understand adult language. Does that state- 
ment sound paradoxical? What I mean is that children don't come 
to understand sentences like “Would you like a drink of juice?" or 
even the meaning of a single word like juice by figuring out the lan- 
guage or by having someone tell them the rules. Children learn be- 
cause initially they can hypothesize the meaning of a statement 
from the situation in which it is uttered. An adult saying “Would 
12. LEARNING ABOUT THE WORLD 
199 
you like a drink of juice?” is usually carrying or indicating a drink 
of juice. This language is situation-dependent speech. From such 
situations a child can hypothesize what might happen the next time 
someone mentions juice. The situation provides the meaning and 
the utterance provides the evidence; that is all a child needs to con- 
struct hypotheses that can be tested on future occasions. Children 
don’t learn language to make sense of words and sentences; they 
make sense of words and sentences in understanding how lan- 
guage is used (Macnamara, 1972). Adults help children to do this 
through “caretaker talk” — sometimes called “mother ese” (and 
“fatherese”) — engaging infants in language use, making language 
easy and useful. We shall see a similar phenomenon in the way 
adults (or other experienced readers) support children in their 
learning to read by reading to them and for them. 
There is an interesting role for the eyes to play in these first expe- 
riences with language. Newson and Newson (1975) noted that the 
sharing of meaning is facilitated by a convergence of gaze. When a 
parent offers an infant a drink of juice, they are probably looking 
not at each other but at the juice that the adult is offering. When a 
parent says “There’s a big dog" to a baby who doesn’t understand 
the word, the gaze of parent and baby is likely to converge on the 
dog. By bringing a possible meaning to the utterance, the infant can 
hypothesize a relationship between the two, and thus test, confirm, 
or modify provisional rules about this relationship — a highly effi- 
cient procedure that will work only if the infant can make sense of 
the purpose of adult language . 
I know of no research on how much spoken language children 
might learn simply by observation. But if a baby can hypothesize 
and test a potential meaning when offered juice, there is no reason 
why the child could not test a similar hypothesis by overhearing 
one adult offer another a cup of coffee, provided the situation is vis- 
ible. The child could again compare probable meaning with utter- 
ance. There are obvious limits to the number of language 
interchanges in which infants are directly involved. It might at least 
seem possible that most infants overhear far more language than is 
actually addressed to them, although again there is no research on 
the issue. And by and large much of this overheard domestic lan- 
guage would be situationally meaningful; it would have purposes 
and outcomes that are both predictable and testable. 
It is in fact the purposes of language, the uses to which it is put, 
that are the key to infant language learning. As the linguist 
Halliday (1973) pointed out, children learn language and its uses 
simultaneously. They don’t learn language, either spoken or writ- 
200 
UNDERSTANDING READING 
ten, which they then use for various purposes. The learning 
comes with the use of language and with the understanding of its 
uses. Language learning is incidental. Children don’t learn about 
language as an abstraction, as an end in itself, but as a means of 
achieving other ends, like getting another drink of juice, learning 
to distinguish cats from dogs, or enjoying a story from a book. 
The basic insight that must enable a child to make sense of speech 
is that its sounds are not random; they are not arbitrarily substi- 
tutable. By this I mean that the sounds of speech make a differ- 
ence — they are there for a purpose. An adult can’t produce the 
sounds “There’s a truck” when the intended meaning is “Let’s go 
for a walk.” 
THREE ASPECTS OF LEARNING 
Learning is the modification or elaboration of what is already 
known, of cognitive structure, our theory of the world. What exactly 
is modified or elaborated? It can be any of the three components of 
the theory: the category system, the rules for relating objects or 
events to categories (sets of distinctive features), or the complex 
network of interrelations among categories. 
Children are constantly required to establish new categories 
in their cognitive structure and to discover the rules that limit 
the allocation of events to a new category. They have to learn that 
not all animals are cats and dogs but that some animals are. 
Children learning to sight-recognize the printed word cat have to 
establish a visual category for that word, just as they must have a 
category for actual cats, distinguished from other categories for 
dogs, and so forth. Skilled readers develop categories for every 
letter of the alphabet and also for every word that can be identi- 
fied on sight, together, possibly, with categories for frequently 
occurring syllabic groups of letters, for rhymes, and for mean- 
ingful segments like plurals and tense markers. This process of 
learning to establish categories involves hypothesizing what are 
the significant differences — the only reason to establish a new 
category is to make a new differentiation in our experience, and 
the learning problem is to find the significant differences that 
should define the category. 
Each category that we distinguish must be specified by at least 
one set of distinctive features. Every time children succeed in 
learning to recognize something new, they must have established 
a new set of distinctive features. But usually they go further and 
establish alternative sets of features for specifying the same cate- 
1 2. LEARNING ABOUT THE WORLD 
201 
gories. They learn that an a, a, or even an A should be categorized 
as the letter “a” just as many different-looking animals must be 
categorized as a cat. Any set of features that will serve to catego- 
rize an object is a criterial set, and alternative sets for the same 
category are functionally equivalent. As children learn, they dis- 
cover more and more ways in which to make the decision that a 
particular object or event should be categorized in a certain way. 
The number of functionally equivalent criterial sets gets larger. 
Learning is also involved in the ability to make use of less and less 
featural information to comprehend text. We ran into many exam- 
ples of the use of functionally equivalent criterial sets of features 
in our discussions of letter and word identification. Most skilled 
readers can identify words that have had large parts (many fea- 
tures) obliterated and can make sense of text that has even more 
features obliterated. All this is possible because we have learned 
to make optimal use of the information that is available, both vi- 
sually and from our acquired knowledge of the language. 
Finally, children constantly learn new interrelationships among 
categories, developing their ability to make sense of language and 
the world. Understanding how words go together in meaningful 
language makes prediction possible, and therefore comprehen- 
sion. These interrelationships are also not taught. But a child can 
learn them by the same process of hypothesis testing. Comprehen- 
sion is the basis of a child's learning to read, but reading, in turn, 
contributes to a child’s growing ability to comprehend by permit- 
ting elaboration of the complex structure of categories, feature 
lists, and interrelationships that constitute every child’s theory of 
the world. 
LEARNING ALL THE TIME 
Learning is continual and effortless, as natural as breathing. A 
child doesn’t have to be especially motivated or rewarded for learn- 
ing. Children will strive to avoid situations where there is nothing 
to learn, just as they will struggle to escape situations where 
breathing is difficult. Inability to learn can be suffocating. 
There’s no need to worry that children who are not constantly 
driven and cajoled will “take the easy way out" and not learn. Young 
children who read the same book twenty times, even though they 
know the words by heart, are not avoiding more “challenging" ma- 
terial in order to avoid learning; they are still learning. It may not be 
until they know just about every word in a book that they can get on 
with some of the more complex aspects of reading, such as testing 
202 
UNDERSTANDING READING 
hypotheses about meaning and learning to use as little visual infor- 
mation as possible. 
Underestimating Learning 
It is because children learn continuously and effortlessly that 
adults generally fail to give them credit for the amount of learning 
that they do. It is a common adult belief that learning is a difficult 
and even painful activity, that it involves grappling with something 
that you don't understand, and therefore necessarily leaves marks 
of effort and strain. But in fact, the sight of a child struggling to 
learn is a clear sign that learning is not taking place, that the child 
is confronted by something incomprehensible. When learning does 
occur, it is inconspicuous. 
Because of this myth that learning is effortful, many adults be- 
lieve that they themselves don't learn often or without strain. They 
regard learning as a struggle to make sense of a textbook or set of 
exercises, not as something that takes place whenever they relax to 
read a magazine or enjoy a movie. But the next day they can relate a 
large part of what interested them in the magazine and recall a sur- 
prising amount of detail from the movie, detail that may stay with 
them for months or years afterward. If we can remember, we must 
have learned, and it is pointless to argue that this wasn't learning 
because there was no conscious effort to remember. 
Children are equipped with a very efficient device that prevents 
their wasting time in situations where there is nothing to learn. 
That device is called boredom, and boredom is something all chil- 
dren want to avoid. A child who is bored in class is not demonstrat- 
ing ill will, inability, or obstinacy; boredom should convey just one 
very clear message for the teacher. There is nothing in the particu- 
lar situation for the child to learn. 
There are two reasons why there might be nothing for a child to 
learn in a particular situation, and hence two reasons for bore- 
dom, that arise from quite different sources. One reason why chil- 
dren might have nothing to learn is very simple — they know it 
already. Children won’t attend to anything they already know. But 
children will also suffer and exhibit the same symptoms of bore- 
dom because they can't make sense of what they are expected to 
learn. Teachers might believe that a certain exercise will improve 
useful knowledge or skills, but unless the learner can see sense in 
the exercise, the instruction is a waste of time. (I know this is hard 
on teachers who are obliged to follow a set curriculum, but learn- 
ing is natural and administrative edicts often aren't.) 
1 2. LEARNING ABOUT THE WORLD 
203 
The Risk and Rewards of Learning 
There is one other reason why children might turn their faces 
against learning, and that is its risk. In order to learn you must 
take a chance. When you test a hypothesis, there must be a possi- 
bility of being wrong. If you are certain of being right, there can be 
nothing to learn because you know it already. And provided there is 
a possibility of being wrong, you learn whether you are right or not. 
If you have a hypothesis about what constitutes a cat, it makes no 
difference whether you say “cat” and are right or say “dog” and are 
wrong. In fact, you often get the most useful information when you 
are wrong because you may be right for the wrong reason, but when 
you are wrong you know you have made a mistake. 
Many children become reluctant to learn because they are afraid 
of making a mistake — consider the relative credit children are 
given in and out of school for being “correct” and for being “wrong.” 
There’s no need for learning to be extrinsically rewarded. The fi- 
nal exquisite virtue of learning is that it provides its own reward. 
Learning is satisfying, as everyone knows. It is part of the totally ful- 
filling and absorbing state of flow described by Csikszentmihalyi 
( 1 990), which we achieve only by losing ourselves in what we are do- 
ing. Deprivation of learning opportunities is boring, and failure to 
learn is frustrating. If a child needs reward or special recognition for 
learning, then there is only one conclusion to be drawn: that the 
child doesn’t see any sense in the activity in the first place. 
THE CONDITIONS OF LEARNING 
I want to elaborate upon what I have said about learning in a rather 
different way. Learning is continuous, a natural state of the brain, 
and children therefore are likely to be learning all the time. There 
can be no other explanation for the enormous amount of unsus- 
pected learning of the conventions of language that takes place. 
What then are the conditions under which these ever-learning 
brains succeed in learning as much as they do? And why is it that 
learning sometimes fails, as it sometimes does for all of us, so that 
something that even the learner wants to master remains un- 
learned? Three constituents seem to determine what is learned, 
when it is learned, and whether indeed learning will take place at 
all. These may be termed demonstrations, which are learning con- 
ditions existing in the world around us; engagement, which is the 
interaction of the learner with a demonstration; and sensitivity , 
the learner’s learning condition (Smith, 1981). 
204 
UNDERSTANDING READING 
Demonstrations 
The first essential constituent of learning is the opportunity to see 
what can be done and how. Such opportunities may be termed 
demonstrations because they literally show a potential learner 
“This is how something is done.” The world continually provides 
demonstrations, through people and through their products, by 
acts and by artifacts. 
Every act is a cluster of demonstrations. Saying to a child 
"Here’s your juice” demonstrates the meaning of the word juice 
and the language with which juice is presented. Saying “There’s a 
big dog” demonstrates that there is a category of objects called 
dogs, that “dog” is the name of that category, and that the animal 
being referred to is a member of that category with all the appropri- 
ate distinctive features. A teacher who stands before a class dem- 
onstrates how a teacher stands before a class, how a teacher talks, 
how a teacher dresses, how a teacher feels about what is being 
taught and about the people being taught. A tired teacher demon- 
strates how a tired teacher behaves; a disinterested teacher dem- 
onstrates disinterest. Enthusiasm demonstrates enthusiasm. The 
fact that children are learning all the time is a ticking bomb in every 
classroom. What kind of reading do children see teachers doing? 
What do teachers demonstrate about their interest in reading? 
Every artifact is a cluster of demonstrations. Every book demon- 
strates how pages are put together, how print and illustrations are or- 
ganized on pages, how words are set out in sentences, and how 
sentences are punctuated. A book demonstrates the appearance and 
meaning of every word in that book. It demonstrates a particular 
genre scheme, discourse structure, and perhaps a story grammar 
too. What do our artifacts in the classroom demonstrate? Is it possi- 
ble that children, and older students, are exposed to demonstrations 
that books can be incomprehensible, that they can be nonsense? 
An important category of demonstrations is self-generated, like 
those we can perform in our imagination. We can try things out in 
the mind — in the world inside rather than in the world around 
us — and explore possible consequences without anyone actually 
knowing what we are doing. How much opportunity do children 
have for such private demonstrations? 
Engagement 
I chose the term engagement deliberately for the productive inter- 
action of a learner with a demonstration, because my image is of 
12. LEARNING ABOUT THE WORLD 
205 
the meshing of gears. Learning occurs when the learner engages 
with a demonstration, so that it becomes, in effect, the learner’s 
demonstration. 
Most people are familiar with the experience of reading a book 
or magazine and stopping suddenly, not because of something 
they didn’t understand, but because their attention was engaged 
by a spelling they didn’t know. They didn’t start to read to have a 
spelling lesson, nor could they have predicted the unfamiliar 
spelling that they actually met, but when they encountered it — 
perhaps a name that they had only previously heard on radio or 
television — they stopped and in effect said, "Ah, so that’s the way 
that word is spelled.” At such a moment we can catch ourselves in 
the act of learning; we have not simply responded to a spelling, we 
have made it a part of what we know. Sometimes it is not a spelling 
that stops us. but a particularly interesting fact or the answer to a 
question that has been puzzling us for some time. 
The second example is similar. Once again we find ourselves 
pausing while we read, this time not because of a spelling or some 
other piece of information, and certainly not for lack of under- 
standing, but simply because we have just read something that is 
particularly well put, an interesting idea appropriately expressed . 
This time we have engaged not with a spelling or an interesting new 
item of information, but with a style, a tone, a register. We are learn- 
ing about language from the way someone else is using it. 
The two examples given were necessarily of situations in which 
we might actually be consciously aware of a learning moment. But 
such moments are rare. Perhaps we catch ourselves engaging 
with a new spelling or idea because it is a relatively infrequent 
event in our lives, because we have learned most of the things we 
want or expect to learn by now. New information or experiences 
can be surprising. But children learning the sounds, meanings, 
and written appearance of scores of new words every day of their 
lives are hardly likely to be stopped, like an adult, by the novelty of 
actually meeting something new. Instead, most of their learning 
must be so constant and casual that it doesn’t intrude into con- 
sciousness. 
Learning by engaging in the demonstrations of others is a par- 
ticularly efficient and economical way for children to learn, be- 
cause it limits the possibility of mistake and uncertainty. You are 
not likely to be in error if you let the demonstrator do your learn- 
ing trials for you. This is learning by conducting experiments, 
where the other person (who can do it) conducts the experiment. 
It is hypothesis testing where the appropriate hypothesis is easily 
206 
UNDERSTANDING READING 
available in the skilled performance of the demonstrator. The act 
of learning becomes vicarious. 
Sensitivity 
What makes the difference between whether we learn or don't learn 
from any particular demonstration? The answer can’t be motiva- 
tion, a grossly overrated factor, especially in schools where it is 
sometimes used to cover a multitude of other possibilities. For a 
start, learning of the kind described in this chapter usually occurs 
in the absence of motivation, certainly in the sense of a deliberate, 
conscious intention. It makes no sense to say an infant is motivated 
to learn to talk or that we are motivated to remember what is in the 
newspaper, unless the meaning of motivation is made so general 
that it can’t be separated from learning. 
On the other hand, motivation doesn’t ensure learning. No mat- 
ter how much they are motivated to spell, or to write fluently, or to 
learn a foreign language, many people still fail to learn these things. 
Desire and effort don’t necessarily produce learning. Indeed, the 
only relevance of motivation to learning that I can see is that it puts 
us in situations where appropriate demonstrations are particu- 
larly likely to occur, and that learning will certainly not take place if 
there is motivation not to learn. 
Closer to the truth is that we learn when we expect to learn, when 
learning is taken for granted. But a conscious expectation is not 
precisely what is required. Infants may take learning to talk for 
granted, but not in the sense of consciously expecting it. Rather, 
what seems to make the difference is absence of the expectation 
that learning will not take place. 
This is how I propose to define sensitivity, the third constituent 
of every learning situation: the absence of expectation that learning 
will not take place, or that it will be difficult. Where does sensitivity 
come from? Every child is born with it. Children don’t need to be 
taught that they can learn; they have this implicit expectation that 
they exhibit in their earliest learning about language and about the 
world — they believe they are omnipotent. Experience teaches chil- 
dren that they have limitations, and unfortunately, experience of- 
ten teaches this unnecessarily. 
WTiy is learning to walk usually so much easier than learning to 
swim? Walking must surely be the more difficult accomplishment. 
Infants have minimal motor coordination and on two tottering feet 
must struggle against gravity. Little wonder walking takes several 
months to master. Swimming, on the other hand, can be learned in 
12. LEARNING ABOUT THE WORLD 
207 
a weekend — if it is learned at all. It is learned when the learner has 
much better motor coordination and is in a supportive element — 
water. And it must be as “natural” as walking. So why the differ- 
ence? Could it be that difficulty and failure are so often anticipated 
with swimming and not with walking? 
Why is learning to talk generally so easy while learning to read is 
sometimes so much harder? The answer can’t be the intrinsic diffi- 
culty of reading. Infants learning to talk start with essentially noth- 
ing; they must make sense of it all for themselves. Despite the 
remarkable speed with which they are usually credited with learn- 
ing about language, it still takes them several years to show any- 
thing approaching mastery. Reading should be learned more 
quickly, as it has so much language understanding to support it. 
And when children do learn to read, whether they learn at 3 years 
of age, or 6, or 10, they learn — in the observation of many teach- 
ers — in a matter of a few weeks. The instruction may last for years, 
but the learning is accomplished in weeks. What is the difference? I 
can only think that with reading an expectation of failure is fre- 
quently communicated to the child. 
The apparent “difficulty" of some learning can’t be explained 
away on the basis of age. Teenagers are expected to learn to drive 
cars — surely as complicated a matter as learning to swim, if not to 
read — and lo, they learn to drive cars. In fact, for anything that in- 
terests us, where the learning is taken for granted, we continue to 
learn throughout our lives. We don’t even realize we are learning, as 
we keep up to date with our knowledge of language, music, astron- 
omy, automotive engineering, spelling, world affairs, video games, 
the television world, or whatever — for the “kind of person” we hap- 
pen to be. 
Engagement takes place in the presence of appropriate demon- 
strations whenever we are sensitive to learning, and sensitivity is 
an absence of expectation that learning will not take place. Sensi- 
tivity is obviously related to two factors I mentioned earlier re- 
garding willingness to engage in critical thinking, namely, 
disposition and authority. Individuals who don't feel competent to 
think critically on particular occasions, because of the way they 
perceive themselves or the way others perceive them, could be 
said to lack sensitivity for critical thinking. If they don’t feel it is 
appropriate or possible for them to behave in a particular way, 
they will also feel that it is inappropriate (and probably impossi- 
ble) for them to learn to behave in those ways. Lacking the dispo- 
sition and authority to learn, they will decline opportunities for 
the necessary engagement. 
208 
UNDERSTANDING READING 
Sensitivity doesn’t need to be accounted for; its absence does. 
Expectation that learning will not take place is itself learned. 
The ultimate irony is that our constant propensity to learn may 
in fact defeat learning; we can learn that particular things are not 
worth learning or are unlikely to be learned. Children are indis- 
criminate in their learning — the ticking bomb in the class- 
room — and they can learn things that they would really do much 
better not learning at all. Learning that something is useless, un- 
pleasant. difficult, or improbable may be devastatingly perma- 
nent in its effect. 
LEARNING— A SOCIAL EVENT 
So far, learning has been discussed as if it were the entire responsi- 
bility of the learner, a matter of individual effort. But this is not the 
case. Whether or not learning takes place usually depends more on 
people around learners than on the learners themselves. Personal 
effort doesn’t guarantee learning, nor does conscious motivation. 
Learners often need do nothing in order to learn. Someone else 
does something, and the learner learns — the vicarious aspect of 
learning. Neither the learner nor the people around the learner 
need know that learning is taking place. 
Family and friends are generally unaware of how much even the 
youngest children learn of spoken language, for example. Infants 
don’t practice talking, they say something — and usually they are 
right first time. They occasionally make mistakes, of course, and 
when they do, grownups regard the mistakes as cute and tell their 
friends. But most of the time parents are unaware that learning is 
taking place, until they suddenly find themselves saying, “Where 
did junior learn to say that?” Then they stop having private conver- 
sations in front of the child. Learning inconspicuously and effort- 
lessly continues into adulthood. How else would we all learn the 
meanings of the scores of thousands of words that we know, and to 
talk the way we do? Where does it all come from? 
The typical absence of evident error might seem to provide prob- 
lems for the hypothesis testing point of view adopted earlier in this 
chapter. If children test hypotheses in order to learn, they must get 
the hypotheses right most of the time. And right or wrong, where do 
the hypotheses come from? 
The answer to all of these questions must be — from other peo- 
ple. Much of what children (and adults) learn, they learn when they 
are interested in something someone else is doing. They learn as if 
they were doing it themselves. 
12. LEARNING ABOUT THE WORLD 
209 
Learning From Other People 
George Miller (1977) recognized the importance of other people in 
the title of his book Spontaneous Apprentices: Children and Lan- 
guage. He argued that infants learn to talk and to understand 
speech by apprenticing themselves to adults or to more competent 
children. And they learn to talk in exactly the same way as the peo- 
ple they apprentice themselves to. Children don’t even learn to talk 
like the people they hear talking most. (Once they get to school, 
children hear their teachers talking more than anyone else, but 
they don’t grow up talking like teachers — unless they are going to 
become teachers themselves.) 
No modeling is involved. This is not a matter of infants saying, “I 
want to be like that person,” and studying and practicing the other 
person’s behavior. Instead, the child seems effortlessly to learn 
what the other person does. The other person is an unwitting sur- 
rogate for the child’s learning. If this is trial-and-error learning, 
other people conduct the trial, and because they can already do 
what they are doing, there are very few errors. 
A baby babbles, someone else puts the utterance into conven- 
tional language — “You want a drink of juice?” — and the child has 
learned something about drinks of juice, without practice, without 
error. One adult says to another, “Pass the salt” — it could be babble 
as far as the infant is concerned — but the adult behavior allows the 
infant to hypothesize the meaning of the utterance. If the salt is 
passed, the child has learned by hypothesis testing, without error 
and without anyone knowing that learning has taken place. 
We learn when we comprehend. (A struggle to learn is always a 
struggle to comprehend.) Other people help us to learn by helping 
us to understand. That is essentially the social nature of learning, 
even when we are learning from books, when it is the author's re- 
sponsibility to facilitate the reader’s comprehension. 
Joining the Spoken Language Club 
An alternative metaphor for explaining how infants learn about 
language (and everything else) is that they join a club (Smith. 
1988). Infants join communities of people they see themselves as 
being like, who accept the infants as being like them, and the in- 
fants learn to be exactly like other members of the club. They learn 
not to be like members of the clubs they don’t belong to. 
A spoken-language club is probably the first club most infants 
join, but it has exactly the same advantages as any other club 
210 
UNDERSTANDING READING 
they might join later in life. First, more experienced members 
disclose the nature of the club's activities. These are the demon- 
strations I have referred to in this chapter. In the spoken lan- 
guage club, members demonstrate to the child what spoken 
language can be used for, how it helps to fulfill intentions in a va- 
riety of ways. 
Second, when new members of the club themselves want to en- 
gage in club activities — when they want to use spoken language to 
fulfill their own intentions — more experienced members of the 
club help. They don't give newcomers instruction from which 
learning is supposed to take place, they provide collaboration. To 
be specific, other members help the infant to say what the infant is 
trying to say, and they help the infant to understand what the infant 
is trying to understand. The learner is totally involved because ev- 
erything centers on the learner’s intentions and interest — this is 
the engagement to which I referred earlier. 
Children finish up talking exactly like their friends, the other 
members of the spoken language club they eventually affiliate 
with. They learn to dress and ornament themselves exactly like 
their friends — like the kind of person they see themselves as be- 
ing. They learn the other club members' ways of perceiving the 
world, their attitudes, their values, their dislikes, their impera- 
tives. They learn a culture — not by practice or by trial and error, 
but by imperceptibly yet inevitably coming to be exactly like the 
kind of person they see themselves as being. In other words, the 
clubs they join become their identity. 
If we see ourselves as members of a club, and the club members 
don’t exclude us, then we can t help becoming like the other mem- 
bers because of the demonstrations and collaboration we receive. 
But if we are rejected by a club, or if we decide to exclude our- 
selves, then we not only fail to become like the club members, we 
often become as different from them as we can be. We lose our 
sensitivity — and it is usually almost impossible to get it back. It is 
as if we don’t want to be mistaken for members of the club — ex- 
cept that none of the learning or failure to learn is under con- 
scious control. Everyone fails, in one way or another, to become 
members of clubs of people who have mastered things like statis- 
tics, automobile engines, computer programming, algebra, iden- 
tifying constellations — or reading or writing. This has nothing to 
do with motivation or effort — the most conspicuous things most 
of us have failed to learn are often things we have been most moti- 
vated to learn, and that we have spent the most "time on task” try- 
ing to learn. 
12. LEARNING ABOUT THE WORLD 
211 
ISSUES 
Despite all the everyday evidence to the contrary, it is still com- 
monly taken for granted that learning is the result of frequency and 
intensity of effort. Many politicians and media authorities — and 
even some self-styled researchers — appear convinced that failure 
to learn indicates a lack of trying on the part of students and their 
teachers. I don't know where they learned this. 
SUMMARY 
Most of what individuals know about language and the world is not 
formally taught. Instead, children develop their theory of the world 
and competence in language by testing hypotheses, experimenting 
in meaningful and purposeful ways with tentative modifications of 
what they know already. Thus the basis of learning is comprehen- 
sion. Children learn continuously, through engagement in dem- 
onstrations that make sense to them, whenever their natural 
sensitivity for learning is undamaged. Learning is a social activity. 
Children learn from what other people do and help them to do. 
Notes to chapter 12 begin on page 299 covering: 
Language learning 
Vocabulary 
Motivation 
r\ 
I Learning About 
Written Language 
The implication of this present chapter can be summed up in very 
few words. The primary role of reading teachers is to ensure that 
children have adequate demonstrations of written language being 
used for meaningful purposes and to help children to fulfill such 
purposes themselves. Where children see little relevance in read- 
ing, then teachers must show that reading is worthwhile. Where 
children find little interest in reading, then teachers must create 
interesting situations. No one ever taught reading to a child who 
wasn't interested in reading, and interest can't be demanded. 
Teachers must themselves be conspicuous users of written lan- 
guage. What applies to children applies also to older students, 
and to adults. 
This book is primarily about reading, and this chapter is primar- 
ily about learning to read. But nothing a child learns about read- 
ing — whether about letters, words, or meaning — will make any 
sense unless the child has an understanding of what can be done 
with written language. Hence the title of this chapter. Every learner 
must know — and be able to trust — that written language can be used 
for worthwhile purposes, that it is not meaningless marks. 
212 
1 3. LEARNING ABOUT WRITTEN LANGUAGE 
213 
LEARNING TO READ BY READING 
Learning to read doesn't require the memorization of letter names, 
or phonic generalizations, or a large vocabulary, all of which are 
taken care of in the course of learning to read, and little of which will 
make sense to a child without experience of reading. Nor is learning 
to read a matter of application to all manner of exercises and drills, 
which can only distract and even discourage a child from the busi- 
ness of learning to read. And finally, learning to read is not a matter 
of relying on instruction, because the underlying skills of read- 
ing — namely, the efficient uses of nonvisual information — can’t be 
explicitly taught. But they can be learned from experience. 
Learning to read is like the cat-and-dog problem. No one can 
teach explicitly the relevant categories, features, and interrelation- 
ships that are involved. Yet children are perfectly capable of solving 
the problems for themselves provided they have the opportunities 
to generate and test their own hypotheses. Learning to read is espe- 
cially like learning spoken language. No one can even begin to ex- 
plain to infants what essential features and conventions of speech 
should be learned, let alone construct a course of study for infants 
to follow; yet even this complex problem is solved by children, 
without any apparent strain or difficulty, provided again that they 
have the opportunity to exercise their innate learning ability and 
are helped to use and understand speech. All that children require 
to master spoken language, both to produce it themselves and 
more fundamentally to comprehend its use by others, is to experi- 
ence language being used in meaningful settings. Children easily 
learn about spoken language when they are involved in its use, 
when it has the possibility of making sense to them. And in the 
same way children will try to understand written language by being 
involved in its use, in situations where it makes sense to them and 
they can generate and test hypotheses. 
No infallible method of instruction will ever be found to direct a 
child’s progress in learning to read. But it is not possible to chart 
the precise course of a child's learning spoken language either (or 
learning the difference between cats and dogs). It is possible to 
specify the conditions under which children will learn to read, and 
these are again the general conditions that are required for learn- 
ing anything — the opportunity to generate and test hypotheses 
(naturally and unconsciously) in a meaningful context. And to reit- 
erate the constant theme, the only way a child can do all this for 
reading is to read. If the question arises of how children can be ex- 
pected to learn to read by reading before they have learned to read. 
214 
UNDERSTANDING READING 
the answer is very simple. At the beginning — and at any other time 
when it is necessary — the reading has to be done for them. Before 
children acquire competence in reading, everything will have to be 
read to them, but as their ability expands they will need only occa- 
sional help. 
One of the beautiful things about written language that makes 
sense is that it increasingly provides crucial assistance to learners. 
Authors can take over teaching children to read. Meaningful writ- 
ten language, like meaningful speech, not only provides its own 
clues to meaning, so that children can generate appropriate learn- 
ing hypotheses, but it also provides the opportunity for tests. If a 
beginning reader is not sure of a likely meaning, the context (before 
and after) can provide clues. And the subsequent context will indi- 
cate whether the child's hypotheses were right or wrong. Reading 
text that makes sense is like riding a bicycle: children don't need to 
be told when they are losing control. 
Let me list the advantages a child gains from reading meaningful 
texts: building vocabulary, understanding the possibilities and 
limitations of letter-sound relationships, developing mediated 
word and meaning identification ability, acquiring speed, avoiding 
tunnel vision, preventing memory overload, relying on sense, ac- 
quiring familiarity with such conventions as the appropriate dis- 
course structure, grammar, and register — in short, increasing 
relevant nonvisual information and gaining experience in using it 
more efficiently. And always the child will be the best guide for 
learning in the most efficient manner, because children will not 
willingly limit their vision, overload memory, or tolerate nonsense. 
Children also will not tolerate not learning, so there is no reason to 
expect that they will be satisfied with what has become simple and 
routine for them. 
It is also easy to list the conditions required for children to take 
advantage of the learning opportunities that reading meaningful 
text provides. There are only four: plentiful access to comprehensi- 
ble and interesting reading material, assistance where needed ( and 
only to the extent that it is required), willingness to take the neces- 
sary risks (anxiety increases the proportion of visual information a 
reader needs), and freedom to make mistakes. 
I have said little about motivation because it is not something that 
can be artificially promoted or maintained, certainly not by means 
of extrinsic “reinforcers” such as irrelevant material rewards, im- 
proved grades, or even extravagant praise. None of these is neces- 
sary for a child to learn spoken language. All the satisfaction that a 
child requires is in the learning itself, in the utility and understand- 
1 3. LEARNING ABOUT WRITTEN LANGUAGE 
215 
ing that result. And the impetus in the first place? Why do children 
set themselves the enormously time-consuming task of learning 
spoken language? Not, I think, in order to communicate; children 
can't understand this use of language until they have mastered 
some. And certainly not to get their material needs fulfilled or to 
control the behavior of others. Children are never so well looked af- 
ter as before they can use language; afterward they can be told to 
wait, to do without, or to do it themselves. I think there can be only 
one reason why children apply themselves to learning spoken lan- 
guage — because it is there, an interesting and functioning part of the 
world around them. They learn when its sense, its utility, and its 
meaningfulness are demonstrated to them. And because language is 
meaningful, because it changes the world and is not arbitrary or ca- 
pricious, not only do children succeed in learning it, but they want 
to learn it. Children will learn anything that is meaningful to them, 
unless the learning becomes too difficult or too costly for them, in 
which case the learning itself becomes meaningless. The child’s sen- 
sitivity for reading is destroyed. 
Children will endeavor to understand and engage in anything 
they see adults doing, provided the adults demonstrate enjoyment 
and satisfaction in doing it. If meaningful written language exists in 
the child’s world and is conspicuously used with satisfaction, then 
the child will strive to join the club; that is in the nature of child- 
hood. There is no need for special explanations about why children 
should want to learn to read, only for why they might come to the 
conclusion that reading is pointless or too costly. 
TWO CRITICAL INSIGHTS 
There are two special insights that children must have in order to 
learn to read. These insights are fundamental, in the sense that 
children who don’t have them are bound to find reading instruc- 
tion nonsensical and won’t therefore succeed in learning to read. 
Yet not only are these insights not taught in school, but much of 
what constitutes formal reading instruction might be seen as con- 
trary to these insights, and thus likely to inhibit them. The insights 
are, first, that print is meaningful, and second, that written lan- 
guage is not the same as speech. 
Insight 1 : Print Is Meaningful 
There is no need to belabor why the insight that print is meaningful 
is an essential precondition for learning to read. Reading is a mat- 
216 
UNDERSTANDING READING 
ter of making sense of print, and meaningfulness is the basis of 
learning. For as long as children see no sense in print, for as long as 
they regard it as arbitrary or nonsensical, they will find no reason 
to attend to print. They won't learn by trying to relate letters to 
sounds. Written language doesn’t work in that way. and it is not 
something that can make any sense to children. 
Research has offered abundant evidence that children may be as 
much immersed in written language as they are in speech, and they 
respond to it with similar intelligence. I am not referring to school 
nor to those overrated books that are supposed to surround and 
somehow inspire some privileged children to literacy. I refer in- 
stead to the wealth of situation-dependent print to be found on ev- 
ery product in the bathroom, on every jar and package in the 
kitchen, in the television guide (and in commercials on television 
and the web), in comics, catalogs, advertising fliers, telephone di- 
rectories, on street signs, storefronts, gas stations, billboards, at 
fast-food outlets, supermarkets, and department stores. All of this 
print is meaningful; it makes a difference. We no more predict ce- 
real in a package labeled detergent than we expect candy in a store 
advertising dry cleaning or a concert in a television program an- 
nounced as football. 
For those not blind to it (and experienced readers often are un- 
seeing in this way) our visual world is an ocean of print, most of it 
(check your supermarket) literally in front of our eyes. Even chil- 
dren who can’t yet read pay attention to this ambient print. I have 
told of a 3V2-year-old boy who obviously couldn't read the words 
luggage and footwear on signs in a department store (because he 
got both of them wrong) but who nevertheless asserted that the 
first said “cases" and the second said “shoes ” (Smith, 1976). Here 
was one child who could bring meaning to print long before he 
could read the actual print, and who therefore had acquired the in- 
sight that differences in print are meaningful. 
There is only one way in which such an insight might be 
achieved, and that is when a child observes print being responded 
to in a meaningful way. At this point, I am not referring to the read- 
ing of books or stories, but to the occasions when a child is told, 
“That sign says, ‘Stop, The word on that door is Boys,'” or “This 
is the jar for cookies.” Television commercials may do the same for 
a child — they not only announce the product’s name, desirability, 
and uniqueness in spoken and written language, but they even 
demonstrate the product at work. And just as with the spoken lan- 
guage of the home, there is a great deal a child might learn from this 
situation-dependent written language by hypothesizing a likely 
13. LEARNING ABOUT WRITTEN LANGUAGE 
217 
meaning and seeing if the hypothesis is confirmed. Children can 
test hypotheses about the meaning of the printed word toys in a 
mall, not because anyone reads it to them, but by ascertaining 
whether the sign does in fact indicate the location of the toy store. 
There is a consistency between the print and its environment. The 
print that normally surrounds children is potentially meaningful, 
and thus provides an effective basis for learning. 
There may be very little meaningful print in school, in the sense 
that it would not be possible to substitute one word for another. A 
teacher writes the words table or chair on the board but could just 
as well write horse or cow. The words in word lists, or the sen- 
tences in many “stories,” could be changed without any child notic- 
ing anything out of place. Teachers may believe there are good 
reasons for a particular exercise or element of instruction, but if 
children can’t see the sense of the enterprise, then it can reason- 
ably be regarded as incomprehensible. A brief list of fundamen- 
tally incomprehensible aspects of reading instruction to which 
children may be exposed would include: 
1 . The decomposition of spoken words to “sounds.” The spo- 
ken word cat, in some contexts, can make sense, but the sounds 
/kuh/, /a/, /tuh/ never do. 
2. The decomposition of written words to letters. The printed 
word cat, in some contexts, can make sense — when it refers to a 
real or imaginary animal with which children can meaningfully 
interact. But the letters c, a, and t are arbitrary visual symbols 
that have nothing to do with anything else in the child’s life. 
3. The relating of letters to sounds. For a child who has no 
idea of reading to be told that some peculiar shapes called let- 
ters — which have no apparent function in the real world — are re- 
lated to sounds that have no independent existence in the real 
world must be jabberwocky. 
4. Meaningless drills and exercises. There are so many candi- 
dates for this category, ranging from deciding which of three 
ducks is facing the wrong way to underlining silent letters in 
words, that I won’t attempt to make a list. Children may learn to 
score high on repetitive and nonsensical tasks (especially if they 
happen to be competent readers), but such a specialized ability 
won’t make readers of them. 
The preceding kinds of activity may. through their very incom- 
prehensibility, make learning to read more complicated, arduous, 
and nonsensical than it need be. It is not until children have begun 
218 
UNDERSTANDING READING 
reading that they have a chance of making sense of such activities 
at all. Children who lack the insight that written language should 
make sense may never achieve it, and children who have the insight 
may be persuaded they are wrong. 
Insight 2: Written Language Is Different From Speech 
The first insight was concerned primarily with written language in 
the form of single words (or small groups of words) like labels and 
signs. These kinds of print function very much like the everyday 
situation-dependent spoken language outside school in that cues 
to meaning (and constraints on interpretation) are provided 
largely by the physical situation in which they occur. Now I want to 
consider context-dependent written language, where constraints 
on substitutability and interpretation are placed not by the physi- 
cal environment but by the syntax and semantics of the text itself. 
As I discussed in chapter 3, the conventions of written and spoken 
language are evidently not the same, and probably for very good 
reason, including the fact that written language has become espe- 
cially adapted for being read. 
Children who expect written language to be exactly the same as 
speech are likely to have difficulty in predicting and comprehend- 
ing its conventions and thus in learning to read. They must be fa- 
miliar with how written language works. Immersion in functional 
language, the possibility of making sense, a plentiful experience, 
and the opportunity to test hypotheses would seem to be just as 
easily met with written language as with speech. In fact, written lan- 
guage might seem to have several advantages, because a number of 
tests can be conducted on the same piece of material, and a second 
hypothesis tried if the first fails. By virtue of its internal consis- 
tency, the text itself can provide relevant feedback about the cor- 
rectness of hypotheses. 
How might children who can’t yet read acquire and develop the 
insight that speech and written language are not the same? Only by 
being read to, or at least by hearing written language read aloud. 
The kind of reading that would most familiarize children with writ- 
ten language is coherent stories, ranging from items in newspapers 
and magazines to traditional fairy tales, ghost and adventure sto- 
ries, history, and myth. All of these types of story are truly written 
language — produced for a purpose in a conventional medium and 
distinguishable from most school texts by their length, sense, and 
semantic and syntactic richness. There is no evidence that it is any 
harder for children to understand complex texts (when they are 
1 3. LEARNING ABOUT WRITTEN LANGUAGE 
219 
read to them or when they can explore them for themselves) than It 
is difficult for children to understand the complex adult speech 
that they hear around them and on television. 
Children at school may not be provided with complex written 
material as part of their reading instruction for the obvious reason 
that they couldn't be expected to read it by themselves. Because 
material in which children are likely to be interested — and from 
which they would be likely to learn — may be too difficult for them 
to read by themselves, less complex material is found or produced 
in the expectation that children will find it “simpler.” And when 
these specially tailored-for-children texts also seem to confound 
beginners, the assumption may be made that the fault lies with the 
children or with their “language development.” 
And indeed, it may be the case that the language of such texts is 
unfamiliar to many children. But this inadequacy need not have its 
roots in the particular kind of spoken language with which the 
child is familiar nor even in the possibly limited experience of the 
child with print. The reason is more likely to be associated with the 
child’s unfamiliarity with the artificial language of school books, 
whether of the truncated “Sam the cat sat on the mat” variety or the 
more florid “Down the hill, hand in hand, skipped Susie and her 
friend." This is also so different from any other form of language, 
spoken or written, that it is probably safest to put it into an exclu- 
sive category of “school language.” 
Such material tends naturally to be unpredictable for many 
children, who consequently have enormous difficulty under- 
standing it and learning to read from it. And ironically, it may be 
concluded that written language is intrinsically difficult for chil- 
dren, who would be better off learning from “spoken language 
written down.” The text is then based on the intuition of a text- 
book writer or classroom teacher about what constitutes spoken 
language — or more complex still, a dialect of that language or 
even children’s language. All of these are problems that would 
confound a professional linguist. The result is quite unlike writ- 
ten language yet has none of the advantages of speech, because it 
will have to be comprehended out of context. Children may learn 
to recite such print, but there is no evidence that it will make them 
readers. Any insight they might have in advance about the nature 
of written language is likely to be undermined, and worse, they 
might become persuaded that the print that they first experience 
in school is a model for all the written language that they will meet 
throughout their lives — a conviction that would be as discourag- 
ing as it is misleading. 
220 
UNDERSTANDING READING 
ON INSTRUCTIONAL METHODS 
I have not said anything about the best (or the worst) programs, 
methods, or materials for teaching children to read. This was in- 
tentional, because the conclusion to which all my analysis, re- 
search. and experience with teachers and with children in schools 
has led is that children don’t learn to read from programs. In par- 
ticular, they can’t learn from the more structured, systematic 
“reading skills” programs where every supposed learning step is 
predetermined for the child; they can’t acquire or maintain the two 
basic insights just discussed. Only people — and written language 
itself — can demonstrate how written language is used. Programs 
can’t anticipate what a child will want to do or know at a particular 
time. They can’t provide opportunities for engagement. And any- 
thing a program teaches that is irrelevant to a child will be learned, 
if it is learned at all, as something that is irrelevant. 
No “method” of teaching will take care of all the contingencies. 
Nor should the development of a foolproof method be expected, 
despite the billions of dollars that have been spent in its pursuit 
and exploitation. Although some methods of teaching reading are 
obviously worse than others (because they are based on very weak 
theories of the nature of reading), the belief that one perfect 
method might exist to teach all children is contrary to all the evi- 
dence about the multiplicity of individual differences that every 
child brings to reading. 
Research is of little help in the selection of appropriate methods. 
Research tells us that all methods of teaching reading appear to 
work for some children in contrived circumstances but that none 
works for all. Some teachers seem to succeed whatever the method 
they are formally believed to employ. We must conclude that the in- 
structional method is not the critical issue. (Researchers recognize 
this point and have to control in their studies for the “variability" 
introduced first by the differing experience of children and second 
by the varying influence of teachers.) It might not be particularly 
unfair to say that many children learn to read — and many teachers 
succeed in helping them — despite the instructional method used. 
The analysis I have made can’t be translated into a system for 
teaching, although it can indeed be translated into an environment 
for learning. In fact, the analysis explains environments in which 
children do learn to read, whether or not there is a program that is 
supposed to be teaching the child to read at the same time. These 
are environments in which at least some written language makes 
sense, and in which an autonomous teacher has a critical role. 
1 3. LEARNING ABOUT WRITTEN LANGUAGE 
221 
Whatever the setting, books and other interesting reading materi- 
als must play a large part in it. Far more learners become readers 
in libraries than in experimental laboratories. 
The Never-Ending Debate 
TWo basic theories of how children learn to read have been con- 
trasted in these pages. One point of view is sometimes termed the 
“natural approach,” “meaning approach,” “psycholinguistic,” “real 
books,” or "whole language.” The opposing point of view, which re- 
lies heavily on phonics and other exercises and drills, is generally 
called the “skills approach,” “mastery,” or “direct instruction.” Be- 
tween the two sides is a gulf. 
Many reading theorists and researchers have taken strong posi- 
tions on one side of this gulf or the other, perpetuating what Jeanne 
Chall (1967) nearly 40 years ago called “The Great Debate.” The 
debate is still unresolved. In Smith (1992) I referred to it as 
“never-ending,” and in the notes, beginning on page 316, 1 discuss 
why “the interminable controversy” is unlikely to end — and why it 
engenders such intense feeling. 
The Role of Computers 
For anyone who believes there are basic skills that children must 
master in order to become readers and writers, and that repeti- 
tious exercises, correction, tests, and grading are essential for 
learning those skills, computers constitute an ideal educational 
technology. With sophisticated graphics, tightly controlled instruc- 
tional sequences and loops, constant testing, immediate feedback 
of results, and ability to document and compare every score, com- 
puter-based literacy programs offer systematic instruction in a 
form that can appeal to everyone. Children enjoy them because 
they make “fun” out of previously tedious ritual, like television car- 
toons. Parents like the computer programs, because the technol- 
ogy is labeled “educational.” Teachers may like them because they 
plot a path for every student and keep them on track. And adminis- 
trators can find the success and control that computer-assisted in- 
struction promises irresistible. 
Such computer programs must be questioned about what they 
demonstrate to learners about literacy. Yet major publishers of 
reading and writing programs, from kindergarten through high 
school, now invest in elaborate software programs. The 
worksheet activities and tests contained in their print materials 
222 
UNDERSTANDING READING 
are now presented on computer screens in endless profusion and 
variety. And the same claims of instructional efficacy are made. 
There is no evidence that such computer programs have suc- 
ceeded in making children literate, and there are no convincing 
theories that they could succeed. Such programs could rapidly 
give children a totally false idea of the purposes and possibilities 
of literacy (Smith, 1986). 
This doesn’t mean that computers have no place in the literacy 
classroom. As word processors, computers have helped the youn- 
gest children to become writers, by assisting them in the physical 
act of writing and also in such writerly activities as drafting, edit- 
ing, and preparing clean and legible copies of their texts. And when 
used in these and a variety of other practical ways — in simulations, 
games, design activities, communication links, drama, art. and 
music — computers seem able to stimulate children to talk more, 
plan more, think more, write more, and read more. The issue is 
not whether computers should be in classrooms, but how they 
should be used. 
TEACHING READING 
As I said at the beginning of this chapter, teachers must demonstrate 
to their students that reading is worthwhile, and create interesting 
reading opportunities. They must themselves be conspicuous users 
of written language. 
Where children have difficulty in reading, teachers must see that 
they are helped to read what they would like to read. In part, this 
assistance can be given by developing the confidence of children to 
read for themselves, in their own way, taking the risk of making 
mistakes and being willing to ignore the completely incomprehen- 
sible. Even bizarre personal interpretations are better than none at 
all; children find out soon enough the mistakes that make a differ- 
ence. But children will also from time to time look for help from 
others, either in answering specific questions or in assisting with 
reading generally. Such reading on behalf of the child can be pro- 
vided by the teacher, an aide, or by other children. 
The Learner's Point of View 
Children themselves must judge whether materials and activities 
are too difficult or too dull. Anything children would not listen to or 
understand if it were read to them is unsuitable material for them 
to be expected to read. A child’s preference is a far better yardstick 
1 3. LEARNING ABOUT WRITTEN LANGUAGE 
223 
than any readability formula, and grade levels have no reality in a 
child’s mind. Teachers need not be afraid that children will engage 
in reading so easy that there is nothing to learn; that would be bor- 
ing. Children learn about reading as long as they read, but they can 
never learn about reading by not reading. 
There is no simple formula to ensure that reading will be com- 
prehensible; no materials or procedures are guaranteed never to 
interfere with a child's progress. Instead, teachers must under- 
stand the factors that make reading difficult, whether induced by 
the child, the teacher, or the task. Examples include the concentra- 
tion on visual detail that will cause tunnel vision; the overloading of 
short-term memory by attention to fragments of text that make lit- 
tle sense; logjams in long-term memory as a child strives to be 
ready to answer questions afterward or to write “reports”; at- 
tempts to sound out words at the expense of meaning; slow read- 
ing; anxiety not to make a mistake; lack of assistance when a child 
needs it for sense or even word identification; or too-insistent “cor- 
rection” that may be irrelevant to the child and that may in the long 
run inhibit the self-correction that is an essential part of learning. 
All of these ways in which reading is made harder can be character- 
ized as limitations on the extent to which children can use 
nonvisual information. 
And conversely, what makes reading comprehensible for chil- 
dren is the teacher’s facilitation of the use of nonvisual informa- 
tion. Not only should a child come into every reading situation with 
relevant nonvisual information — in plain English, with adequate 
prior understanding — but the child must also feel free to use it. A 
child’s fund of knowledge and confidence should be constantly de- 
veloped, but this will occur as a consequence of reading. Not only 
should the teacher try to avoid materials or activities that are non- 
sense to the child, there should be active encouragement for the 
child to predict, to understand, to enjoy. The worst habit for any 
learner is to treat text as if there were no sense to be found in it. 
Where there is a mismatch, where there is little likelihood that a 
child will comprehend the material, then the preference should be 
to change the material rather than to try to change the child. 
For older students, teachers may be reluctant to change mate- 
rial because a certain content is expected to be learned, but they 
still have a choice. Students can’t learn two things at the same 
time; they can’t simultaneously learn to read and to master an un- 
familiar subject matter like history or math. If the teacher's inten- 
tion is to improve reading, then students must have material they 
can easily understand. If the intention is to extend subject matter 
224 
UNDERSTANDING READING 
knowledge — which will in turn make reading easier — then until 
the student can read it with some fluency the subject matter must 
be taught in some other way, by lecture, film, board work, or indi- 
vidual tuition. The two can be taught concurrently — the math 
need not wait for the reading competence any more than the read- 
ing need wait for the math skills — but they can't be learned simul- 
taneously. 
Teaching the Hard Way 
Teachers sometimes try to resolve problems the hard way — for ex- 
ample, in expecting poor readers to improve while they are doing 
less reading than better readers. When children have trouble un- 
derstanding text they may be given isolated word drills, while prob- 
lems with word identification may provoke attention to letter 
identification and sound blends. Actual reading may be postponed 
in favor of phonemic awareness exercises. But letters (and their 
phonic interrelations) are recognized and learned best when they 
are parts of words, and words are recognized and learned more 
easily when they are in meaningful sequences. Good readers tend 
to be good at letter and word identification and at phonic drills, but 
these more specific skills are a consequence, not a cause, of good 
reading (Samuels, 1971). Good readers tend also to understand 
the technical jargon of reading such as letter, word, verb, sen- 
tence, paragraph, but this again is a result of being able to read. 
Practice with definitions doesn’t make readers (Downing & Oliver, 
1973/1974). Knowledge of specialized words is necessary only if 
they are made a focal part of the instruction, if it is necessary for 
children to understand the words in order to be allowed to get on 
with the business of learning to read and write. It is not an essential 
prerequisite for literacy. 
Children may also be confounded by instruction that is as un- 
necessary as it is futile, often as a consequence of a theoretical 
vogue among specialists. When, for example. Noam Chomsky pop- 
ularized transformational linguistics as a technical method of ana- 
lyzing language, many people thought children would not learn to 
read unless they became miniature linguists themselves and chil- 
dren were made to spend a lot of time doing exercises in transform- 
ational grammar that made no apparent difference to their 
language ability. After psychologists became interested in the theo- 
retical notion of distinctive features, there were several efforts to 
teach children the distinctive features of letters, although no one 
could convincingly demonstrate what these features might be. 
13. LEARNING ABOUT WRITTEN LANGUAGE 
225 
Children who had difficulty with the alphabet or with these exer- 
cises were sometimes diagnosed as having poor feature discrimi- 
nation, although they had no reported difficulty with knives and 
forks or dogs and cats. Phonic-based reading programs and mate- 
rials have flourished whenever linguists have become particularly 
interested in the spelling-sound correspondences of language, and 
there have been moves toward teaching “prediction” as if it were 
something foreign to most children’s experience. “Phonological 
awareness” is urged as a necessity for learning to read, with the re- 
sult that children may spend more time attempting to deconstruct 
speech than exploring written language. In all of these cases, con- 
cepts that scientists have found useful as hypothetical constructs 
in their attempts to understand their discipline have become 
something a child must learn as a prerequisite for learning to read. 
(How children learned to read before these concepts were devised 
is not explained. ) There is a growing acknowledgment today of the 
importance of comprehension as the basis of learning, but at the 
same time there is a feeling that comprehension itself must be 
taught, that it can be broken down into a series of “comprehension 
skills” that presumably can be learned without comprehension. 
I am not saying that it is not useful for children to know the al- 
phabet, to build up sight vocabularies, or even to understand the 
relationships between the spelling of words and their sounds and 
(more importantly) their meaning. But all of these are by-products 
of reading that make more sense as reading itself is mastered 
and understood. It is pointless for teacher and child to labor over 
activities that won’t facilitate learning to read and that will become 
easy once reading experience develops. 
It is certainly not the case that teachers should never correct or 
make suggestions. But correction or advice may be offered too 
soon. A child pauses while reading aloud and half the class shouts 
out the next word, although the reader may be thinking about 
something else six words behind or ahead. Problems arise when 
corrections and explanations sap children’s confidence or stop 
them in their tracks for what might be quite extraneous reasons. 
The teacher should always ask, “What is causing the confusion 
here?” Children afraid of being corrected may become afraid of 
speaking, reading, and writing. 
The First Steps 
Children begin to read with the first written word they are able to 
recognize. Nonvisual information is so important that reading 
226 
UNDERSTANDING READING 
potential is enhanced with every expansion of a child’s knowl- 
edge of the world or of spoken language. (But there is no particu- 
lar need for extensive prior knowledge of the world or of spoken 
language for a child to begin to read, just enough to make sense 
of the first print that will be read. Much of the knowledge and 
language skills of fluent readers is again a consequence of liter- 
acy rather than its cause.) 
There is no “best age” for learning to read. Many children have 
learned to read, often spontaneously, as young as 3 years of age 
(Clark, 1976), and it is equally well documented that adult illiter- 
ates can learn in a few months provided their sensitivity has not 
been blunted by years of failure with formal reading instruction 
(Freire, 1972). Many of the early readers who have been studied 
did not have above-average intellect or any particular social or 
cultural privileges; they simply found reading something that was 
useful and straightforward to learn, usually without any particu- 
lar consciousness of what they were doing. Whether a child will 
learn to read is not something that can be determined with refer- 
ence to a calendar or the learner's “mental age." 
Similarly, there is no unique mental condition of "reading 
readiness.” Children are ready to learn to read whenever they 
have a purpose and intelligible opportunity for reading, not in 
terms of settling down to a concentrated period of systematic in- 
struction but in an explorer’s interest in signs and labels, in tele- 
phone directories and catalogs, and in stories. In educational 
contexts, reading readiness is often related more to the form and 
demands of instruction than to reading itself. Obviously, if in- 
struction emphasizes knowledge of letter names, then a child who 
can’t grasp the nature of the alphabet is not ready. If instruction 
requires detailed attention to the sound patterns of a particular 
dialect, then a child who can’t do this is not ready. But none of this 
has anything to do with reading itself. It is difficult to see what 
kind of special physical, emotional, intellectual, or cultural status 
is required for learning to read, except the two basic insights I 
have already discussed. 
Reading should blend smoothly into all the other visual and lin- 
guistic and intellectual enterprises of a learner’s life. There is no 
magical day when a “pre-reader" suddenly becomes a “learner." 
just as there is no landmark day when learning is completed and a 
reader graduates. No one is a perfect reader, and we all continue to 
learn every time we read. 
None of this is to say that all children will easily learn to read; 
there has always been evidence that such is unlikely to be the 
1 3. LEARNING ABOUT WRITTEN LANGUAGE 
227 
case. But failure need not be attributed to dyslexia, a disease that 
only strikes children who can’t read, and that is invariably cured 
when they can read. I have argued that there is nothing unique 
about reading, either visually or as far as language is concerned. 
There are no evident visual defects that are specific to reading, 
but this doesn’t mean that there are no general visual anomalies 
that will interfere with learning to read. The few children who 
have difficulty learning to understand speech, or learning any- 
thing, may also find learning to read difficult. 
But there is no convincing evidence that children who can see 
normally, with or without spectacles, and who have acquired a 
working competence in the language spoken around them, might 
be physically or congenitally incapable of learning to read. It can't 
be denied that some children who seem able and even bright in all 
other respects may fail to learn to read. But there can be other 
reasons for this failure that don’t presuppose any physical dys- 
function on the part of the child. Children don’t learn to read who 
don’t want to, or who see no point in doing so, or who are hostile 
to the teacher, or to the school, or to the social or cultural group to 
which they perceive the teacher and the school as belonging. 
Children don’t learn to read who expect to fail, or who believe that 
learning to read will demand too much effort or stress, or whose 
image of themselves, for whatever reason, is that of a nonreader. 
Children don’t learn to read if they have the wrong idea of the na- 
ture of reading, if they have learned — or have been taught — that 
reading doesn’t make sense. 
TESTS AND STANDARDS 
It is in the context of failure that brief reference should be made to 
the effects on young readers of the current mania for constant test- 
ing and evaluation, and especially on those having difficulty in 
making sense of the way reading is taught. 
Tests are primarily a bureaucratic tool. They are devised and 
dispensed for a variety of administrative and political reasons to 
categorize children and to evaluate teachers. But no reading test 
ever helped a child learn to read. And there is nothing in tests 
themselves to indicate why a child might not be succeeding in 
learning to read. 
Problems arise in two ways. The first is that it is widely believed 
that the content of reading tests (or of comprehension or readiness 
tests) indicates what a child needs to know in order to learn to 
read. This is a fallacy. Tests may provide indicators of what chil- 
228 
UNDERSTANDING READING 
dren are able to do as a consequence of learning to read, so that 
children who are good readers tend to do well on tests (although 
not invariably) and children who are poor readers don’t. But trying 
to teach a child to score well on individual items on a test won't 
teach a child how to read . Counting the number of times a child vol- 
untarily visits the library might be a relatively sensible test indica- 
tor of reading ability, but training the child to visit the library more 
often would not in itself improve reading ability, although it would 
raise the test score. If anything, tests measure how well children 
have been able to make sense of formal reading instruction, to 
work their way through programs. 
The reason for the close affinity of tests and programs is that 
they both treat reading in the same arbitrary and unnatural way. 
Because someone outside the classroom must determine what is 
appropriate for children to do and know (and that can be mea- 
sured), and because for reasons of control and standardization it 
is necessary to break reading down into fragmented and predeter- 
mined sequences, tests and instructional programs both tend to 
become concerned with the same superficial and isolated aspects 
that are supposed to be “components" of reading. 
Thus the first problem with tests is that they give teachers and 
children a distorted idea of the nature of reading and of what must 
be done to teach a child to read (or to satisfy some outside author- 
ity that the child is being taught to read). This is perhaps not too 
much of a handicap for a child who is indeed learning to read, who 
does reasonably well on the tests, and whose exposure to meaning- 
ful reading is not limited as a consequence. But tests can be devas- 
tating for children who don’t do well, partly because they may then 
have opportunities to read meaningfully withdrawn from them (in 
favor of exercises and drills that they have already demonstrated 
they don’t understand) and also because of the inevitable conse- 
quence for their self-esteem. The second major problem with tests 
is that they do nothing positive for the sensitivity of children who 
do badly on them. 
Teachers don’t need “off-the-shelf" tests to discover if their stu- 
dents are learning or if they are confused. Every teacher can tell (or 
should be able to tell) if a child has made progress in reading, just 
by talking with the child and looking at what the child is voluntarily 
reading. Learning to read doesn't inch along one item of informa- 
tion after another; there should be no difficulty in determining 
whether a child has progressed in ability and interest over a period 
of a few weeks. If the method of instruction is such that neither 
teacher nor child can tell whether progress is being made without 
13. LEARNING ABOUT WRITTEN LANGUAGE 
229 
recourse to a standardized test, then the instruction itself is essen- 
tially meaningless. 
The best tests are “homemade,” constructed on the spot to reas- 
sure the teacher that whatever a particular child is supposed to be 
learning at a particular time is making sense. Good teachers do 
this intuitively, and because such tests are a natural part of what- 
ever activity the child is engaged in, they are both relevant and in- 
conspicuous. 
The situations that I have characterized as making reading more 
difficult, and thus likely to interfere with children’s learning to 
read, are so much a fact of life in many classrooms that many 
teachers feel they can do little about them. Tests must be adminis- 
tered; instruction must be directed toward the tests; the language 
arts are arbitrarily and artificially fragmented; children are catego- 
rized and streamed; teachers are held accountable; certain curric- 
ula must be followed; concerns of parents and trustees must be 
assuaged; work must be graded; competition and anxiety are un- 
avoidable. Much of a teacher’s time is necessarily directed to class- 
room management, many activities are engaged in to satisfy 
“standards” or other demands laid down by external sources, and 
few teachers can find the time or the resources to provide an ideal 
learning environment for children all the time. 
A theory of reading won’t change all this (although it might 
provide ammunition for anyone who tries to resist). The kind of 
change that will make a difference in schools won’t come with 
better theories or with better materials or even with better 
informed teachers, but only with individuals taking action toward 
change. The problem of improving reading instruction, in the long 
run, is a political question. But whether teachers can change their 
world or not, they will still be better off to the extent that they 
understand about reading and about how children learn to read. 
Teachers who can’t relieve children of the disruptions of irrelevant 
demands and activities may at least protect them by pointing out 
that the activities have little to do with reading or with the child’s 
learning ability. Children understand that meaningless tasks are 
often given to them simply to keep them occupied and quiet, but 
they are not helped by being led to believe that such tasks are an 
important part of learning to read. 
THE LITERACY CLUB 
There is substantial evidence that children know a great deal about 
literacy before they come to school (Ferreiro & Teberosky, 1982; 
230 
UNDERSTANDING READING 
Goelman, Oberg, & Smith, 1984; Tolchinsky, 2003). They may not 
have learned to read and to write, but they know how literacy is 
used in the community to which they belong. If their family reads 
books and newspapers, they know about books and newspapers. 
If their family consults the television guide, they know about televi- 
sion guides. If they leave each other messages on the refrigerator 
door, they know about those too. If their friends read comics or 
consult catalogues, they know about comics and catalogues. If 
their friends use e-mail and the Internet, they probably know more 
about e-mail and the Internet than their parents do. Children know 
about signs, labels, lists, letters, greetings cards, telephone direc- 
tories, and everything else that might be part of their personal writ- 
ten language environment. They also know roughly how written 
language works, that it consists of symbols written on lines, that it 
is laid out in various conventional ways. Before they can spell, they 
know there are rules of spelling. They have ideas of why people 
read, even before they can read themselves. They pretend to read 
and to write in their games. 
There are no kits of materials or systematic exercises for teaching 
children how the world uses written language. They learn — usually 
without anyone being aware that they are learning — by participating 
in literate activities with people who use written language. It can all 
be summed up in a metaphor: Children learn about reading and 
writing by “joining the literacy club” (Smith, 1988). They are given 
demonstrations of what written language can be used for, and they 
receive collaboration when they become interested in using written 
language themselves. The assistance is usually completely casual, 
when someone points out that an approaching sign says "Stop" or 
“Burgers,” the way one might say to a child “Look, there’s a horse.” 
Someone helps them to read what they are interested in reading and 
helps them to write what they would like to write. 
Membership of the literacy club offers the same advantages as 
the spoken-language club that I discussed in chapter 12, and all 
the other clubs children might join. Children in the literacy club 
have opportunities to see what written language can do, they are 
encouraged and helped to do those things themselves, and they are 
not at risk of exclusion if they make mistakes or display a passing 
lack of interest. They learn to be like the other members of the 
club. (And if they learn from other demonstrations that reading 
and writing are boring activities or that they don't belong to the 
club, then they learn not to be like people who read and write.) 
This doesn’t mean that children are lost to literacy if they have 
not learned about reading and writing out of school. But it becomes 
1 3. LEARNING ABOUT WRITTEN LANGUAGE 
231 
all the more crucial for every child to have the opportunity to be- 
long to the literacy club in school. For children who are not inter- 
ested in reading and writing, it is even more important that 
activities in the classroom be made interesting and accessible for 
them. They need more demonstrations of worthwhile uses of liter- 
acy and more collaboration in engaging in those uses themselves. 
And in any case, children who have joined the club before they 
come to school should not then risk rejection because school defi- 
nitions of literacy or perceptions of literate activities are different. 
There is no need to fear that reading to learners, or writing for 
them, will make them passive and dependent. They won’t always 
expect other people to do their reading and writing for them. No 
child has that much patience. The moment children feel they can 
read or write well enough to do what they want to do for them- 
selves — often long before adults might think they are ready to do 
so — they reject the helping hand. It is no different when children 
learn to ride a bicycle. Children never want to be pushed when they 
can pedal away for themselves. 
Margaret Meek (1988) described how the authors of children’s 
books teach children how to read. These are the authors of the 
books that children love to read, time and time again. The children 
already know the stories before they open the books — or the sto- 
ries are so predictable that they know what is on the next page be- 
fore they turn to it. The children know the story, according to Meek, 
and the author shows them how to read it. 
Spontaneous admission into the literacy club may even explain 
how children succeed in learning to read when subjected to inten- 
sive classroom regimes of phonics and worksheet activities (which 
then get the credit for the achievement). 
Someone must do the learners’ reading for them until they are 
able to read a few things for themselves, and they are ready to learn 
to read by reading. Reading for children need not take long — only 
until they can read enough for authors to take over. Very little ac- 
tual reading ability is required for this to occur, if the right kinds of 
interesting and familiar materials are available (for details of such 
materials, ask a child). Indeed, for a child to know a story in ad- 
vance by heart may be enough to turn the child over to the author. 
What matters is for the learner to be reading known or familiar 
texts like an experienced reader. 
The role of the teacher is to support the reading and writing of all 
children until authors, and the children’s own interest and 
self-perception, ensure their continued membership in the literacy 
club. For teachers who are themselves committed readers and 
232 
UNDERSTANDING READING 
writers, the opportunity to develop new club members should be a 
pleasure as well as a privilege. 
ISSUES 
Every controversy about reading instruction inevitably comes 
down to whether classroom practices should be determined by ( 1 ) 
what teachers know and can see of individual student interest, 
comprehension, and learning, or by (2) the prescriptions of intel- 
lectual and political authorities. Associated with the formalization 
of instruction is the question of whether learning can be guaran- 
teed or even facilitated by assessment procedures that monitor 
and constrain the behavior of teachers and students, with inevita- 
ble discriminatory consequences. 
SUMMARY 
Learning to read depends on two basic insights — that written lan- 
guage is meaningful and that it is different from spoken language. 
Learners must rely on the visual information of print as little as 
possible. Teachers help children learn to read by stimulating inter- 
est and facilitating written language use to a degree that formal in- 
structional programs, with their necessarily limited objectives, 
can’t achieve. Teachers must ensure that all children receive the 
demonstrations and collaboration necessary to maintain member- 
ship in the literacy club. 
Notes to chapter 13 begin on page 302 covering: 
Learning to read 
Free voluntary reading 
Reading to children 
Literacy and schooling 
Teachers and programs 
Metalinguistic awareness 
Dyslexia and learning disabilities 
Evaluation, testing, and standards 
Is there a crisis? 
Looking ahead 
The interminable controversy 
Federally commissioned studies 
Notes 
NOTES TO PREFACE, pp. vii-xii 
Psycholinguistics and Cognitive Science 
The term psycholinguistics as used in the subtitle and other places 
in this book refers to an area of overlap between specialized fields 
of psychology and linguistics, a common ground where psycholo- 
gists (who study human behavior) and linguists (who study lan- 
guage) meet to explore the ways in which human language in its 
various forms is actually learned and used. In reading education, 
however, psycholinguistics became something of a battle cry (or 
term of opprobrium, depending on which side you were on). A 
heavily phonic approach to teaching reading has long been called 
‘the linguistic method,” and "psycholinguistic” became adopted in 
the early 1970s as the emblem of the opposing point of view (which 
has also been called the “meaning” and “whole language” approach 
to teaching reading) . 
There are two radically divergent points of view about the na- 
ture of reading. I have broadly characterized these theoretical 
perspectives as “inside-out” and “outside-in” (Smith, 1979), de- 
pending on whether the control in reading is presumed to origi- 
nate with the reader or with the text. More generally, these 
233 
234 
UNDERSTANDING READING 
opposing positions have become known as “top-down” or “bot- 
tom-up." Top-down is roughly equivalent to my inside-out, imply- 
ing that the reader determines how a text will be approached and 
interpreted. The bottom-up view is outside-in, putting the text in 
charge, with the letters on the page the first and final arbiters of 
the reader’s responses. 
Metaphors frequently betray their origins. Top-down and bot- 
tom-up are computer jargon, and are usually employed in discus- 
sions of reading by “cognitive scientists” (discussed later) who see 
the brain as some kind of computer. Inside-out and outside-in are 
terms reflecting a “constructivist" orientation (see chap. 2 notes), 
where knowledge is regarded as something generated inside the 
learner rather than imported or delivered from the outside. The 
present book, I should perhaps add, is strongly representative of 
the inside-out view. Naturally there have been attempts at compro- 
mise (or at carving out a third position), arguing for eclectic theo- 
ries that are both top-down and bottom-up at the same time 
(discussed later). But no top-downer (or inside-outer) would want 
to claim that reading doesn't involve interaction with a text. Just 
because meaning has to be constructed by the reader doesn't mean 
that any meaning will do. 
Labels can be promiscuous, and the word psycholinguists has 
had a particularly wayward career. In the late 1960s and early 
1970s the term primarily connoted academic studies into the na- 
ture of language and the manner in which infants learn to use it. fo- 
cusing on the critical role of meaning and constructive thought. In 
education, the psycholinguistic perspective implied a similar focus 
on meaningfulness in literacy learning, opposed to packaged 
“skills-based” materials and activities. But publishers quickly pro- 
duced “psycholinguistic materials” (or relabeled old materials as 
psycholinguistic). The principal successor to the psycholinguistic 
perspective in education became known as “whole language." Once 
again, this was originally a term connoting a philosophy of learn- 
ing, opposed to artificial decontextualized exercises and drills. But 
as whole language gained influence and prominence in education, 
the perspective became distorted, the theory became a method, 
and publishers began to produce “whole language materials." A 
few proponents and almost all opponents of whole language regard 
it as a method rather than a philosophy. For many politicians, jour- 
nalists, and producers of structured instructional schemes, whole 
language has become a derogatory expression. 
The term psycholinguistics has also changed its connota- 
tions. Many researchers who originally called themselves psy- 
NOTES 
235 
cholinguists employed “naturalistic” methods — they observed 
how children learned or used language in natural settings, 
rather than experimentally manipulating learning situations. 
Since the mid-1970s, however, numbers of language research- 
ers have joined forces with computer and artificial intelligence 
specialists in a subdiscipline called cognitive science. This new 
breed of psycholinguists frequently has a distinctly experimen- 
tal and prescriptive attitude to instruction. Attacks on whole 
language and other nonprescriptive approaches to reading in- 
struction are usually made in the name of cognitive science, or 
just plain “science.” 
Cognitive science is primarily concerned with abstract theories 
related to the organization of knowledge, especially through lan- 
guage, in humans and in computers. Its influence in how reading 
should be taught is growing — not, I think, because cognitive sci- 
ence has anything significantly new to say about learning and edu- 
cation, but because it is intimately tied to the dominant technology 
of the day. It also claims to be authoritative because it asserts that 
its technological methodology is the only valid one. But because of 
its insistence on particular conceptualizations and procedures, 
cognitive science is also narrow. Cognitive science looks at reading, 
and also reading instruction, from a limited “knowledge and 
skills" perspective. For an excellent insider critique of the limita- 
tions of cognitive science, see Dodwell (2000). 
Although I say that little of substance has changed in reading 
theory in recent decades, I don’t mean that new concepts and theo- 
ries don't make a difference. They do. New concepts are new ways 
of talking, and they encourage a focus on certain matters rather 
than on others. With the growth of cognitive science, for example, a 
profusion of new terms has been introduced for old concepts, such 
as “metacognition” (for “reflection” or “introspection") and “phone- 
mic awareness” (for "listening to the sounds of language”). People 
who would have no trouble understanding the old terms can be 
overly impressed and even confused by the terms that supersede 
them. An unwarranted aura of infallibility may surround the “ex- 
perts” who use the new terms with facility, especially if they claim 
special insight. 
I have retained the word psycholinguistic in the subtitle of this 
book, partly because that was the way Understanding Reading 
was titled when it first came out in 1971 , but also to indicate con- 
straints I have continued to observe. The primary concern is with 
what readers need to know and do in order to make sense of writ- 
ten language. The book doesn’t delve deeply into specific types of 
236 
UNDERSTANDING READING 
literate behavior, genre theory, literary criticism, semiotics, or so- 
cial and cultural aspects of literacy. 
Research 
Research into the nature of reading and reading instruction 
boomed in the early 1970s with the support of enormous federal 
grants aimed at the eradication of illiteracy. The goal was never 
reached, largely because (I would argue) the research and instruc- 
tional efforts were frequently predicated on fragmented and de- 
contextualized outside-in theories of reading and instruction 
(Smith, 1986). There was a second massive infusion of funds 
around the beginning of the present century, not so much to ex- 
tend the understanding of reading and learning to read as to pro- 
mote and assert a particular point of view, namely, that phonics 
and phonemic awareness are the essential basis of instruction. It 
is difficult to find large-scale research coming from a more neu- 
tral point of view, but not difficult at all to find pronouncements 
and even legislation condemning such research, where it exists, 
as being “unscientific.” All of this is discussed in chapter 13 and 
its notes. 
I have obviously not attempted in this volume to summarize all 
of the research done in the name of reading. That would be impos- 
sible. Instead, I have drawn primarily from what helps me to con- 
struct a coherent picture of reading and of learning to read. 
People with an alternative point of view would be similarly selec- 
tive in order to reject my conclusions. I make no attempt to be 
eclectic. 
One aspect of eclecticism is the view that science is an incremen- 
tal activity — that every bit of research is valid and worthwhile, add- 
ing a nugget of truth to an always-growing accumulation of 
knowledge and understanding. Such a view rests on a rosy belief 
that researchers never start from false assumptions and never fin- 
ish with untenable conclusions. All reported findings are sup- 
posed to fit together like pieces of a jigsaw puzzle. Educational 
textbooks are frequently pastiches of this nature, covering im- 
mense amounts of ground without the provision of navigable 
pathways. The alternative point of view is that scientific research 
is based on conceptual paradigms, or ways of seeing the world, 
that frequently conflict. They are subjectively adopted and emo- 
tionally retained. A paradigm is rarely abandoned by its adher- 
ents unless they find it totally worthless and have another to 
replace it with (Kuhn, 1970). The research that is done, the evi- 
NOTES 
237 
dence gathered, and the conclusions reached all depend on the re- 
searcher's beliefs and expectations. 
Another common form of eclecticism, conspicuous in educa- 
tion, attempts to assimilate alternative points of view into estab- 
lished or "official” lines of thinking. Such conceptual dilution is 
frequently found in reviews of research produced by committees 
or by bureaucratic institutions that are not keen to call attention 
to positions different from their own. There are also eclectic ap- 
proaches to teaching reading that entail using a little bit from ev- 
ery prominent theorist or instructional proponent, on the 
undiscriminating principle that every “authority” is probably a 
little bit right. Nonjudgmental approaches to instruction fail to 
recognize that inappropriate theories and methods can mislead 
the people most in need of a reliable and coherent understand- 
ing of the nature of reading, namely, children trying to learn and 
teachers trying to teach them. It is necessary to take a position. 
Some current views about reading and reading instruction must 
be wrong. 
The "Issues” sections in this book acknowledge that there are 
deeply etched and keenly felt disputes over the nature of reading 
that may never be resolved, not to the extent that proponents of 
different points of view will be satisfied. Over the years I have 
come to the conclusion that the most critical need is for teachers 
and students to be able to make their own judgments and deci- 
sions rather than to trust authorities when there are conflicting 
points of view (the choice of authority is itself a decision, of 
course). To do this, one must recognize that profound differences 
exist. Indeed, if one wants to argue that other people are wrong, 
one has to admit the likelihood that one is wrong oneself (or at 
least that someone else might also be right) and find a reason why 
it is possible for two people living in the same world (if they can be 
said to be living in the same world) to have diametrically opposing 
theories about it. 
Acknowledgment 
All my books would be a hazy shadow of themselves without the 
educational insights and editorial acumen of Mary-Theresa 
Smith. I also acknowledge debts to innumerable authors, stu- 
dents, teachers, and friends. I would be delighted to put all their 
names in this book but mortified to leave any out. They know who 
they are. 
238 
UNDERSTANDING READING 
NOTES TO CHAPTER 1, 
THE ESSENCE OF READING, pp. 1-11 
Matters of Interpretation 
Walkerdine (1982) held that language, thinking, and “context” are 
not separate systems but are jointly related to a basic human need 
to interpret all kinds of signs. Urwin (1982) similarly argued that 
learning reflects social interaction, with a semiotic emphasis on 
the importance of “signs.” 
A forgotten German philosopher Hans Vaihinger (1852-1933) ac- 
tually devised a “philosophy of as if" a century ago. He believed that 
in a frustrating and fundamentally unknowable world, people had 
to live by fictions — as if they had free will, as if ethical certainty were 
possible, as if a material world exists — disregarding uncertainties 
and logical contradictions (Vaihinger, 1952). But this pragmatic phi- 
losophy did not licence anyone to believe anything. Shifting from 
“as-if” to “this-is-how-it-must-be" was impermissible. 
A contemporary French philosopher Clement Rosset (1989) 
said that certain words like “reality" and “actual" are “untheoriz- 
able,” because they are the element in which we live. We can de- 
scribe the content of a particular reality, but not explain the 
phenomenon of reality itself. It's as far back as definition goes. "Ex- 
perience," “comprehension," and “learning” — and “reading" — are 
similarly untheorizable words. It makes no more sense to ask 
“What happens in the brain when you comprehend something?" 
than it would be to ask “What happens in the brain when you expe- 
rience reality?” 
On the point that everything is presumed to be natural until the 
opposite is taught, Fernandez-Armesto (2000), in his study of civi- 
lizations, relates stories (possibly apocryphal, but they make the 
point) of an early 18th century Huron Indian who believed that the 
stone streets of Paris were natural rock formations, and of an ab- 
original visitor from St. Kilda of the same period who thought that 
the pillars and arches of a church in Glasgow were the most beauti- 
ful caves he had ever seen (p. 14). 
Alphabets 
The Chinese scholar Lin Yutang succeeded in compiling the first 
practical English-Chinese dictionary in the 1970s only by arbi- 
trarily imposing an “alphabetical order” on 33 basic stroke for- 
mations in Chinese script. The alphabet is preeminently an 
NOTES 
239 
instrument of control, from many points of view. One of the few 
times that public opinion in China was able to modify a national 
policy of Mao Tse-tungwas when he tried to introduce an alphabet 
into Chinese writing based on European and Russian letters 
( Barlow, 1981). The peasants successfully resisted this purely ad- 
ministrative decree on the sound linguistic ground that the new 
script would prevent them from reading traditional ancient Chi- 
nese writings. Coulmas (1992) observed that no purely logo- 
graphic writing system exists. Chinese is morpho-syllabic; each 
character represents a morpheme (meaning) and also a syllable. 
There are thousands of characters, many in disuse, with proba- 
bly at least 5,000 in current use. Coulmas says that Chinese is no 
more difficult to read than English, and perhaps easier. For read- 
ers of classical Chinese it is certainly easier to read a Chinese text 
that is hundreds of years old than an alphabetic one because al- 
phabetic writing systems are tied to pronunciation, which 
changes. 
Writing Systems 
Senner (1989) gave an accessible collection of essays on the ori- 
gin of various writing systems, showing that writing was not the 
invention of one particular group of people and that it appears to 
have had its antecedents in drawing (pictographic) rather than in 
spoken language. Since the development of the alphabetic sys- 
tem, writing has often been modified to correspond more closely 
with spoken language, but at the same time, spoken language has 
always been influenced by the structure and use of written lan- 
guage. Written language tends to change less rapidly than speech 
and therefore has tended to put a brake on spoken language 
change (or to drift further apart from it). Even today, many people 
seem to feel that written language is a particularly pure form of 
language, a model for what spoken language should be like. The 
other origin of written language, pointed out by several of the con- 
tributors to Senner’s volume, is not so much conventional speech 
as mathematics — from keeping track of the calendar and of astro- 
nomical (and astrological) phenomena to bureaucratic and com- 
mercial record keeping. See also Stevenson (1983), Sampson 
(1985), and Coulmas ( 1990). In a broad-ranging review of the evo- 
lution of civilizations all over the world, Fernandez-Armesto 
(2000) noted that “many societies are seen to have confided what 
was memorable, and therefore of lasting value, to oral transmis- 
sion, and to have devised writing systems in order to record rub- 
240 
UNDERSTANDING READING 
bish; fiscal ephemera, merchants’ memoranda" (p. 29). He later 
concluded that writing systems all over the world developed “first 
as a commercial, priestly, and political tool, then as a medium for 
artistic expression ... it was a mundane contrivance developed for 
potters’ marks” (p. 217). 
NOTES TO CHAPTER 2, 
COMPREHENSION AND KNOWLEDGE, pp. 12-30 
Knowledge and Constructivism 
The view that comprehension is a constant state of mental activity 
corresponds to a philosophical and educational theory called 
constructivism, which holds that knowledge is constructed 
rather than passively received or delivered from the outside 
world. Jean Piaget in particular argued that individuals come to 
know the world through action rather than through their senses. 
Action in this sense means mental as well as physical, in the form 
of prediction and invention. The mental activity is sometimes re- 
ferred to as reflective abstraction. One of Piaget’s shorter and 
more accessible books is entitled To Understand Is to Invent 
(Piaget, 1976), a theme picked up by Constance Kamii ( 1985) in 
Young Children Reinvent Arithmetic: Implications of Piaget’s 
Theory. There is another clear and concise outline of the contri- 
butions of Piaget to constructivist theory in Saxe ( 1991 ), an ac- 
count of how poorly educated child candy- sellers on the streets of 
Recife, Brazil, invent their own complex mathematical and trad- 
ing systems. For more general discussions of Piaget and con- 
structivism, see Fosnot (1996). 
Karl Popper proposed that the knowledge accumulated by every 
individual (and every culture) is a record of the problems they have 
had to solve. Popper tends to be fairly heavy going in his technical 
writing (e.g.. Popper, 1973), but his views are expressed more con- 
cisely in an engaging autobiography (Popper, 1976) and even more 
clearly in a biography (Magee, 1973). Bouldingt 1981 ) argued from 
a behavioral scientist’s point of view that human knowledge is a 
special system, unlike other information systems like libraries, 
computers, or the “real world.” Human knowledge constitutes a 
world in itself and is not simply a combination of the "real world” 
and a brain. In other words — as I interpret it — we live in a world 
NOTES 
241 
that we create, rather than in some concrete world that exists inde- 
pendently of us. I argue that reading can provide actual experi- 
ences in real worlds, not mere replicas of experiences in 
“representations" of the world. Our theory of the world is the basis 
of all our reality. 
Yates (1985) analyzed the contents of momentary awareness — 
what we happen to be aware of at any particular instant — and 
found that although fragmentary it is always part of a complete 
world. Our thoughts and perceptions are never unrelated to the 
world as a whole. They are always capable of anticipating or simu- 
lating future events and thus provide a basis for the formulation of 
appropriate action. Such complex awareness, Yates argued, must 
reflect an underlying model of the world — a theory of the world. For 
thinking, see especially McPeck (1981), Bruner (1986), and 
Vygotsky (1978). Smith (1990) covered the same topic. 
Paradoxically, the computer has provided a great impetus to 
many psychologists interested in comprehension, not necessarily 
because the brain is conceptualized as a kind of computer (al- 
though such a notion does underlie some theorizing) but because 
the computer has proved a convenient tool for simulating organi- 
zations of knowledge and memory. Some experimentalists believe 
that theories about human cognition “lack rigor” unless they can 
be simulated on a computer to prove that they are at least feasible, 
but others find such claims constricting if not misleading. I write 
more critically of these matters in the notes to chapter 4. 
Prediction 
Prediction as I have discussed it in this chapter is not a topic that 
has been widely examined, although there is an extensive psycho- 
logical literature on the consequences of expectancy. Neisser 
(1977) explored the notion that perception is based on anticipated 
information from the environment and that our schemes (dis- 
cussed later) are continually restructured through prediction and 
experience. Our cognitive structures are anticipations. Wildman 
and Kling (1978/1979) discussed “semantic, syntactic and spatial 
anticipation” in reading. 
We don't know what we know unless we put it to use in some way, 
for example, by saying something or by imagining saying some- 
thing (Polanyi, 1966). We similarly don’t have direct access to 
thought. We can be aware of a decision that we have made, but at- 
tempts to reconstruct the “process” by which we came to that deci- 
sion are inventions. Normally we are aware of thought only when 
242 
UNDERSTANDING READING 
we find it difficult to make a decision, just as the only time we are 
normally aware of comprehension is when we are confused. Most 
of the time the brain seems able to take care of our affairs very well, 
without our having to become consciously involved. 
Templeton (1991) disputed that prediction plays a central role 
in reading, citing Perfetti (1991), Stanovich (1991), Vellutino 
(1991), and the “sophisticated eye-movement studies” of Rayner 
and Pollatsek (1989) “that show us precisely where beginning and 
mature readers look as they read connected text.” But where read- 
ers look is not necessarily a reliable indicator of what they are 
thinking. Prediction is a frame of mind, an adjustment to a total sit- 
uation (including one’s own interests and purposes), not a specific 
way of threading through a text. People don’t normally have peri- 
ods of incomprehension and uncertainty before making sense of 
everything they do. There is more on this in chapter 5 and its notes. 
Categories 
There is considerable debate about the exact nature of the cate- 
gorical organization of human knowledge. In particular, the 
hard-edged category boundaries implied by descriptions such as 
those given in this chapter have been challenged. It is not always 
the case that something either belongs in a category or it doesn’t. 
Some things have a greater claim to being in a category — they are 
more “typical” — while others have only a tenuous membership, 
regardless of any distinctive features they might possess. Rosch 
(Rosch & Lloyd, 1978), for example, reported that robins are usu- 
ally regarded as more typical birds than chickens, with eagles 
somewhere between. Penguins and emus are much closer to the 
boundaries between categories, which can be quite fuzzy. Rosch 
proposes that there are certain "natural" categories that form 
themselves around prototypical members, like carrots for vegeta- 
bles and football for sports. The prototypical class members pro- 
vide the main features of the category. Other possibilities achieve 
membership of the category to the extent that they share "family 
resemblances” with the prototypes. In a challenging and wide- 
ranging book taking off from Rosch’ s prototype theory, Lakoff 
(1987) rejected classical categorization theories (going back to 
Aristotle) and proposed instead what he called experiential real- 
ism. We perceive the world in terms of holistic “basic level” cate- 
gories and bodily proportions, functions, and purposes, which 
provide metaphors for understanding the world. This is a com- 
pelling and scholarly book, technical in parts but reiterating in- 
NOTES 
243 
sistently and persuasively the creative and imaginative nature of 
human thought. 
Schemes 
It is unusual for isolated cognitive categories to play a significant 
role in human thought or behavior. We normally function on the ba- 
sis of much larger conceptual structures, the schemes or schemas 
that are constructed out of complex and often dynamic organiza- 
tions of categories. The English word schemes is the standard 
term for the various kinds of abstract mental structure that enable 
us to make sense of the world and participate appropriately in it, 
but the area of study is still known as schema theory, perpetuating 
the Latin term introduced by Bartlett (1932). 
One of the pioneers of schema research, Jean Mandler (1984), 
distinguished three broad categories of schemes: (a) scenes, or 
spatially organized knowledge, (b) events, like the scripts and sce- 
narios I have mentioned in this chapter, and (c) stories, which have 
their own “grammars” of plots, characters, settings, episodes, mo- 
tives, goals, and outcomes. Some theorists — myself included — 
would go further to argue that stories are the primary basis of all 
our perception and understanding of the world. The way we per- 
ceive, comprehend, and remember events is in the form of story 
structures that we impose on them, even though events may not 
present themselves to us in such ways. Rumelhart (1980) saw 
schemes as the building blocks of cognition, comparing them with 
the scripts of plays that can be performed by different groups of ac- 
tors in different settings. Comprehension, according to Rumelhart 
and Ortony (1977), is the confirmation of tentative hypotheses 
about what schemes are relevant by finding the “slots” into which 
the details of events fit. 
A more general point of view is provided by Katherine Nelson 
(1986) and her colleagues in many research papers and in a book 
entitled Event Knowledge. They show that children are skilled at 
expressing ideas in “generic,” abstract representations rather than 
as descriptions of concrete events. Asked to describe a birthday 
party, for example, they talk about cakes, games and gifts, about ex- 
pected behavior and events, rather than about specific occurrences 
on a particular occasion. They speak impersonally — “you" get pres- 
ents — about events with no specific location in time and place. They 
tell a story. For arguments and evidence that perception is much too 
fast, rich, and subtle to be constructed from the discrimination and 
analysis of parts, see McCabe and Balzano (1986). 
244 
UNDERSTANDING READING 
The Narrative Basis of Thought 
Continuously and inevitably, we create stories to explain and un- 
derstand the world and our role in it, to remember and to antici- 
pate events, and to create worlds that might not otherwise exist. 
This urge to create narratives is so compelling that we impose 
them on otherwise meaningless situations. We strain to find ob- 
jects in blurred patterns of color and shade (Potter, 1975), we de- 
tect faces and figures in clouds and other amorphous forms, and 
we impose structure on random sequences of letters or numbers 
(Klahr, Chase, & Lovelace, 1983; Restle, 1970). When people are 
shown randomly flashing points of light in a dark room, they see 
dramas, with objects moving to meet or to avoid each other 
(Michotte, 1946). When thought flows freely, in narrative form, 
comprehension, memory, and learning all seem to take care of 
themselves, as I try to show throughout this book. When the flow is 
broken, when comprehension, memory, and learning are manipu- 
lated from the outside, they may seem to be very difficult indeed. 
Often in these contrived situations, there is only boredom and be- 
wilderment. The narrative nature of children's thought has been 
demonstrated by van Dongen (1987). 
Rosen (1986) argued consistently that thought has a narrative 
basis. If he is correct, then reading and writing must be very funda- 
mental human activities. In a chapter entitled “Stories of Stories: 
Footnotes on Sly Gossipy Practices" (Rosen, 1988). he looked at 
how readers creatively change stories in the retelling, in what he 
called “memory as art.” In a study of how children make sense of 
the world through the construction of stories, at home and at 
school, Wells ( 1985) referred to "the guided reinvention of knowl- 
edge.” Bruner (1986) claims we all employ two modes of thought 
that are complementary but irreconcilable — logical and intuitive 
(or the day view and the night view, truth- seeking versus mean- 
ing-making, well-formed arguments versus a good story). He pro- 
poses that the “self" develops through “autobiographical attitudes 
toward oneself,” seen as talking to oneself about oneself, which 
doesn’t come naturally, but requires knowledge of story conven- 
tions and genres (Bruner & Weisser, 1991). Salmon (1985) pro- 
posed three common metaphors for life — a card game (we re all 
dealt different hands to play as best we can), a natural cycle (birth, 
growth, death, regeneration), and a story (which provides everyone 
with an identity). Sadoski (1983) demonstrated that imagery im- 
proves both comprehension and recall of stories, and Black, Free- 
man, and Johnson-Laird (1986) showed that the more plausible 
NOTES 
245 
we find a tale, the more we are likely to understand and remember 
it. In other words, we understand and remember best when we can 
engage our imagination. Other research would no doubt show that 
drama, excitement, personal relevance, and familiar settings and 
characters are conducive to increased comprehension and recall 
(unless they are so familiar that they are boring). 
Thinking 
An intriguing book stressing the creative and constructive nature 
of thought and also the brain’s narrative mode of functioning is 
Jerome Bruner’s (198 6) Actual Minds, Possible Worlds. In Mind in 
Society, Lev Vygotsky (1978) underlined the social nature of 
thought, which he sees as internalized action. (Bruner would be 
more likely to see action as externalized thought.) 
In a review article on research into metacognition, Bransford, 
Stein, and Vye (1982) observed that less successful students fail 
to activate knowledge that can help them to understand and re- 
member new information. One might argue whether being unable 
to “activate knowledge” (which in less exotic language means to 
make sense) is a cause of failure or simply a description of the 
condition such students find themselves in. The researchers also 
say that less successful students are less able to assess their own 
level of comprehension. But one can hardly be unaware of 
whether one is comprehending or confused — that is like asserting 
that we might be unaware of whether or not we feel hungry. Of 
course, we can believe we understand something when we are in 
error, just as we might feel hungry without needing food. But to 
make a mistake is not a failure of metacognition, of being out of 
touch with our own thought processes. It is simply a matter of be- 
ing wrong. 
NOTES TO CHAPTER 3, 
SPOKEN AND WRITTEN LANGUAGE, pp. 31-54 
Surface Structure and Deep Structure 
There is one notable exception to the statement that surface struc- 
ture is the part of language that exists physically and can be mea- 
sured in the world around us. That exception is the private subvocal 
246 
UNDERSTANDING READING 
speech that we “hear in our heads” when we talk to ourselves or "lis- 
ten to ourselves" reading silently. It would be a mistake to believe 
that such an inner voice is deep structure, or that it is some special 
and (to ourselves) observable kind of raw thought. Subvocal speech 
is just as much a product of thought as overt speech, with the only 
difference being that it is uttered for our own benefit rather than 
anyone else’s. Subvocal speech could be uttered aloud, just as the 
voice we hear in "silent” reading could be made audible to others. 
The inner voice is surface structure, with all the surface structure 
characteristics of vocabulary and grammar (albeit a little tele- 
graphic at times). There must still be a deep structure underlying 
the utterances of the inner voice, a deep structure of meaning, which 
doesn’t consist of sequences of words and sentences but of intangi- 
ble concepts, interrelationships, and propositions. (There is more 
on subvocalization in chap. 9 and its notes.) 
McNeill (1985), reflecting on the nature and origins of language, 
saw parallels between speech and gesture in function, develop- 
ment, use, and even loss with different kinds of aphasia. In a long 
and rather abstruse article entitled “Against Definitions," Fodor, 
Garrett, Walker, and Parkes (1980) argued that sentences are not 
understood by recovering definitions of words. We don’t under- 
stand the statement that someone is a bachelor by understanding 
that he is an unmarried man. "Bachelor” and “unmarried man" are 
not representations of each other but rather alternative represen- 
tations of the same underlying meaning. Anderson and Ortony 
(1975) also show that the interpretation of a sentence, its "mental 
representation," is always much richer than the words in the text 
literally entail. 
Gibbs (1984) argued against the notion that sentences usually 
have “literal meanings” in the context in which they occur (a view 
that he said dominates theories of language). He cited experimen- 
tal evidence that listeners don’t necessarily “compute" the literal 
meaning of an utterance before understanding it. Golden and 
Guthrie (1986) showed that ninth graders respond quite differ- 
ently to the same short story, in the way in which they empathize 
with particular characters or react to events in the text, depending 
on their prior beliefs about what is right and natural. It is. of 
course, unlikely that any two people could ever experience any 
complex series of events, written or "real,” in the same way or come 
away with the same understanding of what took place. 
My very general use of the terms surface structure and deep 
structure should not be taken to relate explicitly to any particu- 
lar linguistic theory, although the distinction between physical 
NOTES 
247 
manifestations of language and meaning is a prevailing view. 
Chomsky (1957), for example, in his original generative trans- 
formational grammar, employed the term surface structure to 
refer not to sound itself, and certainly not to writing, but rather 
as the abstract level at which “input to the phonological system” 
was realized. Similarly, Chomsky’s deep structure was never 
meaning, but rather “input to an underlying semantic system" 
that itself required transformation and interpretation. Obvi- 
ously, sound (or writing) and meaning are much further apart 
than even the surface structures and deep structures that 
Chomsky talked about. Furthermore, there are a great many 
controversies between Chomsky and other linguists and philos- 
ophers about the connotations of these terms, and use of them 
has changed radically over the years. Chomsky’s thoughts on 
language are always significant and — when accessible to the 
layperson — interesting. Chomsky’s clearest exposition is still 
his classic (1959) attack on the behaviorist theorizing of B. F. 
Skinner. For a fascinating debate between Piaget and Chomsky 
see Piattelli-Palmarini (1980). 
Semiotics 
Relatively comprehensible introductions to semiotics generally 
are provided by Davis (1991), Suhor (1992) and Fosnot (1996). 
The educator who has tried the most to make semiotics relevant 
to an understanding of reading and writing is Jerome Harste 
(see Harste, Woodward, & Burke, 1984). One aspect of semiotics 
to which Deely (1982) and Harste drew attention is concerned 
with major forms of logic and their relevance to thinking and 
learning generally. Two of the three forms are widely known, if 
not always well understood. The first is deduction , when conclu- 
sions about specific instances are drawn from general princi- 
ples and procedures. Mathematics and formal logic are 
examples of deductive reasoning. The second category, induc- 
tion, relates to the inferring of general principles from specific 
instances, for example through the “scientific method" of hy- 
pothesis testing. (The “deduction" done by detectives is usually 
induction.) Less well known yet possibly more interesting is the 
third category, abduction, when a new rule or explanation is hy- 
pothesized from a particular result or state of affairs. Such cre- 
ative thinking is not normally considered part of either logic or 
science, but it may better characterize much of human thought, 
including that involved in reading and writing. 
248 
UNDERSTANDING READING 
Discourse and Genre 
Stein (1992) provided a useful but technical volume on reading 
and writing from a discourse analysis point of view. It contains a 
significant paper by Chafe (1992) noting the central role of “dis- 
placed consciousness" in reading, with the reader taking different 
roles and viewpoints. Critical discourse theory combines dis- 
course analysis with social theory, examining, for example, why 
children from poorer homes and families often fail to do well at 
school. For an example, and an outline of critical discourse theory, 
see Rogers (2002) and Cadeiro-Kaplan (2002). The latter is a spe- 
cial issue of the journal Language Arts devoted to critical analyses 
of the literary curriculum and the language used to talk about liter- 
acy. The classic work on language in different socioeconomic con- 
texts remains Ways with Words (Heath, 1983). 
Halliday and his students in Australia have focused attention on 
the role of genre in writing, which they believe should be explicitly 
taught, leading to a considerable educational controversy in that 
country (Cairney, 1992); there is a summary in Smith ( 1994). The 
term genre originally referred to different types of writing, such as 
comedy, tragedy, epic. More recently the term came to refer to dif- 
ferent kinds of media — newspapers, periodicals, novels, nonfic- 
tion — all of which have their own conventions. Current emphasis is 
on complete settings, including such matters as conversations, in- 
terrogations, and classroom procedures. The core is always struc- 
ture or organization — of the text alone or of the situation. Littlefair 
(1991) found British students' knowledge of genre after three or 
four years of study was directly related to the amount and breadth 
of the reading they did. There are some interesting discussions of 
genre and related topics in Cazden (1992). 
Text Organization and Comprehension 
Texts that are difficult to understand because of the way they are 
written have been referred to as “inconsiderate” (Armbruster & An- 
derson, 1984). There is no doubt that texts can often be improved. 
Beck, McKeown, Omanson, and Pople (1984) revised two basal 
reader stories to make them more coherent without altering their 
plots, enhancing the comprehension of both skilled and less 
skilled readers. The length of the two stories was increased from 
782 and 811 words to 900 and 957 words, but comprehension 
went up by a grade level for both second- and third-grade readers. 
Readability formulas based on word counts and sentence length 
NOTES 
249 
have been generally discredited {e.g., MacGinitie, 1984; Krashen, 
2002c). “Simplifying" reading material by fragmenting it into short 
sentences can greatly interfere with comprehension and recall. See 
also Slater (1985). 
Many studies have demonstrated that texts are better under- 
stood and remembered if readers (of all ages and ability) are famil- 
iar with the relevant story grammars. McGee (1982) showed that 
third- and fifth-grade readers were aware of text structures and 
used them in recall, even the poor readers (although they did so 
less well than better readers, who might reasonably be expected to 
have had greater familiarity with stories). Mandler and Goodman 
(1982) found that reading speed dropped when story structures 
weren’t congruent with readers’ expectations and also that second 
sentences were read faster than the first sentences in chapters, 
when readers had gained an idea of what the chapter was about. 
Anderson and Pearson (1984) theorized about the relevance of 
schemes in reading comprehension. See also Grimes (1975) and 
Applebee (1977). 
Piper ( 1 987) suggested cautions about teaching students the ex- 
plicit structures of story grammar, which he saw as no different 
from more traditional modes of analysis which have not fared well 
in education. Consciousness of structure doesn’t necessarily pro- 
mote understanding, he argued. In a careful review, Taylor (1992) 
observed that concern with text structure has required students to 
write “hierarchical summaries” and draw diagrams and “maps” of 
texts — “a whole new skill to learn.” Beers (1987) argued that 
schema theory — and cognitive science generally — are inappropri- 
ate approaches to reading because of their underlying “machine 
metaphor.” Johnson-Laird, Byrne, and Schaeken (1992) asserted 
that the brain is not a logic machine but a sense-making device. 
Durkin (1981) looked critically at the significance of the “new in- 
terest” in comprehension for education. Efforts to systematize in- 
struction have led to a widespread view that comprehension is a 
process, the opposite of which is ignorance. Comprehension, like 
thinking, is seen as a set of skills or procedures which can and must 
be taught. A typical example is the analysis of Pearson. Roehler, 
Dole, and Duffy (1992), which breaks comprehension down into a 
set of seven strategies based on schema theory (Anderson & 
Pearson, 1984). These strategies — supposedly “used differentially” 
by “expert” and “novice” readers — include: searching for connec- 
tions between what they know and new information in text; “moni- 
toring the adequacy of their models of text meaning”; taking steps to 
“repair faulty comprehension” when there is failure to understand; 
250 
UNDERSTANDING READING 
distinguishing important from less important ideas in text; synthe- 
sizing information in and between texts; drawing inferences during 
and after reading for a “full, integrated understanding”; and asking 
conscious and unconscious questions of themselves, authors, and 
texts. Garner ( 1992) discussed “metacognition and self-monitoring 
strategies” as ways of getting students to ask themselves the ques- 
tions that teachers would ask them. 
It might be argued that comprehension is the basis rather than a 
consequence of the previous strategies and that they all depend on 
prior knowledge (including knowledge of the kinds of things that 
can be done with texts). There is an enormous amount of research 
to show the importance of prior knowledge in reading, which 
ought, perhaps, to be considered self-evident. No one, as far as I 
know, has ever proposed an opposing point of view, although some 
instructional methodologies pay little attention to what learners 
may not comprehend or already know — what they may find either 
confusing or boring. 
Robinson, Faraone, Hittleman, and Unrah (1990) reviewed 
comprehension research and instruction since 1 783 — from the ex- 
pectation that comprehension occurs spontaneously to the sys- 
tematic teaching of strategies and self-monitoring techniques. 
Cairney (1990) criticized “traditional comprehension practices" in 
schools as mainly directive and question based. There is a growing 
recognition of the role of inference in comprehension, usually dis- 
cussed either in terms of skills or as internal representations — for 
example, Kintsch (1988) and McNamara, Miller, and Bransford 
(1991). There are substantial and largely nontechnical discus- 
sions of "models of the mind” from both “scientific” and “philo- 
sophical” points of view in Mohyeldin Said, Newton-Smith. Viale, 
and Wilkes (1990). Fodor, an erudite linguistic philosopher prolifi- 
cally involved in many controversies related to language and 
thought, combines densely technical writing with an engaging 
lightheartedness, for example, a chapter entitled “Fodor's Guide to 
Mental Representation: The Intelligent Auntie's Vade-Mecum" in 
Fodor (1990). 
Recondite contemporary theories like deconstructionism and 
intertextuality (which tend to take all meaning out of the text and 
away from authors and readers) were critically discussed by 
Eagleton (1983) in a readably acerbic review of literary theories 
from phenomenology to poststructuralism. In an extended pas- 
sage on reading, Eagleton analyzed the immense amount of inter- 
pretation and prior knowledge that a reader, usually quite 
unconsciously, must bring to bear just to get started on a novel , ar- 
NOTES 
251 
guing that “the text itself is really no more than a series of ‘cues' to 
the reader” (p. 76). For more on deconstructionism, focusing par- 
ticularly on Bahktin’s theory of intertextuality — the idea that the 
meaning of any text is determined by the complex and shifting 
meanings of all other texts — see Lodge (1990). Bloome and 
Egan-Robertson (1993) endeavored to make such theorizing rele- 
vant to classrooms. Finally, there is growing interest in interest as 
a factor in reading. Hidi, Baird, and Hildyard (1982) showed that 
interest can interfere with comprehension because it diverts atten- 
tion; children will not pay extended attention to “trivial informa- 
tion” in texts unless it is interesting, in which case they will 
remember it. 
Some Technical Terms 
One might think that there could not be too much complication 
about the fact that the basic elements of language are sounds. The 
word bed , for example, is made up of three distinctive sounds /b/, 
/el, and /d/ (it is a useful convention that the sounds of language are 
printed between oblique // strokes). With a few perverse excep- 
tions, each sound of the language is represented by a particular let- 
ter of the alphabet, so the number of alternative sounds in English 
must be about 26. Unfortunately, none of the preceding statements 
is correct. 
English has rather more functionally different sounds than it 
has letters in the alphabet, about 45. These sounds have the spe- 
cial name phonemes. A variety of letters can represent a single 
phoneme, and a variety of phonemes can be represented by a sin- 
gle letter or letter combination. It is necessary to be tentative in 
making statements about the total number of phonemes because 
it depends on who is talking and when. All dialects have roughly 
the same number of phonemes, but not always the same ones, so 
that words that are individually distinguishable in some dialects, 
such as guard and god, may not be distinguishable in others un- 
less in a meaningful context. We often think we make distinctions 
between words when in fact we don't — redundancy in context is 
usually sufficient to indicate which alternative we intend. Many 
literate speakers don’t have phonemes to distinguish among 
Mary, marry, and merry, or cot, caught, and court. Say these 
words one at a time and ask a listener to spell what you have just 
said. You may find that the listener can't observe all the differ- 
ences you think you are making. Phonemes often drop out of ca- 
sual or colloquial speech. 
252 
UNDERSTANDING READING 
A phoneme is not so much a single sound as a collection of 
sounds, all of which sound the same. If that description seems 
complicated, a more formal definition will not appear much 
better — a phoneme is a class of closely related sounds constituting 
the smallest unit of speech that will distinguish one utterance from 
another. For example, the /b / at the beginning of the word bed dis- 
tinguishes it from words like fed and led and red, the /e/ in the mid- 
dle distinguishes bed from bad, bide, and bowed, and the /d / 
distinguishes the word from such alternatives as bet and beg. So 
each of the three elements in bed will serve to distinguish the word 
from others, and each also is the smallest unit that can do this. 
Each is a significant difference. It doesn't matter if the /b/ pro- 
nounced at the beginning of bed is a little different from /b/ at the 
beginning of bad, or if the /b/ in bed is pronounced in different 
ways on different occasions. All the different sounds that I might 
make that are acceptable as the sound at the beginning of bed and 
bad qualify as being the same phoneme. A phoneme is not one 
sound, but a variety of sounds any of which is acceptable to listen- 
ers as making the same contrast. The actual sounds that are pro- 
duced are called phones, and the sets of “closely related" phones 
that all serve as the same phoneme are called allophones of each 
other (or of the particular phoneme). Allophones are sounds that 
the listener learns to treat as equivalent and to hear as the same. 
When electronic equipment is used to analyze sounds heard as 
the same, quite marked differences can be found, depending on 
the sound that follows them. For example, the /d/ in dim is basi- 
cally a high-pitched rising sound, while its allophone at the begin- 
ning of doom is much lower pitched and falling (Liberman. 
Cooper, Shankweiler, & Studdert- Kennedy, 1957). A tape re- 
corder will confirm that the two words have no /d/ sound in com- 
mon. If they are recorded, it is impossible to cut the tape in order 
to separate the /im/ or Zoom/ from the /d/. Either one is left with a 
distinct /di/ or /doo / sound, or else the /d/ sound disappears alto- 
gether, leaving two quite different kinds of whistle. Other pho- 
nemes behave in equally bizarre ways. If the first part of the tape- 
recorded word pit is cut and spliced at the front of the final /at 1 of a 
word such as sat or fat, the word that is heard is not pat, as we 
might expect, but cat. The /k/ from the beginning of keep makes 
top when joined to the /op/ from cop and makes poop when com- 
bined with the /oop/ from coop. 
There is a simple way to demonstrate that the sounds that we 
normally hear as the same can be quite different. Say the word pin 
into the palm of your hand and you will feel a distinct puff of air on 
NOTES 
253 
the /p/; however, the puff is absent when you say the word spin. In 
other words, the /p/ in pin is not the same as the /p/ in spin — and 
both are different from the /p/ in limp. If you now pay careful atten- 
tion to the way you say the words, you can probably detect the dif- 
ference. Usually the difference is ignored, because it is not 
significant. Other word pairs provide a similar demonstration — 
for example, kin and skin or team and steam. You may also be able 
to detect a difference between /k / in cool and /k/ in keen, a differ- 
ence that is allophonic in English and phonemic in Arabic, or in the 
A/ at the beginning and end of level. Japanese listeners often have 
difficulty in distinguishing between English words such as link and 
rink because there is no contrast at all between A/ and AY in their 
language. In short, a phoneme is not something present in the sur- 
face level of spoken language — it is something that the listener con- 
structs. We don’t hear different sounds when we are listening to 
speech, but instead we hear significant differences, phonemes in- 
stead of phones. It could be argued that all the discrete sounds that 
are supposed to be discriminable in the continuous flow of speech 
are false analogies created by a bias toward visibly discriminable 
elements of writing — the alphabet, in other words. 
The preceding technical distinctions may be further illustrated 
by reference to writing, where a comparable situation holds. Just 
as the word sound is ambiguous in speech, because it can refer to a 
phone or a phoneme, so the word letter is ambiguous in writing. 
We call a one letter of the alphabet, as distinguished from b, c, d. 
and so forth, but we also talk about a. A, a, and so on as being let- 
ters, although they all in a way represent the same letter. In the first 
case, the letter of the alphabet a is really a category name for a vari- 
ety of written symbols such as a. A, a. The 26 category names for 
the letters in the English alphabet may be called graphemes, the 
written symbols (which are innumerable in their various forms) 
may be called graphs, and the graphs that constitute alternatives 
for a single grapheme are known as allographs. 
Linguists make several other distinctions along the same lines. 
A morpheme is the smallest meaningful part of a word. A word 
may consist of one or more morphemes, some “free” like farm or 
like because they can occur independently, and some “bound” like 
-er (meaning someone who does something) and -s (meaning plu- 
ral) or un- (meaning negative), which have to be joined to a free 
morpheme. Thus farmers is three morphemes, one free and two 
bound, and so is unlikely. Different morphs may represent the 
same morpheme — thus for plurality we can have not only -s but 
-es. -en, and a lot of quite odd forms like the vowel change in 
254 
UNDERSTANDING READING 
man-men or even nothing at all, as in the singular and plural 
sheep. Morphs that constitute the same morpheme are called allo- 
morphs. Meaning itself may be considered in the form of elements 
sometimes called sememes. Bachelor, for example, comprises 
sememes related to maleness, age, marriage, and negation. Words 
in a dictionary — sometimes referred to as “lexical entries" — may 
similarly be termed lexemes. 
Speech, Writing, and "Language" 
Confusion may be caused when reading is referred to as a "lan- 
guage process.” Of course reading is a matter of language. The 
problem arises when the word language is used synonymously 
with the word speech, with reading regarded as some kind of an- 
cillary or parasitical process rather than a language activity in its 
own right. When Perfetti (1985), for example, argued for the “lan- 
guage” basis of reading, he was asserting that reading alphabetic 
text depends on "phonological awareness” of sound patterns in 
speech. 
The priority given to speech is clearly inappropriate (although 
not always recognized to be so) in the case of deaf language. Deaf 
signing is at least as rich, flexible, and expressive as spoken lan- 
guage — although its development may be handicapped by efforts 
to anchor deaf language in speech. For powerful as well as moving 
discussions of the language and thought of deaf people, see Sacks 
(1989) and the references in that volume. On the other hand, the 
word language is often used metaphorically, for example, in refer- 
ences to “computer language" and “body language." 
More About Words 
A number of my observations about the ambiguity of words are de- 
rived from the work of the linguist Fries (1945), who calculated 
that the 500 most common words of our language have an average 
of 28 distinct dictionary meanings each. Miller ( 1951 ) pointed out 
that the 50 most common English words constitute 60% of talk 
and 40% of writing. A mere seven words do 20% of the work of Eng- 
lish — the. of, and, a, to, in, and is. The 10 most common French 
words — a, de, dans, sur, et, ou, que, ne, pas, and y — constitute 
25% of that language. 
NOTES 255 
NOTES TO CHAPTER 4, 
INFORMATION AND EXPERIENCE, pp. 55-71 
Measuring Information and Uncertainty 
A brief introduction to a few elementary calculations employed in 
information theory will allow some numbers to be put to the rate at 
which readers are able to deal with visual information from the 
eyes (our “channel capacity"). The same techniques also permit 
quantification of the uncertainty or redundancy of letters and 
words of English (or any other language) in various circumstances. 
It is necessary to be a little circumlocutionary in putting actual 
figures to information and uncertainty because although both are 
measured with respect to alternatives, the measure is not simply 
the number of alternatives. Instead, information is calculated in 
terms of a unit called a bit, which always reduces by half the uncer- 
tainty on a particular occasion. Thus the card player who discovers 
that an opponent’s strongest suit is red (either diamonds or hearts) 
gets one bit of information, and so does a child trying to identify a 
letter who is told that it comes from the second half of the alphabet. 
In the first case two alternatives are eliminated (the two black 
suits), and in the second case 13 alternatives are removed (and 13 
still remain). In both cases the proportion of uncertainty reduced 
is a half, and therefore the amount of information received is re- 
garded as one bit. The uncertainty of a situation in bits is equal to 
the number of times a “yes or no” question would have to be asked 
and answered to eliminate all uncertainty if each answer reduced 
uncertainty by a half. Thus there are two bits of uncertainty in the 
card-playing example because two questions will remove all doubt, 
for example: Ql. Is it a black suit? Q2. If yes, is it clubs? (If no, is it 
hearts?), or Ql. Is it spades or diamonds? Q2. If yes, is it spades? 
(If no, is it hearts?) 
You can see it doesn’t matter how the questions are posed, pro- 
vided they permit a yes-no answer that will eliminate half the alter- 
natives. The final qualification is important. Obviously, a single 
question such as “Is it clubs?” will eliminate all alternatives if the 
answer is “yes,” but will still leave at least one and possibly more 
questions to be asked if the answer is “no.” The most efficient way 
of reducing uncertainty when the answer can only be “yes" or “no" 
is by a binary split, that is, by partitioning the alternatives into two 
equal sets. In fact the word bit, which may have sounded rather 
colloquial, is an abbreviation of the words binary digit, or a num- 
ber representing a choice between two alternatives. 
256 
UNDERSTANDING READING 
The uncertainty of the 26 letters of the alphabet lies somewhere 
between four and five bits. Four bits of information (four ques- 
tions) will allow selection among 16 alternatives, not quite 
enough, with the first bit reducing this number to 8, the second to 
4, the third to 2, and the fourth to 1 . Five bits will select among 32 
alternatives, slightly too many, with the first eliminating 16 and 
the other four removing the rest of the uncertainty. In brief, x bits 
of information will select among 2' alternatives. Two bits will se- 
lect among 2 2 = 4, three bits among 2 3 = 8. four bits among 2 4 = 
16, and soon. One question settles only 2 1 = two alternatives, and 
no questions at all are required if you have only one alternative to 
begin with (2° = 1 ). Twenty bits (“twenty questions") are theoreti- 
cally sufficient to distinguish among 2 20 = 1 ,048,576 alternatives. 
There is a mathematical formula that shows that the theoretical 
uncertainty of 26 letters of the alphabet is almost precisely 4.7 
bits, although, of course, it is not easy to see how one could ask 
just 4.7 questions. (The formula is that the uncertainty of a: alter- 
natives is log 2 x, which can be looked up in a table of logarithms to 
the base 2: log 2 26 = 4.7, since 26 = 2 4 7 .) Can you calculate the 
uncertainty in a deck of 52 playing cards? Since 52 is exactly 
twice 26, the uncertainty of the cards must be one bit more than 
that of the alphabet, or 5.7 bits. 
Measuring Redundancy 
It will be helpful to pursue the matter of redundancy a little more 
deeply, partly because of the importance of the concept of redun- 
dancy to reading, but also because the discussion of how bits of 
uncertainty or information are computed contained an oversim- 
plification that can now be rectified. We consider two aspects of 
redundancy, termed distributional and sequential. 
Distributional redundancy is associated with the relative prob- 
ability that each of the alternatives in a particular situation can oc- 
cur. There is less uncertainty when some alternatives are more 
probable than others. And because there is less uncertainty when 
alternatives are not equally probable, there is redundancy. The 
very fact that alternatives are not equally probable is an additional 
source of information that reduces the uncertainty of the set of al- 
ternatives as a whole. Redundancy that occurs because the proba- 
bilities of alternatives are not equally distributed is therefore 
called distributional redundancy. 
Uncertainty is greatest when every alternative has an equal 
chance of occurring. Consider a coin-tossing game where there 
NOTES 
257 
are only two alternatives, head or tail, and there are equal 
chances of a head or a tail turning up. The informativeness of 
knowing that a particular toss of the coin produced a head (or a 
tail) is one bit, because whatever the outcome, uncertainty is re- 
duced by a half. But now suppose that the game is not fair, and 
that the coin will come down head 9 times out of 10. What is the 
uncertainty of the game now (to someone who knows the coin’s 
bias)? The uncertainty is hardly as great as when the odds were 
50-50, because then there was no reason to choose between head 
and tail, while with the loaded coin it would be foolish knowingly 
to bet tail. By the same token, there is likely to be far less informa- 
tion on being told the outcome of a particular toss of the loaded 
coin. Not much uncertainty is removed if one is told that the coin 
has come down head because that is what was expected all the 
time. In fact, the informativeness of a head can be computed to be 
about 0.015 bit, compared with 1 bit if the game were fair. It is 
true that there is much more information in the relatively unlikely 
event of being told that a toss produced a tail — a total of 3.32 bits 
of information compared with the one bit for a tail when heads 
and tails are equally probable — but we can expect a tail to occur 
only once in every 10 tosses. The average amount of information 
available from the loaded coin will be nine tenths of the 0.015 bit 
of information for head and one tenth of the 3.32 bits of informa- 
tion for tail, which when totaled is approximately 0.35 bit. The 
difference between the 1 bit of uncertainty (or information) for the 
50-50 coin, and the 0.35 bit for the 90-10 loaded coin, is the dis- 
tributional redundancy. 
The statement that the uncertainty of letters of the English lan- 
guage is 4.7 bits is perfectly true for any situation involving 26 
equally probable alternatives — for example, drawing a letter from 
a hat containing one instance only of each of the 26 letters of the al- 
phabet. But the letters of English don’t occur in the language with 
equal frequency; some of them, such as e, t, a, o, i, n , s, occur far 
more often than others. In fact, e occurs about 40 times more often 
than the least frequent letter, z. Because of this inequality, the aver- 
age uncertainty of letters is somewhat less than the maximum of 
4.7 bits that it would be if the letters all occurred equally often. The 
actual uncertainty of letters, considering their relative frequency, is 
4.07 bits, with the difference of about 0.63 bit being the distribu- 
tional redundancy of English letters, a measure of the prospective 
informativeness that is lost because letters don’t occur equally of- 
ten. If letters were used equally often, we could achieve the 4.07 
bits of uncertainty that the 26 letters currently have with a little 
258 
UNDERSTANDING READING 
over 16 letters. We could save ourselves about 9 letters if we could 
find a way to use the remainder equally often. 
Like letters, words also have a distributional redundancy in 
English. One of the oldest and still least understood findings in ex- 
perimental psychology concerns the “word-frequency effect," that 
more common words of our language can be identified on less vi- 
sual information than less frequent words (Broadbent, 1967; 
Broadbent & Broadbent, 1975; Howes & Solomon, 1951). Compu- 
tations of the distributional redundancy of letters and words in 
English are contained in Shannon ( 1951 ). 
Sequential redundancy exists when the probability of a letter 
or word is constrained by the presence of surrounding letters or 
words in the same sequence. For example, the probability that the 
letter H will follow T in English words is not 1 in 26 (which would 
be the case if all letters had an equal chance of occurring in any 
position) nor about 1 in 17 (taking distributional redundancy 
into account) , but about 1 in 8 (because only eight alternatives are 
likely to occur following T in English, namely, R, H, or a vowel). 
Thus the uncertainty of any letter that follows T in an English 
word is just three bits (2 3 = 8). The average uncertainty of all let- 
ters in English words is about 2.5 bits (Shannon, 1951 ). The dif- 
ference between this average uncertainty of 2.5 bits and a 
possible uncertainty of 4.07 bits (after allowing for distributional 
redundancy) is the sequential redundancy of letters in English 
words. An average of 2.5 bits of uncertainty means that letters in 
words have a probability of about 1 in 6 instead of 1 in 26. This 
figure, of course, is only an average computed over many read- 
ers, many words, and many letter positions. There is not a pro- 
gressive decline in uncertainty from letter to letter, from left to 
right, in all words. An English word beginning with q, for exam- 
ple, has zero uncertainty about the next letter u, an uncertainty of 
about two bits for the four vowels that can follow the u. and then 
perhaps four bits of uncertainty for the next letter, which could be 
one of over a dozen alternatives. Other words have different un- 
certainty patterns, although in general the uncertainty of any let- 
ter goes down the more other letters in the word are known, 
irrespective of order. Because of constraints in the spelling pat- 
terns of English — due in part to the way words are pro- 
nounced — there is slightly more uncertainty at the beginning of a 
word than at the end, with slightly less in the middle (Bruner & 
O’Dowd, 1958). 
The orthographic (spelling) redundancy of print to which I have 
referred comprises both distributional and sequential redundancy 
NOTES 
259 
for letters within words, while syntactic (grammar) and semantic 
(meaning) redundancy are primarily sequential redundancy be- 
tween words. “Predictive” software on computers makes use of the 
sequential redundancy of both letters and words, in combination 
with a “virtual keyboard” on the monitor screen where letters (or 
other symbols) need only be touched or otherwise indicated in or- 
der for choices to be made. In contrast to the fixed array of keys on 
a “real” keyboard, the characters on a virtual keyboard can be re- 
placed almost instantly, from letter to letter, so that the writer is 
confronted only by those items most likely to be required . If a k has 
already been selected, for example, only those letters likely to come 
next will be presented (like l, n, r or a vowel), reducing clutter and 
the time required for a decision to be made. Once one or more 
words have been selected, the software may be able to suggest en- 
tire words that are likely to come next, so that the writer need make 
only one choice, out of very few alternatives, to select an entire 
word. Predictive software can do more than anticipate letters and 
words based on their frequency in language as a whole — it can re- 
flect probabilities of words and sequences of words in particular 
subject areas and even be “trained” to anticipate the probable word 
choices of individual writers. 
Limitations of Information Theory 
The first practical application of information theory was in mea- 
suring the efficiency of communication systems like telephone 
lines and radio links. The measure was related to the proportion 
of words emitted by the “transmitter" at one end of a communica- 
tion channel that the “receiver” at the other end could correctly 
identify. Information theory had a brief but spectacular decade of 
influence in psychology in the 1950s and 1960s, primarily due to 
the erudition of George Miller (1956) in an article that referred to 
limits on human “capacity for processing information.” For many 
psychologists this was a new and seductive way of talking about 
the brain. 
Information theory became influential at a time when reading 
was primarily considered to be a matter of identifying letters and 
words. The text could be regarded as a transmitter, the reader as a 
receiver, and the visual system as a communication channel. The 
efficiency or “capacity” of this channel could be computed from the 
proportion of letters and words that the reader correctly identified 
under various conditions. The perspective and techniques were 
useful theoretically and enabled all kinds of interesting compari- 
260 
UNDERSTANDING READING 
sons to be made. They helped to demonstrate in a quantifiable way 
that the visual system has limitations — we can’t see everything that 
is in front of our eyes. The theory also offered some useful con- 
cepts, notably that of redundancy. 
However, information theory itself has severe limitations, with 
respect to texts and to readers. It can measure "information" in its 
own narrow terms of reducing uncertainty among known sets of 
alternatives, but it can't say how meaningful a text is, or how 
much understanding there might be. Outside the experimental 
laboratory, readers usually read for meaning rather than for in- 
formation — or they get their information enveloped in meaning. 
And when readers do “receive information." it is usually not in the 
limited sense of information theory. Statements like “More than a 
thousand kinds of brown algae exist” may be informative, but 
there is no way of saying how much uncertainty they reduce. Re- 
dundancy in spelling patterns can be calculated, but not redun- 
dancy in short stories. Information theory loses its utility once we 
get inside the head. 
Besides, information doesn’t seem to be what the brain is pri- 
marily concerned with (Smith, 1983a). The brain deals with un- 
derstanding rather than information. Either information 
becomes understanding as part of our interconnected theory of 
the world, or it remains an isolated fact, at best potential mean- 
ing, like an item in an encyclopedia. Information can be derived 
from experience in the same way that vitamins can be obtained 
from food, but information is no more experience than vitamins 
are food. The semantic complication is compounded by the fact 
that in contrast with the narrow sense in which the word informa- 
tion is employed in information theory, its general use these days 
is totally undiscriminating, and therefore meaningless. Every- 
thing is information — the content of every book, journal, and tele- 
vision program, the entire educational curriculum, anything on a 
computer, even junk mail. 
Rosenblatt (1978), in her distinction between readingfor infor- 
mation and reading for experience, didn’t use the word informa- 
tion in either the narrow information theory sense or the catch-all 
general sense. To her, information meant “facts.” Perhaps be- 
cause the terms information and experience have such broad 
general uses, she actually employed quite uncommon terms for 
what she wanted to explain. Informational reading she described 
as efferent, meaning “carrying outward" from the text, and the al- 
ternative she termed aesthetic , implying involvement in the text 
through the senses. 
NOTES 
261 
Computers and People 
In the jargon of cognitive science (see notes to Preface), informa- 
tion, whether in humans or computers, is always “processed” 
rather than “understood,” Almost everything now is a process in 
the educational research literature — and in disquisitions on teach- 
ing as well . It is rare to read of unadulterated reading, writing, com- 
prehension, learning, or teaching; instead there is the reading 
process, the writing process, the comprehension process, the 
learning process, and the teaching process. And according to my 
dictionary, the word process has broad mechanical connota- 
tions — it entails a succession of actions or operations in a specific 
or prescribed sequence (much like the manner in which a com- 
puter is programmed). Among cognitive scientists, reading and 
comprehension are now both “text processing.” Writing is “word 
processing.” Thinking is "ideas management" or “the organization 
of knowledge.” 
Perhaps to underline the growing influence of computer-based 
ways of thinking, the word information is being superseded by 
data (a plural word that, like criteria and media, is commonly 
used in the singular), and the word knowledge by database. Pro- 
cess may be losing ground to procedures, an artificial intelligence 
term, and even to instructions, which are what computers run on. 
Schemas and scenarios are seen as procedures rather than as nar- 
ratives. “Procedural” knowledge is contrasted with “propositional” 
knowledge. Larsen ( 1986) even argued that there is a need for “pro- 
cedural literacy," which is not the same thing as “computer liter- 
acy,” but rather the ability to produce and understand sequences of 
instructions for the organization of knowledge. 
It can be argued that cognitive science has become an elaborate 
behavioristic stimulus-response system, despite the fact that it 
calls itself cognitive. Individuals are perceived as totally under the 
control of an environment, which is the source of all data or infor- 
mation. There are no intentions, feelings, or values except for what 
something outside the system has put into it. This is not my argu- 
ment, but that of the foremost exponent of behaviorism, B. F. Skin- 
ner ( 1 985 ) , who claimed that the only difference between cognitive 
science and his own theory is in the language that is used. 
The discussion of cognitive science is inseparable from discus- 
sion of the relationship of computers and the brain, a complex 
philosophical topic. Computers don’t do anything unless told what 
to do and how to do it, so they are not like the human brain unless 
you believe that is what humans are like. Computers may do things 
262 
UNDERSTANDING READING 
that we can't do — like calculate thousands of prime numbers, or 
scrutinize complex landscapes, or generate elaborate de- 
signs — but nevertheless they are following procedures that hu- 
mans give them (or that the computers develop as a consequence of 
earlier procedures that humans have given them). And there is no 
guarantee that the procedures computers follow are those used 
when humans make mathematical calculations, study a land- 
scape, or paint a picture; they almost certainly are not. A computer 
doesn’t have plans and intentions, except those it is given. It doesn’t 
have wishes or feelings or values. These characteristics may be 
simulated on computers, but this doesn't give them human charac- 
teristics. Computers don’t have experiences. They don’t under- 
stand sarcasm, irony, or affection. They don’t understand 
anything. Programming a computer isn’t the same thing as teach- 
ing a person, and learning certainly isn’t the same thing as being 
programmed. Computers are said to have learned when they per- 
form differently. Human beings don’t necessarily behave differ- 
ently as a consequence of learning and can change their behavior 
without learning anything. 
NOTES TO CHAPTER 5, 
BETWEEN EYE AND BRAIN, pp. 72-94 
Vision and Information 
It is difficult to avoid referring to bursts of neural energy in the op- 
tic nerve as “information” or “messages” that the eyes send to the 
brain about the world. But both terms can be misleading, with 
their implication that the eyes know something that they try to 
communicate to the brain. It might be more appropriate to refer to 
the neural impulses that travel between eye and brain as clues to a 
world forever concealed from direct inspection. No scientist or phi- 
losopher can say what the world is “really like," because everyone's 
perception of the world — even when mediated by microscopes or 
telescopes, by photographs or x-rays — still depends on the sense 
the brain can make of neural impulses that have come through the 
dark tunnel between eye and brain. We can no more see the image 
of the world that falls on the retina than we can see the nerve im- 
NOTES 
263 
pulses that the retina sends to the brain. The only part of vision of 
which we can ever be aware is the final sensation of seeing, con- 
structed within the brain. 
That there is a limit to how much print can be identified at any 
one time, varying according to the use a reader can make of re- 
dundancy. is not exactly a recent discovery. The illustration in 
this chapter of how much can be identified from a single glance 
at a row of random letters, random words, and meaningful se- 
quences of words is derived directly from the early researches of 
Cattell (1885, republished 1973), Erdmann and Dodge (1898), 
and Dodge (1900). Descriptions of many similar experimental 
studies were included in a remarkably insightful book by Huey 
(1908, republished in 1967), which remains the only classic in 
the psychology of reading. A good deal of recent research on per- 
ception in reading is basically replication of early studies with 
more sophisticated equipment; nothing has been demonstrated 
that controverts them. Yet the pioneer research was neglected by 
experimental psychologists for almost a century and is still 
widely unknown in education, partly because behaviorism in- 
hibited psychologists from studying “mental phenomena” and 
partly because "systematic” or “scientific” piecemeal ap- 
proaches to reading instruction have concentrated on decoding 
and word attack — the “tunnel vision" extreme — at the expense of 
comprehension. 
An excellent historical survey of eye-movement research was 
provided by Paulson and Goodman (1999), who took pains to give 
credit to “valid, reliable and high-quality work” done in the first 
decades of the 20th century, and even earlier. Huey (1908), for ex- 
ample, established that the first fixation in a line of text was not 
necessarily on the first word, nor was the last fixation necessarily 
the last word. Individual words were frequently skipped (from 
30% to 80% of the time, depending on the familiarity of the lan- 
guage, material and genre to the reader), and some of the fixations 
were regressions. Two of the most prolific researchers, Judd and 
Buswell ( 1922), showed that context was the prime factor in de- 
termining the meaning of words in normal reading. Paulson and 
Goodman noted that a contemporary habit of restricting research 
reviews to the most recent 5 years not only leads to ignorance of 
an important knowledge base, but also conceals current misun- 
derstandings or misrepresentations of original work. Their re- 
view was published on the Internet, perhaps appropriate for a 
study that begins with pioneer research conducted with home- 
made devices on tabletops. 
264 
UNDERSTANDING READING 
The mathematics of information theory (chap. 4 notes) shows 
that readers identifying just four or five random letters, a couple of 
random words, or a meaningful sequence of four or five words in 
one glance are analyzing the same amount of visual information 
each time. The differences among the three conditions must be at- 
tributed to the varying amounts of nonvisual information that 
readers are able to contribute, related to distributional and se- 
quential redundancy within the print. The random letter condition 
suggests that the limit for a single glance (the equivalent of a second 
of processing time) is about 25 bits of information, based on a max- 
imum of 5 letter identifications of about 5 bits of uncertainty each 
(5 bits = 32 alternatives). In random sequences of letters, of 
course, there is no distributional or sequential redundancy that a 
reader can utilize. That 25 bits per second is indeed a general limit 
on the rate of human information processing was concluded by 
Quastler ( 1956) from studies not only of letter and word identifica- 
tion but of the performance of piano players and “lightning calcula- 
tors” as well. 
How is it possible then to identify two random words, consist- 
ing on the average of 4.5 letters each, with just 25 bits of visual in- 
formation? Nine or 10 letters at 5 bits each would seem to require 
closer to 50 bits. But as I pointed out in the chapter 4 notes, be- 
cause of distributional and sequential redundancy the average 
uncertainty of letters in English words is about 2.5 bits each 
(Shannon, 1951), the total average uncertainty for letters in two 
random words would be something under 25 bits. From a differ- 
ent perspective, random words taken from a pool of 50,000 alter- 
natives would have an uncertainty of between 15 and 16 bits each 
(2 15 = 32,768, 2' 6 = 65,536), but due to distributional redun- 
dancy among words — and because unusual words are unlikely to 
be employed in reading studies — we can probably again accept 
the estimate of Shannon that the average uncertainty of English 
words without syntactic or semantic constraints (sequential re- 
dundancy) is about 12 bits per word. So whether we look at the 
random word condition from the point of view of letter uncer- 
tainty in words (about 2.5 bits per letter) or of isolated word un- 
certainty (about 12 bits per word) the result is still that the reader 
is making the identification of about 9 or 10 letters or two words 
with roughly 25 bits of visual information. The fact that both the 
number of letters identified and the effective angle of vision dou- 
ble in the random word condition compared with the four or five 
letters that can be perceived in the random letter condition re- 
flects the use that the reader can make of redundancy. In other 
NOTES 
265 
words, the viewer in the isolated word condition contributes the 
equivalent of 25 bits of nonvisual information to enable twice as 
much to be seen in a single glance. 
In meaningful passages of English, there is considerable se- 
quential redundancy among the words themselves. Speakers and 
authors are not free to choose any word they like whenever they 
please, at least not if they expect to make any sense. From statisti- 
cal analyses of long passages of text and also by a “guessing game” 
technique in which people were actually required to guess letters 
and words, Shannon calculated that the average uncertainty of 
words in meaningful sequences was about 7 bits (a reduction 
over words in isolation of about a half) and that the average uncer- 
tainty of letters in meaningful sequences was only slightly over 1 
bit (again reducing by half the uncertainty of letters in isolated 
words). On this basis, one would expect viewers in the meaningful 
word sequences condition to see twice as much again compared 
with isolated words, which is of course the experimental result. A 
phrase or sentence of four or five words can be seen in one glance, 
a total of 20 letters or more. This is four times as much as can be 
seen in the random letter condition, but still on the basis of the 
same amount of visual information: 4 or 5 random letters at 
about 5 bits each, or 20 letters in a meaningful sequence at just 
over 1 bit each. Put in another way, when reading sequences of 
meaningful words in text the reader can contribute at least three 
parts nonvisual information (in the form of prior knowledge of re- 
dundancy) to one part visual information so that four times as 
much can be perceived. 
Other analyses of uncertainty and redundancy in English are in- 
cluded in Garner (1962, 1974) and in Miller, Bruner, and Postman 
(1954), with the latter using carefully constructed “approximations 
to English." McNeill and Lindig (1973) demonstrated that how 
much listeners perceive in spoken language also depends on what 
they are attending to — individual sounds, syllables, or entire words. 
The Rate of Visual Decision Making 
Many years after the first demonstrations of what can be seen in a 
single glance, other tachistoscopic studies showed that the limits 
can’t be attributed to the amount of visual information that the eye 
can gather from the page, nor to viewers’ forgetting letters or words 
already identified before they can report them. Rather, the bottle- 
neck occurs as the brain labors on what is transiently a consider- 
able amount of raw visual information, organizing “seeing” after 
266 
UNDERSTANDING READING 
the eyes have done their work. Perceptual decision making takes 
time, and there is a limit to how long visual information sent back 
by the eyes remains available to the brain. 
Viewers in the kind of tachistoscopic experiment that I have de- 
scribed often feel that they have potentially seen more than they are 
able to report. The brief presentation of visual information leaves a 
vaguely defined “image" that fades before they are fully able to at- 
tend to it. The validity of this observation was established by an ex- 
perimental technique called partial recall (Sperling, 1960) in 
which viewers are required to report only 4 letters out of a presen- 
tation of perhaps 12 so that the required report is well within the 
limits of short-term memory (see chap. 6). However, the viewers 
don't know which 4 letters they must report until after the visual 
presentation, so they must work from visual information that re- 
mains available to the brain after its source has been removed 
from before the eyes. At one stroke the experimental technique 
avoids any complication of memory by keeping the required report 
to a small number of items, while at the same time testing whether 
in fact viewers have information about all 1 2 items for a brief time 
after the eyes’ work is complete. 
The experimental technique involves presenting the 12 letters 
in three rows of 4 letters each. Very soon after the 50-msec pre- 
sentation is ended, a tone is sounded. The viewer already knows 
that a high tone indicates that the letters in the top row are to be 
reported, a low tone calls for the report of the bottom row. and an 
intermediate tone indicates the middle line. When this method of 
partial recall is employed, viewers can normally report back the 4 
required letters, indicating that for a short while at least they have 
access to raw visual information about all 12 letters. The fact that 
viewers can report any 4 letters, however, doesn't indicate that 
they have identified all 12, but simply that they have time to iden- 
tify 4 before the visual information fades. If the cue tone is delayed 
more than half a second after the end of the presentation the num- 
ber of letters that can be reported falls off sharply. The “image" is 
raw visual information that decays by the time about 4 letters 
have been identified. 
Other evidence that visual information remains available in a 
sensory store for about a second and that the entire second is re- 
quired if a maximum of four or five letters is to be identified has 
come from the masking studies described in this chapter (e.g., 
Averbach & Coriell, 1961; Smith & Carey, 1966). Ifa second visual 
array is presented to the eye before the brain has finished identify- 
ing the maximum number of letters that it can from the first input 
NOTES 
267 
of visual information, then the amount reported from the first pre- 
sentation declines. The second input of visual information erases 
information from the first presentation. However, it can be as dis- 
ruptive for reading if visual information reaches the eye too slowly 
as it is if too fast. Kolersand Katzman (1966), Newman (1966), and 
Pierce and Karlin (1957) showed an optimum rate of about six pre- 
sentations a second for receiving visual information about individ- 
ual letters or words; at faster rates the brain can’t keep up and at 
slower speeds there tends to be a greater loss of earlier items 
through forgetting. These studies and the earlier calculations of 
Quastler (1956) all tend to support the view that the “normal read- 
ing rate” of between 200 and 300 words a minute is an optimum; 
slower reading disrupts comprehension. 
Eye Movements in Reading 
Rayner (1997) summarized decades of his own and other re- 
search into what can be seen as the eye moves from one fixation to 
another. He confirmed the strong influence of context on where 
the eye lands in each fixation, usually on content words. Words 
are often skipped, depending on their length, frequency, and pre- 
dictability. Readers are very accurate at regressing to a point in 
the text where comprehension problems occurred. Rayner cited 
numerous studies showing that the perceptual span can extend to 
1 5 letter spaces to the right of the fixation point and 4 to the left 
(the reverse in languages that are read from right to left). No use- 
ful information comes from the line below the line fixated on. The 
span is not fixed, but varies with the experience of the reader and 
the predictability of the text. There is a “preview benefit” for 
words not clearly seen around the focal area of gaze, mainly from 
beginning letters. McClelland and O’Regan (1981) showed that 
this benefit is increased if readers are able to make predictions 
about what they will see. 
The generalizability of conclusions from eye movement studies 
is limited because under laboratory conditions viewers often have 
no control over where they fix their gaze or of how long fixations 
may last. The experimenter makes these decisions for them. In 
many tachistoscopic and eye-movement studies, viewers are not 
even free to move their heads, which are constrained by chin rests 
or even bite plates. Yet head movements are a conspicuous part of 
normal reading. To avoid head movement constraints, some read- 
ing researchers have gone to another extreme by providing helmets 
equipped with electronic technology for their viewers to wear while 
268 
UNDERSTANDING READING 
looking at a computer screen. Many computer-based experiments 
also involve bizarre reading conditions, such as words that may 
change during a saccade or the presence of irrelevant words or 
lines of x's just outside the foveal (sharp focus) area (Rayner & 
Pollatsek, 1987). Important areas of the text may be removed, for 
example, when the letter in the center of the reader’s field of vision 
is automatically obscured on every fixation (Rayner, 1992). The 
consequence that reading speed may be slowed by 50% in such cir- 
cumstances is cited as demonstrating that readers need the infor- 
mation in the obliterated letter, although it could be argued that it 
is the distraction itself that causes the disruption. For other repre- 
sentative eye-movement studies see Rayner and Pollatsek (1989) 
and Stanovich (1991). 
Duckett (2002) also challenged claims that good readers look 
at each successive word in turn as they read. In a close study of six 
first-grade beginning readers, reading aloud from computer 
screens, he found that they did not fixate on every word of a sim- 
ple illustrated story, they spoke words that they did not fixate on. 
and they did not always fixate serially left to right. Fixations were 
selected strategically where, when and for as long as needed for 
them to get the sense of the story. Paulson (2002) showed that pre- 
diction overrides evidence as readers strive to make sense of the 
content of their fixations. They change the text rather than change 
their mind, supporting the view that reading is a constructive ac- 
tivity, depending on what is “behind the eyes." Kucer and Tuten 
(2003) examined the miscues of 24 fluent adult readers and 
found that the majority of miscues were influenced by the sense, 
syntax, and style of the author. The miscues paralleled those of 
young developing readers. 
It might be asked why the most general and most efficient fixa- 
tion rate in reading seems to be about four fixations a second when 
the information from a single glance persists for a second or more 
and at least 1 second is required for analyzing all the information 
that can be acquired from a single fixation. Why do readers make 
so many fixations when they can see four or five words at a glance? 
There has not been a good research answer to this question, but 
my conjecture is that the brain is less concerned with squeezing the 
last bit of information from each fixation than with receiving a 
smooth inflow of selected visual information as it builds up a co- 
herent understanding of the text. Successive words provide useful 
anchorages for the eyes as the brain does its work. Obviously, read- 
ers must already have a good idea of where to fixate if their gaze 
rests primarily on content words. 
NOTES 
269 
Interesting analogies may be drawn between eye movements and 
hand movements. The top speed of movement for eye and for hand 
are roughly similar, and, like the eye, the hand moves faster when it 
moves over a greater distance. The hand performs the same kind 
of activity as the eye. It moves precisely and selectively to the most 
useful position, and it starts “picking up” only when it has arrived. 
But although the hands and eyes of children may move almost as 
fast and accurately as those of adults, they can’t always be used as 
efficiently. Children lack the experience of the adult and may not 
know so precisely what they are reaching for. 
On Seeing Backward 
Moyer and Newcomer (1977) demonstrated that reversals are 
caused by inexperience with directionality rather than by percep- 
tual deficits. Stanovich (1982a, 1982b) noted that normal readers 
may also make orientation errors (“reading or writing backward”) 
that are not a cause of reading difficulties. 
Hemispheric Specialization 
A basic article about functional differences in the cerebral hemi- 
spheres and about the occasionally bizarre consequences of their 
being superficially disconnected was provided by one of the earli- 
est researchers in the area, Sperry ( 1 968). Discussions of the rela- 
tionship between the hemispheres and language are included in 
several chapters in Caplan ( 1980). Research leading to the conclu- 
sion that hemispheric specialization should not be considered an 
explanation of reading difficulties in children is reported in Naylor 
(1980), Young and Ellis (1981), and in the chapter by Bryden in 
Underwood (1978). 
NOTES TO CHAPTER 6, 
BOTTLENECKS OF MEMORY, pp. 95-109 
Theories of Memory 
One of the earliest and most coherent attempts to distinguish 
short-term and long-term characteristics of memory was by Nor- 
270 
UNDERSTANDING READING 
man ( 1969), revised and expanded into a comprehensive analysis 
of the processes and contents of memory (Norman, 1976). There 
are useful reviews by Kintsch (1982), Baddeley (1992), and 
Schneider and Shiffrin ( 1977), who particularly emphasized the 
relationship between short-term memory and attention. Lewis 
( 1979) critically reviewed the short-term and long-term memory 
distinction, proposing that most forgetting is retrieval failure 
rather than storage loss and suggesting instead an active and in- 
active memory distinction with active memory part of the greater 
inactive one (similar to my Fig. 6.2). 
Tulving ( 1985a, 1985b) proposed that there are three different 
memory systems, which he termed episodic, semantic, and pro- 
cedural. He associated each with a different kind of conscious- 
ness (or absence of consciousness). Tulving's basic memory 
system is procedural: it is also the most primitive, the only one 
that animals have. It is also the only one of the three systems that 
can be completely independent of the other two. Procedural mem- 
ories require overt action to become established and are not ac- 
cessible to consciousness. Tulving called this condition anoetic 
(literally, “without knowledge”). We can never be aware of what we 
know procedurally (except by actually doing something, possibly 
in the imagination). Such a memory system may be fundamental, 
but it is not trivial. It is probably the aspect of our memory con- 
taining the “rules” of language, which are not learned consciously 
(Krashen, 1985). Tulving’s semantic memory is a subset within 
the procedural, and it makes possible representations of states of 
the world not perceptually present (i.e., which we can imagine). 
Semantic memory includes facts — but not in any particular or- 
der. Most people, for example, know that both John F. Kennedy 
and Charles de Gaulle are dead, but they can t immediately say 
who died first. Semantic memory "describes" events and situa- 
tions for us — it brings them to consciousness. Tulving calls this 
noetic. Finally, episodic memory, which is nested within the se- 
mantic system, is “self-knowing." or autonoetic. It is our aware- 
ness of the order or sequence of events, the only conscious form of 
memory that includes temporal relationships. Tulving stressed 
what few cognitive psychologists would dispute these days, that 
the quality of a particular memory depends on the manner and 
circumstances in which it was originally learned. 
Another distinction frequently drawn is between recognition 
and reproduction memory. It is usually (but not always) easier to 
recognize a face or a place (or an object or symbol) than it is to 
NOTES 
271 
draw it. We recognize correct spellings easier than we can pro- 
duce them. At any age we can understand more of language than 
we can produce ourselves. This should not be taken as implying 
that we have two entirely different kinds of memory — that we go to 
one “store” for recognition and another for reproduction. It is not 
that we have a collection of pictures (or “images”) in the brain that 
we can refer to for recognition. Mental images themselves have to 
be constructed, and we can usually recognize faces and other 
things more easily than we can imagine them. There is probably a 
simple explanation — that we usually need to produce less detail 
for recognition than for reproduction. To recognize a face, or even 
a word, we may need to see only a part of it, but reproduction 
means that we have to generate it all, without omission or error. 
For factual matters, a similar distinction is frequently made be- 
tween recognition and recall memory. We may be able to agree 
that a certain actor starred in a particular movie, yet be quite un- 
able to think of the actor’s name if asked who the star of the movie 
was. Once again, it is probably more complex cognitively to con- 
struct or to complete what we think is a true statement than sim- 
ply to recognize the statement as true when it is produced by 
someone else (see Anderson, 1980). In an article entitled “Good 
Morning, Mr ... er,” Burton (1992) examined how it might be that 
we can recognize a face without being able to recall anything else 
about a person and that we may remember almost everything else 
about someone — their occupation, nationality, where they live — 
but still not get their name. 
The idea that memory is constructive, or reconstructive, rather 
than a simple recall of original information, also has a long history 
in psychology, with its own classic by Bartlett (1932). See Smith 
(1990) for an extended discussion of the role of imagination in 
learning, comprehension, and thinking as well as memory, and 
Morris (1988) for descriptions of our remarkable memory capac- 
ity for things we are interested in, like sport scores. Reber (1989) 
showed that there are massive amounts of implicit learning, with- 
out awareness, and Bahrick and Hall (1991) demonstrated that 
the longer the period something is studied or experienced, the lon- 
ger it is remembered. We remember names of people we know for a 
few years better than names we have known for a few weeks. People 
who do math in college can remember for half a century what they 
would soon forget from high school. Foreign language vocabulary 
fades within 3 years of one college semester, but over 60% is re- 
tained 25 years later with 5 college semesters of study. 
272 
UNDERSTANDING READING 
Chunking 
The two apparent bottlenecks of memory — the limited capacity of 
short-term memory and the slow entry into long-term memory — 
can both be circumvented by the strategy known as chunking, or 
organization of information into the most compact (most meaning- 
ful) unit. For example, it is easier to retain and recall the sequence 
of digits 1491625364964 when they are recognized as the first 
eight square numbers, or the letters JFMAMJJASOND as the ini- 
tials of the months of the year, than to try to remember either se- 
quence as a dozen or so unrelated elements. But it is a mistake to 
think that we normally perceive first and chunk afterward; we no 
more read the letters h,o,r.s, and e, which we then chunk into the 
word horse, than we perceive a particular nose, ear, eye, and 
mouth, which we then chunk into a friend’s face. Chunking re- 
search and instruction both tend to get things backward, starting 
with arrays of ostensibly unrelated elements that the individual is 
supposed to group together in some meaningful way. In practice, 
prior knowledge and expectation of the larger grouping lead to the 
perception of elements in a chunked manner — if we recognize a 
word, we don’t see the individual letters. The size or character of a 
chunk is determined by what we are looking for in the first place. A 
remarkable report of how a college student increased his short- 
term memory span from 7 to 79 digits with 230 hours of chunking 
practice over 20 months is contained in Ericsson, Chase, and Fal- 
con (1980). 
Imagery 
One important and common means of chunking is to employ imag- 
ery to remember; there is a substantial literature demonstrating 
the unsurprising fact that our recall of particularly graphic sen- 
tences that we have heard or read is more likely to be related to 
scenes that we imagine from descriptions provided by the words 
than to the words themselves (Barclay, 1973; Sachs, 1974). But al- 
though we remember some sequences of words in terms of the pic- 
tures they conjure up, we also often remember scenes or pictures 
in terms of their descriptions. We recall a scene of birds flying over 
a town but not whether they were seagulls or pigeons, nor how 
many there were. There is nothing remarkable about any of this; 
We naturally try to remember in the most efficient manner possi- 
ble. If a scene is easiest to remember, or most efficiently remem- 
bered, in terms of a description, because perhaps we are interested 
NOTES 
273 
in particular things rather than in the scene as a whole, then the 
memorization will proceed accordingly. Not only is our recall influ- 
enced by the way we learned or perceived in the first place, but the 
manner of memorization will tend to reflect the most probable way 
in which we shall want to recall or use the information in the fu- 
ture. Sadoski, Goetz, and Kansinger (1988) reported that readers 
spontaneously generate images as they read. Long, Winograd, and 
Bridge (1989) agreed, especially when the text is interesting and fa- 
cilitates image creation, which improves comprehension, memory, 
thinking — and enjoyment. Wilson, Rinck, McNamara, Bower, and 
Morrow (1993) added that readers will — when the text permits it 
and especially if a task requires it — construct mental models (like 
the floor plans of buildings) as they read about them. 
Children's Memory 
There is no evidence that children have poorer or less well-devel- 
oped memories than adults. Simon (1974) argued that children 
have the same memory capacity as adults but don’t chunk as effi- 
ciently; however, there is probably an adult bias behind the notion 
of chunking "efficiently.” We tend to chunk— or to perceive and re- 
member in rich meaningful units — that which is most rich and 
meaningful to us. Recall of strings of unrelated letters and digits, 
which is the test by which children are usually judged to have 
memories inferior to those of adults, is not the most meaningful 
of tasks, for children especially. The number of digits a child can 
repeat after a single hearing increases from an average of 2 at the 
age of 2 Vi to 6 at the age of 10 (and 8 for college students). But 
rather than suppose that children’s memory capacity grows with 
their height and weight, one can argue that the younger children 
have had little experience, and see little sense, in repeating se- 
quences of numbers, especially before they have become accus- 
tomed to using the telephone. The memory span of adults can be 
magically increased by teaching them little tricks or strategies, 
for example, to remember strings of numbers not as single digits 
(2. 9. 4, 3, 7, 8 ...) but as two-digit pairs (29, 43, 78 ...). Practice 
improves performance on any memory task but doesn't seem to 
improve memory beyond the particular skill into unrelated areas 
or activities. The best aid to memory for anyone of any age is a gen- 
eral understanding of the structure and purpose behind the re- 
quired memorization. If chess pieces are arranged as part of an 
actual game, skilled players can recall the layout of most or all of 
the pieces on aboard after just a couple of glances although begin- 
274 
UNDERSTANDING READING 
ners can remember the positions of only a few pieces. But if the 
pieces are organized randomly, then the skilled player can re- 
member no more than the beginner. 
NOTES TO CHAPTER 7, 
LETTER IDENTIFICATION, pp. 110-124 
Recognition Versus Identification 
G. Mandler ( 1980) made a similar distinction to that made in this 
chapter between identification (putting a name to something) and 
recognition (deciding that something is familiar). He proposed a 
general theory of word recognition relevant to chapter 9 of this 
book. Benton (1980) discussed the remarkable human ability to 
recognize patterns (in this case, faces) years after perhaps only a 
single partial glimpse. 
Theories of Pattern Recognition 
Pinker (1984) gave a basic but technical examination of contempo- 
rary theories of visual perception, including template, feature, and 
other more complex models, and also a discussion of the nature 
and role of imagery. Pinker pointed out that there are problems 
with all theories — a horse, for example, would seem to consist of 
too many lines and curves to be easily recognized by "features" 
alone, yet it is inadequate to say that a horse must therefore be rec- 
ognized by body parts like hooves, because the parts themselves 
would have to be recognized by features. The current alternative is 
to rely on “massively parallel" models, which search concurrently 
for numbers of features and for interrelationships among them. 
On the other hand, perhaps to show that pattern recognition 
should not be complicated and mystified out of proportion. Blough 
(1982) showed that pigeons can easily be taught to distinguish the 
letters of the alphabet. When shown a particular letter on one side 
of a screen, they had to distinguish it from two incorrect alterna- 
tives elsewhere on the screen by pecking at the correct alternative. 
The pigeons, which were hungry, were rewarded with 3 seconds of 
eating mixed grain for every 4 successive correct answers they gave 
in 2,700 test trials over 4 days. When they made mistakes, the 
NOTES 
275 
birds demonstrated the same confusions as humans, for example, 
C-G-S, M-N-W, and D-O-Q. 
It is not suggested that we have visual analyzers that function 
solely to collect information about letters of the alphabet. Informa- 
tion used in letter identification is received from analyzers involved 
in many visual activities, of which those concerned with reading 
are only a small part. The same analyzers might contribute infor- 
mation in other circumstances to the identification of words, dig- 
its, geometric forms, faces, automobiles, or any other set of visual 
categories, as well as to the apprehension of meaning. The brain 
makes a variety of specialized uses of very general receptor sys- 
tems; thus statements can be made about analyzers “looking for” 
alphabetic features without the implication that a benign provi- 
dence has “prewired” us to read the alphabet. We all have a “biolog- 
ical inheritance” that enables us to talk and read, to ride a bicycle 
and play the piano, not because of some specific genetic design, but 
because spoken and written languages, bicycles and pianos, were 
progressively developed by and for human beings with precisely 
the biological equipment that humans are born with. For an early 
attempt to specify possible features of English letters, see Gibson 
(1965). 
Incidentally, it is just as appropriate to talk about distinctive fea- 
tures of speech as it is to refer to distinctive features of written lan- 
guage. In fact, the feature model for letter identification that was 
developed in the 1960s was inspired by a feature theory of speech 
perception published in the 1950s (Jakobson & Halle, 1956). In 
both theories a physical representation, acoustic or visual, is 
scanned for distinctive features which are analyzed in terms of fea- 
ture lists that determine a particular categorization and perceptual 
experience. The number of physical features requiring to be dis- 
criminated will depend on the percipient’s uncertainty and other 
sources of information about the language (redundancy) that can 
be utilized. 
Just as the basic elements of the written or printed marks on a 
page are regarded as distinctive features smaller than letters, so 
elements smaller than a single sound are conceptualized as dis- 
tinctive features of speech. Distinctive features of sounds are usu- 
ally regarded as components of the process by which a phoneme 
is articulated, such as whether or not a sound is voiced (whether 
the vocal chords vibrate as for /b/, /d/, /g/ compared with /p/, /t/. 
/k/), whether the sound is nasal { like /m/ and /n/), the sound's du- 
ration , and the position of the tongue. Each distinctive feature is a 
significant difference, and the discrimination of any one feature 
276 
UNDERSTANDING READING 
may eliminate many alternatives in the total number of possible 
sounds (the set of phonemes). Every feature cuts the set of alter- 
natives in a different way, so that theoretically a total of only six 
distinctive features could be more than enough to distinguish 
among 40 alternative phonemes (2 6 = 64). There are many analo- 
gies between the distinctive features of print and those of speech. 
The total number of different features is presumed to be much 
smaller than the set of units that they differentiate (26 for letters, 
about 40 for sounds). The number of features suggested for pho- 
nemes is usually 12 or 13 (note again the redundancy). Phonemes 
can be confused in the same manner as letters, and the more 
likely two sounds are to be confused with each other, the more 
distinctive features they are assumed to share. Some sounds, 
such as Pol and /d/, which probably differ in only one feature, are 
more likely to be confused than /b/ and IX], which differ in perhaps 
two, and /t / and /v/, which may differ in three features. Spoken 
words may also differ by only a single feature. Ban and Dan, 
which have only a single feature's difference, should be rather 
more likely to be confused than ban and tan, and much more 
likely than tan and van; experimental evidence suggests that as- 
sumptions of this kind are correct (Miller & Nicely. 1955). 
The perception of speech is no less complex and time-consum- 
ing than that of reading; what we hear is the end product of a deci- 
sion-making procedure that leads lo the identification (the 
categorization) of a sound or word or meaning prior to the percep- 
tual experience. We rarely “hear" words and then identify them; the 
identification must precede the hearing, otherwise we would just 
hear noise. And we don’t hear distinctive features of sound any 
more than we see distinctive features of written language: the unit 
that we are aware of discriminating is determined by the sense that 
the brain is able to make, the kind of question it is asking. Usually 
we are aware only of meaning for both spoken and written lan- 
guage. Occasionally we may attend to particular words, but in spe- 
cial circumstances we may become aware of the surface structure 
phonemes or letters. The features themselves evade our awareness 
completely. 
NOTES 111 
NOTES TO CHAPTER 8, 
WORD IDENTIFICATION, pp. 125-137 
Template and Feature-Analytic Theories 
In a long and technical summary of template and feature analytic 
theories of shape recognition, Hummel and Biederman (1992) 
stressed the need for “structural descriptions” that include rela- 
tionships among parts rather than just point-by-point compari- 
sons. The authors say that enormous numbers of units and 
connections are required to “bind” all the parts of any figure to- 
gether into possible wholes that can be recognized from different 
angles. Their solution for computer recognition of a simple figure 
(a cone on a rectangular block) involves a seven-layered network of 
activating or inhibiting cells, with the sensitivity of each varying ac- 
cording to experience. Obviously, any attempt to specify (or teach) 
distinctive features and their relationships among letters or words 
is bound to be an oversimplification. 
Letter Identification in Words 
It has long been known that readers can make use of redundancy 
among distinctive features in words. Smith (1969) projected let- 
ters or words at such a low intensity that there was barely any con- 
trast with the background on which they were shown, and then 
increased the contrast slowly, gradually making more and more vi- 
sual information available until observers were able to make iden- 
tifications. Under this procedure, viewers are not constrained by 
time or memory limitations and may choose to make either word 
or letter identifications with the information available at any mo- 
ment. They typically identify letters within words before they say 
what an entire word is, although the entire word may still be identi- 
fied before any of its letters could be identified in isolation (see also 
Wheeler, 1970). This finding is not inconsistent with the classic evi- 
dence that words can be identified before any of their component 
letters in isolation, but it does make clear that words are not recog- 
nized all-or-none “as wholes” but by analysis of their parts. The se- 
quential redundancy among features that exists within word 
configurations permits identification of letters on fewer features 
than would be required if they were presented in isolation. The let- 
ter ft. for example, requires fewer features to identify if presented 
in the sequence hat than if presented alone, even if the reader iden- 
tifies the ft before the at. The additional information that enables 
278 
UNDERSTANDING READING 
the earlier letter identification to be made in words is based on or- 
thographic redundancy in the spelling of words, reducing the un- 
certainty of letters from over 4 bits (26 alternatives) to less than 3 
(about 7 alternatives), as discussed in the notes to chapter 4. Even 
if a reader has not discriminated sufficient features in the second 
and third positions of the configuration hat to identify the letters 
at, there is still some featural information available from those po- 
sitions, which, when combined with nonvisual information about 
featural redundancy within words, permits identification of the let- 
ter in the first position on minimal visual information. 
There is other evidence that although words are identified “as 
wholes,” in the sense that featural information from all parts may 
be taken into account in their identification, they are by no means 
identified on the basis of the familiarity of their shape or contour. 
Examples were given earlier in this chapter of the ease with which 
quite unfamiliar configurations like rEaDiNg could be read. Entire 
passages printed in these peculiar configurations can be read 
about as fast as normal text (Smith, Lott, & Cronnell, 1969). In fact 
if the size of the capital letters is reduced slightly so that they don’t 
interfere with the discriminability of the lower case letters, for ex- 
ample, rEaDiNg, then there is no difference at all in the rates at 
which such words and normal text are read. The facility with which 
we can read passages of handwriting when individual letters and 
even words would be indecipherable on their own is further evi- 
dence that reading doesn’t depend on letter identification. 
Graphic support for the assertion that we attend to features 
rather than letters in words is provided by a recent snippet on the 
Internet. I haven’t been able to trace the original source, but the text 
itself illustrates the point that it makes. It reads, "Acocdrnig to an 
elgnsih unviesitry sutdy the oredr of letetrs in a word dosen’t 
mttaer, the olny thnig thta's iopmrantt is that the frsit and lsat 
ltteer of eevry word is in the crrecot ptoision. The rset can be 
jmbueld and one is stlil able to raed the txet wiohtut dclftfuiiy." 
It is obviously an oversimplification to talk about the relative 
“discriminability” of individual letters of the alphabet or to as- 
sume that letters difficult to identify when standing alone must be 
difficult to perceive when in words. This argument applies espe- 
cially to the issue of "reversals” of letter pairs like b and d, which 
are particularly bothersome to some children, to adults when 
perception is difficult, and also, apparently, to pigeons (Blough. 
1982). Reversals were specifically considered in chapter 5. The 
amount of visual information required to identify a letter has rela- 
tively little to do with the physical characteristics of the actual 
NOTES 
279 
configuration but depends much more on the reader’s experience 
and the context in which the letter occurs. And precisely the same 
kind of argument applies to words. Children learning to read can 
often identify words in context that they can’t identify in isolation 
(Pearson & Studt, 1975). It is misleading to talk of children’s 
word identification ability in terms of their “sight vocabulary” or 
word attack skills. 
Use of Redundancy by Children 
There is no evidence that children need to be trained to seek or use 
redundancy in any way; perception naturally involves the use of 
prior knowledge and the youngest children demonstrate ability to 
limit uncertainty by eliminating unlikely alternatives in advance. 
Studies with young readers have found ability to use sequential re- 
dundancy very early (Lott & Smith, 1970). First-grade children 
who had had a limited amount of reading instruction in kindergar- 
ten showed themselves able to identify letters in words on less vi- 
sual information than when letters only were presented. For 
children in fourth grade the difference between the information on 
which letters were identified in words and that on which the same 
letters were identified in isolation was equal to that of skilled adult 
readers, indicating that for familiar three-letter words at least, 
fourth graders could make as much use of sequential featural re- 
dundancy as adults. Krueger, Keen, and Rublevich (1974) subse- 
quently confirmed that fourth-grade children may be as good as 
adults in making use of redundancy among letter sequences in 
words and nonwords. 
Distributional Redundancy Among Words 
The sequential redundancy that exists among words in text — 
which has a critical role in making reading possible — is discussed 
in the notes to chapter 10. But there is also a distributional redun- 
dancy among words, reflecting the obvious fact that some words 
are used far more often than others. The distributional redun- 
dancy of English words has not been formally calculated but is 
probably related to the maximum theoretical uncertainty of be- 
tween 15 and 16 bits for a set of about 50,000 alternatives and the 
actual 12-bit uncertainty of isolated words computed by Shannon 
(1951) and discussed in the notes to chapter 4. Distributional re- 
dundancy among words complicates experimental studies of word 
identification because more frequent words are usually identified 
280 
UNDERSTANDING READING 
faster, more accurately, and on less visual information than less 
frequent words (see p. 258). 
NOTES TO CHAPTER 9, 
PHONICS AND MEDIATED WORD IDENTIFICATION, 
pp. 138-155 
A Defining Moment 
I’ll begin by reviewing the meaning of some terms that are widely but 
not always consistently used by reading theorists and researchers: 
phones, phonology, phonetics, phonemes, and phonics. 
The natural sounds of speech are as turbulent and intermingled 
as the waters of a rushing stream. Nevertheless linguists and other 
scientists attempt to isolate basic elements of spoken language. 
The smallest unit they isolate is called a phone, which is not any- 
thing that anyone can normally detect or reproduce. Phones are 
like atoms; they don’t exist in isolation and are modeled more by 
caricatures than actual replicas. The specialized study of phones 
and their production is called phonology or phonetics, the con- 
cern of phonologists or phoneticians — pairs of terms that are usu- 
ally synonymous. Some educators believe that foreign language 
students need to know something about phonology; a few even 
think beginning readers need knowledge of this kind (although 
generally they don’t mean what they say — their belief is that begin- 
ning readers need to know something about phonemes). 
Phonemes are perceptual rather than physical phenomena, ab- 
stract composites of phones. They are defined as the smallest units 
of sound that differentiate one spoken word from another. Thus 
the initial sounds of tip and dip are phonemes, and so are the final 
sounds of see and say. The /XJ phone at the beginning of tip is not 
the same as the /t / phone at the beginning of top (the following 
sound makes a difference to both of them), but because these two 
different /XJ phones don’t differentiate one word from another — 
perceptually they are the same — they are not regarded as different 
phonemes. The number of phones in most languages exceeds 100, 
but the number of phonemes is much fewer — about 45 in English, 
depending on the analyzer and the dialect. Neither phones nor 
phonemes exist in written language, nor do they correspond di- 
NOTES 
281 
rectly with the letters of alphabetic written languages. To refer as I 
have done to a phone or phoneme as /t/ is conventional but may be 
misleading. The sound represented by the symbol /t/ is a product of 
the vocal system; it is not a letter and should not be confused with 
one. Phonologists and phoneticians use an extended cast of char- 
acters to denote sounds, such as e, c[> and fe. which at least don’t 
look as if they have anything to do with reading. 
Phonics is concerned with correspondences between phonemes 
and graphemes, which have nothing to do with speech but are 
characters in writing systems. Phonics is not a scientific study at 
all, but a method of instruction, specifically concerned with teach- 
ing children relationships between graphemes and phonemes (or 
between letters and sounds), the utility of which is discussed 
throughout chapter 9. Phonics is sometimes incorrectly referred 
to as phonetics, just as “phonemic awareness” may also be called 
“phonological awareness.” It can all be very complicated. 
The Relevance of Phonics 
The analysis of the relationship between the spelling of written 
words and the sounds of speech is mainly derived from the work 
of a group of researchers (Berdiansky, Cronnell, & Koehler, 
1969) associated with the Southwest Regional Laboratory 
(SWRL), a federally sponsored research and development center 
in California. Researchers at SWRL and many other federally 
funded institutions have continued to analyze the maze of spell- 
ing-sound correspondences in English, and to devise instruc- 
tional programs to teach these correspondences to children in 
the expectation that it will make them better readers and spell- 
ers. The 58th in the series of SWRL technical reports (Rhode & 
Cronnell, 1977) provided an analysis of a 10.000-word lexicon 
that they consider to be the basis of a kindergarten through 
sixth-grade communication skills program. This time, words of 
three or more syllables are included, a total of 27% of the 
10,000. Ninety-nine grapheme units are distinguished. 77 re- 
lated to 225 spelling-sound correspondence “rules” and 22 to 
32 “exceptions.” Eighty of the rules are associated with 48 con- 
sonant grapheme units. 111 with the six primary vowels 
(a/e/i/o/u/y), and 34 with 23 secondary vowel units (ai/au/etc.). 
Computer programs that convert written to spoken language do 
so not by reliance on phonic rules (the last resort) but by storing 
the complete sounds of thousands of written words. And even 
then such devices have difficulty selecting the appropriate pro- 
282 
UNDERSTANDING READING 
nunciation of the 200 or more homographs in English — com- 
mon words like wind, tear, read, and live — which require 
syntactic and semantic knowledge to disentangle them. 
The supposed value of phonics as a useful method of instruc- 
tion has long been challenged. Clymer (1963a) told of problems of 
trying to teach a student a list of phonic generalizations when the 
student kept pointing out exceptions. Clymer analyzed 45 gener- 
alizations and concluded that 18 might be useful — and they de- 
pended on local dialect. For the most common “rule" — that when 
two vowels go walking, the first does the talking — he found 309 
conforming words (bead) and 377 nonconforming (chief), and 
politely concluded that many commonly taught generalizations 
were of limited value. Elsewhere in the same issue he commented 
that because children can learn to read under a particular set of 
materials doesn’t mean they should (Clymer, 1963b). Johnston 
(2001 ) reanalyzed Clymer’s results and claimed that some broad 
generalizations are more applicable when broken down into spe- 
cific letters — but then they are more complex to learn. She con- 
cludes that words are recognized and remembered more on the 
basis of patterns than on rules about sounds. See also Johnston 
(2000/2001) and Krashen (2002). 
The classic volume on the relevance of phonics instruction is the 
frequently cited Learning to Read: The Great Debate (Chall, 
1967), although the discussion is more about how reading is 
taught than how it is learned, and the conclusions may not be 
thought to follow inevitably from the evidence presented. The con- 
troversy, often reduced to a question of “code emphasis” (phonics 
exercises) versus “whole language” (meaningful texts), is discussed 
in chapter 13 and its notes. 
Phonological and Phonemic Awareness 
The mediated identification of written words through the blending 
(aloud, silently, or subconsciously) of sounds supposedly repre- 
sented by their letters is frequently referred to as phonological 
recoding, or sometimes phonological decoding. The fact that 
many children have difficulty learning phonics, and difficulty 
learning to read despite intensive instruction in phonics, has led 
many researchers favoring the phonological recoding view to as- 
sert that such children suffer a deficiency related to spoken lan- 
guage perception. (It is not uncommon for difficulties children 
might experience in learning to read to be attributed to "deficits" on 
their part, rather than to externally produced factors such as con- 
NOTES 
283 
fusion, depressed interest, or learning the wrong thing. ) According 
to a widespread view, generally attributed to Isabelle Liberman and 
her colleagues at the Haskins Laboratories (Liberman & 
Liberman, 1992), the problem with such children is that they are 
unable to identity in spoken words the discrete sounds — or “pho- 
nological segments” — represented in writing by the letters of the al- 
phabet. This ability to analyze hypothetical sounds has nothing to 
do with the understanding of speech, but is necessary for reading, 
according to this point of view. 
The term phonemic awareness has gradually taken over from 
phonological awareness in recognition of the fact that the concern 
is with the identification of abstract phonemes rather than actual 
phones in sound. A major problem with all these analyses of the 
phonemic structure of spoken words is that the “units” are com- 
pletely arbitrary and hypothetical — they don’t exist as distinct ele- 
ments in spoken language, in either the perception or the 
production of speech (Liberman & Liberman, 1992). Children 
find it difficult to understand the statement that bet consists of 
three elements and almost impossible to understand that best 
has four (at least until they are readers). The separate sounds that 
letters are supposed to represent and that beginning readers are 
supposed to be “aware of” are fictions. Sounds are “coartic- 
ulated” in the production of speech — at any moment in saying the 
word bet the speaker might be physically producing part of the 
/b/, /e/ and /t/ simultaneously. It is impossible to separate the three 
sounds electronically — or by trying to snip independent sections 
out of a tape recording. The “internal structure” of speech can’t be 
displayed. 
So although children may in some sense be aware of the fact that 
pairs of words rhyme, or begin with the same sound, or are “the 
same” or “different” in some way, to expect them to isolate the 
sounds in speech is a purely artificial task, based solely on the con- 
ventions of written (alphabetic) language. The code emphasis be- 
comes not so much a matter of finding speech structures in 
writing, but of writing structures in speech — not something that is 
likely to be very easy or meaningful for children before they can 
read, and of limited utility afterward. Phonemic awareness might 
be regarded as another skill, like phonics, that children may be re- 
quired to learn as a separate subject from reading. It all falls apart 
if reading is perceived to be the interpretation of written language 
directly, rather than through speech. 
Another prolific and influential proponent of the view that read- 
ing proceeds on the basis of continual grapheme-phoneme analy- 
284 
UNDERSTANDING READING 
sis, and that phonics and phonemic awareness skills should be 
taught directly, is Ehri (1997). Despite concessions that “sight 
word reading" might be necessary for “irregular” spellings like 
none, calf, break, and prove, and demonstrations that individual 
words are identified immediately even if the reader intends to ig- 
nore them (because the words are printed over pictures that the 
readers are instructed to focus on and identify), she maintained 
that grapheme-phoneme connections are made and used “auto- 
matically," which means subconsciously, without any observable 
indication (p. 167). All this is accomplished by a "connection-form- 
ing mechanism." Ehri also talked of the importance of “practicing 
reading” to meet unfamiliar words and add them to the lexicon (p. 
179). Theorists with a skills orientation often talk of children 
“practicing reading" rather than simply “reading.” When I first en- 
countered the expression I was tempted to practice laughing. 
More articles with a similar approach are included in Lehr and 
Osborn (1994), who entitle their volume Reading, Language and 
Literacy: Instruction for the Twenty-First Century (without ex- 
plaining exactly why instruction had to change with the advent of 
the present century). A typical chapter in the volume is by Adams 
(1994), who wrote, “The value of phonics instruction has been 
demonstrated with sobering consistency across literally hundreds 
of studies” (p. 3), without citing most of these studies, and subse- 
quently admitted, “The phonics advantage documented by this re- 
search is neither awesomely large nor comfortingly reliable" (p. 4). 
Only 38 of these “hundreds” of studies made the cut for the Na- 
tional Reading Panel report (discussed in the chapter 13 notes). 
Wolf and Katzir-Cohen (2001 ) reported that research into "pho- 
nological processes” has been insufficient to explain fluent reading 
and why it breaks down. Their view was that explicit instruction is 
required “to link phonological, orthographic, semantic and mor- 
phological processes to sub-lexical and word-level subskills” (p. 
229), all of which must become automatic. The view that phonetic 
recoding takes place “automatically," at a subconscious level, even 
in fluent reading, is an almost Freudian theory that observable be- 
havior is determined by specific processes so deep in our uncon- 
scious that they are undetectable. It is like saying that in our 
unconscious we still put our feet on the ground when we cycle be- 
cause we did so a few times when we were learning to ride. 
In a footnote in their book on Phonological Skills and Learning 
to Read, Goswami and Bryant (1990) admitted coming to the "un- 
comfortable conclusion" that phonological awareness doesn't 
seem to be of much use to children learning to read (p. 46). An ex- 
NOTES 
285 
tensive review of research had demonstrated “that there is very lit- 
tle direct evidence that children who are learning to read do rely on 
letter-sound relationships to help them read words” (p. 46). On the 
other hand, they noted “a great deal of evidence that these young 
children take easily and naturally to reading words in other ways 
. . . they either recognize the word as a pattern or remember it as a 
sequence of letters” (p. 46). Goswami and Bryant’s findings have 
not deterred others from employing the concept of phonological 
awareness in their theories of reading and pronouncements about 
instruction, often while citing the work of Goswami and Bryant but 
disregarding their 1990 conclusion — see Rieben and Perfetti 
(1991), Brady and Shankweiler (1991), and Gough, Ehri, and 
Treiman (1992) for numerous examples. The latter volume even 
includes a chapter by Goswami and Bryant ( 1992) themselves, cit- 
ing their 1990 publication but not their uncomfortable conclusion. 
They still advocate teaching phonological awareness because chil- 
dren attempting to identify new words by analogy make extensive 
use of rhyme (see later discussion). Even in Finnish, a language 
that is supposed to have almost perfect sound-symbol correspon- 
dence for its alphabetic writing system, teachers have found that a 
heavy emphasis on phonics instruction may confuse children 
about the nature of reading and cause comprehension problems 
later (Korkeamaki & Dreher, 1993). 
Scholes and Willis (1991) found that phoneme deletion — an- 
swering questions like “What is left if you remove the /k / sound 
from cat?” — can’t be done by nonreaders: it requires literate 
knowledge of alphabetical writing. Read, Yun-Fei, Hong-Yin, and 
Bao-Qing (1986) deduced from studies with adult Chinese read- 
ers that ability to manipulate “speech sounds” depends on know- 
ing alphabetic writing. Liberman and Liberman (1992) 
themselves estimated that perhaps as many as 75% of children 
will “discover the alphabetic principle ... no matter how unhelpful 
the instruction” (p. 345). It might be wondered whether the re- 
maining children are simply the victims of unhelpful instruction. 
As stated in the main text (p. 146), “phonemic awareness” refers 
to a supposed ability to divide spoken words (or artificial words) 
into discrete sounds, represented by letters of the alphabet, alone 
or in combination. This is the complex system of "sound-spelling 
relationships” analyzed in chapter 9. However, the flow of speech is 
not broken down into discrete words (see p. 34), and the sounds 
that constitute speech are complexly produced and interrelated by 
various parts of the entire vocal apparatus (vocal cord, throat, 
mouth, tongue, nasal passages) and can't be isolated into discrete 
286 
UNDERSTANDING READING 
units, either by people or by electronic equipment. In other words, 
phonemes are artifacts, invented to correspond with the letter pat- 
terns of written words. They can only be detected by people who 
can read, or who have been specially trained with a limited set of 
sounds for a limited array of letters, and have no conceivable role 
in reading. 
Why does phonemic awareness survive as a concept? Because it 
sounds reasonable to anyone believing that reading is a matter of 
decoding print to speech (the fascination with the alphabet), be- 
cause it seems to account for the failure of phonics instruction, and 
because contrived “research” into the impossible relationship can 
be claimed to be “scientific” (and other research “unscientific”). 
Krashen (1999b) reviewed 15 studies of phonemic awareness 
training and found it had greater effect on tests of phonemic aware- 
ness than on tests of real words and reading comprehension. He 
has also reported that many children and adults with low phone- 
mic awareness learn to read, as a result of massive amounts of in- 
teresting and comprehensible reading (Krashen, 2001a, 2001b, 
2001c). In a detailed and impassioned critique of what she called 
the “Spin Doctors of Science,” Taylor ( 1998) condemned many as- 
pects of studies claiming to support phonemic awareness and 
massive phonics, from the statistical and sampling techniques in- 
volved to conclusions drawn from data presented, not to mention 
use made of those conclusions in media sound bites, and by politi- 
cians and the publishing industry. 
Finally, it might be fair to consider what children think about 
phonemic awareness training. Castiglioni-Spalten and Ehri 
(2003), in a routine report on the effects of various kinds of word- 
segmentation training with 5-year-old children, noted a marked re- 
luctance on the part of the children to cooperate or to pay attention. 
In a pilot study the experimenter, who followed a “fully prescribed 
and clearly scripted” training procedure, recorded “several off- 
task and resistence behaviors committed by students: refusing to 
use the mirror (used in part of the training): leaving their seats 
without permission; playing with the blocks by building a tower, 
house, or train; throwing the blocks on the floor; talking about ex- 
traneous topics; interacting with others in the room; and express- 
ing reluctance to finish the instruction" (p. 36). During the actual 
study, “students rarely committed such behaviors more than twice 
because the experimenter discouraged them” (p. 43) — by remind- 
ing students that she would report back to their teacher about how 
well they did . . . and also by using a screen to isolate the children 
from distractions in the room (p. 36.) It would be hard to find a 
better description of induced boredom. 
NOTES 
287 
Word Identification by Analogy 
The argument in the chapter that unfamiliar words may often be 
identified by analogy doesn’t entail that phonological recoding is 
necessary for analogies to be made. Analogies can be made on the 
basis of visual characteristics — the beginning of medication looks 
like the beginning of medical. Goswami and Bryant (1990; see also 
Goswami, 1986, 1990) held to the view that learners confronted by 
words they are unable to identify immediately frequently attempt to 
do so by piecing together the known sound of the initial consonant 
or consonant cluster (the “onset”) with the rhyming sound of the rest 
of the syllable of a similarly spelled word. The unknown word teak , 
for example, is likely to rhyme with the known word beak and fright 
with night. The view is usually referred to as “onset and rime” (em- 
ploying the archaic spelling to better exemplify the basic assertion 
that words that rhyme often share similar spelling patterns). The 
onset-rime unit is seen as intermediate in size between individual 
phones and syllables, more predictable from its spelling than a sin- 
gle phoneme. See also Treiman (1992) and Treiman, Goswami, and 
Bruck (1990) for reviews of studies of children’s abilities to detect 
rhymes, alliteration, and other partial aspects of spoken words. 
However, Savage (2001) reevaluated claims relating rhymed 
sounds and reading, and concluded that the issue must remain 
controversial; any apparent correlations may be due to ortho- 
graphic analogies (i.e., the visual structure of words) rather than 
the sounds. Goswami responded in the same issue (Goswami. 
2001). See also Macmillan (2002). Christensen (1997) reviewed 
studies of relationship of onset and rhyme and phonemes, and 
concluded that "Overall, the picture of learning to read that 
emerges from this study is one of diversity and complexity. 
Children appear to acquire a range of phonological skills and liter- 
acy-related knowledge in a variety of ways.” (p. 357). Like many 
publications, this article talks about the identification of new 
words in the context of “learning to read,” that is, effects attribut- 
able to different kinds of instruction, rather than with what chil- 
dren learn as a consequence of reading, independently of whatever 
instructional regime they may be put into. 
Spelling 
Goswami and Bryant (1990), whose uncomfortable conclusion 
that phonological awareness may not help children learn to read 
has been noted, nevertheless proposed that such instruction 
should be continued because it will help them to spell. Unfortu- 
288 
UNDERSTANDING READING 
nately, children who spell words the way they are pronounced, as 
most children do naturally at the beginning (Read, 1971), spell 
poorly. Spelling by rule is also not an efficient strategy for spelling 
most common words (Brown, 1970). Although clues may be ob- 
tained from known words with similar meanings, the main re- 
quirement for good spelling is to remember individual spellings 
(Smith, 1994), a requirement that is not relevant to helping chil- 
dren learn to read. Gillooly ( 1973) argued that there is no justifi- 
cation for attempts to change the current spelling of English, 
which he said increases reading speed and is nearly optimum for 
learning to read. Venezky ( 1999) gave a scholarly analysis of the 
complexity and history of American English spelling, skewed by 
his reiterated belief that some “rules” must be taught. 
Spelling is conspicuous, and probably the only aspect of writ- 
ing that most people feel competent to pass judgment on, so er- 
rors are treated almost as antisocial behavior. As a consequence 
of this sensitivity, there is a widespread perception that most peo- 
ple (including often ourselves) spell badly. But in a review article 
entitled “How Well Do People Spell?” Krashen (1993a) reported 
that in self-generated writing, college freshmen achieved an accu- 
racy of from 97.9% to 99.8% of the words written, although soci- 
ety demands perfection. He commented that it is unlikely that 
such high levels of competence occur from formal instruction; 
ability to recite rules doesn’t make students any better spellers, 
nor does having errors pointed out. He concluded that the ability 
can only come from reading. 
In an earlier review. Krashen and White (1991) went back almost 
a century to reanalyze two classic studies (Rice, 1897; Cornman, 
1902) of the relation between spelling instruction and proficiency 
between Grade 3 and Grade 8, both of which found little difference 
as a result of instruction. Research through the 20th century sug- 
gested nothing different. The analysis raised the same doubts in 
the contemporary researchers that prompted the original stud- 
ies — whether the direct teaching of spelling is worth the trouble. 
NOTES 
289 
NOTES TO CHAPTER 10, 
THE IDENTIFICATION OF MEANING, pp. 156-177 
Effects of Meaningful Context 
Meaningfulness clearly has a substantial role in facilitating the 
identification of words in reading, reducing their uncertainty 
from at least 12 bits (the equivalent of 4,096 equiprobable alter- 
natives) for words in isolation to fewer than 8 bits (256 alterna- 
tives) for words in context (chap. 4 notes). It is really irrelevant to 
talk of letters at this stage — letters are not normally a concern 
when meaningful text is read. But as a yardstick, it is interesting 
to recall from the chapter 4 discussion that the uncertainty of let- 
ters falls from 4.7 bits to scarcely 1 bit when context is meaning- 
ful, enabling perception of four times as much of a line of print. 
Not only can twice as many words be identified in a single glance 
when they are in a meaningful context, thus overcoming bottle- 
necks of information processing and memory, but problems of 
ambiguity and the gulf between surface structure and meaning 
are removed by the prior elimination from consideration of un- 
likely alternatives. Context has its effect because it contributes in- 
formation that reduces the uncertainty of individual words 
through sequential redundancy ; it places constraints on what 
each individual word might be. Sequential redundancy is usable 
only if reflected in the prior knowledge, or nonvisual information, 
that the reader can bring to bear. That is why I stress that the con- 
text must be meaninqful, with all the relative connotations of that 
word. If a particular context is not comprehensible to a reader, or 
if for one reason or another the reader is reluctant to take advan- 
tage of it, then the context might just as well be nonsense, a ran- 
dom arrangement of marks on the page. 
Meaningful context as I have been using the term exercises its 
constraints on word occurrence in two ways, syntactic and seman- 
tic. These are two types of restriction on the particular words an au- 
thor can select — or a reader predict — at any time. (There are other 
constraints on authors, such as the limited set of words a reader 
might be expected to understand. ) Choice of words is always limited 
by what we want to say (semantics) and how we want to say it (syn- 
tax). I have not tried to separate the effects of syntax and semantics 
in this discussion for the simple reason that I have not found a good 
way to do so. The theoretical analysis of chapter 2 argued that the 
two are inseparable — that without meaning it is pointless to talk 
about grammar. (Also as outlined in chap. 2, words can also be con- 
290 
UNDERSTANDING READING 
strained by more general situational contexts; for example, there is 
a very small and highly predictable set of alternative words likely to 
occur on a toothpaste tube.) 
There is no shortage of research demonstrating the powerful 
facilitatory effect of meaningful context on word identification. In- 
deed. there is no evidence to the contrary. But I must reiterate that 
reading doesn’t usually involve or rely on word identification. 
Meaningful context makes reading for meaning possible and word 
identification unnecessary. The fact that meaningful context makes 
individual words easier to identify is basically as irrelevant as the 
fact that individual letters are easier to identify in words; the re- 
search evidence is merely a demonstration of the effect of meaning- 
ful context. Even when reading aloud is involved, so that the 
accurate identification of words is required, the prior apprehension 
of meaning is an important prerequisite. Reading aloud is difficult if 
prior comprehension is limited, and if word identification is given 
priority there will be interference with comprehension (Howe & 
Singer, 1975). Sharkey and Mitchell (1985) showed that word rec- 
ognition in meaningful contexts is frequently minimal and not very 
predictable, especially when the context is itself a predictable 
“script” — when it is a familiar narrative, in other words. Levy ( 1 978) 
reported that changing the wording but not the sense of written lan- 
guage did not affect a short-term memory task — meanings, not spe- 
cific words, are retained whenever possible, even for brief periods. 
Sometimes context leads us astray, evidence again for the potency of 
meaning and prediction. Carpenter and Daneman ( 1981 ) gave sev- 
eral examples of “garden path” texts where context misleads rather 
than facilitates, such as “Cinderella could not go to the ball. There 
were tears in her dress." Crowder and Wagner (1991) show that 
readers use meaning from other words to help identify specific 
words in text; see also Goldsmith-Phillips (1989). 
Experienced readers read familiar text at the highest possible 
level of comprehension, phrases if possible, then words — this ap- 
plies to Chinese and Japanese logographic readers as well as to 
English language alphabetic readers (Tao & Healy, 2002). Experi- 
enced readers of both systems make more small unit errors (like 
letters in English text) in familiar than unfamiliar larger units. We 
all attend to the broader picture rather than the unimportant de- 
tail. Masonheimer. Drum, and Ehri ( 1 984) observed that children 
may identify labels correctly even when a letter has been changed 
(like “Xepsi” for “Pepsi”). They reported this with some concern, 
because unlike the children, they were looking for mastery of let- 
ters, not of entire words or sense. 
NOTES 
291 
Context and Prediction 
Some theorists have argued that context and prediction play a 
smaller role in reading than proposed in this book. But they tend to 
use the terms more narrowly than I do. For example, Stanovich 
(1992) concluded from eye-movement studies that “context ef- 
fects” are minimal because visual information rapidly blurs out- 
side the foveal area in the center of the field of view. Earlier, 
Stanovich ( 1986) asserted that poor readers depend more on pre- 
diction than fluent readers. But I see context in terms of a reader’s 
understanding at any given point in the text of what it is about, en- 
hanced by what has already been seen behind and sometimes 
ahead of where the eyes happen to fall. Such advance knowledge, 
primarily nonvisual rather than visual, facilitates reading by re- 
ducing the reader’s uncertainty. There is a tendency to equate pre- 
diction with "guessing,” a term I would never use to describe what 
fluent readers do. My definition of prediction — the prior elimina- 
tion of unlikely alternatives — is precisely what makes experienced 
readers so effective when reading texts whose language and subject 
matter are familiar to them. Once again, everything hinges on the 
purposes of the reader and on the reader’s freedom to be flexible 
and selective. 
Various studies have demonstrated the readiness of children to 
make use of context in early reading (if they are so permitted), for 
example, Klein, Klein, and Bertino (1974), Golinkoff ( 1975/1976), 
Doehring (1976), and McFarland and Rhodes (1978). Rosinski, 
Golinkoff, and Kukish (1975) concluded that meaning is irresist- 
ible to children; it can interfere with performance on a task be- 
cause they can’t ignore it. Such studies also tend to show that use of 
context and reading ability increase together. This correlation is of- 
ten attributed to the fact that better readers can make more use of 
context; less often, the possibility is considered that use of context 
makes better readers. Studies of children’s misreadings tend to 
highlight the important fact that many of the errors, especially 
those made by better readers, preserve the meaning of the context 
and also that errors that make a difference to meaning are often 
subsequently corrected by children who are reading for meaning 
(and therefore are not errors that need be a great cause for con- 
cern). Children who read more literally, perhaps because of an em- 
phasis on “accuracy” during instruction, may, however, make 
nonsensical errors without being aware of them. 
The paradigm example of a child learning to read by meaning 
alone, without any possibility of decoding to sound, must be that 
292 
UNDERSTANDING READING 
of Helen Keller. Henderson (1976) reported the case of a deaf 
child who learned to recognize 4.400 printed words in 9 months 
at the age of 6 by relating them to manual signs in a meaningful 
context, basically in response to the child’s own spontaneous in- 
quiries. Ewoldt (1981) demonstrated that deaf children read in 
the same meaningful way as hearing children. 
"Dual Process" 
The most popular assumption about comprehension in experi- 
mentally based research on reading is centered on the notion that 
there is an “internal lexicon” in the head — a kind of mental dictio- 
nary — where the meanings of words are stored. Comprehension is 
achieved when definitions of the words that we read are “looked 
up” in the internal lexicon. The concept is a metaphor, not an expla- 
nation. It fails to explain how the internal definitions are under- 
stood and doesn’t allow for selectivity, prediction, and all the other 
things that readers normally do. The idea actually precludes cer- 
tain explanations — for example, that comprehension might on oc- 
casion be a pictorial or sensory image or a physical response, that 
comprehension is a state rather than a process, or that meaning is 
brought to language. It papers over the profundity of understand- 
ing. Yet the notion that such a lexicon actually exists is often taken 
for granted, and it is simply assumed that words are understood 
by being identified and “routed” to the internal lexicon. This is a 
great theoretical convenience, as it permits discussion of hypothet- 
ical “routes” between eye and understanding without any consider- 
ation of the nature of the terminus. Understanding can be ignored 
simply as “lexical access.” Among many discussions of the internal 
lexicon, see Rieben and Perfetti ( 1991 ). Miller and Fellbaum ( 1991 ) 
discussed the enormous problems of constructing an actual lexi- 
con of English — their database of 46,000 entries includes 26 dif- 
ferent hierarchical relationships for nouns, up to 10 levels deep 
with at least three different categories of distinguishing fea- 
tures — parts, attributes, and functions. Nevertheless such seman- 
tic decomposition is popular (see also Miller & Johnson-Laird. 
1975). Anotorious example is McCawley's ( 1968) “Kill = cause be- 
come not alive,” ridiculed together with all theories of innate gram- 
mar by Robinson (1975), who decried what he calls "linguistic 
atomism.” Fodor ( 1981 ) said most words are not decomposable. If 
people know 50,000 word meanings, there must be close to 
50,000 “primitive concepts." 
NOTES 
293 
Dual process theorists see two ways of getting from visual infor- 
mation to the internal lexicon — direct (which means an unmedi- 
ated loop between word and meaning, similar to the general point 
of view of the present book but not necessarily in the same direc- 
tion) and by phonological recoding. The idea is that the lexicon 
contains not only meanings but also spellings and pronunciations, 
information about syntactic functions, and possibly featural de- 
scriptions of various kinds. All these are alternative ways in which 
the lexicon can be entered, both for recognition and for production 
(writing and spelling). 
Most researchers say both routes are possible but put different 
emphasis on each. Some theorists, like Gough (e.g., Gough & 
Walsh, 1991), argue that only a phonological route to the lexicon 
exists: that reading is a process of decoding the “cipher” of English 
spelling. Ehri (1992) claimed that access is through the spelling 
patterns of words, without necessarily recoding into phonological 
form (which nonetheless usually becomes available with recogni- 
tion of meaning). Ehri called this the visual-phonological route, 
which is used even for “direct access." She cited evidence that chil- 
dren “learning to read by sight” find msk easier to learn and re- 
member as the word “mask” than an arbitrary spelling like uhe, 
though nonreaders don’t. Some orthographic utilization in word 
identification is obvious, for example, in distinguishing homo- 
phones like their from there. The term orthographic structure (lit- 
erally, visual structure) is usually taken to refer specifically to 
spelling, although it can also refer to the shape or featural detail of 
words as a whole and to arrangements of letters, individually or in 
clusters, seen as visual configurations, not as guides to pronuncia- 
tion. Walters, Komoda, and Arbuckle (1985) offered experimental 
evidence that phonological recoding plays a very small part in 
skilled reading and that it is unnecessary unless there are detailed 
memory demands. It is often assumed that decoding (or recoding) 
written language into speech will automatically provide meaning 
for written words. But spoken words must themselves be inter- 
preted and they can be even more ambiguous than written words. 
Including a phonological loop in reading constitutes an additional 
step, not a shortcut, for understanding. 
294 
UNDERSTANDING READING 
NOTES TO CHAPTER 11, 
READING, WRITING, AND THINKING, pp. 178-193 
Comprehension and Thinking 
Reading and thinking are fundamentally inseparable, especially 
when reading is discussed or researched under the heading of 
comprehension. Vygotsky (1978) defined thinking as “internalized 
action,” and reading might be regarded as “internalized experi- 
ence.” In a special “literacy" issue of Harvard Educational Review, 
Scribner and Cole (1978) made the important point that literacy 
doesn’t change the basic way in which people reason. In any aspect 
of thinking, what we know already — our “prior knowledge" — is ob- 
viously an important factor in what we can accomplish. Tierney 
and Cunningham ( 1984) discussed the importance of building up 
and “activating” background knowledge prior to reading. Basically 
this is common sense — the more we know about a topic before 
reading, the more we understand. But there is a decidedly mecha- 
nistic tone about theorizing that talks of the provision, utilization, 
and activation of any aspect of thought, which in practice can lead 
teachers to spend more time on preparation for reading than on 
reading itself, although reading is a major source of prior knowl- 
edge. Undergraduates given scenic photographs to look at and de- 
scriptive paragraphs to read remembered a number of pictures 
and paragraphs a week later, but also remembered (with a high de- 
gree of confidence) photographs they had not seen — the images 
they had created from paragraphs they had read (Intraub & 
Hoffman, 1992). 
Kimmel and MacGinitie (1984) showed that children may per- 
severate with inappropriate hypotheses while reading (and pre- 
sumably in other circumstances). This is not necessarily the fault 
of the readers, however. Children can often cope well with para- 
graphs or entire texts where the “main idea” is at the beginning. But 
many school texts are not written in this way. Kimmel and 
MacGinitie show that school texts often begin with examples, anal- 
ogies, and even refutations, and make their point clear only at the 
end. Nicholson and Imlach ( 1981 ) found that prior knowledge and 
text could compete when children were required to answer ques- 
tions about their reading. Eight-year-olds could incorrectly impose 
their own expectations on narrative, but on the other hand they 
could also “be assailed by every word in the paragraph" — a phrase 
first used in an important article by Thorndike (1977). Paradoxi- 
cally, O’Brien and Myers (1985) demonstrated that comprehen- 
NOTES 
295 
sion difficulty could occasionally improve recall, because readers 
spent more time looking back. 
Comprehension in reading doesn’t necessarily take place imme- 
diately or all at once. Samuels (1979) and O’Shea, Sindelar, and 
O’Shea ( 1985) showed, not surprisingly, that reading the same text 
more than once improves fluency, comprehension, and memory, 
especially for "poor readers” or for difficult texts. The effect is par- 
ticularly pronounced if readers are cued to read for comprehen- 
sion rather than for accuracy. This research underlines an 
important general point: It is usually more effective to read a text 
quickly, more than once, than to plod through it slowly once only. 
Initial “skimming,” and even browsing through an entire book by 
glancing at occasional pages, adds to prior knowledge and facili- 
tates subsequent efforts to make sense of the entire text. 
Reading Speed 
Carver (1985) criticized studies that claim to demonstrate fast rates 
of reading. He asserted that "comprehension” is rarely adequately 
defined or measured in such studies and held that unless readers 
comprehend the author’s thoughts on a sentence-by-sentence basis 
(a procedure to which he gives the special name of rauding ), then 
“skimming” rather than reading is taking place. Demonstration of 
rauding under laboratory conditions involves tests of recall of detail 
that constitute great impositions on memory. Not surprisingly. 
Carver’s experimental subjects fail to meet such a criterion at 
speeds of more than 600 words a minute. But it can be argued that 
no one reading in normal circumstances would ever try to remem- 
ber the detail of every sentence of a novel or even of a business letter. 
Inability to remember detail doesn’t mean that a book was not com- 
prehended or even that every sentence in that book was not compre- 
hended at the appropriate time. Carver’s own studies showed that 
“speed readers” were able to write an adequate 1 00-word summary 
of a 6,000-word text after perusing it for 4 minutes. Carver ( 1992) 
subsequently reported that college students typically read to memo- 
rize at 138 words per minute (wpm), to study for a multiple-choice 
test at 299 wpm, to “comprehend complete thoughts in sentences” 
at 300 wpm, to skim at 450 wpm, and to scan for a target word at 
600 wpm. He refers to these different reading rates as gears and 
says that “shifting” flexibility is needed. 
Saenger (1991) discussed how silent reading facilitates rapid 
reading and the scanning of text, subordinating the constraints of 
the text to the aims and biases of the reader. The purpose of the 
296 
UNDERSTANDING READING 
spaces between words in alphabetic writing, Saenger argued, is to 
facilitate silent reading. British researchers Harding, Beech, and 
Sneddon ( 1985) found that the reason more proficient readers aged 
from 5 to 9 years old appeared to “process larger units of informa- 
tion” was that they read faster and therefore had less of a memory 
handicap. Potter ( 1984), using a technique that delivers individual 
words at a controlled rate to readers (Forster, 1970), found that col- 
lege students preferred a rate of 360 words a minute. At 720 words a 
minute, almost all the words could be read but ideas “seem to pass 
through the mind without being adequately retained” — on a sen- 
tence-by-sentence basis, at least. At 960 to 1 ,680 words a minute, 
most viewers felt they could not see most of the words or under- 
stand individual sentences, although they could be shown to have 
acquired some understanding. Potter, Kroll, Yachzel, and Harris 
( 1980) corroborated that it is memory that takes time in reading. In- 
creasing the rate of reading from 180 to 600 words a minute left 
comprehension unaffected but reduced memory for detail. 
Comprehension and Context 
There is some research suggesting that “poor" readers use context 
more than fluent readers (Stanovich, 1980, 1981, 1986; 
Stanovich, Cunningham, & Feeman, 1984; Perfetti & Roth, 1981; 
Perfetti, Goldman, & Hogaboam, 1979), contrary to the position 
argued in this book that inexperienced readers use less nonvisual 
information. The resolution of this apparent paradox may be that 
when reading is difficult, all readers need context more, and as 
Thompson (1981) pointed out, “good and poor readers” reading 
the same text are not doing equivalent reading — one is reading easy 
material and the other difficult. In the experimental conditions, in- 
experienced readers are forced to rely on context and every other 
source of available help. But there is a related factor. Experi- 
menters, especially those with a cognitive science orientation, typi- 
cally define “context” as a few words on the page on either side of a 
“target word” in contrived situations that emphasize word identifi- 
cation or memory. There is much more to nonvisual information 
than adjacent words on a page. Nonvisual information includes all 
of a reader's relevant prior knowledge, plus understanding of the 
text as a whole. In fact, adjacent words on the page should be con- 
sidered visual information — they are not “context" so much as ad- 
ditional features to be analyzed if individual word identification is 
emphasized and difficult. And the less nonvisual information a 
reader can bring to bear, the more visual information, in the form 
NOTES 
297 
of distinctive features from the text, needs to be identified. In such 
circumstances, readers may need supplementary features outside 
the boundaries of target words in order to identify those words. 
The eyes may appear to focus on individual words during reading 
(Just & Carpenter, 1980), but they have to be focused somewhere. 
The particular focal point doesn’t necessarily indicate that words 
are being identified one at a time. 
Benefits of Reading 
Krashen (1993a) provided an excellent and concise summary of 
the benefits of “free voluntary reading” with regard to reading abil- 
ity, comprehension, vocabulary, grammar, spelling, writing, sec- 
ond language learning, attitude toward school, career choice, 
public esteem, and self-esteem. He discussed the significance and 
value of every kind of reading that learners will voluntarily engage 
in, including comic books and “romances.” New Zealand re- 
searcher Warwick Elley (1989) showed that vocabulary increases 
with reading — or with listening to stories. Elley (1992) reviewed 
reading instruction and achievement in 32 countries — mostly 
from 1,500 to 3,000 students in each, tested by researchers in 
their own country on ability to understand narrative, expository 
text, and “documents” (like charts, maps, and lists of instruc- 
tions). Factors that consistently differentiated high and low scoring 
countries were large school and classroom libraries, frequent si- 
lent reading, and story reading aloud by the teacher. Age of begin- 
ning instruction (up to age 7), and even instruction in a language 
unfamiliar to the learner (English in Singapore), did not make a 
difference. Torrance and Olson (1985) reported that good readers 
speak in more complex utterances and use a wider range of meta- 
linguistic words related to thinking and language. For discussions 
of how literacy can make a difference to an individual's knowledge 
of “language as an object,” see Olson and Torrance (1991), particu- 
larly the chapter by Scholes and Willis (1991) arguing that learning 
to read promotes ability to talk and an understanding of grammar. 
West, Stanovich, and Mitchell (1993) questioned individuals wait- 
ing alone in an airport waiting area — 111 who were “reading 
recreationally” for 10 continuous minutes and 106 others who did 
not read at all for 10 continuous minutes (with almost equal num- 
bers of females and males in each group). They found significant 
differences in vocabulary size and “cultural knowledge" in favor of 
the readers, who also tended to be somewhat older and to have had 
more educational experience. West, Stanovich, and Mitchell also 
298 
UNDERSTANDING READING 
discussed evidence that there are much greater opportunities for 
learning new words from reading than from conversation or 
watching television — popular magazines, for example, providing 
three times the number of opportunities. Reitsma (1983) reported 
that experienced beginning readers can learn the “graphemic 
structure” (spelling) of new words as well as their meaning with lit- 
tle practice. 
Eckhoff (1983) showed how children’s reading influences what 
they write. If their primers contained “stories” in which each short 
sentence was on a separate line, children wrote their own stories in 
the same way. If their reading was richer and more conventional, so 
was their writing. Calkins ( 1980) found that children learned more 
about punctuation from their reading than from instruction and 
used more punctuation as a consequence. The children also 
adopted stylistic features of the texts they read, such as beginning 
sentences with “And” or ending them with “too.” For a general argu- 
ment that children learn to write by reading, see Smith (1994). 
Conversely, Tierney and Shanahan ( 1991 ) noted that children who 
write more are more enthusiastic and competent readers. 
For the richness and diversity of the roles literacy plays in indi- 
vidual lives (without detracting from the overriding significance of 
oral language and experience generally), see Heath (1986). 
Csikszentmihalyi (1990) originated a concept of “flow” to denote 
the uniquely satisfying experience of being deeply and effortlessly 
involved in an activity to the extent of forgetting oneself, time, and 
everything else. Reading, he said, is perhaps the most often men- 
tioned flow activity in the world. Nell (1988) described the marked 
physiological effects that reading can have, both arousing and re- 
laxing. 
On the other hand, it can be also argued that literacy is often 
overrated (Pattanayak, 1991; Smith, 1989). Nonliterates should 
not be regarded as “problems” to be cured, eradicated, or declared 
war on, nor should they be held responsible for social and eco- 
nomic crises of society. Graff (1987a. 1987b) provided compendi- 
ous discussions of the history of literacy, the misunderstandings 
that surround it, and its social roles and functions, including a crit- 
ical review of the “crisis” approach to literacy and alarms about de- 
clining standards or abilities. He asserted that the efficacy of 
literacy in improving an individual's life is a myth; literacy is a tech- 
nology, or set of techniques, not an agent of change for individuals 
or societies. The “oral world” (of which writing is an extension 
rather than a replacement) is of continuing importance. In a more 
concise presentation of his views, Graff (1986) concluded that as 
NOTES 
299 
far as the current state of the world is concerned, “Literacy is nei- 
ther the major problem, nor is it the main solution.” 
NOTES TO CHAPTER 12, 
LEARNING ABOUT THE WORLD, pp. 194-211 
Language Learning 
Clark and Hecht ( 1983) reviewed research showing how compre- 
hension precedes the production of language. Bridges, Sinha, 
and Walkerdine (1981) demonstrated how infants figure out the 
intentions of their mothers, taking into account the circum- 
stances in which their mothers are talking, in order to under- 
stand what they are saying. Nelson (1985) proposed that children 
learn primarily through being involved in meaningful events with 
adults. The reasoning for why learning and comprehension are 
the same, the consequence of engagement, demonstrations, and 
sensitivity, is in Smith (1998). Krashen (2003) has long main- 
tained that learning (which he terms acquisition ) is a subcon- 
scious process, effortless and involuntary; we are not usually 
aware of the knowledge we acquire. His “comprehension hypothe- 
sis” proposes that language acquisition takes place when we “un- 
derstand messages” (not necessarily messages addressed to us 
personally). Brown and Palincsar (1989) demonstrated that 
“comprehension-learning” is social; children learn how to under- 
stand texts through discussion with other readers. Gallistel 
( 1990) argued that children learn language by analogy rather than 
by rules, relating language they don’t know to language they do 
know. See also Fosnot (1996). 
Vocabulary 
In extensive research, Anglin (1993) found that children knew 
10,000 words by their sixth birthday, and learned 4,000 new roots 
between Grade 1 and Grade 5, leading to a further 14,000 new 
words. The bulk of this expansion was done by “morphological 
problem solving” — breaking new words down into component 
morphemes. See also Nagy and Scott (2000) and E. Clark (1993). 
In an intensive study, Carey ( 1978) estimated that 6-year-olds have 
mastered (to some degree) an average of 14,000 words, noting that 
300 
UNDERSTANDING READING 
"this massive vocabulary growth seems to occur without much 
help from teachers.” According to Carey ( 1 978), there is first a “fast 
mapping" when a child hypothesizes a probable general meaning 
for a new word. A gradual process of refining and adding “partial 
knowledge" of meaning ensues on successive encounters, with 
from 4 to 10 encounters required for word learning to be “com- 
plete.” Rice (1990) described “QUID’ — quick incidental learning — 
with fast mapping often occurring when the word is first encoun- 
tered, followed by a long period of refinement. The initial partial 
comprehension of a new word includes both a meaning and a syn- 
tactic role; it doesn’t require explicit assistance from adults. In one 
12-minute exposure to an animated television program, 5-year- 
olds on average "picked up” the meanings of 5 test words such as 
gramophone, nurturant, viola, makeshift, malicious, and artisan 
that occurred seven times each. Three-year-olds picked up an av- 
erage of 1.5 of these words. Mandler ( 1992) analyzed how babies 
move from preverbal perceptions of different kinds of object to 
conceptions that can become language. Markman (1992) exam- 
ined the assumptions that 2-year-olds appear to make in learning 
words concerning “whole objects," taxonomic relations, and mu- 
tual exclusivity. The word dog is likely to mean the whole animal, 
not a part of it or its color; the word is likely to apply to similar ob- 
jects (a cat might be a dog, but not an umbrella), and objects will 
not have more than one name. 
Nagy and Herman ( 1987) recalibrated earlier research results 
and calculated an annual growth in vocabulary between Grade 3 
and Grade 12 of from 2,000 to nearly 4,000 words, and a me- 
dian of 3,500. Nagy, Osborn. Winsor, and O’Flahavan (1994) an- 
alyzed 10,000 new words an average fifth-grade reader might 
encounter in a year (an avid fifth-grade reader might encounter 
several times that number), finding that more than half were 
clearly related to words already known, but 1 ,000 were “truly 
new.” Skilled readers deal with these new words on the basis of 
context, phonics, or structural analysis, meaning familiarity 
with parts of words, including grammatical and meaningful 
units like s and ed. Structural analysis also helps spelling, for 
example, with segments like ceive, which has no clear meaning 
on its own but is part of several words. The authors note that in 
comparison with phonics, very little attention has been paid to 
structural analysis, which they say begins in early childhood, 
both to understand and coin new words, obviously without in- 
struction. Nevertheless, the authors propose including word 
analysis as part of reading instruction — without recognizing that 
NOTES 
301 
when outsiders take charge of what readers are able to under- 
stand and learn for themselves, interventions intended to be 
helpful often become obstacles. 
Children required to learn specific new words in contrived 
experimental contexts — typically artificial words embedded in 
half a dozen unrelated sentences — usually find the task diffi- 
cult. McKeown (1985), for example, required 30 fifth graders to 
figure out the meaning of artificial words in short sentences 
like “Eating lunch is a narp thing to do,” and found that good 
readers could accomplish the task, but not poor readers (who 
presumably had difficulty reading the sentences in the first 
place). She concluded that “the meaning-acquisition process" 
is complex and difficult, even for high-ability readers. But this 
may be a case of what might be called the Laboratory Fallacy, 
which asserts that children who have difficulty on artificially 
contrived learning (or comprehension or memory) tasks will 
have similar difficulty in all situations. Schatz and Baldwin 
(1986) showed that “context cues” are not usually reliable pre- 
dictors of word meaning — not in experimental situations where 
texts are brief and assembled in a contrived manner. The more 
naturalistic approach of Nagy and his colleagues that I have just 
cited suggests a different conclusion, although the technique of 
embedding unfamiliar words in minimal context is common in 
laboratories and classrooms. 
McKeown (1993) showed that conventional dictionary defini- 
tions hardly facilitate learning word meanings. Instruction often 
focuses on words and meanings that are either too common or 
too uncommon to require attention, and examples given are of- 
ten misleading. McKeown suggested alternatives that focus on 
describing in familiar language the circumstances in which the 
word is usually (“prototypically”) used. Thus instead of defining 
morbid as “not healthy or normal.” she proposed “showing a 
great interest in horrible gruesome details, especially about 
death”; and for transitory , instead of “passing soon or quickly, 
lasting only a short time” (which led to student sentences like 
“the train was transitory”), she suggested "describes a mood or 
feeling that only lasts a short time.” McKeown observed that 
such changes still don’t make the traditional method of looking 
up a word and writing a sentence a good way of building vocabu- 
lary. Miller and Fellbaum (1991) offered a detailed discussion of 
complex problems of definitions and dictionaries, and 
Schwanenflugel (1991) provided an interesting edited volume 
on the psychology of word meanings. 
302 
UNDERSTANDING READING 
Motivation 
The role of motivation in learning should be clarified, especially 
because failure to learn is frequently attributed to lack of motiva- 
tion. However, learning continually takes place in the absence of 
conscious motivation — for example, the effortless growth of vo- 
cabulary. And the presence of motivation doesn’t guarantee learn- 
ing. We have all failed to learn things we have been highly 
motivated to learn, on which we may have expended considerable 
effort and “time on task.” Learning normally depends not on ef- 
fort but on the demonstrations, collaboration, engagement, and 
sensitivity discussed in this chapter. Interest and expectation of 
learning are better predictors of learning than overt motivation. 
At best, motivation has the beneficial effect of putting learners 
into situations where demonstrations and collaboration are 
likely to be found. And of course, anyone motivated not to learn, 
or anticipating failure, is likely to find the expectation fulfilled. 
NOTES TO CHAPTER 13, 
LEARNING ABOUT WRITTEN LANGUAGE, pp. 212-232 
Learning to Read 
Some of the most significant research into children’s developing 
understanding of literacy has been done not by educators or psy- 
chologists — who tend to look at individuals in isolation, or in a per- 
sonal relationship to “knowledge” — but by sociologists and 
anthropologists. The edited volume of Goelman. Oberg. and Smith 
(1984) contains summaries and reviews of research into the social 
basis of preschool literacy by workers in a number of disciplines 
and also outlines the prevailing methodologies. The primary 
method of research is not experimental, but observational. 
Ethnographic is the technical term for such research, also called 
“naturalistic research,” and (by Yetta Goodman, 1978, 1980) “kid 
watching." Two of the main findings of this research are that chil- 
dren in all cultures develop insights into the forms and functions of 
written language before school and that these insights are based on 
meaning and use. The research also shows that learning about 
reading can't be separated from learning about writing, and about 
how written language is used. 
NOTES 
303 
The research has also made clear that children don’t need to be 
economically privileged or the recipients of special kinds of in- 
structional support in order to learn about reading and writing. 
Ferreiro (1978, 1985), for example, who demonstrated how 3- 
and 4-year-olds gain insights about letters, words, and sentences, 
did much of her work in the slums of Mexico City with children 
whose parents were illiterate. The fact that early readers have not 
necessarily had advantages is also made by M. Clark ( 1 976) in her 
classic book on Young Fluent Readers , many of whom came from 
large poor families and were not “good risks" for reading instruc- 
tion in school. Research shows that children raised in mid- 
dle-class homes with well-educated parents generally do well in 
school, while other children tend to be behind when they start 
school and stay behind. Individuals — students and parents — are 
usually held responsible for this . But Neuman and Celano (2001) 
claimed that ecological research, which looks at contexts larger 
than family settings, shows that the environment of lower income 
families in and around Los Angeles tends to be strikingly lacking 
in places for reading, reading material, labels, logos, and books at 
home, in preschool, in local libraries, and in public library 
branches. There is social isolation and there are unequal oppor- 
tunities. Duke (2000) found the same imbalance in the Greater 
Boston area. Not only were there more books and other materials 
in richer classrooms, but a greater proportion were displayed 
and made available for use. In math, higher socioeconomic status 
children studied “architectural shapes” while lower did mindless 
workbook tasks. Unequal library opportunities are also stressed 
by Krashen (1995). There are innumerable studies showing dif- 
ferences between the schools and schooling of poor and better off 
children, from the streets around the schools and their hallways 
and classrooms to materials and learning opportunities made 
available. See especially Savage Inequalities: Children in Amer- 
ica’s Schools (Kozol, 1991). 
In a poignant article entitled Literacy at Calhoun Colored 
School 1892-1945, Willis (2002) examined the philosophy behind 
the education of African American children at a private independ- 
ent boarding school in Alabama, which tried sincerely to raise 
skills levels of students while avoiding the power that full literacy 
might bring. The author commented that it is fashionable today to 
look to science, medicine, and psychology for explanations when 
the literacy development of any group differs from the mainstream, 
and to look for methods that will work for all. When these methods 
fail, students, their families, their communities, and their language 
304 
UNDERSTANDING READING 
are blamed. Willis called for recognition of the relationship be- 
tween power and knowledge as a primary factor in the way all stu- 
dents are educated and their performance assessed. 
Clay (1992/1993) reported from extensive experience that the 
best preparation for literacy is opportunities for conversation. She 
referred specifically to bilingual schools in New Zealand con- 
cerned with the survival of the Maori language, and gave several 
other international examples where children read to in preschool 
had improved reading and speaking abilities in later years. If no 
books are available, competent storytellers can prepare children 
for literacy learning. Clay noted that children under 5 years of age 
are “amazingly good" at learning and losing languages, depending 
on the opportunities they have to use them. The least complicated 
entry into literacy learning is to begin to read and write in the lan- 
guage already spoken. For further discussion of language policies 
and literacy learning in multilingual situations, see Elley and 
Mangubhai (1983). 
Proficiency in oral language is not a prerequisite for learning to 
read. Rottenberg and Searfoss (1992) showed that hearing-im- 
paired children in a mixed preschool class learn about literacy in 
the same ways hearing children do— without direct instruction — 
and also learn about the hearing world. Hartman and Kretschmer 
(1992) reported that hearing-impaired teenage students also learn 
about reading by talking and writing about what they read. For a 
summary of research demonstrating that children with “special 
needs” don’t learn to read and to write differently from other learn- 
ers and don’t require special kinds of instruction (although they 
may need more time), see Truax and Kretschmer (1993). 
Routman (2003) asserted that teachers should gradually hand 
over responsibility for literacy development to learners, and ex- 
plained how this involves demonstrations by the teacher, demon- 
strations with students, guided practice, and independent practice. 
Free Voluntary Reading 
Krashen (1993a) is an energetic advocate of “free voluntary read- 
ing” and reading for pleasure, in any genre including comic books 
and romances. He claimed that most people are able to read and 
write — there is no “literacy crisis” — but many don’t do so very well, 
not for want of instruction but because of lack of experience. He de- 
plored the fact that many public and school libraries are closed or 
starved for funds that are relatively plentiful for more structured 
kinds of reading instruction and tests. Krashen and another popu- 
NOTES 
305 
lar free reading crusader, Jim Trelease, are so anxious to see chil- 
dren have the maximum opportunity and encouragement to read 
that they proposed that school libraries follow the model of com- 
mercial bookstores and facilitate eating and drinking among the 
books (Trelease & Krashen, 1996). They preemptively responded 
to predictable objections. 
Von Sprecken and Krashen (2002) found that contrary to popular 
stereotypes, there is little evidence of a decline in interest in recre- 
ational reading during adolescent years, or of negative attitudes to 
reading (although both occur with school-related reading). Reading 
is also a powerful incentive for reading. Following a suggestion by 
Trelease that one positive reading experience — one “home run” 
book — is enough to establish someone as a reader, Von Sprecken, 
Jiyong, and Krashen (2000) questioned 214 fourth-grade students 
in three elementary schools in the Los Angeles area. Over half the 
students said they became interested in reading after just one book, 
which they could name (including “scary books,” comics, and popu- 
lar series). Cho and Krashen (1994) found that three Korean and 
one Mexican female speakers, aged from 21 to 35, all studying Eng- 
lish as a second language, became “addicted” to the easy reading of a 
romance series highly popular with high school female students, 
voluntarily reading up to 23 volumes in a month, with dramatic 
gains in vocabulary, ability, and confidence to converse in English. 
Commenting on the fact that many studies have shown a strong 
relationship between poverty and reading scores, Krashen (2002) 
drew on research to show that children from low-income families 
have very little access to books, in school and out, compared with 
children from high-income families. He argued that “readability” 
systems that direct children to certain books limit their choices of 
books they might understand and enjoy. Elley and Mangubhai 
(1983) reported success in Third World countries with a “book 
flood” program — based on the assumption that children can over- 
come the disadvantages of inadequate exposure to reading and 
poor motivation if their classrooms are flooded with high-interest 
illustrated reading books and the teacher helps them to read to- 
gether. An updated, expanded discussion of widespread adoption 
in Fiji, Niue, Singapore, and South Africa was presented in a paper 
by Elley (1996). 
Reading to Children 
Reading to children is frequently recommended, although it is not 
always made clear what exactly the practice is expected to achieve. 
306 
UNDERSTANDING READING 
Obviously, reading to children may interest them in stories (or 
whatever else is read to them) and also may demonstrate the inter- 
est and utility that other people find in reading. But every occasion 
when a child is read to can also be a reading (and writing) lesson, 
an opportunity to learn more about the conventions and purposes 
of written language. Taylor and Strickland (1986) described many 
family story-reading sessions and their benefits to children. 
Durkin (1984) studied 23 children of “average intelligence” who 
had frequently transferred schools, who became “successful read- 
ers” (reading above “grade level" by Grade 5). She found that the 
children had supportive parents who liked stories and read to 
them. Dombey (1988) also discussed children’s movement into the 
nature of stories and into reading from hearing a story read. 
Eldredge (1990) showed that children learn more when they are 
helped to read — and therefore can read material that would other- 
wise be too difficult for them. 
Heath and Thomas ( 1 984) provided a fascinating case study of a 
teenaged, unemployed high school dropout mother of two (coau- 
thor Charlene Thomas) herself learning to read in the course of 
helping her children learn to read. Another of Shirley Brice Heath’s 
many insightful contributions is an article entitled "What No Bed- 
time Story Means” (Heath, 1982d). For more on Vygotsky and edu- 
cation, see Moll (1990), 
Literacy and Schooling 
The fact that children often learn so much before school and that 
cultural influences are so important doesn't release schools from 
responsibility or provide convenient justification for failures of in- 
struction. If children have not received adequate environmental 
support for embarking upon literacy, then schools must provide it. 
The ethnographic research shows clearly the collaborative condi- 
tions under which learning to read and write takes place. If parents 
fail to read to children, it is all the more important that teachers 
read to them. 
Sulzby (1985) discussed how kindergarten children, some as 
young as 4Vfe, begin to make sense of stories, from commenting on 
pictures to telling a story that gains more and more fidelity to the 
text. There was always a story behind the children's com- 
ments — and the children’s versions of the story always made 
sense. For a general, research-based review of many aspects of lit- 
eracy and school, see the edited volume by Raphael (1986). 
NOTES 
307 
Intelligence has never been found to be an important factor in 
learning to read, although reading appears to contribute signifi- 
cantly to intelligence. Stanovich, Cunningham, and Feeman (1984) 
found only a low relationship between intelligence and reading abil- 
ity in first-grade children. The correlation was higher by fifth grade, 
and the researchers attributed the increase to "reciprocal causa- 
tion.” They also found that children who quickly learned to read 
continued to read well through life, indicating the importance of 
avoiding obstacles, irrelevancies, and confusions in a child’s early 
experiences with literacy. 
Unsuspected and incidental influence of instruction on chil- 
dren’s behavior has been well documented. Research on how chil- 
dren learn to read rarely remains uncontaminated by the 
influence that unrelated instruction has already had on them. 
Barr ( 1972, 1974) wrote seminal papers on the effect of instruc- 
tion on children’s reading. Holdaway (1976) commented on the 
importance of self-correction for children learning to read and on 
the risk that some instructional techniques take this responsibil- 
ity away from them. Eckhoff (1983) was cited in the previous 
chapter notes for demonstrating that what children read is re- 
vealed in their writing. Juel and Roper/Schneider (1985) re- 
ported that what children read also shows up in how they read. In 
particular, texts made up of “decodable” regular words produced 
children whose main strategy in reading was to sound out unfa- 
miliar words. DeFord (1981) reached a similar conclusion. 
MacGinitie and MacGinitie (1986) argued that an emphasis on 
"mechanics” in the primary grades teaches students not to read, 
and they noted a deemphasis on extended writing, literature, and 
“content-rich reading” in high school. When students have difficul- 
ties with a text, teachers respond (or ask other students to re- 
spond) rather than bringing students back to the text. In a 
book-length analysis of British practices, Hull (1985) showed how 
the written and spoken language used in reading instruction and in 
content areas at all grade levels is frequently incomprehensible to 
students. He is particularly scathing when textbook or examina- 
tion questions include words and phrases that students don’t un- 
derstand (because of their ambiguity or vacuous definitions) and 
complaints are then made that the students can't read. Hull’s book 
was published in Britain. His occasional use of Anglicisms like 
“lower school" (junior high) and “sixth form” ( 1 2th grade) will illus- 
trate for North American teachers his point that one doesn’t have to 
be ignorant or learning disabled to be confused by unfamiliar uses 
of common words. MacGinitie ( 1984), in an article entitled “Read- 
308 
UNDERSTANDING READING 
ability as a Solution Adds to the Problem,” reported that attempt- 
ing to “simplify” texts by making them conform to formulas 
restricting word and sentence length can make them more difficult 
to read. Furness and Graves (1980) demonstrated experimentally 
that emphasis on accuracy (in oral reading) can actually reduce 
comprehension, just as an emphasis on correct spelling will in- 
hibit children’s writing. Paradoxically, both emphases result in less 
learning of what is supposed to be the object of the correction. 
Children reluctant to make mistakes will rarely venture beyond 
what they know already. Hiebert (1983) took a critical look at abil- 
ity grouping. 
Salmon and Claire (1984) made a 2-year study of four compre- 
hensive (secondary) schools in Britain and found that classroom 
collaboration, both socially and in “learning,” between teachers 
and students and among students, resulted in better student un- 
derstanding of the curriculum. Chambers, Jackson, and Rose 
( 1993) conducted a careful analysis and evaluation of a large-scale 
collaborative reading project in five inner-city schools in Britain. 
They found that new abilities were developed by teachers as well as 
by students as a result of reflective reading, group work, and lots of 
talk and discussion. Significantly, although all students in the pri- 
marily 8- to 13-year age group gained in reading ability compared 
with similar students in other schools, by far the greatest gains 
were made by students identified as “low achievers” — an unusual 
and important consequence. The evaluation is discussed in more 
detail in Gorman, Hutchison, and Trimble (1993). 
Willinsky (1990) distinguished a view of literacy as a set of skills 
taught piecemeal in educational institutions from literacy as a so- 
cial process, “making something of the world.” He described many 
attempts to “reshape the classroom” from a place of formal in- 
struction to a place where literacy is learned through experience, 
and stressed the importance of both individual voice and collabo- 
rative effort — among students and among teachers — rather than 
the following of timetables and agendas of others. Even where re- 
searchers have recognized the importance of active participation 
in literacy by the learner, he said, it is adopted as an extension of 
technological models of instruction. Courts ( 1991 ) contained a bit- 
ter attack on the skills approach to the teaching and testing of liter- 
acy (and thinking). See also Edelsky ( 1991 ) and Myers (1992) for 
the importance of “contexts,” in school and outside, for literacy. 
Mikulecky (1982) compared reading in school that was sup- 
posed to prepare students for the workplace and reading in vari- 
ous occupations. He found that students read less often and less 
NOTES 
309 
competently than most workers on the job, though the students 
read easier material to less depth. 
Books are often overrated in research and practice. Books have 
no unique or essential properties for literacy learning, and they are 
not always the easiest texts to read. Newspapers and magazines of- 
ten contain material that can attract the attention of the smallest 
children, depending upon their interests and mood. Krashen 
(1987) reviewed the research on comics, showing that they can be 
highly productive materials for developing reading interests and 
ability, and also that comics frequently have rich vocabularies and 
conceptual content. 
Teachers and Programs 
“Whole language” is the instructional philosophy that reflects 
most consistently the view that meaning and “natural language” 
are the basis of literacy learning. Kenneth and Yetta Goodman 
are the theorists most closely identified with the origin and de- 
velopment of the whole language approach. Many TAWL 
(Teachers Applying Whole Language) groups have been estab- 
lished, especially in the United States and Canada. K. Goodman 
(1986) summed up the whole language view: “Many school tradi- 
tions seem to have actually hindered language development. In 
our zeal to make it easy, we’ve made it hard ... primarily by 
breaking whole (natural) language up into bite-size, but abstract 
little pieces. We took apart the language and turned it into words, 
syllables, and isolated sounds. Unfortunately, we also post- 
poned its natural purpose — the communication of mean- 
ing — and turned it into a set of abstractions, unrelated to the 
needs and experiences of the children we sought to help” ( p . 7 ) . 
In Report Card on Basal Readers, Goodman, Shannon, Free- 
man, and Murphy (1988) criticized from a whole language point 
of view unnatural language and behaviors in the production and 
use of basal readers. Basal readers, they said, demonstrate a lack 
of trust in teachers, who are consequently “deskilled.” Basal au- 
thor Baumann (1992) responded with a defense of the books and 
a denial that they deskill teachers. See also Shannon (1989, 
1993). Huck (1992) described how concern with whole language 
has led to a growth of “literature-based" teaching, summarizing 
the approach and its advantages. McGee ( 1992) also provided a 
historical review of the “literature -based reading revolution” 
showing that younger readers are capable of quite sophisticated 
responses to stories. See also Morrow (1992). Meek (1988), and 
310 
UNDERSTANDING READING 
McMahon (1992). The term "literature” in this context usually re- 
fers to books written especially for children. 
"Reading Recovery” is an early intervention tutoring program 
devised by Clay ( 1 985 ) , in which children who have not “caught on” 
to reading after a year spend about 3 months getting extra 
help — reading entire books but also taking “temporary instruc- 
tional detours” at relevant times — in “concentrated encounters” 
with specially trained teachers; see also Pinnell (1989) and a chap- 
ter by Clay and Cazden in Cazden (1992). Wasik and Slavin (1993) 
included a useful detailed description and review of Reading Re- 
covery and three other tutoring programs. 
For social aspects of literacy and learning, see Ong (1982), Le- 
vine (1986), and several chapters in Olson, Torrance, and Hildyard 
(1985). Meek ( 1982, 1984) was insightful on social aspects of liter- 
acy instruction. See also Meek, Armstrong, Austerfield, Graham, 
and Plackett (1983) regarding a not always encouraging and suc- 
cessful struggle to help adolescents to read. Atwell (1987) also ex- 
amined teaching reading and writing to adolescents. An excellent 
compendium of research in reading and writing (termed “compre- 
hension and composition”) is Squire (1987). Hedley and Baratta 
(1985) included important articles on reading, learning, and 
thinking generally, including ethnographic research. Tuman 
(1987) examined social attitudes toward reading and reading in- 
struction. 
Krashen (1985) examined the evidence and arguments that both 
spoken and written language are learned only by comprehension. 
He also examined the emotional blocks that can stand in the way of 
such learning. In “Learning to Read at Forty-Eight." Yatvin (1982) 
shared insights gained in belated efforts to learn to read Hebrew. 
In addition to their interests (what they like to read about), chil- 
dren have their preferences — to read alone or in groups, for long or 
short periods, at particular times of day, with or without supervi- 
sion. The best way to find out is to ask them or to observe them. A 
formalized characterization of “reading styles” and “learning 
styles" has been constructed by Carbo, Dunn, and Dunn (1987). 
For a discussion of this, and of other matters concerning instruc- 
tion and teacher behavior from a generally whole language point of 
view, see Weaver (1988). 
In the introductory chapter to the book What Research Has to 
Say About Reading Instruction, in a paragraph entitled 
“Teachers, You Have a Lot to Learn," Otto (1992) discussed dilem- 
mas facing teachers, including the fact that there is already more 
reading research than a practitioner could sort out in a lifetime 
NOTES 
311 
(with more coming faster than anyone could read it). But much of 
the research and argumentation is repetitious; it is not so difficult 
for teachers to find out what the research — and the issues — are 
about. The real dilemma is having to come to a decision. Kutz 
(1992) offered brief notes and a helpful bibliography on research 
by teachers, finding their own answers to questions. 
Metalinguistic Awareness 
The notes to the previous chapter included references related to 
the general role of metacognition in learning — the awareness of 
one’s own thought processes. Metalinguistic awareness is meta- 
cognition specifically related to linguistic matters, particularly (in 
the case of reading and writing) to the nature of written language. It 
is not clear that such awareness plays an important role in learn- 
ing, or, indeed, that such awareness can take place until after 
learning has occurred. It is difficult to see how terms like “letter.” 
“word,” and “sentence” can have any meaning to anyone who can’t 
read. Many people able to converse fluently can’t say what the dif- 
ference is between nouns and verbs, or active and passive sen- 
tences, not to mention verbalizing the complexities of 
transformational grammars and the conventions of cohesion. Nev- 
ertheless, some theorists not only feel that metalinguistic under- 
standing is essential for learning to read, but they define learning 
to read in terms of such understanding. 
Ehri ( 1985) emphasized spelling instruction as a way to make 
children aware of the relation between words they hear and words 
they see. Metacognitive and metalinguistic theorists typically em- 
phasize relationships of sound and spelling in reading rather than 
meaning. They also stress the role of cognitive strategies. A. 
Brown (1982), for example, discussed "self-regulatory strategies" 
that contribute to “learning how to learn from reading,” including 
predicting, planning, checking, and monitoring knowledge of one’s 
own abilities. She cited extensive documentation that better read- 
ers are more efficient and effective at such tasks, although there is 
always the chicken-and-egg problem of whether the competence 
produces better readers or that experienced readers naturally gain 
more competence. She notes that teachers treat good and poor 
readers differently. Shannon (1984, 1985) reviewed a mass of re- 
search indicating that schoolchildren who are identified as poor 
readers, or likely to be poor readers, do less reading, less interest- 
ing reading, more difficult reading, more exercises, and receive 
less assistance than other children. Hiebert (1983) looked at the 
312 
UNDERSTANDING READING 
consequences of ability grouping and also found that more able 
readers get the most meaningful reading with an emphasis on 
meaning, while the others received a greater emphasis on accuracy 
(which of course makes reading more difficult). See also Stanovich 
(1986) on the “Matthew effect” (that the rich get richer) in reading. 
Stanovich acknowledged that the main reason good readers get 
better is the differential treatment they receive in school (but as a 
believer with Gough and Hillinger [ 1 980] that reading is an “unnat- 
ural act,” he favored “surgical strikes” of specific skill instruction 
for children with problems). Metacognitive instruction tends to be- 
come a matter of more questions in preset sequences, or more 
drills, before reading actually begins, according to Langer ( 1982a, 
1982b). 
Dyslexia and Learning Disabilities 
Staller (1982) reviewed research on the relationship between 
neurological impairment and reading disability and concluded 
that it is not yet possible to relate dyslexic behavior to specific 
neurological correlates. Berninger and Colwell (1985) studied 
241 children between ages 6 and 12 identified as having no prob- 
lem, a possible problem, or a definite problem in reading — and 
could find no support for the use of neurodevelopmental and edu- 
cational measures in the diagnosis of specific learning disabili- 
ties. Dorman (1985) criticized research that defines or diagnoses 
dyslexia, because there is no agreement on what would constitute 
neurological evidence of neurological dysfunction in relation to 
reading. He concluded that “insistence upon the inclusion of [cen- 
tral nervous system] dysfunction in the definition and diagnosis 
of dyslexia seems to be putting the cart before the horse, in the 
sense that the neurological basis of any or all developmental read- 
ing disorders remains hypothetical.” Lipson and Wixson (1986) 
also reviewed reading disability research and concluded that it 
must “move away from the search for causative factors within the 
reader and toward the specification of the conditions under 
which different readers can and will learn.” Rather than accepting 
deficit explanations for reading problems, Wong ( 1982) proposed 
metacognitive factors, such as inadequate self-monitoring and 
self- questioning (in other words, attention to the meaning of what 
is read). Calfee (1983) said that dyslexia ought to be viewed as a 
problem with the development of the mind (i.e., experience) 
rather than a disease of the brain. Vellutino (1987) found no evi- 
dence that dyslexia is due to a visual deficit, nor to support 
NOTES 
313 
remediation based on exercises to improve visual perception. He 
added. “In any case, not enough is yet known about how the brain 
works to enable anyone to devise activities that would have a di- 
rect and positive effect on neurological functions responsible for 
such basic processes as visual perception, cross-modal transfer 
and serial memory.” Instead he recommends lots of assisted 
reading. 
Johnston (1985) examined three case studies of adult reading 
disability and found “overwhelming feelings of inadequacy and 
confusion,” anxiety, rational and irrational use of self-defeating 
strategies, conflicting motives, and inappropriate attributions of 
cause and blame, all going back to the individual’s earliest read- 
ing experiences. Clay (1979) suggested that children become 
reading failures by learning the wrong things. Downing ( 1 977) ar- 
gued that society creates reading disabilities, for example, by arti- 
ficially establishing “critical periods” for learning and through 
inappropriate expectations and stereotypes. Graham (1980) 
showed that “learning disabled” readers may have the same word 
recognition skills as children who are succeeding. Individual dif- 
ferences always occur, of course. Some people will find it harder 
to learn to read than others, but not because they have special 
“problems.” Bryant and Impey (1986) concluded that there is no 
essential difference between dyslexics and normal readers — dys- 
lexics simply are at the lower end of the scale of readers. They do 
nothing different from normal readers except they have more dif- 
ficulty. Medical researchers Shaywitz. Escobar, Shaywitz, 
Fletcher, and Makuch (1992) also proposed that dyslexia and 
“learning disabilities” may not be medical conditions but simply 
the lower end of a normal distribution of abilities. For a discus- 
sion of dyslexia from a phonological coding point of view, see 
Rack, Snowling, and Olson (1992). In a careful examination of 30 
years of research into reading failure, Hamill and McNutt ( 1981 ) 
could find no relationship to intelligence, perceptual or motor 
abilities, reasoning, or even affective factors, and only a marginal 
correlation with spoken-language ability. 
The consequences of reading failure are rarely adequately 
stressed. Reichardt ( 1977) talked of children “playing dead or run- 
ning away” from reading situations, defensive reactions among 
“reading handicapped students” whose physiological responses to 
reading ranged from complete apathy to hypertension. More gen- 
erally, Seligman (1975) discussed “learned depression" (due to 
lack of control over failures) and “learned laziness” (due to lack of 
control over rewards), both consequences of situations where the 
314 
UNDERSTANDING READING 
learner lacks control and understanding. Coles (1987) warned of 
the misuse, overuse, and dangers of the term “learning disabled.” 
For descriptions of how “at-risk” students and their teachers can 
learn from each other in linguistically and culturally diverse class- 
rooms, see Heath, Mangiola, Schachter, and Hull (1991) and Truax 
and Kretschmer (1993). See also Atwell (1991). 
Despite all the conjectures, it should not be surprising that no 
one has actually succeeded in finding a specific reading center or 
system in the architecture of the brain. Literacy has not existed as 
a cultural phenomenon long enough for the brain to develop spe- 
cialized reading processes. One might just as meaningfully sug- 
gest areas of the brain dedicated to cycling or sewing. It has been 
known for over 100 years that one side of the brain (9 times out of 
10 the left) is essential for the production and comprehension of 
language (except in infants, whose brains are more labile). But 
this asymmetry doesn’t mean that only one side of the brain is in- 
volved in language. All thinking activities involve the entire brain. 
Different people obviously have different preferences — some like 
listening to music and others would rather look at pictures — but 
this doesn’t mean that they lack parts of their brains or are “domi- 
nated” by one side of them. Looking at pictures, listening to mu- 
sic, and reading and writing involve experience, knowledge, and 
feelings and can’t be restricted to one area of the brain. It is a naive 
interpretation of neurological, language, and learning research to 
imagine that reading can be learned only with the left side of the 
brain or with the right. 
Evaluation, Testing, and Standards 
Testing, evaluation, and the setting of “standards” exercise great 
control over teachers in classrooms and can have lasting conse- 
quences on learners, driving the curriculum and emphasizing skills 
that can be taught and tested by prescription. Although many teach- 
ers express alarm over evaluation and standardization, especially 
when imposed on them (and when results are publicized), a number 
of prominent teachers and teacher organizations have become asso- 
ciated with such efforts. For example, the International Reading As- 
sociation and the National Council for Teachers of English joined 
forces with the federally funded Center for the Study of Reading in 
October 1992 to develop national standards (in the United States) 
for reading, the language arts, and English. Although acknowledg- 
ing diversity, it was expected that the standards would define “a 
common core of what is valuable in the teaching and learning of 
NOTES 
315 
English" and reduce the “contention in the discipline,” for example, 
over the role of phonics and the value of recommended reading lists. 
It was not explained how such disagreements might be caused by an 
absence of standards among teachers. 
Hiebert and Calfee (1992) discussed the assessment of literacy, 
from standardized tests to portfolios, noting the growing public de- 
mand for assessment to drive instruction and policy decisions. 
They referred to concern about the deprofessionalization of teach- 
ers and discouragement of students and concluded that teachers 
are the best judges concerning instructional decisions. In a book en- 
titled Learning Denied , Taylor (1991) related a dramatic example of 
the disruption that can occur in a child’s life — in addition to the anx- 
ieties and conflicts of the teacher and the parents — as a result of 
testing and “early diagnosis.” Giroux ( 1 992) discussed how the test- 
ing movement ignores successful local knowledge in favor of trained 
“leadership,” marginalized teachers, and standardization. 
Santa (1999/2000), then president of the International Reading 
Association (IRA), eloquently pointed out that teachers assess stu- 
dents daily as part of teaching, but not for passing judgments. The 
term “assessment” has become distorted and misused to make 
high-stakes decisions on the basis of a single standardized test 
score, although Santa doubted whether the tests are valid indica- 
tors of anything. She added that fear of flunking a test is a child's 
greatest fear next to the death of a member of the family. Her re- 
marks follow a position statement by the IRA Board of Directors 
(1999) expressing the “central concern . . . that testing has become 
a means of controlling instruction as opposed to a way of gathering 
information to help students become better readers.” Problems of 
"teaching in a world focused on testing" was the concern of 
Buckner (2002) in a journal issue dedicated to the effects of high- 
stakes testing on learners, teachers, and the development of liter- 
acy. Shannon ( 1996) said (in his title) that he is “mad as hell" over 
the use of standards and tests. So also was Ohanian (1999). 
Is There a Crisis? 
In The Manufactured Crisis: Myths. Fraud, and the Attack on 
America's Public Schools, Berliner and Biddle (1995) reviewed 
years of criticism, official and unofficial, on "failures” of American 
education, based on “evidence” that they say was either unavailable 
or misleadingly overgeneralized. None of the charges could be sup- 
ported. they claimed, and blamed, among other things, industrial- 
ists worried about overseas competition, a long-established 
316 
UNDERSTANDING READING 
tradition of “school-bashing," scapegoating of educators to divert 
attention from social problems, self-interest of some government 
officials, and irresponsible actions of the media (p. 7). In The Liter- 
acy Crisis: False Claims and Real Solutions, McQuillan (1998) 
also documented how an alleged “literacy crisis” is fabricated. 
Looking Ahead 
A discussion of the future of books, the future of literacy, and the 
future of education would fill any number of books. Interesting ar- 
ticles that happened to cross my desk on one day recently are 
Sutherland-Smith (2002) on changes in reading from page to 
screen. Miller, DeJean, and Miller (2002) comparing literacy by 
teachers with integrated [electronic] learning systems, and 
Trushell (2002) on the future of the book (warning that “prophecy 
is a mug’s game”). Bolter ( 1991 ) said that readers will have to learn 
to read electronic texts and “hypertexts" (networks of subtexts) in 
different ways from linear print-reading, following routes that the 
writers might not even have imagined. The notion of what consti- 
tutes readers and writers will change, and also the idea of "autono- 
mous texts” that exist independently of readers and writers, with 
clearly defined beginnings and ends. I would merely state that 
whatever opportunities, possibilities, and demands descend on 
readers in the future, the act of reading, as described and dis- 
cussed, will continue to be the same. And although learning will in- 
evitably take place in different circumstances, learning will also be 
the same. My advice to anyone wanting to anticipate and prepare 
for the future is to read extensively, evaluate objectively, and think 
all the time. 
The Interminable Controversy 
In the previous edition of Understanding Reading (Smith, 1994b), 
I contended that the “never-ending debate" between advocates of 
whole language and those of direct instruction (or “code empha- 
sis”) would never end because it is not something that research or 
experimentation will resolve, or indeed that partisans are ever 
likely to change their minds about. The disagreement is fundamen- 
tally a dispute over whether teachers can and should be trusted to 
teach and learners trusted to learn. Ttoo of the original protago- 
nists in the dispute, Jeanne Chall and Kenneth Goodman, were 
still vigorously contending with each other after 30 years until 
Chall died in 1999. Their many adherents on each side, with Good- 
NOTES 
317 
man and his wife Yetta still active, continue the struggle even more 
strenuously today as the issues have become more political. 
Chall, who launched the term “The Great Debate” in the subti- 
tle of her classic 1967 book Learning to Read, was fully aware of 
the gulf between the two points of view. In Chall (1992/1993) she 
declared, "Whole language proponents tend to view learning to 
read as a natural process, developing in ways similar to language 
[she means spoken language]. Therefore, like language, most 
whole language proponents say it is not necessary to teach read- 
ing directly. Direct instruction models, on the other hand, view 
reading as needing to be taught, and taught systematically” (p. 8). 
She cites the existence of many illiterate individuals as evidence 
that reading is not “natural” and needs to be taught, adding: “Gen- 
erally, direct instruction models favor the systematic teaching 
and learning of the relationships of sounds and symbols. This 
goes under many names — phonics, decoding, phonological 
awareness, word analysis, word attack, phonetic analysis, 
sound-symbol relations, etc.” (p. 8). 
In the same publication, Goodman (1992/1993) began by as- 
serting that whole language is “a broad, fundamental revolution in 
education” and that opposition to it is politically based. He said 
that reducing whole language “to a method of reading which is sim- 
ply the opposite of an approach that stresses phonics trivializes 
the broad nature of this revolution” (p. 8). Because he didn’t regard 
phonics as the issue, Goodman declined to debate it and instead 
focused on “why Jeanne Chall and others choose to frame the revo- 
lution in education over whether to use whole language or phonics 
in teaching reading.” He said that “her words and her name are 
used by those outside the research community — including the far 
right — who focus on reading as a simple means of attacking public 
education. They quote Chall to support their claim that whole lan- 
guage is a conspiracy to deprive children of literacy” (p. 9). Good- 
man provided examples of attacks on whole language, in the media 
and in Congress, that often mistakenly conflate it with “look-say” 
and other “methods” of teaching reading, blaming it for illiteracy, 
confusion in classrooms, and even a serious public health prob- 
lem, and frequently citing Chall as the authority for the condemna- 
tion of whole language and the desirability of phonics. 
Advocates of whole language tend to see phonics and direct in- 
struction as rigid, mindless, authoritarian, unfeeling, and unnec- 
essary procedures, and the whole language philosophy is seen by 
its opponents as unrealistic, unscientific, romantic, and anarchis- 
tic idealism, threatening to the maintenance of standards and 
318 
UNDERSTANDING READING 
teacher accountability. Goodman unwittingly provided a limitless 
supply of ammunition to his detractors in the title of one of his ear- 
liest publications when he characterized reading as "a 
psycholinguistic guessing game” (Goodman, 1967). It is frequently 
cited, always pejoratively. For a critical review of the rhetoric of 
whole language, see Moorman, Blanton, and McLaughlin (1992), 
and for a critical description of the ways in which the term “whole 
language” is misunderstood and misused in classrooms, see 
Dudley-Marling and Dippo (1991). 
As an example, Perfetti (1985) characterized the whole lan- 
guage or “psycholinguistic approach" as a whole word approach, 
which is misleading if not incorrect because “whole word ap- 
proach” typically refers to an instructional method of teaching 
words in isolation, in lists or on flashcards, which is far from the 
whole language position. Beck and Juel (1992) presented typical 
arguments for an early need to decode — as a “tool" of reading, 
which they saw as an alternative to “word attack" or “look-say” 
methods of learning by repetition. They called on “substantial re- 
search support” to assert that “letters correspond to the sounds 
in spoken language” — giving as examples cat and fat. Similar 
views by Ehri and her colleagues have already been cited on p. 
284. Despite the fact that studies of this kind are usually labeled 
"reading,” they normally don’t extend beyond the identification of 
words in isolation. In a spirited attack on whole language from the 
phonological awareness point of view. Liberman and Liberman 
(1992) consistently (and not untypically) compared reading with 
talking, contending that the former is unnatural and difficult 
while the latter is natural and easy. Gill (1992) made similar argu- 
ments. But if any comparison between spoken and written lan- 
guage is to be made, it should be between the “receptive" or 
“understanding” behaviors of reading and listening or the "pro- 
ductive” behaviors of writing and speaking. The manner in which 
infants learn to understand print should be compared with the 
manner in which they learn to understand the spoken language 
that surrounds them — both are based on bringing sense to mean- 
ingful situations, and very little additional effort is required to use 
the eyes rather than the ears. Using the vocal apparatus is another 
matter, and whether it is easier to learn to reproduce the pronun- 
ciation of a word than its spelling is a moot point, but it has noth- 
ing to do with teaching reading. In another impassioned assault. 
Thompson (1992) claimed that whole language lacks evidence 
and has a philosophy of abandoning students. He objected to the 
enthusiasm that permeates whole language. Miller (1991) offered 
NOTES 
319 
a friendly but pointed appraisal of the “whole language band- 
wagon,” observing evangelism, polemics, exploitation, deifica- 
tion, and vagueness and noting the problem of finding a brief 
answer for people wondering what whole language is. 
From the opposite point of view, Carbo (1988) criticized many of 
Chalks analyses of research findings in an article entitled “De- 
bunking the Great Phonics Myth,” noting that despite a 20-year 
emphasis on phonics in American classrooms, at the time of writ- 
ing the United States rated 49th in literacy out of 159 members of 
the United Nations and that New Zealand, which taught reading 
through a whole language approach, was ranked first. Chall (1989) 
was a response. 
Weaver (1994) provided a compendious text discussing whole 
language theory and practice, from the point of view of teachers and 
parents. She said that extensive phonics is worst for children in 
lower reading groups (or lower socioeconomic groups) and that 
most recommendations by researchers (like Chall, 1967) acknowl- 
edge that systematic phonics is not superior to other approaches 
and should always be part of more general meaning-based instruc- 
tion. The hyperbole usually comes from “interpreters” of the re- 
search, with a crusading agenda or links to complex and expensive 
commercial programs. Like Weaver. Krashen (2002a) argued that 
whole language is misrepresented and underrated in comparisons. 
See also his Three Arguments Against Whole Language and Why 
They Are Wrong (Krashen, 1999a). 
In an article entitled “Captives of the Script: Killing Us Softly 
with Phonics,” Meyer (2002) described problems of a teacher re- 
quired to teach reading as a series of episodes of non words and 
non sentences: “The mandated program is so oriented to precise- 
ness that her students are less willing to take risks as readers and 
writers.” Children wonder why they are taught to read such 
“words” as supermand, shoolbun , and reced. The word "script” 
also arose in an article by Dudley -Marling and Murphy ( 200 1 ) , dis- 
cussing difficulties many teachers have in teaching the language 
arts in the way they think most fitting and productive. The authors, 
experienced researchers and editors, argued that the professional 
autonomy of teachers is threatened by overregulation and the de- 
regulation of schooling. They listed and briefly critiqued the most 
popular commercial reading programs and the scripts they pro- 
vide “to keep all teachers on the same page." Pressure to adopt 
“scripted programs” from university-based researchers and mar- 
ket-based developers inhibited the improvisational skills and ex- 
perience of teachers. 
320 
UNDERSTANDING READING 
Phonics as a remedial reading technique is being introduced to 
secondary students, reported Los Angeles Times staff writer 
Duke Helfand (2002). He said 35,000 middle and high school stu- 
dents in the Los Angeles Unified School District lacking second- 
or third-grade reading skills were being given 2 hours a day of 
mandated phonics instruction, normally taught to 6-year-olds, in 
place of music and art. Classics like Romeo and Juliet were being 
replaced by books with big pictures, large print, and sentences 
like “Dad had a sad lad.” Most students disliked the program and 
the classes, feeling that they were being treated as being “dumb," 
“retarded," or “little kids.” Meanwhile, administrators were en- 
thusiastic. They regarded it as “the last chance” for many stu- 
dents. Success was claimed when students scored well on timed 
tasks reading a list of words like at, bat, sat, and Sam. 
In view of the extensive and sometimes indiscriminate manner in 
which a few “classic” or “important” publications are repeatedly 
cited on both sides of the debate, it may be relevant to note 
Cronbach’s (1992) comment that large numbers of citations don’t 
imply that an intended message has been widely understood, and 
Fiske and Campbell’s ( 1992) observation that citations don’t solve 
problems. Vincente and Brewer (1993) remarked that theoretical 
bias leads to misrepresentations in the reading and citing of original 
research — a bias that is increased as secondary sources are used. 
I conclude this section with a personal note. Many of the argu- 
ments in successive editions of this book have been generally sup- 
portive of the whole language philosophy, and I am frequently 
identified as a proponent of whole language. Nevertheless, I have 
never called myself a whole language person, nor have I ever of- 
fered a blanket testimonial for a whole language approach. I 
should perhaps explain my reluctance to embrace the label. (1)1 
dislike all labels and slogans, which I regard as open invitations to 
abandon reflective thought. Once someone says or does something 
that is identified as whole language, there is no need to question 
further, the person is pigeon-holed. (2) I dislike all "methods” of 
teaching reading, and whole language, despite the best intentions 
of many of its adherents, has become a method. Experts talk about 
how to “do” whole language, whole language materials are pro- 
duced, both for instruction and for assessment, and teachers look 
for whole language hints that they can introduce into their class- 
rooms. Whole language becomes something that is done, not a way 
of thinking about children and learning. (3) Labels get stolen. Peo- 
ple can call themselves “whole language teachers” while doing 
things totally alien to the underlying whole language philosophy. 
NOTES 
321 
Worse, they can encourage or persuade other teachers to engage in 
certain practices without understanding what they are doing. My 
preference in every case is for demonstrations, observations, de- 
bate, and critical thought. 
Federally Commissioned Studies 
In the 1980s the unending debate focused on a federally commis- 
sioned report entitled Becoming a Nation of Readers (Anderson, 
Hiebert, Scott, & Wilkinson, 1985), whose conclusions favored 
heavy early phonics instruction. The report was castigated in the fol- 
lowing manner by the editor of the International Reading Associa- 
tion’s annual review of reading research ( Weintraub, 1 986) : “There’s 
no guarantee that the big name is synonymous with quality. Even 
when well-funded and headed by a blue ribbon committee, a sup- 
posedly comprehensive review may be narrowly based, consider- 
ably less than comprehensive, and biased in its election of what is 
included and what is excluded .... I happen to concur that a very se- 
lective body of literature was included and some rather critical re- 
search excluded” (p. vi). 
The book Beginning to Read, by Marilyn Jager Adams (1990), 
was commissioned by the U.S. Department of Education follow- 
ing a request by Congress for an examination of the role of pho- 
nics in reading instruction. It is frequently referred to by both 
sides of the debate. Adams acknowledged the difficulty and un- 
naturalness of breaking down speech into separate sounds and 
relating these sounds to spellings, how difficult and laborious 
phonics and word recognition instruction is for children before 
they are experienced readers, and observed specifically that 
“spelling-sound relationships are not the basis of reading skills 
and knowledge” (p. 10) — but ended by “concluding” that the 
“symbol-sound system should be taught explicitly and early, to- 
gether with phonemic awareness training.” Dorothy Strickland 
and Bernice Cullinan, two members of a panel set up by the Cen- 
ter for the Study of Reading to advise the author, felt constrained 
to declare polite but firm disagreement in an Afterword published 
with the volume. They were particularly concerned by the empha- 
sis on phonics, the references to children who have not yet begun 
to receive instruction as “pre-readers” rather than “emergent 
readers,” the selection of studies, interpretation of research, and 
the amount of research that took place in decontextualized situa- 
tions. A similar controversy followed another federally commis- 
sioned report by Snow, Burns, and Griffin (1998) entitled 
322 
UNDERSTANDING READING 
Preventing Reading Difficulties in Young Children , which pre- 
sented similar conclusions. 
Possibly the mother of all educational conflicts was ignited by 
the report of the National Reading Panel (2000) — set up by the Na- 
tional Institute of Child Health and Human Development — whose 
main publication was entitled Teaching Children to Read: An Evi- 
dence Based Assessment of the Scientific Research Literature on 
Reading and Its Implications for Reading Instruction. The terms 
“evidence based” and “scientific research” in the title are signifi- 
cant. The report is voluminous, but there is a 35-page summary. 
The conclusions once again are a catalog of recommendations 
stressing direct instruction of phonics and phonemic awareness 
skills. Any attempt to examine the content of the report and of op- 
posing arguments in detail would simply be a further repetition of 
the interminable debate. But two unique characteristics of the re- 
port have wider implications. The first was the panel's decision to 
partition all research on reading into two categories, scientific — 
which in practice means supportive of the phonics point of view — 
and unscientific — which means everything else. The distinction 
has extensive consequences ranging from the funding of literacy re- 
search to the professional status of individual teachers and profes- 
sors of education. The distinction was used to justify the panel's 
own decision that of over 100,000 research reports that came be- 
fore them, only 428 warranted close attention, and of these only 38 
were used as a basis for their conclusions. The second unique 
characteristic was the panel's political stance. The recommenda- 
tions were prescriptive, and quickly mandated by federal and 
many state administrations as the sole basis of instruction for stu- 
dents in school and teachers at faculties of education, as well as for 
participants in inservice training. 
Vehement criticism of the report and its conclusions followed 
immediately, starting with the only experienced reading teacher on 
the panel, Joanne Yatvin. She wrote a minority report that was not 
published or referred to in the summary but subsequently ap- 
peared in an academic journal under the title Babes in the Woods: 
The Wanderings of the National Reading Panel (Yatvin, 2002). 
Yatvin believed the members of the panel "lost their integrity" be- 
cause government agencies at all levels are using the “science" of a 
flawed report to support changes in school instruction and teacher 
education. The titles of some related publications succinctly indi- 
cate the tenor of their contents: Misreading Reading: The Bad Sci- 
ence that Hurts Children (Coles, 2000); Literacy as Snake Oil: 
Beyond the Quick Fix (Larson, 2001): Resisting Reading Man- 
NOTES 
323 
dates: How to Triumph With the Truth (Garan, 2002); Reading 
Between the Lines (Metcalf, 2002); and The Politics of Phonics 
(Paterson, 2002). Articles by Yatvin, Garan, Paterson, and others 
are printed or reprinted in Big Brother and the National Reading 
Curriculum: How Ideology Trumped Evidence (Allington, 2002). 
My personal view of the panels is that they share two common 
but misguided characteristics: ( 1 ) a fixation on the alphabet and on 
letter-sound relationships, and (2) a conviction that children will 
only learn when given explicit instruction. Others might add strong 
elements of commercial, political, or fundamentalist bias. Do I 
have a better idea? I think this is a case where ultimate truth is un- 
attainable. But anyone who objectively studies language and ob- 
serves children can see that characteristics 1 and 2 are both 
wrong. It is not difficult to see error. On the other hand, I would not 
presume to speculate on "what really happens in the head.” Scans 
of various kinds may show activity in different regions of the brain 
at different times, but they no more explain reading than they ex- 
plain consciousness, attention, awareness, or any other mental 
state. Some scientists (real scientists) believe we may never resolve 
such issues because we don’t (and perhaps can’t) ask the right 
questions; our organized perceptions of gross events in the physi- 
cal world don’t relate to quantum events in the physical brain in 
any way that we can comprehend (see Cohen & Schooler, 1997; 
Dodwell, 2000). The brain can never understand itself. This does- 
n’t mean that we are helpless victims of ignorance or that anything 
goes; it means that we should avoid the pursuit of fictions and re- 
spect what can be unambiguously observed about the behavior, ca- 
pacities, and feelings of people. Such an attitude won’t end the 
interminable controversy, but it could take us beyond it. 
This page intentionally left blank 
Glossary 
This list doesn’t define words as a dictionary does, but rather indi- 
cates the general way that certain terms are employed in this book. 
Figures in parentheses indicate the chapter (or chapter notes) in 
which a term was first used or primarily discussed. Terms in ital- 
ics appear elsewhere in the glossary. 
Aesthetic reading: Reading done primarily for experience; contrasted 
with efferent reading (4). 
Artificial intelligence: The study of systems designed to emulate human 
language and thought with computer technology (Preface). 
Basics: See skills learning. 
Cat and dog problem: Example of the fact that distinctive features can- 
not be explicitly taught but must be learned by the testing of hypothe- 
ses (12). 
Categories: See cognitive categories. 
Category interrelationships: The various ways in which cognitive cate- 
gories can be combined as a basis for prediction or action (2). 
Channel capacity: Limit to the amount of information that can pass 
through any part of an information-processing system (4). 
Cognition: A particular organization of knowledge in the brain, or the re- 
organization of such knowledge (2). See cognitive structure. Also 
called thinking. 
325 
326 
UNDERSTANDING READING 
Cognitive categories: Prior decisions to treat some aspects of experience 
as the same, yet as different from other aspects of experience: the con- 
stantly developing framework of cognitive structure (2). 
Cognitive questions: The specific information sought by the brain to 
make a decision among alternatives: the range of a prediction (2). 
Cognitive science: An area of common concern in psychology, linguis- 
tics, and the design of computer systems, related to the manner in 
which knowledge can be acquired, stored, retrieved, and utilized (Pref- 
ace). See also artificial intelligence. 
Cognitive structure: The totality of the brain’s organization of knowl- 
edge; everything an individual knows (or believes) about the world. 
Comprises cognitive categories, feature lists, and category interrela- 
tionships. Also referred to as long-term memory, the theory of the 
world, and prior knowledge (2). 
Comprehension: The interpretation of experience; relating new infor- 
mation to what is already known; asking cognitive questions and be- 
ing able to find answers to them; a normal state, the absence of 
confusion (2). 
Constraint: The exclusion or reduced probability of certain alternatives: 
the mechanism of redundancy (4). 
Context: The setting, physical or linguistic, in which words occur and 
that places constraints on the range of alternatives that these words 
might be (3). 
Context-dependent language: Spoken or written language coherent 
within itself and not related to the concurrent physical situation in 
which it occurs (3). 
Conventions: Arbitrary or accidental forms of behavior made meaning- 
ful by mutual understanding of and respect for their use and implica- 
tions (3). 
Criterial set: A set of distinctive features within a feature list that per- 
mits an identification to be made on minimal information from a given 
set of alternatives (7). 
Criterion level: The amount of information an individual requires to 
make a particular decision, varying with the perceived uncertainty 
of the situation and the perceived risk and cost of making a mistake 
(4). 
Decoding to sound: The view that reading is accomplished by transforming 
print into actual or subvocalized (implicit) speech through the exercise of 
spelling-sound correspondences. See also phonological recoding (9). 
Deep structure: The meaningful aspect oflanguage: the interpretation of 
surface structure (3). 
Demonstrations: Displays, by people or artifacts, of how something is 
done (12). 
Direct instruction: An educational philosophy based on the belief that 
learning takes place most effectively when learners are told in spe- 
cific detail what they should learn and are monitored closely to en- 
sure that they do so (13). Frequently contrasted with whole 
language. 
GLOSSARY 
327 
Discourse structure: Conventions concerning the organization of lan- 
guage, for example, turn-taking and interruption in speech, and para- 
graphing and repetition in texts (3). 
Distinctive features: Significant differences among visual (or acoustic) 
patterns, that is, differences that make a difference. For reading, any 
aspect of visual information that permits distinctions to be made 
among alternative letters, words, or meanings (7). See also/eature list. 
Distributional redundancy: Reduction of uncertainty because alterna- 
tives are not equally probable (4). May exist in letters or words. See 
also featural redundancy. 
Efferent reading: Reading done primarily for information; contrasted 
with aesthetic reading (4). 
Engagement: The interaction of a brain with a demonstration ; the act of 
learning (12). 
Ethnographic research: Observation of behavior in natural contexts; 
nonintrusive research; also termed “naturalistic research”; contrasted 
with controlled experimentation (13). 
Event knowledge: Hypothesized mental representations of patterns of 
behavior in specific events; clusters of related expectations. See 
scheme (2). 
Expectations: See prediction. 
Featural redundancy: In reading, redundancy among the distinctive 
features of print as a consequence of constraints on letter or word oc- 
currence (7). 
Feature analysis: A theory of pattern recognition proposing that visual 
configurations such as digits, letters, or words are identified by the 
analysis of distinctive features and their allocation to feature lists ; in 
contrast to template theory (7). 
Feature list: A cognitive specification or “set of rules” for particular 
combinations of distinctive features that will permit identification 
in reading (7). 
Feedback: Information that permits a decision whether an hypothesis is 
right or wrong (7). 
Fixation: The pause for the selection of visual information as the gaze 
rests at one place in the text between saccades (5). 
Functional equivalence: Specification of the same cognitive category by 
two or more feature lists (7). 
Genre: Schemes for conventional structures of written language (3). 
Grammar: See syntax. 
Grapheme: A letter of the alphabet, one of 26 alternatives (9). 
Grapheme unit: Single graphemes, or combinations of two or more let- 
ters of the alphabet that function as single graphemes (9). 
Hypothesis: A tentative modification of cognitive structure ( cognitive 
categories, feature lists, or category interrelationships ) that is tested 
as a basis for learning (12). 
Identification: In reading, a cognitive decision among letter, word, or 
meaning alternatives based on the analysis of selected visual informa- 
tion in print (7). 
328 
UNDERSTANDING READING 
Immediate meaning identification: The comprehension of language 
without the prior identification of words (10). 
Immediate word identification: The identification of a word on sight, 
without information from another person and without the prior identi- 
fication of letters or letter combinations within the word (8). 
Information: Any property of the physical environment that reduces un- 
certainty, eliminating or reducing the probability of alternatives 
among which a perceiver must decide (4). 
Learning: The modification or elaboration of cognitive structure, specifi- 
cally the establishment of new or revised cognitive categories, category 
interrelationships, or feature lists (12). 
Letter-sound correspondence: See spelling-sound correspondence. 
Lexical access: Computer-derived metaphor for making sense of 
words in reading or speech through reference to an internal lexicon 
(3). 
Lexicon: A hypothesized mental store of knowledge about words, includ- 
ing their sound, spelling, and meaning. See lexical access (3). 
Long-term memory: The totality of an individual's knowledge and be- 
liefs, including summaries of past experience and ways of interacting 
with the world (2). 
Mastery learning: See skills learning. 
Meaning: A relative term; the interpretation that a reader places on text 
(the answer to a cognitive question). Alternatively, the interpretation 
an author or third party expects a reader to place on text. The product 
of comprehension (2). 
Meaningful (ness): In reading, a text that is relevant to a reader's pur- 
pose, expectations, and understanding (2). 
Mediated meaning identification: An inferior alternative to immediate 
meaning identification, attempting to derive meaning by the prior 
identification of words (10). 
Mediated word identification: A less efficient alternative to immediate 
word identification, requiring analysis of letters or letter combina- 
tions within the word (9). 
Memory: See sensory store, short-term memory, and long-term memory. 
Metacognition: Thought about one s own thinking, understanding, or 
learning (2). 
Metalanguage: Language about language (3). 
Metalinguistic awareness: Understanding of metalanguage, notably 
the way aspects of spoken and written language may be discussed in 
reading instruction (13). 
Noise: A signal that reduces no uncertainty (4). 
Nonvisual information: Prior knowledge "behind the eyes" that reduces 
uncertainty in advance and permits identification decisions with less 
visual information (5). 
Orthography: Spelling; the arrangement of letters in words (9). 
Perception: Identification decisions made by the brain; subjective 
awareness of these decisions (2). 
GLOSSARY 
329 
Phoneme: One of about 45 discriminable categories of significantly dif- 
ferent speech sounds in English. Other languages have different sets of 
roughly the same number of phonemes (3). 
Phonetics: The scientific study of the sound structure of speech; has 
nothing to do with reading (9). See phonics. 
Phonics: Reading instruction based on the assumption that reading is 
decoding to sound and requires learning spelling-sound correspon- 
dences (9). Sometimes erroneously referred to as phonetics. 
Phonological awareness: Ability to detect sounds in speech that are sup- 
posed to be represented by the letters of the alphabet. Also termed pho- 
nemic awareness (9). 
Phonological recoding: Transforming written words to sound in order 
to understand their meaning (as opposed to understanding written 
words directly) (9). 
Prediction: The prior elimination of unlikely alternatives; (in reading) 
the remaining set of alternatives among which an identification deci- 
sion will be made from selected visual information in print (2). 
Prior knowledge: Relevant knowledge already possessed that reduces 
uncertainty in advance and facilitates identification decisions (2). 
Also referred to as cognitive structure and nonvisual information. 
Procedural knowledge: Knowledge of integrated sequences of behavior; 
see also propositional knowledge, event knowledge (2). 
Propositional knowledge: Knowledge in the form of internalized state- 
ments (such as facts, proverbs, formulas); also referred to as declara- 
tive knowledge (as opposed to procedural knowledge) (2). 
Psycholinguistics: An area of common concern in psychology and lin- 
guistics studying how individuals learn and use language (Preface). 
Redundancy: Information that is available from more than one source. 
In reading, may be present in the visual information of print, in the or- 
thography, the syntax, the meaning, or in combinations of these 
sources. Redundancy may be distributional or sequential. Redun- 
dancy must always reflect nonvisual information ; prior knowledge on 
the part of the reader permits redundancy to be used (4). 
Regression: An eye movement ( saccade ) from right to left along a line or 
upward on a page (in English and similar writing systems) (5). 
Saccade: Movement of the eyes as the gaze moves from onefixation to an- 
other in reading (5). 
Scenario: A generalized mental representation of conventional patterns 
of behavior in specific situations. See also scheme (2). 
Scheme: A generalized mental representation of complex patterns of be- 
havior or events; also referred to as schema, plural schemata. See 
also scenario, script, event knowledge (2). 
Script: A generalized mental representation of conventional behavior on 
specific occasions (2). 
Semantics: The meaningful aspect of language; the study of this aspect (2). 
Sensitivity: The prior learning state of the brain; readiness for engage- 
ment: absence of expectation that learning will not occur (12). 
330 
UNDERSTANDING READING 
Sensory store: In vision, the very brief retention of visual information 
while identification decisions are made; also called the visual image 
( 6 ). 
Sequential redundancy: Reduction of uncertainty attributable to con- 
straints on the number or relative probability of likely alternatives in 
context (8, 10); may exist among letters or words and as a consequence 
of a reader’s or listener’s expectations. See also featural redundancy. 
Short-term memory: The limited and constantly changing content of 
what is attended to at any particular moment (6). 
Significant difference: A difference in the physical properties of an event 
that forms the basis of an identification decision (7). 
Situation-dependent language: Spoken or written language referring to 
and made meaningful by the concurrent physical situation in which it 
occurs (3). 
Skills learning: The view that learning takes place most effectively when 
learners are systematically taught and rigorously tested in "the basics" 
of what is to be learned — in literacy, for example, the alphabet, phon- 
ics, and rules of spelling and punctuation (13). Known also as direct 
instruction and mastery learning. 
Specification: A constantly changing outline in a reader's (or writer’s) 
mind about the structure and content of a text; the basis of prediction 
in reading (2). 
Spelling-sound correspondence: The co-occurrence of a particular let- 
ter or group of letters in a written word and the assumed sound of the 
same part of the word in speech (9). 
Surface structure: The physical properties of language: for reading — vi- 
sual information (3). 
Syntax: The manner in which words are organized in meaningful lan- 
guage; also referred to as "grammar” (2). 
Tachistoscope: A projector or other viewing device with a shutter or 
timer controlling the presentation of visual information for brief peri- 
ods of time (5). 
Template theory: A theory of pattern recognition that visual configura- 
tions such as digits, letters, or words can be identified by comparison 
with prestored representations or templates in the brain: in contrast to 
featural analysis (7). 
Text: A meaningful (or potentially meaningful) instance of written lan- 
guage; can range from a word to an entire book. 
Theory: In science, a summary of a scientist’s past experience, the basis 
for interpreting new experience and for predicting future events (2). 
Theory of the world: The brain’s theory : also known as cognitive struc- 
ture and long-term memory (2). 
Transformational grammar: Part of the theory of the world of every lan- 
guage user: the bridge between deep structure and surface structure (3). 
Uncertainty: The amount of information required to make an identifica- 
tion decision, determined by the number of alternative decisions that 
could be made, the perceived probability of each alternative, and the 
individual’s criterion level for making the decision (4). 
GLOSSARY 
331 
Understanding: Sec comprehension. 
Visual image: See sensory store. 
Visual information: In reading, information that is available to the brain 
through the eyes from the surface structure of print, for example, from 
the ink marks on a page (3). 
Whole language: An educational movement based on the belief that lan- 
guage learning takes place most effectively when learners are engaged 
collaboratively in meaningful and purposeful uses of language, as op- 
posed to exercises, drills, and tests (13). Sometimes referred to as the 
naturalistic approach or (misleadingly) as child-centered learning, 
and known in Britain as real books. Frequently contrasted with direct 
instruction. 
This page intentionally left blank 
References 
Adams, Marilyn Jager (1990). Beginning to Read: Learning and 
Thinking About Print. Cambridge, MA: MIT Press. 
Adams, Marilyn Jager ( 1 994). Phonics and beginning reading instruction. 
In Fran Lehr and Jean Osborn (Eds.), Reading , Language and Liter- 
acy: Instructionfor the Twenty-First Century. Hillsdale, NJ: Lawrence 
Erlbaum Associates. 
Allington, Richard L. (Ed.) (2002). Big Brother and the National Reading Cur- 
riculum: How Ideology Trumped Evidence. Portsmouth, NH: Heinemann. 
Anderson, John R. (1980). Cognitive Psychology and Its Implications. 
San Francisco: Freeman. 
Anderson, Richard C., Elfrieda H. Hiebert, Judith A. Scott, and Ian A. G. 
Wilkinson (1985). Becoming a Nation of Readers: The Report of the 
Commission on Reading. Washington, DC: National Academy of Edu- 
cation. 
Anderson, Richard C. and Andrew Ortony (1975). On putting apples into 
bottles: A problem in polysemy. Cognitive Psychology, 7, 167-180. 
Anderson, Richard C. and R David Pearson (1984). A schema-theoretic 
view of basic processes in reading comprehension. In Pearson, E David. 
Rebecca Barr, Michael L. Kamil, and Peter Mosenthal (Eds.), Handbook 
of Reading Research (Vol. 1). New York: Longman. 
Anglin, Jeremy M. (1993). Vocabulary development: A morphological 
analysis. Monoqraphs for the Society for Research in Child Develop- 
ment. #238, 58(10). 
Applebee, Arthur N. (1977). A sense of story. Theory Into Practice. 16. 
342-347. 
333 
334 
UNDERSTANDING READING 
Armbruster, Bonnie B. and T. H. Anderson (1984). Content area text- 
books. In Anderson, Richard C.. Jean Osborn, and Robert J. Tierney 
(Eds.), Learning to Read in American Schools: Basal Readers and 
Content Texts. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Atwell, Nancie ( 1987). In the Middle: Reading, Writing and Learning with 
Adolescents. Portsmouth, NH: Heinemann. 
Atwell, Nancie (1991). Side by Side: Essays on Teaching to Learn. 
Portsmouth, NH: Heinemann. 
Averbach, E. and A. S. Coriell ( 1961 ). Short-term memory in vision. Bell 
Systems Technical Journal, 40, 309-328. 
Baddeley, Alan (1992). Working memory. Science, 255, 556-559. 
Bahrick, Harry R and Lynda K. Hall ( 1991 ). Lifetime maintenance of high 
school mathematics content. Journal of Experimental Psychology: 
General, 120(1), 20-33. 
Barclay, J. R. (1973). The role of comprehension in remembering sen- 
tences. Cognitive Psychology, 4, 229-252. 
Barlow, John A. (1981). Mass line leadership and thought reform in 
China. American Psychologist, 36(3), 300-309. 
Barr, Rebecca C. ( 1972). The influence of instructional conditions on word 
recognition errors. Reading Research Quarterly, 7, 509-529. 
Barr, Rebecca C. (1974). The effect of instruction on pupil reading strate- 
gies. Reading Research Quarterly, 10, 555-582. 
Bartlett, Frederick C. (1932). Remembering: A Study in Experimental 
and Social Psychology. London: Cambridge University Press. 
Baumann, James F. (1992). Basal reading programs and the deskilling of 
teachers: A critical examination of the argument. Reading Research 
Quarterly, 27(4), 390-398. 
Beck, Isabel L. and Connie Juel (1992). The role of decoding in learning to 
read. In Samuels. S. Jay and Alan E. Farstrup (Eds.), What Research 
Has to Say About Reading Instruction (2nd ed.). Newark, DE: Interna- 
tional Reading Association. 
Beck, Isabel L., Margaret G. McKeown, Richard C. Omanson, and Martha 
T. Pople (1984). Improving the comprehensibility of stories: The effects 
of revisions that improve coherence. Reading Research Quarterly. 
19(3), 263-277. 
Beers, Terry ( 1987). Commentary: Schema-theoretic models of reading: Hu- 
manizing the machine. Reading Research Quarterly. 22(3), 369-377. 
Benton, Arthur L. (1980). The neuropsychology of facial recognition. 
American Psychologist, 35(2), 176-186. 
Berdiansky, Betty, Bruce Cronnell, and John A. Koehler ( 1969). Spelling- 
Sound Relations and Primary Form-Class Descriptions for Speech- 
Comprehension Vocabularies of 6-9 Year-Olds. Inglewood, CA: South- 
west Regional Laboratory for Educational Research and Development. 
Technical Report No. 15. 
Berliner, David C. and Bruce J. Biddle (1995). The Manufactured Crisis: 
Myths, Fraud, and the Attack on America’s Public Schools. Reading, 
MA: Perseus. 
REFERENCES 
335 
Berninger, Virginia Wise and Sarah O. Colwell (1985). Relationships be- 
tween neurodevelopmental and educational findings in children aged 6 
to 12 years. Pediatrics, 75, 697-702. 
Black, Alison, Paul Freeman, and Philip N. Johnson-Laird (1986). Plausi- 
bility and the comprehension of text. British Journal of Psychology, 
77(1), 51-62. 
Bloome, David and Ann Egan-Robertson (1993). The social construction 
of intertextuality in classroom reading and writing lessons. Reading 
Research Quarterly, 28(4), 304-333. 
Blough, DonaldS. (1982). Pigeon perception ofletters of the alphabet. Sci- 
ence, 218, 397-398. 
Bolter, Jay David (1991). Writing Space: The Computer, Hypertext, and 
the History of Writing. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Boulding. Kenneth E. (1981). Human knowledge as a special system. Be- 
havioral Science, 26, 93-102. 
Brady, Susan A. and Donald R Shankweiler (1991). Phonological Pro- 
cesses in Literacy: A Tribute to Isabelle Y. Liberman. Hillsdale, NJ: 
Lawrence Erlbaum Associates. 
Bransford, John D., Barry S. Stein, and Nancy J. Vye (1982). Helping stu- 
dents learn how to learn from written texts. In Singer, M. H. (Ed. ), Com- 
petent Reading, Disabled Reader: Research and Application. 
Hillsdale. NJ: Lawrence Erlbaum Associates. 
Bridges, Allayne, Chris Sinha, and Valerie Walkerdine ( 1981). The devel- 
opment of comprehension. In Wells, Gordon (Ed.), Learning Through 
Interaction: The Study of Language Development. Cambridge, UK: 
Cambridge University Press. 
Broadbent, Donald E. (1967). The word-frequency effect and response 
bias. Psychological Review, 74, 1-15. 
Broadbent, Donald E. and Margaret H. R Broadbent (1975). Some further 
data concerning the word frequency effect. Journal of Experimental 
Psychology: General, 104, 297-308. 
Brown, Ann L. ( 1982). Learning how to learn from reading. In Langer, Ju- 
dith A. and M. Trika Smith-Burke (Eds.), Reader Meets Author/Bridging 
the Gap. Newark, DE: International Reading Association. 
Brown. Ann L. and Annemarie S. Palincsar (1989). Guided, cooperative 
learning and individual knowledge acquisition. In Resnick, Lauren 
(Ed.), Knowing, Learning, and Instruction: Essays in Honor of Robert 
Glaser. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Brown. H. Douglas (1970). Categories of spelling difficulty in speakers of 
English as a first and second language. Journal of Verbal Learning and 
Verbal Behavior, 9, 232-236. 
Brown. Roger and David McNeill ( 1966). The “tip of the tongue” phenome- 
non. Journal of Verbal Learning and Verbal Behavior. 5(4), 325-337. 
Bruner, Jerome S. (1986). Actual Minds. Possible Worlds. Cambridge. 
MA: Harvard University Press. 
Bruner, Jerome S. and D. O’Dowd ( 1958). A note on the informativeness of 
parts of words. Language and Speech, 1 , 98-101 . 
336 
UNDERSTANDING READING 
Bruner, Jerome and Susan Weisser ( 1991 ). The invention of self: Autobi- 
ography and its forms. In Olson, David R. and Nancy Torrance (Eds.), 
Literacy and Orality. Cambridge, UK: Cambridge University Press. 
Bryant, Peter and Lawrence Impey (1986). The similarities between normal 
readers and developmental and acquired dyslexics. Cognition, 24, 1 21-137. 
Buckner, Aimee (2002). Teaching in a world focused on testing. Language 
Arts, 79(3), 212-215. 
Burton, Mike (1992). Good morning, Mr ... er. New Scientist, 1331 1806). 
39-41. 
Cadiero-Kaplan, Karen. (2002). Literacy ideologies: Critically engaging 
the language arts curriculum. Language Arts, 78(5), 372-381. 
Cairney, Trevor H. (1990). Teaching Reading Comprehension: Meaning 
Makers at Work. Milton Keynes: Open University. 
Cairney, Trevor (1992). Mountain or mole hill: The genre debate viewed 
from “down under." Reading, 36, 23-29. 
Calfee. Robert C. (1983). The mind of the dyslexic. In Hartwig, Elinor Linn 
(Ed.), Annals of Dyslexia, Vol. 33. Baltimore: Orton Dyslexia Society. 
Calkins, Lucy (1980). When children want to punctuate: Basic skills be- 
long in context. Language Arts, 57, 567-573. 
Caplan, David (Ed.) (1980). Biological Studies of Mental Processes. Cam- 
bridge, MA: MIT Press. 
Carbo, Marie (1988). Debunking the great phonics myth. Phi Delta 
Kappan, 70(3), 226-240. 
Carbo, Marie, Rita Dunn, and Kenneth Dunn (1987). Teaching Students 
to Read Through Their Individualized Learning Styles. Reston, VA: 
Prentice-Hall. 
Carey, Susan (1978). The child as word learner. In Halle. Morris, J. 
Breslin, and George A. Miller ( Eds.), Linguistic Theory and Psychologi- 
cal Reality. Cambridge, MA: MIT Press. 
Carpenter, Patricia A. and Marcel A. Just (1975). Sentence comprehen- 
sion: A psycholinguistic processing model of verification. Psychological 
Review, 82, 45-73. 
Carpenter, Patricia A. and Meredyth Daneman (1981). Lexical retrieval 
and error recovery in reading: A model based on eye fixations. Journol 
of Verbal Learning and Verbal Behavior, 20. 137-160. 
Carver, Ronald P ( 1985). How good are some of the world's best readers? 
Reading Research Quarterly, 20(4), 389-419. 
Carver, Ronald R (1992). Reading rate: Theory, research and practical im- 
plications. Journal of Reading, 36(2), 84-95. 
Castiglioni-Spalton, M. and Linnea C. Ehri (2003). Phonemic awareness 
instruction: Contribution of articulatory segmentation to novice begin- 
ners’ reading and writing. Scientific Studies of Reading, 7( 1 ). 25-52. 
Cattell, James McKeen (1973). Ueber die Zeit der Erkennung und 
Benennung von Schriftzeichen, Bildern und Farben (On the time re- 
quired for recognizing and naming letters and words, pictures and 
colors.] Philosophische Studien (1885), 2, 635-650. Translated and 
reprinted in Poffenberger, A. T. (Ed.). James McKeen Cattell, Man of 
Science, 1860-1944 (Vol. 1). New York: Arno Press. 
REFERENCES 
337 
Cazden, Courtney B. ( 1992). Whole Language Plus: Essays on Literacy in 
the United States and New Zealand. New York: Teachers College Press. 
Chafe, Wallace (1992). Immediacy and displacement in consciousness 
and language. In Stein, Dieter (Ed.), Cooperating with Written Texts: 
The Pragmatics and Comprehension of Written Texts. Berlin: Mouton 
de Gruyter, pp. 231-255. 
Chall, Jeanne S. (1967). Learning to Read: The Great Debate. New York: 
McGraw-Hill. 
Chall. Jeanne S. ( 1989). Learning to read: The great debate 20 years later: 
A response to “Debunking the great phonics myth.” Phi Delta Kappan. 
70(7), 521-538. 
Chall. Jeanne S. ( 1992/1993). Research supports direct instruction mod- 
els. In Point/Counterpoint: Whole language versus direct instruction 
models. Reading Today, 10(3), 8-10. 
Chambers, Maryl, Adrian Jackson, and Mary Rose (1993). Children De- 
veloping as Readers: The Avon Collaborative Reading Project: Bris- 
tol, UK: County of Avon. 
Cho, Kyung-Sook and Stephen D. Krashen (1994). Acquisition of vocabu- 
lary from the Sweet Valley Kids series: Adult ESL acquisition. Journal 
of Reading, 37(8), 662-667. 
Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton. 
Chomsky, Noam (1959). Review of Verbal Learning (by B. F. Skinner). 
Language, 35, 26-58. 
Chomsky, Noam and Morris Halle (1968). Sound Pattern of English. New 
York: Harper and Row. 
Christensen, Carol A. (1997). Onset, rhymes, and phonemes in learning to 
read. Scientific Studies of Reading, 1(4), 341-358. 
Clark, Eve V. (1993). The Lexicon in Acquisition. Cambridge, UK: Cam- 
bridge University Press. 
Clark, Eve V and Barbara Frant Hecht (1983). Comprehension, production, 
and language acquisition. Annual Review of Psychology, 34, 325-349. 
Clark. Margaret M. ( 1976). Young Fluent Readers. London: Heinemann. 
Clay, Marie (1985). The Early Detection of Reading Difficulties. 
Portsmouth, NH: Heinemann. 
Clay, Marie ( 1992/1993). Language policy and literacy learning. Reading 
Today, 10(3), 3-4. 
Clay, Marie M. (1979). Theoretical research and instructional change: A 
case study. In Resnick, Lauren B. and Phyllis A. Weaver (Eds.), Theory 
and Practice of Early Reading. Hillsdale, NJ: Lawrence Erlbaum Asso- 
ciates. 
Clymer, Theodore (1963a). The utility of phonic generalizations in the pri- 
mary grades. The Reading Teacher. 16. 252-258. 
Clymer, Theodore (1963b). Does "can” mean “should”? The Reading 
Teacher, 16. 217. 
Cohen. Jonathan D., and Schooler, Jonathan W. (Eds.) (1997). Scientific 
Approaches to Consciousness. Mahwah, NJ: Lawrence Erlbaum As- 
sociates. 
Coles, Gerald (1987). The Learning Mystique. New York: Pantheon. 
338 
UNDERSTANDING READING 
Coles, Gerald (2000). Misreading Reading: The Bad Science That Hurts 
Children. Portsmouth, NH: Heinemann. 
Cornman, O. (1902). Spelling in the Elementary School. Boston, MA: 
Ginn. 
Coulmas, Florian (1990). The Writing Systems of the World. Oxford: 
Blackwood. 
Coulmas, Florian (1992). On the relationship between writing system, 
written language, and text processing. In Stein, Dieter (Ed.), Cooper- 
ating with Written Texts: The Pragmatics and Comprehension of 
Written Texts. Berlin: Mouton de Gruyter. 
Courts, Patrick L. (1991). Literacy and Empowerment: The Meaning 
Makers. New York: Bergin and Garvey. 
Cronbach, Lee J. ( 1992). Four Psychological Bulletin articles in perspec- 
tive. Psychological Bulletin. 112(3), 389-392. 
Crowder, Robert G. and Richard K. Wagner (1991). The Psychology of 
Reading: An Introduction. New York: Oxford University Press. 
Csikszentmihalyi, Mihaly ( 1990). Flow — The Psychology of Optimal Ex- 
perience. New York: Harper & Row. 
Davis, Steven (Ed.) ( 1991 ). Pragmatics: A Reader. Oxford: Oxford Univer- 
sity Press. 
Deely, John ( 1982). Introducing Semiotics. Bloomington: Indiana Univer- 
sity Press. 
DeFord, Diane E. (1981). Literacy: Reading, writing and other essentials. 
Language Arts, 58(5), 652-658. 
Dodge, R. (1900). Visual perceptions during eye movement. Psychological 
Review. VII. 454-465. 
Dodwell, Peter (2000). Brave New Mind: A Thoughtful Inquiry into the 
Nature and Meaning of Mental Life. New York: Oxford University 
Press. 
Doehring, Donald G. (1976). Acquisition of rapid reading responses. 
Monographs of the Society for Research in Child Development. 41(2). 
No. 165. 
Dombey, Henrietta (1988). Stories at home and at school. In Lightfoot, 
Martin and Nancy Martin (Eds.), The Word for Teaching is Learning: 
Essays for James Britton. London: Heinemann. 
Dorman, Casey ( 1985). Defining and diagnosing dyslexia: Are we putting 
the cart before the horse? Reading Research Quarterly. 20(4). 
505-508. 
Downing, John (1977). How society creates reading disability. Elemen- 
tary School Journal, 77. 274-279. 
Downing, John (1979). Reading and Reasoning. New York: 
Springer-Verlag. 
Downing, John and Peter Oliver (1973/1974). The child's conception of "a 
word." Reading Research Quarterly, 9(4), 568-582. 
Duckett, Peter (2002). New insights: Eye fixations and the reading pro- 
cess. Talking Points. 13(2). 16-21. 
Dudley-Marling, Curt and Don Dippo ( 1991 ). The language of whole lan- 
guage. Language Arts. 68. 548-554. 
REFERENCES 
339 
Dudley-Marllng, Curt and Sharon Murphy (2001). Changing the way we 
think about language arts. Language Arts, 78(6), 574-578. 
Duke, Nell K. (2000), For the rich it’s richer: Print experiences and envi- 
ronments offered to children in very low- and very high-socioeconomic 
status first-grade classrooms. American Educational Research Jour- 
nal, 37(2), 441-478. 
Dunn-Rankin, Peter ( 1968). The similarity of lower case letters of the Eng- 
lish alphabet. Journal of Verbal Learning and Verbal Behavior, 7, 
990-995. 
Durkin, Dolores (1981). What is the value of the new interest in reading 
comprehension? Language Arts, 58(1), 23-43. 
Durkin, Dolores (1984). Poor black children who are successful readers: 
An investigation. Urban Education, 19, 53-76. 
Eagleton, Terry (1983). Literacy Theory: An Introduction. Minneapolis: 
University of Minnesota Press. 
Eckhoff, Barbara (1983). How reading affects children’s writing. Lan- 
guage Arts, 60(5), 607-616. 
Edelsky, Carole (1991). With Literacy and Justice for All: Rethinking the 
Social in Language and Education. Bristol, PA: Falmer. 
Ehri, Linnea C. ( 1985). How orthography alters spoken language compe- 
tencies in children learning to read and spell. In Downing, John and 
Renate Valtin (Eds.), Language Awareness and Learning to Read. 
New York: Springer-Verlag. 
Ehri, Linnea C. (1992). Reconceptualizing the development of sight word 
reading and its relationship to recoding. In Gough, Philip B., Linnea C. 
Ehri, and Rebecca Treiman (Eds.), Reading Acquisition. Hillsdale, NJ: 
Lawrence Erlbaum Associates. 
Ehri, Linnea C. (1997). Sight word learning in normal readers and dys- 
lexics. In Blachman, Benita (Ed.), Foundations of Reading Acquisi- 
tion and Dyslexia: Implications for Early Intervention. Mahwah, NJ: 
Lawrence Erlbaum Associates. 
Eldredge, J. Lloyd ( 1990). Increasing the performance of poor readers in 
the third grade with a group-assisted strategy. Journal of Educational 
Research, 84(2), 69-77. 
Elley, Warwick (1989). Vocabulary acquisition from listening to stories. 
Reading Research Quarterly, 24, 174-187. 
Elley, Warwick B. (1992). How in the World Do Students Read? The IEA 
Study of Reading Literacy. The Hague: International Association for the 
Evaluation of Educational Achievement. 
Elley, Warwick B. (1996). Raising literacy levels in third world countries: 
A method that works. Paper presented at the World Conference on Lit- 
eracy, Philadelphia, March 1996. 
Elley, Warwick B. and Francis Mangubhai (1983). The impact of reading 
on second language learning. Reading Research Quarterly, 1 9, 53-67. 
Erdmann, B. and R. Dodge (1898). Psychologische Untersuchungen 
ueber das Lesen auf Experimenteller Grundlage. Halle: Niemeyer. 
Ericsson, K. Anders, William G. Chase, and Steve Falcon (1980). Acquisi- 
tion of a memory skill. Science, 208, 1 181-1 182. 
340 
UNDERSTANDING READING 
Ewoldt, Carolyn (1981). A psycholinguistic description of selected deaf 
children reading in sign language. Reading Research Quarterly, 17(1), 
58-89. 
Fernandez-Armesto, Felipe (2000). Civilizations: Culture, Ambition, and 
the Transformation of Nature. New York: Free Press. 
Ferreiro, Emilia ( 1978). What is written in a written sentence? A develop- 
mental answer. Journal of Education, 160(4), 25-39. 
Ferreiro, Emilia (1985). Literacy development: A psychogenetic perspec- 
tive. In Olson, David R., Nancy Torrance, and Angela Hildyard (Eds.). 
Literacy, Language and Learning: The Nature and Consequences of 
Reading and Writing. Cambridge. UK: Cambridge University Press. 
Ferreiro, Emelia and Ana Teberosky (1982). Literacy Before Schooling. 
Exeter, NH: Heinemann. 
Fiske, Donald W. and Donald T. Campbell (1992). Citations do not solve 
problems. Psychological Bulletin, 112(3), 393-395. 
Fodor, Jerry A. (Ed.) (1981). Representations. Cambridge, MA: MIT 
Press. 
Fodor, Jerry A. (1990). A Theory of Content and Other Essays. Cam- 
bridge, MA: MIT Press. 
Fodor, Jerry A., Merrill F. Garrett, E. Walker, and C. H. Parkes (1980). 
Against definitions. Cognition, 8, 263-367. 
Forster, Kenneth I. ( 1970). Visual perception of rapidly-presented word se- 
quences of varying complexity. Perception and Psychophysics. 8, 
215-221. 
Fosnot, Catherine TWomey (Ed.) (1996). Constructivism: Theory. Per- 
spectives and Practice. New York: Teachers College Press. 
Fraisse, Paul (1984). Perceptual processing of words and drawings. In 
Sarris, Victor and Allen Parducci (Eds.), Perspectives in Psychological 
Experimentation: Toward the Year 2000. Hillsdale, NJ: Lawrence 
Erlbaum Associates. 
Freire, Paulo (1972). Pedagogy of the Oppressed. New York: Herder and 
Herder. 
Fries, Charles C. (1945). Teaching and Learning English as a Foreign 
Language. Ann Arbor: University of Michigan Press. 
Furness, David W. and Michael F. Graves ( 1980). Effects of stressing oral 
reading accuracy on comprehension. Reading Psychology . 2( 1), 8-14. 
Gallistel, Charles R. (1990). The Organization of Learning. Cambridge, 
MA: MIT Press. 
Garan, Elaine M. (2002). Resisting Reading Mandates: How to Triumph 
With the Truth. Portsmouth, NH: Heinemann. 
Garner, Ruth (1992). Metacognition and self-monitoring strategies. In 
Samuels, S. Jay and Alan E. Farstrup (Eds.), What Research Has to 
Say About Reading Instruction (Second edition). Newark, DE: Interna- 
tional Reading Association. 
Garner, Wendell R. (1962). Uncertainty and Structure as Psychological 
Concepts. New York: Wiley. 
Garner, Wendell R. (1974). The Processing of Information and Structure. 
Hillsdale, NJ: Lawrence Erlbaum Associates. 
REFERENCES 
341 
Gibbs, Raymond W. (1984). Literal meaning and psychological theory. 
Cognitive Science , 8, 275-304. 
Gibson, Eleanor J. (1965). Learning to read. Science , 148, 1066-1072. 
Gill, J. Thomas (1992). Development of word knowledge as it relates to 
reading, spelling and instruction. Language Arts, 69, 444-453. 
Gillooly, William B. ( 1973). The influence of writing system characteristics 
on learning to read. Reading Research Quarterly, 8(2), 167-199. 
Giroux. Henry (1992). Educational leadership and the crisis of demo- 
cratic government. Educational Researcher, 21 (4), 4-11. 
Goelman, Hillel, Antoinette A. Oberg, and Frank Smith (Eds.) (1984). 
Awakening to Literacy. Portsmouth. NH: Heinemann. 
Golden, Joanne M. and John T. Guthrie (1986). Convergence and diver- 
gence in reader response to literature. Reading Research Quarterly, 
21 ( 4), 408-421. 
Goldsmith-Phillips, J. ( 1989). Word and context in reading development: 
A test of the interactive-compensatory hypothesis. Journal of Educa- 
tional Psychology, 81 , 299-305. 
Golinkoff, Roberta Michnick (1975/1976). A comparison of reading com- 
prehension processes in good and poor readers. Reading Research 
Quarterly, 11, 623-659. 
Goodman, Kenneth S. ( 1967). Reading: A psycholinguistic guessing game. 
Journal of the Reading Specialist, 6, 126-135. 
Goodman, Kenneth S. ( 1969). Analysis of oral reading miscues: Applied 
psycholinguistics. Reading Research Quarterly, 5(1). 9-30. 
Goodman, Kenneth S. (1986). What’s Whole in Whole Language? Rich- 
mond Hill, ON: Scholastic. 
Goodman, Kenneth S. ( 1992/1993). Gurus, professors, and the politics of 
phonics, in Point/ Counterpoint, Whole language versus direct instruc- 
tion models. Reading Today. 10(3), 8-10. 
Goodman, Kenneth S., Patrick Shannon, Yvonne S. Freeman, and Sharon 
Murphy (1988). Report Card on Basal Readers. Katonah. NY: Richard 
C. Owen. 
Goodman, Yetta M. (1978). Kid watching: An alternative to testing. Na- 
tional Elementary Principal, 57(4), 41-45. 
Goodman, Yetta M. (1980). The roots of literacy. In Douglass, Malcolm E 
(Ed.), Proceedings: Claremont Reading Conference, 44th Annual 
Yearbook. Claremont, CA. 
Goody, Jack and Ian Watt ( 1972). The consequences of literacy. In Gigliolo. 
P R (Ed.), Language and Social Context. London: Penguin. 
Gorman, Tom, Dougal Hutchison, and John Trimble ( 1993). Reading in 
Reform: The Avon Collaborative Reading Project: Slough. UK: Na- 
tional Foundation for Educational Research in England and Wales. 
Goswami, Usha C. (1986). Children’s use of analogy in learning to read: A 
developmental study. Journal of Experimental Child Psychology, 42, 
73-83. 
Goswami, Usha C. ( 1990). Phonological priming and orthographic analo- 
gies in reading. Journal of Experimental Child Psychology, 49, 
323-340. 
342 
UNDERSTANDING READING 
Goswami, Usha C. (2001 ). Rhymes are important: A comment on Savage. 
Journal of Reading Research, 24( 1 }, 19-30. 
Goswami. Usha C. and Peter Bryant (1990). Phonological Skills and 
Learning to Read. Hove, UK: Lawrence Erlbaum Associates. 
Goswami. Usha C. and Peter Bryant (1992). Rhyme, analogy, and chil- 
dren’s reading. In Gough. Philip B., Linnea C. Ehri, and Rebecca 
Treiman (Eds .) Reading Acquisition. Hillsdale, NJ: Lawrence Erlbaum 
Associates. 
Gough, Philip B., Linnea C. Ehri, and Rebecca Treiman (Eds.) (1992). 
Reading Acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Gough. Philip B. and M. L. Hillinger (1980). Learning to read: An unnatu- 
ral act. Bulletin of the Orton Society, 30, 180-196. 
Gough, Philip B. and Margaret A. Walsh (1991). Chinese, Phoenicians, and 
the orthographic cipher of English. In Brady, Susan A. and Donald R 
Shankweiler (Eds.), Phonological Processes in Literacy: A Tribute to 
Isabelle Y. Liberman. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Graff, Harvey J. (1986). The Legacies of Literacy: Continuities and Contra- 
dictions in Western Culture and Society. In de Castell, Suzanne. Allan 
Luke, and Kieran Egan (Eds.), Literacy, Society and Schooling: A 
Reader. New York: Cambridge University Press. 
Graff, Harvey J. (1987a). The Labyrinths of Literacy: Reflections on Liter- 
acy Past and Present. Philadelphia: Falmer. 
Graff, Harvey J. (1987b). The Legacies of Literacy: Continuities and Con- 
tradictions in Western Culture and Society. Bloomington: Indiana Uni- 
versity Press. 
Graham, Steven ( 1980). Word recognition skills of learning disabled chil- 
dren and average students. Reading Psychology, 2(1), 23-33. 
Grimes, Joseph E. (1975). The Thread of Discourse. The Hague: Mouton. 
Haber, Lyn R., Ralph N. Haber, and Karen R. Furlin (1983). Word length 
and word shape as sources of information in reading. Reading Re- 
search Quarterly, 18(2), 165-189. 
Haber, Ralph N. and Lyn R. Haber ( 1981 ). The shape of a word can specify 
its meaning. Reading Research Quarterly, 16(3), 334-345. 
Halliday, Michael A. K. (1973). Explorations in the Functions of Lan- 
guage. London: Arnold. 
Hamill, Donald D. and Gaye McNutt (1981). The Correlates of Reading. 
Austin, TX: Pro-Ed. 
Harding, Leonora M., John R. Beech, and William Sneddon (1985). The 
changing pattern of reading errors and reading style from 5 to 1 1 years 
of age. British Journal of Educational Psychology, 55, 45-52. 
Hardyck, C. D. and L. F Petrinovich (1970). Subvocal speech and compre- 
hension level as a function of the difficulty level of reading material. 
Journal of Verbal Learning and Verbal Behavior, 9, 647-652. 
Harste, Jerome C., Virginia A. Woodward, and Carolyn L. Burke (1984). 
Language Stories and Literacy Lessons. Portsmouth. NH: Heinemann. 
Hartman. Maria and Robert E. Kretschmer (1992). Talking and writing: 
Deaf teenagers reading Sarah, Plain and Tall. Journal of Reading. 
36(3), 174-180. 
REFERENCES 
343 
Havelock, Eric A. (1976). Origins of Western Literacy. Toronto: Ontario 
Institute for Studies in Education. 
Heath. Shirley Brice (1982). What no bedtime story means: Narrative 
skills at home and at school. Language in Society. 1 1 , 49-76. 
Heath, Shirley Brice (1983). Research currents: A lot of talk about noth- 
ing. Language Arts. 60(4), 999-1007. 
Heath, Shirley Brice (1986). The functions and uses of literacy. In de 
Castell, Suzanne, Allan Luke, and Kieran Egan (Eds.), Literacy. Soci- 
ety and Schooling: A Reader. New York: Cambridge University Press. 
Heath, Shirley Brice, Leslie Mangiola, Sandra R. Schachter, and GlyndaA. 
Hull (Eds.) (1991). Children of Promise: Literate Activity in Linguis- 
tically and Culturally Diverse Classrooms. Urbana, IL: National Coun- 
cil of Teachers of English. 
Heath, Shirley Brice and Charlene Thomas (1984). The achievement of 
pre-school literacy for mother and child. In Goelman, Hillel, Antoinette 
A. Oberg, and Frank Smith (Eds.), Awakening to Literacy. Exeter, NH: 
Heinemann. 
Heckenmueller, E. G. (1965). Stabilization of the retinal image: A review of 
method, effects and theory. Psychological Review, 63, 157-169. 
Hedley, Carolyn and Anthony N. Baratta (Eds.) (1985). Contexts of Read- 
ing. Norwood, NJ: Ablex. 
Helfand, Duke (2002). Teens Get a Second Chance at Literacy Education. 
Los Angeles Times, July 21, 2002. 
Henderson, John M. (1976). Learning to read: A case study of a deaf child. 
American Annals of the Deaf, 121 , 502-506. 
Hidi, Suzanne, William Baird, and Angela Hildyard (1982). That’s impor- 
tant but is it interesting? T\vo factors in text processing. In Flammer, Au- 
gust and Walter Kintsch (Eds.), Discourse Processing. Amsterdam: 
North Holland. 
Hiebert, Elfrieda H. (1983). An examination of ability grouping for reading 
instruction. Reading Research Quarterly, 18(3), 231-255. 
Hiebert. Elfrieda H. and Robert E. Calfee ( 1992). Assessing literacy: From 
standardized tests to portfolios and performances. In Samuels, S. Jay 
and Alan E. Farstrup (Eds.), What Research Has to Say About Reading 
Instruction (2nd ed. ). Newark, DE: International Reading Association. 
Holdaway, Don (1976). Self-evaluation and reading development. In 
Merritt, John (Ed.), New Horizons in Reading. Newark, DE: Interna- 
tional Reading Association. 
Howe, M. J. A. and Linda Singer (1975). Presentation variables and stu- 
dents’ activities in meaningful learning. British Journal of Educational 
Psychology, 45, 52-61. 
Howes. D. H. and R. L. Solomon (1951). Visual duration threshold as a 
function of word-probability. Journal of Experimental Psychology, 41 , 
401-410. 
Huck. Charlotte S. (1992). Literacy and literature. Lanquaae Arts, 69. 
520-526. 
Huey, Edmund Burke ( 1908). The Psychology and Pedagogy of Reading . 
New York: Macmillan. Reprinted Cambridge, MA: MIT Press (1968). 
344 
UNDERSTANDING READING 
Hull, Robert ( 1985). The Language Gap: How Classroom Dialogue Fails. 
London: Methuen. 
Hummel, John E. and Irving Biederman (1992). Dynamic binding in a neu- 
ral network for shape recognition. Psychological Review. 99(3), 
480-517. 
Hunt, Earl and Franca Agnoli ( 1991). The Whorfian hypothesis: A cogni- 
tive psychology perspective. Psychological Review, 98(3), 377-389. 
International Reading Association Board of Directors (1999/2000). Posi- 
tion statement. Reading Today, December 1999/January 2000, 37. 
Intraub, Helen and James E. Hoffman (1992). Reading and visual mem- 
ory: Remembering scenes that were never seen. American Journal of 
Psychology, 105(1), 101-114. 
Jakobson, Roman and Morris Halle ( 1956). Fundamentals of Language. 
The Hague: Mouton. 
Johnson-Laird, Philip N., Ruth M. J. Byrne, and Walter Schaeken ( 1992). 
Propositional reasoning by model. Psychological Review, 99(3), 
418-439. 
Johnston, Francine R, (2000/2001). Spelling exceptions: Problems or 
possibilities? The Reading Teacher, 54(4), 372-373. 
Johnston, Francine R. (2001). The utility of phonic generalizations: Let's 
take another look at Clymer's conclusions. The Reading Teacher, 
55(2), 132-143. 
Johnston, Peter H. (1985). Understanding reading disability: A case-study 
approach. Harvard Educational Review. 55(2), 153-177. 
Judd, C. H., and G. T. Buswell ( 1922). Silent Reading: A Study of the Var- 
ious Types. Chicago, IL: University of Chicago Press. 
Juel, Connie and Diana Roper/Schneider (1985). The influence of basal 
readers on first grade reading. Reading Research Quarterly, 20(2), 
134-152. 
Just, Marcel A. and Patricia A. Carpenter (1980). A theory of reading, from 
eye fixadons to comprehension. Psychological Review. 87(6), 329-354. 
Kamii, Constance K. ( 1985). Young Children Reinvent Arithmetic: Impli- 
cations of Piaget's Theory. New York: Teachers College Press. 
Kimmel, Susan and Walter H. MacGinitie (1984). Identifying children who 
use a perseverative text processing strategy. Reading Research Quar- 
terly, 19(2), 162-172. 
Kintsch, Walter (1982). Memory for text. In Flammer, August and Walter 
Kintsch. Discourse Processing . Amsterdam: North Holland. 
Kintsch, Walter (1988). The use of knowledge in discourse processing: A 
construction-integration model. Psychological Review, 95, 163-182. 
Klahr, D., W. G. Chase, and E. A. Lovelace (1983). Structure and process in 
alphabetic retrieval. Journal of Experimental Psychology: Learning. 
Memory and Cognition, 9, 462-477. 
Klein, Helen Altman. Gary A. Klein, and Mary Bertino ( 1974). Utilization of 
context for word identification in children. Journal of Experimental 
Psychology, 1 7 , 79-86. 
Kolers, Paul A. ( 1966). Reading and talking bilingually. American Journal 
of Psychology , 79. 357-376. 
REFERENCES 
345 
Kolers, Paul A. (1967). Reading is only incidentally visual. In Goodman, 
Kenneth S. and James T. Fleming (Eds.), Psycholinguistics and the 
Teaching of Reading. Newark, DE: International Reading Association. 
Kolers, Paul A. and M. T. Katzman ( 1966). Naming sequentially presented 
letters and words. Language and Speech, 9(2), 84-95. 
Korkeamaki, Riitta-Liisa, and Mariam Jean Dreher (1993). Finland, pho- 
nics, and whole language: Beginning reading in a regular letter-sound 
correspondence language. Language Arts, 70, 475-482. 
Kozol, Jonathan (1991). Savage Inequalities: Children in America's 
Schools. New York: Harper Collins. 
Krashen, Stephen D. (1985). Inquiries and Insights. Hayward, CA: 
Alemany Press. 
Krashen, Stephen D. ( 1987). Comic Book Reading and Language Devel- 
opment. Victoria, BC: Abel Press. 
Krashen, Stephen D. (1993a). How well do people spell? Reading Im- 
provement, 31, 9-20. 
Krashen, Stephen D. (1993b). The Power of Reading: Insights from the 
Research. Englewood, CO: Libraries Unlimited. 
Krashen, Stephen D. (1995). School libraries, public libraries, and the 
NEAP reading scores. School Library Media Quarterly, 23, 235-238. 
Krashen, Stephen D. (1999a): Three Arguments Against Whole Lan- 
guage and Why They Are Wrong. Portsmouth, NIL Heinemann. 
Krashen, Stephen D. (1999b). Training in phonemic awareness: Greater 
on tests of phonemic awareness. Perceptual and Motor Skills, 89. 
412-416. 
Krashen, Stephen D. (2001a). The testing movement and delayed gratifi- 
cation. Educators for Urban Minorities, 2(1), 48-50. 
Krashen, Stephen D. (2001b). Low RA. can read O.K. Practically Primary. 
6(2), 17-19. 
Krashen, Stephen D. (2001c). Does “pure” phonemic awareness training 
affect reading comprehension? Perceptual and Motor Skills, 93, 
356-358. 
Krashen, Stephen D. (2002). The lexile framework: The controversy con- 
tinues. CSLA Journal, 25(2), 29-31. 
Krashen, Stephen D. (2003). Explorations in Language Acquisition and 
Use. Portsmouth, NH: Heinemann. 
Krashen, Stephen D. and Howard White (1991). Is spelling acquired or 
learned? A re -analysis of Rice ( 1897) and Cornman (1902). ITL Review 
of Applied Linguistics, 91-92, 1-48. 
Krueger, Lester E., Robert H. Keen, and Bella Rublevich (1974). Letter 
search through words and nonwords by adults and fourth-grade chil- 
dren. Journal of Experimental Psychology , 102(5), 845-849. 
Kucer, Stephen B. and Jenny Tuten (2003). Revisiting and rethinking the 
reading process. Language Arts, 80(4), 284-290. 
Kuhn, Thomas (1970). The Structure of Scientific Revolutions. Chicago: 
University of Chicago Press. 
Kutz, Eleanor (1992). Teacher research: Myths and realities. Language 
Arts, 69, 193-197. 
346 
UNDERSTANDING READING 
Lakoff, George ( 1987). Women, Fire, and Dangerous Things: What Cate- 
gories Reveal About the Mind. Chicago: University of Chicago Press. 
Langer, Judith A. ( 1982a). Facilitating text processing: The elaboration of 
prior knowledge. In Langer, Judith A. and M. Trika Smith-Burke 
(Eds.), Reader meets author/Bridging the gap. Newark, DE: Interna- 
tional Reading Association. 
Langer, Judith A. (1982b). Reading, thinking, writing ... and teaching. 
Language Arts, 59(4), 336-341 . 
Larsen, Steen F. (1986). Procedural thinking, programming and computer 
use. In Hollnagel. E., G. Mancini. and D. Woods (Eds.), Intelligent Deci- 
sion Aids in Process Environments. Berlin: Springer-Verlag. 
Larson, Joanne (Ed.) (2001). Literacy as Snake Oil: Beyond the “ Quick 
Fix. ” New York: Lang. 
Lehr, Fran and Jean Osborn (eds.) (1994). Reading, Language and Liter- 
acy: Instruction for the Twenty-First Century. Hillsdale, NJ: Erlbaum. 
Levine, K. (1986). The Social Context oj Literacy. London: Routledge and 
Kegan Paul. 
Levy, Betty Ann (1978). Speech analysis during sentence processing: 
Reading and listening. Visible Language, 12, 81-102. 
Lewis, Donald J. (1979). Psychobiology of active and inactive memory. 
Psychological Bulletin, 86, 1054-1083. 
Liberman, Alvin M„ Franklin S. Cooper, Donald R Shankweiler. and M. 
Studdert-Kennedy (1957). Perception of the speech code. Psychologi- 
cal Review, 54, 358-368. 
Liberman, Isabelle Y. and Alvin M. Liberman ( 1992). Whole language ver- 
sus code emphasis: Underlying assumptions and their implications for 
reading instruction. In Gough, Philip B., Linnea C. Ehri and Rebecca 
Treiman (Eds .) Reading Acquisition, Hillsdale, NJ: Lawrence Erlbaum 
Associates. 
Lipson, Marjorie Y. and Karen K. Wixson (1986). Reading disability re- 
search: An interactionist perspective. Review of Educational Re- 
search, 56(1), 111-136. 
Littlefair, Alison B. ( 199 1 ). Reading All Types of Writing: The Importance 
of Genre and Register for Reading Development. Milton Keynes: Open 
University. 
Lodge, David (1990). After Bakhtin: Essays on Fiction and Criticism. 
London: Routledge. 
Long, Shirley A., Peter N. Winograd, and Connie A. Bridge (1989). The ef- 
fects of reader and text characteristics on imagery reported during and 
after reading. Reading Research Quarterly, 24(3), 353-372. 
Lott, Deborah and Frank Smith (1970). Knowledge of intra-word redun- 
dancy by beginning readers. Psychonomic Science. 19(6). 343-344. 
MacGinitie, Walter (1984). Readability as a solution adds to the problem. 
In Anderson, Richard C., Jean Osborn, and Robert J. Tierney (Eds.), 
Learning to Read in American Schools: Basal Readers and Content 
Texts. Hillsdale, NJ: Lawrence Erlbaum Associates. 
MacGinitie, Walter and Ruth K. MacGinitie ( 1986). Teaching students not 
to read. IndeCastell. Suzanne, A. Luke, and Kieran Egan (Eds.). Liter- 
acy, Society and Schooling: A Reader. New York: Cambridge Univer- 
sity Press. 
REFERENCES 
347 
Mackworth, Norman H. ( 1 965). Visual noise causes tunnel vision. Psycho- 
nomic Science, 3, 67-68. 
Macmillan, Bonnie M. (2002). Rhyme and reading. A critical review of the 
research methodology. Journal of Research in Reading, 25(1), 4-42. 
Macnamara, John (1972). Cognitive basis of language learning in infants. 
Psychological Review, 79(1), 1-13. 
Magee, Bryan (1973). Popper. London: Fontana. 
Mandler, George (1980). Recognizing: The judgment of previous occur- 
rence. Psychological Review, 87(3), 252-271. 
Mandler, George (1985). Cognitive Psychology: An Essay in Cognitive 
Science. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Mandler. Jean Matter (1984). Stories, Script, and Scenes: Aspects of 
Schema Theory. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Mandler, Jean Matter (1992). How to build a baby: II. Conceptual primi- 
tives. Psychological Review, 99(4), 587-604. 
Mandler, Jean Matter and Marsha S. Goodman ( 1 982). On the psychologi- 
cal validity of story structure. Journal of Verbal Learning and Verbal 
Behavior, 21, 507-523. 
Markman, Ellen M. (1992). Constraints on word learning: Speculations 
about their nature, origins and domain specificity. In Gunnar, Megan R. 
and Michael Maratsos (Eds.), Modularity and Constraints in Lan- 
guage and Cognition (The Minnesota Symposia on Child Psychology. 
Vol. 25). Hillsdale, NJ: Lawrence Erlbaum Associates. 
Marshall, John C. and E Newcombe ( 1966). Syntactic and semantic errors 
in paralexia. Neuropsychologia, 4. 169-176. 
Masonheimer, Patricia E., Priscilla A. Drum, and Linnea C. Ehri (1984). 
Does environmental print identification lead children into word read- 
ing? Journal of Reading Behavior, 16(4), 257-272. 
Mathews, Mitford M. (1966). Teaching to Read: Historically Considered. 
Chicago: University of Chicago Press. 
McCabe, Viki and Gerald J. Balzano (Eds.) (1986). Event Cognition: An 
Ecological Perspective. Hillsdale, NJ: Lawrence Erlbaum Associates. 
McCawley, James D. (1968). The role of semantics in a grammar. In 
Eamon Bach and R. T. Harms (Eds.), Universals in Linguistic Theory. 
New York: Holt. Rinehart and Winston. 
McClelland, J. L. and J. K. O’Regan (1981). Expectations increase the ben- 
efit derived from parafoveal visual information in reading words aloud. 
Journal of Experimental Psychology: Human Perception and Perfor- 
mance, 7(3), 634-644. 
McFarland, Carl E. and Deborah H. Rhodes (1978). Memory for meaning 
in skilled and unskilled readers. Journal of Experimental Child Psy- 
chology, 25, 199-207. 
McGee, Lea M. ( 1982). Awareness of text structure: Effects on children’s 
recall of expository text. Reading Research Quarterly, 17(4). 
581-590. 
McGee. Lea M. ( 1992). Exploring the literature -based reading revolution. 
Language Arts, 69, 529-537. 
McKeown, Margaret G. (1985). The acquisition of word meaning from con- 
text by children of high and low ability. Readinq Research Quarterly, 
20(4). 482-496. 
348 
UNDERSTANDING READING 
McKeown, Margaret G. (1993). Creating effecting definitions for young 
word learners. Reading Research Quarterly, 28(1), 16-31. 
McMahon, Susan I. ( 1992). Bookclub: A case study of a group of fifth grad- 
ers as they participate in a literature based reading program. (IRA Out- 
standing Dissertation Award for 1991-1992). Reading Research 
Quarterly, 27(4), 292-293. 
McNamara, Timothy R, Diane I. Miller, and John D. Bransford (1991). 
Mental models and reading comprehension. In Barr. Rebecca, Michael 
L. Kamil, Peter Mosenthal, and R David Pearson (Eds.). Handbook of 
Reading Research. New York: Longman. 
McNeill, Daniel (1998). The Face: A Natural History. Boston, MA: Little 
Brown. 
McNeill, David ( 1967). Developmental psycholinguistics. In Smith, Frank 
and George A. Miller (Eds.), The Genesis of Lanquaqe. Cambridge, 
MA: MIT Press. 
McNeill, David (1985). So you think gestures are nonverbal? Psychologi- 
cal Review, 92(3), 350-371. 
McNeill, David and Karen Lindig (1973). The perceptual reality of pho- 
nemes, syllables, words and sentences. Journal of Verbal Learning 
and Verbal Behavior. 12(4), 419-430. 
McPeck, John E. (1981). Critical Thinking and Education. Oxford: Mar- 
tin Robertson. 
McQuillan, Jeff ( 1998). The Literacy Crisis: False Claims and Real Solu- 
tions. Portsmouth, NIL Heinemann. 
Meek, Margaret (1982). Learning to Read. London: Bodley Head. 
Meek, Margaret (1984). Speaking of shifters. In Meek, Margaret and Jane 
Miller (Eds.), Changing English: Essays for Harold Rosen. London: 
Heinemann. 
Meek, Margaret ( 1988). How texts teach what readers learn. In Lightfoot, 
Martin and Nancy Martin (Eds.), The Word for Teaching is Learning: 
Essays for James Britton. London: Heinemann. 
Meek, Margaret, Stephen Armstrong, Vicky Austerfield, Judith Graham, 
and Elizabeth Plackett (1983). Achieving Literacy: Longitudinal 
Studies of Adolescents Learning to Read. London: Routledge and 
Kegan Paul. 
Metcalf, Stephen (2002). Reading Between the Lines. The Nation, 28 January. 
Meyer. Richard J. (2002). Captives of the script: Killing us softly with pho- 
nics. Language Arts. 79(6), 452-461. 
Michotte, A. (1946). La Perception de la Causalite. Louvain: Institut 
Superieur de Philosophic. 
Mikulecky, Larry (1982). Job literacy: The relationship between school 
preparation and workplace actuality. Reading Research Quarterly. 
17(3), 400-419. 
Miller, George A. (1951). Language and Communication. New York: 
McGraw-Hill. 
Miller, George A. (1956). The magical number seven, plus or minus two: 
Some limits on our capacity for processing information. Psychological 
Review, 63, 81-92. 
REFERENCES 
349 
Miller, George A. (1965). Some preliminaries to psycholinguistics. Ameri- 
can Psychologist, 20, 15-20. 
Miller. George A. (1977). Spontaneous Apprentices: Children and Lan- 
guage. New York: Seabury. 
Miller, George A., Jerome S. Bruner, and Leo Postman (1954). Familiarity 
of letter sequences and tachistoscopic identification. Journal of Ge- 
netic Psychology, 50, 129-139. 
Miller, George A. and Christine Fellbaum (1991). Semantic networks of 
English. Cognition, 41, 197-229. 
Miller, George A. and Philip N. Johnson-Laird (1976). Language and Per- 
ception. Cambridge, MA: Belknap Press. 
Miller, George A. and Patricia E. Nicely (1955). An analysis of perceptual 
confusions among some English consonants. Journal of the Acoustical 
Society of America, 27, 338-353. 
Miller, Larry (1991). The whole language path: Postholes and signposts. 
Reflections on Canadian Literacy, 9(2), 123-127. 
Miller, Larry, Jillian DeJean, and Rebecca Miller (2002). The literacy cur- 
riculum and use of an integrated learning system. Journal of Research 
in Reading, 23(2), 123-135. 
Mohyeldin Said, K. A., W. H. Newton-Smith, R. Viale, and Kathleen V. 
Wilkes (Eds.) (1990). Modelling the Mind. Oxford: Clarendon. 
Moll, Luis C. (Ed.) (1990). Vygotsky and Education: Instructional Impli- 
cations and Applications of Sociohistorical Psychology. Cambridge, 
UK: Cambridge University Press. 
Moorman, Gary B., William E. Blanton, and Thomas M. McLaughlin 
( 1992). The rhetoric ofwhole language. Parts 1 and 2. Reading Psychol- 
ogy, 13(2) iii-xiv and 13(3) iii-xiv. 
Morris. R E. ( 1 988). Expertise and everyday memory. In Gruneberg, M. M. . 
R E. Morris, and R. N. Sykes (Eds.), Practical Aspects of Memory: Cur- 
rent Research and Issues, Vol. I: Memory in Everyday Life. Chichester, 
UK: Wiley. 
Morrow, Lesley Mandel (1992). The impact of a literature -based program 
on literacy achievement, use of literature, and attitudes of children from 
minority backgrounds. Reading Research Quarterly, 27(3), 251-275. 
Moyer. Sandra B. and Phillis L. Newcomer (1977). Reversals in reading di- 
agnosis and remediation. Exceptional Children, 43, 424-429. 
Myers. Jamie ( 1992). The social contexts of school and personal literacy. 
Reading Research Quarterly, 27(4), 296-333. 
Nagy, William E. and Patricia A. Herman ( 1987). Breadth and depth of vo- 
cabulary knowledge: Implications for acquisition and instruction. In 
McKeown, Margaret G. and M. E. Curtis (Eds.), The Nature of Vocabu- 
lary Acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Nagy, William E., Patricia A. Herman, & Richard C. Anderson (1985). 
Learning words from context. Reading Research Quarterly, 20(2). 
233-253. 
Nagy. William E., Jean Osborn, Pamela Winsor and John O'Flahavan 
(1994). Structural analysis: Some guidelines for instruction. In Fran 
Lehr and Jean Osborn (Eds.), Reading, Language and Literacy: In- 
350 
UNDERSTANDING READING 
structlon for the Twenty-First Century. Hillsdale, NJ: Lawrence 
Erlbaum Associates. 
Nagy, William E. and Judith A. Scott (2000). Vocabulary processes. In 
Barr, Rebecca, M. Kamil. Peter Mosenthal, & R David Pearson (Eds.), 
Handbook of Reading Research: Vol. Ill (pp. 269-284). Mahwah. NJ: 
Lawrence Erlbaum Associates. 
National Reading Panel. (2000). Teaching Children to Read: An Evidence 
Based Assessment of the Scientific Research Literature on Reading 
and Its Implications for Reading Instruction. Washington, DC: Na- 
tional Institute of Child Health and Human Development. 
Naylor, Hilary ( 1980). Reading disability and lateral asymmetry: An infor- 
mation-processing analysis. Psychological Bulletin. 87(3). 531-545. 
Neisser, Ulric (1977). Cognition and Reality. San Erancisco: Freeman. 
Nell, Victor (1988). The psychology of reading for pleasure: Needs and 
gratifications. Reading Research Quarterly, 23(1). 6-50. 
Nelson, Katherine (1985). Making Sense: Development of Meaning in 
Early Childhood. New York: Academic Press. 
Nelson, Katherine (1986). Event Knowledge. Hillsdale. NJ: Lawrence 
Erlbaum Associates. 
Neuman, Susan B. and Donna Celano (2001). Access to print in low-in- 
come and middle-income communities: An ecological study of four 
neighborhoods. Reading Research Quarterly, 36(1), 8-26. 
Newman, Edwin B. ( 1966). Speed of reading when the span of letters is re- 
stricted. American Journal of Psychology, 79, 272-278. 
Newson, John and Elizabeth Newson (1975). Intersubjectivity and the 
transmission of culture: On the social origins of symbolic functioning. 
Bulletin of the British Psychological Society, 28, 437-446. 
Nicholson, Tom and Robert Imlach (1981 ). Where do their answers come 
from? A study of the inferences which children make when answering 
questions about narrative stories. Journal of Reading Behavior, 13(2), 
111-129. 
Norman, Donald A. (1969). Memory and Attention: An Introduction to 
Human Information Processing. New York: Wiley (2nd ed. 1976). 
O’Brien, Edward J. and Jerome L. Myers (1985). When comprehension 
difficulty improves memory for text. Journal of Experimental Psychol- 
ogy: Learning, Memory and Cognition, 11(1), 12-21. 
Ohanian, Susan (1999). One Size Fits Few: The Folly of Educational 
Standards. Portsmouth. NH: Heinemann. 
Olson, David R. (1977). From utterance to text: The bias of language in 
speech and writing. Harvard Educational Review, 47(3), 257-281. 
Olson, David R., Nancy Torrance, and Angela Hildyard (Eds.) (1985). Lit- 
eracy. Language and Learning: The Nature and Consequences of 
Reading and Writing. Cambridge. UK: Cambridge University Press. 
Ong, Walter J. (1982). Orality and Literacy: The Technologizing of the 
Word. London: Methuen. 
O'Shea, Lawrence J., Paul T. Sindelar, and Dorothy J. O’Shea (1985). The 
effects of repeated readings and attentional cues on reading fluency and 
comprehension. Journal of Reading Behavior, 17(2), 129-142. 
REFERENCES 
351 
Otto, Wayne (1992). The role of research in reading instruction. In 
Samuels, S. Jay and Alan E. Farstrup (Eds.), What Research Has to 
Say About Reading Instruction (2nd ed.). Newark, DE: International 
Reading Association. 
Paterson, Frances A. (2002). The politics of phonics. Journal oj Curricu- 
lum and Supervision, 15(3). Reprinted in Allington (2002). 
Pattanayak, D. R ( 199 1 ). Literacy: An instrument of oppression. In Olson, 
David R. and Nancy Torrance (Eds.), Literacy and Orality. Cambridge, 
UK: Cambridge University Press. 
Paulson, Eric J. (2002). Are oral reading word omissions and substitutions 
caused by careless eye movements? Reading Psychology, 23(1), 45-66. 
Paulson, Eric J. and Kenneth S. Goodman (1999). Influential studies in eye- 
movement research. Reading Online, <http://www.readingonline.org/ 
past/past_index.asp?HREF-/critical/ACT.html>. Posted January 1999. 
Pearson, R David, Laurel R. Roehler, Janice A. Dole, and Gerald G. Duffy 
(1992). Developing expertise in reading comprehension. In Samuels. S. 
Jay and Alan E. Farstrup (Eds.), What Research Has to Say About 
Reading Instruction (2nd ed.). Newark, DE: International Reading As- 
sociation. 
Pearson, R David and Alice Studt (1975). Effects of word frequency and 
contextual richness on children’s word identification abilities. Journal 
of Educational Psychology, 67(1), 89-95. 
Perfetti, Charles A. (1985). Reading Ability. New York: Oxford University 
Press. 
Perfetti, Charles A. ( 199 1 ). The psychology, pedagogy, and politics of read- 
ing. Psychological Review, 2. 70-76. 
Perfetti, Charles A., S. R. Goldman, and T. W. Hogaboam (1979). Reading 
skill and the identification of words in discourse context. Memory and 
Cognition, 4, 273-282. 
Perfetti, Charles A. and S. Roth (1981). Some of the interactive processes 
in reading and their role in reading skill. In Lesgold, Alan M. and 
Charles A. Perfetti (Eds.), Interactive Processes in Reading. Hillsdale, 
NJ: Lawrence Erlbaum Associates. 
Piaget, Jean (1976). To Understandls tolnvent. New York: Penguin Books. 
Piattelli-Palmarini, Massimo (Ed.) (1980). Language and Learning: The 
Debate Between Jean Piaget and Noam Chomsky. Cambridge, MA: 
Harvard University Press. 
Pierce, J. R. and J. E. Karlin ( 1957). Reading rates and the information rate 
of a human channel. Bell Systems Technical Journal, 36, 497-516. 
Pillsbury, W. B. (1897). A study in apperception. American Journal of Psy- 
chology, 8, 315-393. 
Pinker, Steven (1984). Visual cognition: An introduction. Cognition. 18, 
1-63. 
Pinker, Steven (1999). Words and Rules: The Ingredients of Language. 
London: Weidenfeld and Nicolson. 
Pinnell, Gay-Su (1989). Success for at-risk children in a program that 
combines reading and writing. In Mason. Jana M. (Ed.). Reading and 
Writing Connections. Boston: Allyn and Bacon. 
352 
UNDERSTANDING READING 
Piper, David (1987). Teaching story grammar: Some reasons for caution. 
Canadian Journal of English Language Arts, 10( 1 ), 30-37. 
Polanyi, Michael (1966). The Tacit Dimension. Garden City, NY: 
Doubleday. 
Popper, Karl R. (1973). Objective Knowledge: An Evolutionary Approach. 
Oxford: Clarendon. 
Popper, Karl R. (1976). Unended Quest: An Intellectual Autobiography. 
London: Fontana/Collins. 
Potter, MaryC. ( 1975). Meaning in visual search. Science, 187, 965-966. 
Potter, MaryC. ( 1984). Rapid serial visual presentation (RSVP): A method 
for studying language processing. In Kieras, David E. and Marcel A. 
Just (Eds.), New Methods in Reading Comprehension Research. 
Hillsdale, NJ: Lawrence Erlbaum Associates. 
Potter, Mary C., J. E Kroll. B. Yachzel, and C. Harris (1980). Comprehen- 
sion and memory in rapid sequential reading. In Nickerson, R. (Ed.). 
Attention and Performance VIII. Hillsdale. NJ: Lawrence Erlbaum As- 
sociates. 
Pritchard, R. M. (1961 ). Stabilized images on the retina. Scientific Ameri- 
can, 204(6), 72-78. 
Quastler, Henry (1956). Studies of human channel capacity. In Cherry, 
Colin (Ed.), Information Theory. London: Butterworths. 
Rack, John E, Margaret J. Snowling, and Richard K. Olson (1992). The 
nonword reading deficit in developmental dyslexia: A review. Reading 
Research Quarterly, 27(1), 28-53. 
Raphael, Taffy E. (Ed.) (1986). The Contexts of School Based Literacy. 
New York: Random House. 
Rayner, Keith (1992). Eye Movements and Visual Cognition: Scene Per- 
ception and Reading. New York: Springer-Verlag. 
Rayner, Keith (1997). Understanding eye movements in reading. Scientific 
Studies of Reading, 1(4), 317-339. 
Rayner, Keith and Alexander Pollatsek (1987). Eye movements in read- 
ing: A tutorial review. In Coltheart, Max (Ed.), Attention and Perfor- 
mance XIII: The Psychology of Reading. Hove, UK: Lawrence 
Erlbaum Associates. 
Rayner, Keith and Alexander Pollatsek (1989). The Psychology of Read- 
ing. Englewood Cliffs, NJ: Prentice-Hall. 
Read, Charles ( 1971 ). Pre-school children's knowledge of English phonol- 
ogy. Harvard Educational Review, 41(1), 1-34. 
Read, Charles, Zhang Yun-Fei, Nie Hong-Yin. and Ding Bao-Qing (1986). 
The ability to manipulate speech sounds depends on knowing alpha- 
betic writing. Cognition, 24, 31-44. 
Reber, Arthur S. ( 1989). Implicit learning and tacit knowledge. Journal of 
Experimental Psychology: General, 118, 219-235. 
Reichardt, Konrad W. (1977). Playing dead or running away: Defense 
mechanisms during reading. Journal of Reading, 20, 706-71 1 . 
Reitsma, Pieter ( 1983). Printed word learning in beginning readers. Jour- 
nal of Experimental Child Psychology. 75. 321-339. 
Restle, E { 1970). Theory of serial pattern learning. Psychological Review. 
77, 481-495. 
REFERENCES 
353 
Rhode, Mary and Bruce Cronnell ( 1977). Compilation of a Communica- 
tion Skills Lexicon Coded with Linguistic Information. Los Alamitos, 
CA: Southwest Regional Laboratory for Educational Research and De- 
velopment, Technical Report No. 58. 
Rice, J. (1897). The futility of the spelling grind. Forum, 23, 163-172, 
409-419. 
Rice, Mabel L. (1990). Preschoolers’ QUIL: Quick Incidental Learning of 
Words. In Conti-Ramsden. Gina and Catherine E. Snow (Eds.), Children's 
Language (Vol. 7). Hillsdale, NJ: Lawrence Erlbaum Associates. 
Rieben, Laurence and Charles A. Perfetti (1991). Learning to Read: Basic Re- 
search and Its Implications. Hillsdale, NJ : Lawrence Erlbaum Associates. 
Robinson, H. Alan, Vincent Faraone, Daniel R. Hittelman, and Elizabeth 
Unrah ( 1990). Reading Comprehension Instruction 1 783-1 987: A Re- 
view of Trends and Research. Newark, DE: International Reading As- 
sociation. 
Robinson, Ian (1975). The New Grammarians' Funeral. Cambridge, UK: 
Cambridge University Press. 
Rogers. Rebecca (2002). Between contexts: A critical discourse analysis of 
family literacy, discursive practices, and literate subjectivities. Reading 
Research Quarterly, 37(3), 248-277. 
Rosch, Eleanor and B. B. Lloyd (1978). Cognition and Categorization. 
Hillsdale, NJ: Lawrence Erlbaum Associates. 
Rosen, Harold (1986). The importance of story Language Arts, 63(3), 
226-237. 
Rosen. Harold (1988). Stories of stories: Footnotes on sly gossipy prac- 
tices. In Lightfoot, Martin and Nancy Martin (Eds.), The Word for 
Teaching is Learning: Essaysfor James Britton. London: Heinemann. 
Rosenblatt, Louise M. (1978). The Reader: The Text: The Poem. Carbon- 
dale: Southern Illinois University Press. 
Rosenblatt, Louise M. ( 1980). What facts does this poem teach you? Lan- 
guage Arts, 57(4). 386-394. 
Rosinski, Richard R., Roberta Michnick Golinkoff, and Karen S. Kukish 
( 1975). Automatic semantic processing in a picture-word interference 
task. Child Development, 46(1), 247-253. 
Rosset, Clement (1989). Reality and the untheorizable. In Kavanagh. 
Thomas M. (Ed. ), The Limits of Theory. Stanford. CA: Stanford Univer- 
sity Press. 
Rottenberg, Claire J. and Lyndon W. Searfoss ( 1992). Becoming literate in 
a preschool class: Literacy development of hearing-impaired children. 
Journal of Reading Behavior, 24(4), 463-479. 
Routman, Regie (1994). Invitations: From Literature to Literacy. 
Portsmouth, NH: Heinemann. 
Rumelhart, David E. ( 1 980). Schemata: The building blocks of cognition. 
In Spiro. Rand J., Bertram C. Bruce, and William E Brewer (Eds.), The- 
oretical Issues in Reading Comprehension: Perspectives from Cogni- 
tive Psychology, Linguistics, Artificial Intelligence, and Education. 
Hillsdale, NJ: Lawrence Erlbaum Associates. 
Rumelhart, David E. and Andrew Ortony (1977). The representation of 
knowledge in memory. In Anderson, Richard C. , Rand J. Spiro, and Wil- 
354 
UNDERSTANDING READING 
liam E. Montague (Eds.), Schooling and the Acquisition of Knowledge. 
Hillsdale, NJ: Lawrence Erlbaum Associates. 
Sachs, Jacqueline S. (1974). Memory in reading and listening to dis- 
course. Memory and Cognition, 2(1A), 95-100. 
Sacks, Oliver (1989). Seeing Voices: A Journey Into the World of the Deaf 
Berkeley: University of California Press. 
Sadoski, Mark ( 1983). An exploratory study of the relationships between 
reported imagery and the comprehension and recall of a story. Reading 
Research Quarterly , 19(1), 110-121. 
Sadoski, Mark, Ernest M. Goetz, and S. Kansinger (1988). Imagination in 
story response: Relationships between imagery, affect and structural 
importance. Reading Research Quarterly, 23, 320-336. 
Saenger, Paul (1991). The separation of words and the physiology of read- 
ing. In Olson, David R. and Nancy Torrance (Eds.), Literacy and 
Orality. Cambridge, UK: Cambridge University Press. 
Salmon. Phillida (1985). Living in Time: A New Look at Personal Devel- 
opment. London: Dent. 
Salmon, Phillida and Hilary Claire (1984). Classroom Collaboration. 
London: Routledge and Kegan Paul. 
Sampson, Geoffrey (1985). Writing Systems: A Linguistic Introduction. 
London: Hutchinson. 
Samuels, S. Jay (1971). Letter-name versus letter-sound knowledge in 
learning to read. The Reading Teacher, 24, 604-608. 
Samuels, S. Jay (1979). The method of repeated readings. The Reading 
Teacher, 32, 403-408. 
Santa, Carol Minnick ( 1999/2000). President’s message: High stakes, low 
benefits: An open letter to U.S. Secretary of Education Richard Riley. 
Reading Today, December 1999/January 2000. 37. 
Savage, Robert (2001). A re-evaluation of the evidence for orthographic 
analogies: A reply to Goswami (1999). Journal of Reading Research, 
24(1), 1-19 
Saxe, Geoffrey B. (1991 ). Culture and Cognitive Development: Studies in 
Mathematical Understanding. Hillsdale. NJ: Lawrence Erlbaum Asso- 
ciates. 
Schatz, Elinore Kress and R. Scott Baldwin (1986). Context cues are unre- 
liable predictors of word meanings. Reading Research Quarterly, 
21 (4). 439-454. 
Schneider, Walter and Richard M. Shiffrin (1977). Controlled and auto- 
matic human information processing: I. Detection, search and atten- 
tion. Psychological Review, 84(1), 1-66. 
Scholes, Robert J. and Brenda J. Willis (1991). Linguists, literacy and the 
intensionality of Marshall McLuhan's Western man. In Olson. David R. 
and Nancy Torrance (Eds.), Literacy and Orality. Cambridge. UK: 
Cambridge University Press. 
Schwanenflugel, Paula J. (Ed.). (1991). The Psychology of Word Mean- 
ings. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Scribner, Sylvia and Michael Cole (1978). Literacy without schooling: 
Testing for intellectual effects. Harvard Educational Review. 48(4). 
448-461. 
REFERENCES 
355 
Seligman, Martin E. P ( 1975). Helplessness : On Depression, Develop- 
ment and Death. San Francisco: Freeman. 
Senner, Wayne M. (Ed.) (1989). The Origins of Writing . Lincoln: University 
of Nebraska. 
Shallice, Tim and Elizabeth K. Warrington (1975). Word recognition in a 
phonemic dyslexic patient. Quarterly Journal of Experimental Psy- 
chology, 27, 187-199. 
Shannon, Claude E. (1951). Prediction and entropy of printed English. 
Bell Systems Technical Journal, 30, 50-64. 
Shannon, Patrick (1984). Mastery learning in reading and the control of 
teachers and students. Language Arts, 61(5), 484-493. 
Shannon, Patrick ( 1985). Reading instruction and social class. Language 
Arts, 62(6], 604-613. 
Shannon, Patrick (1989). Broken Promises: Reading Instruction in 
Twentieth Century America. Granby, MA: Bergin and Garvey. 
Shannon, Patrick (1993). Critique of false generosity: A response to 
Baumann. Reading Research Quarterly, 28(1), 8-14. 
Shannon, Patrick (1996). Mad as hell. Language Arts, 73(1), 14-19. 
Sharkey, Noel E. and D. C. Mitchell (1985). Word recognition in a func- 
tional context: The use of scripts in reading. Journal of Memory and 
Language, 24(2), 253-270. 
Shaywitz, S. E., M. Escobar, M. Shaywitz, J. Fletcher, and R. Makuch 
(1992). Evidence that dyslexia may represent the lower tail of a normal 
distribution of reading ability. New England Journal of Medicine, 326. 
145-150. 
Simon, Herbert A. (1974). How big is a chunk? Science, 183, 482-488. 
Skinner, B. F. ( 1985). Cognitive science and behaviorism. British Journal 
of Psychology, 76(3), 291-301. 
Slater, Wayne H. ( 1985). Revising inconsiderate elementary school exposi- 
tory text: Effects on comprehension and recall. In Niles, Jerome A. and 
Rosary V Lalik (Eds.), Issues in Literacy: A Research Perspective. 
Rochester, NY: National Reading Conference, 34th Yearbook. 
Smith, Frank (1969). The use of featural dependencies across letters in 
the visual identification of words. Journal of Verbal Learning and Ver- 
bal Behavior, 8, 215-218. 
Smith, Frank (1976). Learning to read by reading: A brief case study. Lan- 
guage Arts, 53(3), 297-299. 
Smith, Frank ( 1979). Conflicting approaches to reading research and in- 
struction. In Resnick, Lauren B. and Phyllis A. Weaver (Eds.), Theory 
and Practice of Early Reading (Vol. 2). Hillsdale, NJ: Lawrence 
Erlbaum Associates. 
Smith, Frank (1981). Demonstrations, engagement and sensitivity: A re- 
vised approach to language learning. Language Arts, 58(1), 103-1 12. 
Smith, Frank (1983a). Essays into Literacy. Portsmouth, NH: 
Heinemann. 
Smith, Frank (1983b). Reading like a writer. Language Arts, 60(5). 
558-567. 
Smith, Frank (1986). Insult to Intelligence. New York: Arbor House. Re- 
printed Portsmouth, NH: Heinemann, 1998. 
356 
UNDERSTANDING READING 
Smith, Frank (1988). Joining the Literacy Club. Portsmouth, NH: 
Heinemann. 
Smith, Frank ( 1989), Overselling literacy. Phi Delta Kappan, 70(5), 352-359. 
Smith. Frank (1990). To think. New York: Teachers College Press. 
Smith, Frank (1992). Learning to read: The never-ending debate. Phi 
Delta Kappan. 73(6), 432—441. 
Smith, Frank (1993). Whose Language? What Power? New York: 
Teachers College Press. 
Smith, Frank (1994). Writing and the Writer (2nd ed.). Hillsdale, NJ: Law- 
rence Erlbaum Associates. 
Smith, Frank (1994b). Understanding Reading (Fifth edition). Hillsdale, 
NJ: Lawrence Erlbaum Associates. 
Smith, Frank (1998). The Book of Learning and Forgetting. New York: 
Teachers College Press. 
Smith, Frank (2003). Unspeakable Acts. Unnatural Practices: Flaws 
and Fallacies in “Scientific” Reading Instruction. Portsmouth, NH: 
Heinemann. 
Smith, Frank and Peter Carey ( 1966). Temporal factors in visual informa- 
tion processing. Canadian Journal oj Psychology, 20(3), 337-342. 
Smith, Frank, Deborah Lott, and Bruce Cronnell (1969). Effect of type size 
and case alternation on word identification. American Journal of Psy- 
chology, 82(2), 248-253. 
Snow, Catherine E., M. S. Burns, and R Griffin (1998). Preventing Reading 
Difficulties in Young Children. Washington, IOC: National Academy Press. 
Sperling, George ( 1960). The information available in brief visual presen- 
tations. Psychological Monographs, 74(1 1), Whole No. 498. 
Sperry, R. W. (1968). Hemisphere disconnection and unity in conscious 
awareness. American Psychologist, 23, 723-733. 
Squire, James R. (Ed.) (1987). The Dynamics of Language Learning . Ur- 
bana, IL: National Conference on Research in English. 
Staffer, Joshua (1982). Neurological correlates of reading failure. In 
Singer, Martin H. (Ed.), Competent Reader, Disabled Reader: Re- 
search and Application. Hillsdale. NJ: Lawrence Erlbaum Associates. 
Stanovich, Keith E. ( 1980). Toward an interactive-compensatory model of 
individual differences in the development of reading fluency. Reading 
Research Quarterly. 15, 32-71. 
Stanovich, Keith E. (1981). Attentional and automatic context effects in 
reading. In Lesgold, Alan M. and Charles A. Perfetti (Eds.), Interactive 
Processes in Reading. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Stanovich, Keith E. (1982a). Individual differences in the cognitive pro- 
cesses of reading: I. Word decoding. Journal of Learning Disabilities, 
15, 485—493. 
Stanovich, Keith E. (1982b). Individual differences in the cognitive pro- 
cesses of reading: II: Text-level processes. Journal of Learning Disabil- 
ities. 15. 549-554. 
Stanovich, Keith E. (1986). Matthew effects in reading: Some conse- 
quences of individual differences in the acquisition of literacy. Reading 
Research Quarterly. 21(4), 360-407. 
REFERENCES 
357 
Stanovich, Keith E. (1991). Cognitive science meets beginning reading. 
Psychological Science, 2(70). 77-81. 
Stanovich, Keith E. (1992). Speculations on the causes and consequences 
of individual differences in early reading acquisition. In Gough, Philip 
B., Linnea C. Ehri, and Rebecca Treiman (Eds.) Reading Acquisition. 
Hillsdale, NJ: Lawrence Erlbaum Associates. 
Stanovich, Keith E., Anne E. Cunningham, and Dorothy J. Feeman (1984). 
Intelligence, cognitive skills, and early reading progress. Reading Re- 
search Quarterly, 19(3), 278-303. 
Stein, Dieter (Ed.) ( 1992). Cooperating with Written Texts: The Pragmatics 
and Comprehension of Written Texts. Berlin: Mouton de Gruyter. 
Stevenson, V. (Ed.) (1983). Words: The Evolution of Written Languages. 
London: Methuen. 
Suhor, Charles (1992). Semiotics and the English language arts. Lan- 
guage Arts, 69, 228-230. 
Sulzby, Elizabeth (1985). Children’s emergent reading of favorite story 
books: A developmental study. Reading Research Quarterly, 20(4). 
458-481. 
Sutherland-Smith, Wendy (2002). Weaving the literacy Web: Changes in 
reading from page to screen. The Reading Teacher, 55(7), 662-669. 
Tao, Liang and Alice F. Healy (2002). The unitization effect in reading Chi- 
nese and English text. Scientific Studies of Reading, 6(2), 167-197. 
Taylor, Barbara M. (1992). Text structure, comprehension and recall. In 
Samuels, S. Jay and Alan E. Farstrup (Eds.), What Research Has to 
Say About Reading Instruction (2nd ed.). Newark, DE: International 
Reading Association. 
Taylor, Denny (1991). Learning Denied. Portsmouth, NH: Heinemann. 
Taylor, Denny (1998). Beginning to Read and the Spin Doctors of Sci- 
ence: The Political Campaign to Change America’s Mind About How 
Children Learn to Read. Urbana, IL: National Council of Teachers of 
English. 
Taylor, Denny and Dorothy S. Strickland (1986). Family Storybook Read- 
ing. Portsmouth, NH: Heinemann. 
Templeton, Shane (1991). New trends in an historical perspective: The 
“what” and “why” of skills instruction in literacy. Language Arts, 68, 
590-595. 
Thompson, G. Brian. (1981). Toward a theoretical account of individual 
differences in the acquisition of reading skill. Reading Research Quar- 
terly, 15(4), 596-599. 
Thompson, Richard A. (1992). A critical perspective on whole language. 
Reading Psychology, 13(2), 131-155. 
Thorndike, E. ( 1 977). Reading as reasoning: A study of mistakes in para- 
graph reading. Journal of Educational Psychology. 9. 77-110. 
Thorndike, E. L. and I. Lorge (1944). The Teacher’s Word Book of 30,000 
Words. New York: Teachers College. 
Tierney, Robert J. and J. W. Cunningham (1984). Research on teaching 
reading comprehension. In Pearson, R David (Ed.), The Handbook of 
Reading Research. New York: Longman. 
358 
UNDERSTANDING READING 
Tierney, Robert J. and T. Shanahan ( 1991 ). Research on the reading-writ- 
ing relationships: Interactions, transactions and outcomes. In Barr, 
Rebecca, Michael L. Kamil, Peter Mosenthal, and P David Pearson 
(Eds.), Handbook of Reading Research, Vol II. New York: Longman. 
Tolchinksy. Liliana (2003). The Cradle of Culture and What Children 
Know About Writing and Numbers Before Being Taught. Mahwah, NJ: 
Lawrence Erlbaum Associates. 
Torrance, Nancy A. and David R. Olson (1985). Oral and literate compe- 
tencies in early school years. In David R. Olson, Nancy A. Torrance, and 
Angela Hildyard (Eds.), Literacy, Language and Learning: The Nature 
and Consequences of Reading and Writing. Cambridge, UK: Cam- 
bridge University Press. 
Treiman, Rebecca ( 1992). The role of intrasyllabic units in learning to read 
and spell. In Gough, Philip B., Linnea C. Ehri, and Rebecca Treiman (Elds.), 
Reading Acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates. 
Treiman, Rebecca, Usha C. Goswami, and Maggie Bruck (1990). Not all 
nonwords are alike. Implications for reading development and theory. 
Memory and Cognition, 18, 559-567. 
Trelease, Jim and Stephen D. Krashen (1996). Eating and reading in the li- 
brary. Emergency Librarian, 23(5), 27. 
Truax, Roberta R. and Richard R. Kretschmer, Jr. (1993). Finding new 
voices in the process of meeting all the needs of children. Language 
Arts, 70(7), 592-601. 
Trushell, John (2002). Editorial introduction: The future of the book? 
Journal of Research in Reading, 23(2 ) 103-109. 
Tulving, Endel ( 1985a). How many memory systems are there? American 
Psychologist, 40, 385-398. 
Tulving, Endel (1985b). Memory and consciousness. Canadian Journal 
of Psychology , 25, 1-12. 
Tuman, Myron C. (1987). A Preface to Literacy: An Inquiry into Peda- 
gogy, Practice, and Progress. Tuscaloosa: University of Alabama Press. 
Underwood, Geoffrey (Ed.) ( 1978). Strategies of Information Processing. 
London: Academic Press. 
Urwin, Cathy ( 1982). The contribution of nonvisual information systems 
and language to knowing oneself. In Beveridge, Michael (Ed.), Children 
Thinking Through Language. London: Arnold. 
Vaihinger, Hans (1952). The Philosophy of “As-If. ” London: Routledge and 
Kegan Paul. 
van Dongen, Richard ( 1987). Children's narrative thought, at home and at 
school. Language Arts, 64(1), 79-87. 
Vellutino, Frank R. (1987). Dyslexia. Scientific American. 256(3), 34-41. 
Vellutino, Frank R. (1991). Has basic research in reading increased our 
understanding of developmental reading and how to teach reading? 
Psychological Science, 2(70), 81-83. 
Venezky, Richard L. (1967). English orthography: Its graphical structure 
and its relation to sound. Reading Research Quarterly, 2, 75-106. 
Venezky, Richard L. (1970). The Structure of English Orthography. The 
Hague: Mouton. 
REFERENCES 
359 
Venezky, Richard L. (1999). The American Way of Spelling: The Struc- 
tures and Origins of American English Orthography. New York: 
Guilford. 
Vincente, Kim J. and William F. Brewer ( 1 993). Reconstructive remember- 
ing of the scientific literature. Cognition, 46, 101-128. 
Von Sprecken, Debra, Jiyoung Kim, and Stephen Krashen (2000). The 
home run book: Can one positive reading experience create a reader? 
School Library Journal, 23(2), 8-9. 
Vygotsky, LevS. (1978). Mind in Society: The Development of Higher Psy- 
chological Processes. Cambridge, MA: Harvard University Press. 
Walkerdine, Valerie (1982). From context to text: A psychosemiotic ap- 
proach to abstract thought. In Beveridge, Michael (Ed.), Children 
Thinking Through Language. London: Arnold. 
Walters, Gloria S., Melvin K. Komoda, and Tannis Y. Arbuckle (1985). The 
effects of concurrent tasks on reading: Implications for phonological 
recoding. Journal of Memory and Language, 24(1), 27-45. 
Wasik, Barbara A. and Robert E. Slavin (1993). Preventing early reading 
failure with one-to-one tutoring: A review of five programs. Reading Re- 
search Quarterly , 28(2), 178-200. 
Weaver, Constance (1988). Reading Process and Practice: From Socio- 
Psycholinguistics to Whole Language. Portsmouth, NH: Heinemann. 
Weaver, Constance (1994). Reading Process and Practice: From Socio- 
Psycholinguistics to Whole Language. Portsmouth, NH: Heinemann. 
Weintraub, Sam ( 1 986). The fuzzy area of literature reviews. In Weintraub, 
Sam, Helen K. Smith, Nancy L. Roser, Walter J. Moore, Michael W. 
Kibby, Kathleen S. Jongsma, and Peter L. Fisher (Eds.), Summary of 
Investigations Relating to Reading: July 1, 1984 to June 30, 1985. 
Newark, DE: International Reading Association. 
Wells, Gordon (1985). Preschool literacy-related activities and success in 
school. In Olson, D. R., Nancy Torrance and Angela Hildyard (Eds.), Lit- 
eracy, Language and Learning: The Nature and Consequences of 
Reading and Writing. Cambridge, UK: Cambridge University Press. 
West, Richard E, Keith E. Stanovich, and Harold R. Mitchell (1993). Read- 
ing in the real world and its correlates. Reading Research Quarterly, 
28(1), 34-50. 
Wheeler, D. D. (1970). Processes in word recognition. Cognitive Psychol- 
ogy. 1 , 59-85. 
Whorf, Benjamin Lee (1956). Language. Thought and Reality. Cam- 
bridge, MA: MIT Press. 
Wildman, Daniel and Martin Kling (1978/1979). Semantic, syntactic and 
spatial anticipation in reading. Reading Research Quarterly, 14(2), 
128-164. 
Willinsky, John ( 1990). The New Literacy: Redefining Reading and Writ- 
ing in the Schools. London: Routledge. 
Willis, Arlette Ingram (2002). Literacy at Calhoun Colored School 
1892-1945. Reading Research Quarterly, 37(1). 8-45. 
Wilson, Stephanie Gray, Mike Rinck, Timothy R McNamara, Gordon H. 
Bower, and Daniel G. Morrow (1993). Mental models and narrative 
360 
UNDERSTANDING READING 
comprehension: Some qualifications. Journal oj Memory and Lan- 
guage, 32, 141-154. 
Wolf, Maryanne and Tami Katzir-Cohen (200 1 ). Reading fluency and its in- 
tervention. Scientific Studies of Reading, 5(3), 21 1-239. 
Wong, Bernice Y. L. (1982). Understanding learning disabled students' 
reading problems: Contributions from cognitive psychology. Topics in 
Learning and Learning Disabilities, 2, 43-50. 
Yates, Jack (1985). The content of awareness is a model of the world. Psy- 
chological Review, 92(2), 249-284. 
Yatvin, Joanne (1982). Learning to read at forty-eight. Language Arts, 
59(8), 822-828. 
Yatvin, Joanne (2002). Babes in the woods: The Wanderings of the Na- 
tional Reading Panel. Phi Delta Kappan, 83(5), 364-369. Reprinted in 
Allington (2002). 
Young, Andrew W. and Andrew W. Ellis (1981). Asymmetry of cerebral 
hemispheric functioning in normal and poor readers. Psychological 
Bulletin, 89( 1), 183-190. 
Author Index 
A 
Adams, Marilyn Jager, 284, 321 
Agnoli, Franca, 20 
Allington, Richard L., 323 
Anderson, John R., 271 
Anderson, Richard C., 170, 190, 
246, 249, 321 
Anderson, T. H., 248 
Anglin, Jeremy M., 299 
Applebee, Arthur N.. 249 
Arbuckle, Tannis Y., 293 
Armbruster, Bonnie M., 248 
Armstrong, Stephen, 310 
Atwell, Nancie, 310, 314 
Austerfield, Vicky, 310 
Averbach. E.. 266 
B 
Baddeley, Alan, 270 
Bahrick, Harry B., 271 
Baird, William, 251 
Baldwin, R, Scott, 301 
Balzano, Gerald J., 243 
Bao-Qing, Ding, 285 
Baratta, Anthony N., 310 
Barclay, J. R., 272 
Barlow, John A.. 239 
Barr, Rebecca C., 307 
Bartlett, Frederick C., 14, 48, 243, 
271 
Baumann, James F , 309 
Beck. Isabel L., 248. 318 
Beech. John R.. 296 
Beers, Terry, 249 
Benton, Arthur L., 274 
Berdianski, Betty, 143, 281 
Berliner, David C., 315 
Berninger, Virginia Wise, 312 
Bertino, Mary, 291 
Biddle, Bruce J., 315 
Biederman, Irving. 277 
Black, Alison. 244 
Blanton, William E., 318 
Bloome, David. 251 
Blough. Donald S., 274, 278 
Bolter. Jay David. 316 
Boulding, Kenneth E., 240 
Bower, Gordon H., 273 
Brady, Susan A., 285 
Bransford, John D., 245, 250 
361 
362 
UNDERSTANDING READING 
Brewer. William E , 320 
Bridge, Connie A., 273 
Bridges. Allayne, 299 
Broadbent, Donald E., 258 
Broadbent, Margaret H. R, 258 
Brown. Ann L., 299, 311 
Brown, H. Douglas, 288 
Brown, Roger, 101 
Bruck, Maggie, 287 
Bruner, Jerome S., 127, 241, 244. 
245, 258, 265 
Bryant, Peter, 284, 285, 287, 313 
Buckner, Aimee, 315 
Burke, Carolyn L., 247 
Burns, M. S., 321 
Burton, Mike, 271 
Buswell, G. T., 263 
Byrne, Ruth M. J., 249 
C 
Cadeiro-Kaplan, Karen. 248 
Cairney, Trevor H., 248, 250 
Calfee, Robert C., 312, 315 
Calkins, Lucy, 298 
Campbell, Donald T., 320 
Caplan, David, 269 
Carbo, Marie, 310, 319 
Carey, Peter, 266 
Carey, Susan, 299, 300 
Carpenter, Patricia A., 290, 297 
Carver, Ronald, 295 
Castiglioni-Spalton, M., 286 
Cattell, James McKeen, 263 
Cazden, Courtney B., 248, 310 
Celano, Donna, 303 
Chafe, Wallace, 248 
Chall. Jeanne, 221. 282, 317. 319 
Chambers, Maryl, 308 
Chase, William G., 244, 272 
Cho, Kyung-Sook, 305 
Chomsky, Noam, 146 
Christensen, Carol A.. 287 
Claire, Hilary, 308 
Clark, Margaret M., 226, 303 
Clark, Eve V, 299 
Clay, Marie M„ 304, 310, 313 
Ciymer, Theodore, 282 
Cohen, Jonathan D., 323 
Cole, Michael, 294 
Coles, Gerald, 314. 322 
Colwell. Sarah O., 312 
Cooper, Franklin S., 252 
Coriell, A. S., 266 
Cornman, O., 288 
Coulmas, Florian, 239 
Courts, Patrick L.. 308 
Cronbach, Lee J., 320 
Cronnell, Bruce, 143, 278. 281 
Crowder, Robert G., 290 
Csikszentmihalyi, Mihalyi, 298 
Cunningham, Anne E., 296. 307 
Cunningham, J. W.. 294 
D 
Daneman, Meredyth, 290 
Davis, Steven, 247 
Deely, John, 247 
DeFord, Diane E.. 307 
DeJean, Jillian, 316 
Dippo, Don, 318 
Dodge, R.. 263 
Dodwell, Peter, 235, 323 
Doehring, Donald G., 291 
Dole. Janice A., 249 
Dombey, Henrietta, 306 
Dorman, Casey. 312 
Downing, John, 52, 224, 313 
Dreher. Mariam Jean, 285 
Drum, Priscilla A., 290 
Duckett, Peter. 268 
Dudley-Marling, Curt. 318. 319 
Duffy, Gerald G.. 249 
Duke, Nell K., 303 
Dunn. Kenneth. 310 
Dunn. Rita. 310 
Dunn-Rankin, Peter, 122 
Durkin, Dolores, 249. 306 
E 
Eagleton, Terry, 250 
Eckhoff, Barbara, 298. 307 
Edelsky, Carole, 308 
Egan-Robertson, Ann. 251 
Ehri, Linnea C.. 284. 285. 286. 290. 
293. 311 
Eldredge. J. Lloyd. 306 
Elley, Warwick B.. 297, 304. 305 
Ellis. Andrew W., 269 
Erdmann, B., 263 
Ericsson. K. Anders. 272 
Escobar, M., 313 
Ewoldt. Carolyn, 292 
AUTHOR INDEX 
363 
F 
Falcon, Steve, 272 
Faraone. Vincent, 250 
Feeman, Dorothy J., 296, 307 
Fellbaum, Christine, 292, 301 
Fernandez-Armesto, Felipe, 238, 
239 
Ferrelro, Emilia, 229, 303 
Fiske, Donald W., 320 
Fletcher, J., 313 
Fodor, Jerry A., 246, 250, 292 
Forster, Kenneth I., 296 
Fosnot, Catherine Ttoomey, 240, 
247, 299 
Fraisse, Paul, 131 
Freeman, Paul, 244 
Freeman, Yvonne S., 309 
Freire, Paulo, 226 
Fries, Charles C., 36, 254 
Furlin, Karen R., 130 
Furness, David W., 308 
G 
Gallistel, Charles R., 299 
Garan, Elaine ML, 323 
Garner, Ruth, 250 
Garner, Wendell R., 265 
Garrett, Merrill E., 246 
Gibbs, Raymond W., 246 
Gibson, Eleanor J., 275 
Gill, J, Thomas, 318 
Gillooly, William B., 288 
Giroux, Henry, 315 
Goelman. Hillel, 230, 302 
Goetz, Ernest M.. 273 
Golden, Joanne M., 246 
Goldman, S. R., 296 
Goldsmith-Phillips, J., 290 
Golinkoff, Roberta Michnick, 29 1 
Goodman, Kenneth S., 160, 263, 
309. 317, 318 
Goodman, Marsha S., 249 
Goodman, Yetta, 302 
Goody, Jack. 45 
Gorman, Tom, 308 
Goswami, Usha C., 284, 285, 287 
Gough, Philip B., 285, 293, 312 
Graff. Harvey J., 298 
Graham, Judith, 310 
Graham, Steven, 313 
Graves, Michael F., 308 
Griffin, R, 321 
Grimes, Joseph E., 249 
Guthrie, John T., 246 
H 
Haber, Lyn R., 130 
Haber, Ralph N., 130 
Hall. Lynda K., 271 
Halle, Morris, 146, 275 
Halliday, Michael A. K„ 199 
Hamill, Donald D., 313 
Harding, Leonora M., 296 
Hardyck, C. D., 167 
Harris, C., 296 
Harste. Jerome C., 247 
Hartman, Maria, 304 
Havelock, Eric A., 45 
Healy, Alice F., 290 
Heath, Shirley Brice, 248, 298, 306, 
314 
Hecht, Barbara Frant, 299 
Heckenmueller, H. G.. 83 
Hedley, Carolyn. 310 
Helfand, Duke, 320 
Henderson, John M., 292 
Herman, Patricia A., 170, 300 
Hidi, Suzanna, 251 
Hiebert, Elfrieda H„ 190, 308. 311, 
315, 321 
Hildyard. Angela. 251, 310 
Hillinger. M. L.. 312 
Hittleman, Daniel R.. 250 
Hoffman, James E., 294 
Hogaboam, T, W., 296 
Holdaway, Don, 307 
Hong-Yin, Nie, 285 
Howe. M. J. A„ 290 
Howes, D. H., 258 
Huck, Charlotte S., 309 
Huey, Edmund Burke, ix, 166, 263 
Hull, Glynda A,. 314 
Hull, Robert, 307 
Hummel, John E., 277 
Hunt, Earl. 20 
Hutchison. Dougal. 308 
I 
Imlach. Robert, 294 
Impey, Lawrence, 313 
International Reading Association 
Board of Directors. 315 
364 
UNDERSTANDING READING 
Intraub, Helen. 294 
J 
Jackson, Adrian, 308 
Jakobson, Roman, 275 
Jiyong, Kim, 305 
Johnson-Laird, Philip N., 244, 249, 
292 
Johnston, Francinc R., 282 
Johnston, Peter H., 313 
Judd, C. H„ 263 
Juel, Connie, 307, 318 
Just, Marcel A., 297 
K 
Kamii, Constance K., 240 
Kansinger, S., 273 
Karlin, J. E., 267 
Katzir-Cohen, Tami, 284 
Katzman, M. T., 267 
Keen, Robert H., 279 
Kimmel, Susan, 294 
Kintsch, Walter, 250, 270 
Klahr, D., 244 
Klein, Gary A., 291 
Klein, Helen Altman, 291 
Kling, Martin, 241 
Koehler, John A., 143, 281 
Kolers, Paul A„ 75, 160, 267 
Komoda, Melvin K., 293 
Korkeamaki, Ritta-Liisa, 285 
Kozol, Jonathan, 303 
Krashen, Stephen D., 249, 270. 282, 
286, 288, 297, 299, 303, 
304, 305, 309, 310, 319 
Kretschmer. Richard R., 304, 314 
Kretschmer, Robert E., 304 
Kroll, J. E. 296 
Krueger, Lester E.. 279 
Kucer, Stephen B., 268 
Kuhn, Thomas, 236 
Kukish, Karen S., 291 
Kutz, Eleanor. 311 
L 
Lakoff, George, 242 
Langer, Judith A., 312 
Larsen, Steen F., 261 
Larson, Joanne, 322 
Lehr, Fran. 284 
Levine, K., 310 
Levy, Betty Ann. 290 
Lewis, Donald J., 270 
Liberman, Alvin M., 252, 283, 285, 
318 
Liberman, Isabelle A., 283, 285, 
318 
Lindig, Karen, 265 
Lipson, Marjorie Y., 312 
Littlefair, Alison B., 248 
Lloyd, B. B., 242 
Lodge, David, 251 
Long, Shirley A., 273 
Lorge. I., 136 
Lott, Deborah, 278, 279 
Lovelace, E. A., 244 
M 
MacGinitie. Ruth K.. 307 
MacGinitie, Walter H., 249. 294. 
307 
Mackworth, Norman H.. 79 
Macmillan. Bonnie M.. 287 
Macnamara, John. 199 
Magee, Bryan. 240 
Makuch. R . 313 
Mandler, George. 107. 274 
Mandler, Jean Matter. 243. 249, 300 
Mangiola. Leslie, 314 
Mangubhai, Francis. 304. 305 
Markman. Ellen M., 300 
Marshall, John C., 161 
Masonheimer, Patricia E.. 290 
Mathews. Mitford M., 5 
McCabe, Viki. 243 
McCawley, James D.. 292 
McClelland. J, L.. 267 
McFarland. Carl E.. 291 
McGee. Lea M., 249. 309 
McKeown, Margaret G.. 248. 301 
McLaughlin, Thomas M.. 318 
McMahon, Susan I.. 310 
McNamara. Timothy R. 250. 273 
McNeill. Daniel. 3 
McNeill, David. 101. 168. 246, 265 
McNutt, Gaye, 313 
McPeck, John E.. 28. 241 
McQuillian. Jeff. 316 
Meek, Margaret. 231. 309. 310 
Metcalf, Stephen. 323 
AUTHOR INDEX 
365 
Meyer, Richard J., 319 
Michotte, A., 192, 244 
Mikulecky. Larry, 308 
Miller, Diane A., 250 
Miller, George A., 33, 127, 209, 254, 
259, 265, 276, 292, 301 
Miller, Larry, 316, 318 
Miller, Rebecca, 316 
Mitchell, D, C., 290 
Mitchell. Harold R., 297 
Mohyeldin Said, K. A., 250 
Moll, Luis C., 306 
Moorman, Gary B., 318 
Morris, R E., 271 
Morrow, Daniel G., 273 
Morrow. Lesley Mandel, 309 
Moyer, Sandra B., 269 
Murphy, Sharon, 309, 319 
Myers, Jamie, 308 
Myers, Jerome L., 294 
N 
Nagy, William E„ 170, 299, 300 
National Reading Panel, 322 
Naylor, Hilary, 269 
Neisser, Ulric, 241 
Nell, Victor, 298 
Nelson, Katherine, 22, 243, 299 
Neuman, Susan B., 303 
Newcombe, F., 161 
Newcomer, Phillis L., 269 
Newman, Edwin B., 267 
Newson, Elizabeth, 199 
Newson, John, 199 
Newton-Smith, W. H., 250 
Nicely, Patricia E.. 276 
Nicholson, Tom, 294 
Norman, Donald A., 270 
O 
O’Brien, Edward J,, 294 
O'Dowd, D., 258 
O’Flahavan, John, 300 
O’Regan, J. K„ 267 
O’Shea, Dorothy J., 295 
O’Shea, Lawrence J., 295 
Oberg, Antoinette A., 230, 302 
Ohanian, Susan, 315 
Oliver. Peter, 224 
Olson. David R.. 45, 297, 310 
Olson, Richard K., 313 
Omanson, Richard C., 248 
Ong, Walter J., 310 
Ortony, Andrew, 243, 246 
Osborn, Jean, 284, 300 
Otto, Wayne, 310 
P 
Palincsar, Annemarie S., 299 
Parkes, C. H., 246 
Paterson, Frances A., 323 
Pattanayak, D. E., 298 
Paulson, Eric J., 263, 268 
Pearson, R David, 249, 279 
Perfetti, Charles A., 242, 254, 285, 
292, 296, 318 
Petrinovich, L. F., 167 
Piaget. Jean, 240 
Piattelli-Palmarini, Massimo, 247 
Pierce, J. R., 267 
Pillsbury. W. B.. 127 
Pinker, Steven, 131, 146, 274 
Pinnell, Gay-Su, 310 
Piper, David, 249 
Plackett, Elizabeth, 310 
Pollatsek, Alexander, 242, 268 
Polyani, Michael, 241 
Pople, Martha T., 248 
Popper, Karl R., 179, 240 
Postman, Leo, 127, 265 
Potter, Mary C., 244. 296 
Pritchard, R. M., 83 
Q 
Quastler, Henry. 264, 267 
R 
Rack, John R, 313 
Raphael, Taffy E., 306 
Rayner, Keith. 242, 267, 268 
Read, Charles, 285, 288 
Reber, Arthur S., 271 
Reichardt, Konrad W., 313 
Reitsma, Pieter, 298 
Restle, E. 244 
Rhode, Mary. 281 
Rhodes, Deborah H.. 291 
Rice, J., 288 
Rice, Mabel L., 300 
366 
UNDERSTANDING READING 
Rieben, Lawrence, 285, 292 
Rinck, Mike, 273 
Robinson, H, Alan. 250 
Robinson, Ian, 292 
Roehler, Laurel R., 249 
Rogers, Rebecca, 248 
Roper/Schneider, Diana, 307 
Rosch, Eleanor, 242 
Rose, Mary, 308 
Rosen, Harold, 244 
Rosenblatt, Louise, 55, 56, 68, 260 
Rosinski, Richard R., 291 
Rosset, Clement, 238 
Roth, S., 296 
Rottenberg, Claire J., 304 
Routman, Regie, 304 
Rublevich, Bella, 279 
Rumelhart. David E., 243 
S 
Sachs, Jacqueline B., 272 
Sacks, Oliver, 254 
Sadoski, Mark, 244, 273 
Saenger. Paul, 295 
Salmon, Phillida, 244, 308 
Sampson, Geoffrey, 239 
Samuels, S. Jay, 224, 295 
Santa, Carol Minnick, 315 
Savage, Robert, 287 
Saxe, Geoffrey B., 240 
Schachter, Sandra R., 314 
Schaeken, Walter, 249 
Schatz. Elinore Kress, 301 
Schneider, Walter, 270 
Scholes, Robert J., 285, 297 
Schooler, Jonathan W., 323 
Schwanenflugel, Paula J., 301 
Scott, Judith A., 190, 299, 321 
Scribner, Sylvia, 294 
Searfoss, Lyndon W., 304 
Seligman, Martin E. R, 313 
Senner, Wayne M., 239 
Shallice, Tim, 161 
Shanahan, T., 298 
Shankweiler, Donald R, 252, 285 
Shannon, Claude E., 258, 264, 279 
Shannon, Patrick, 309, 311 
Sharkey, Noel E., 290 
Shaywitz, M., 313 
Shaywitz, S. E.. 313 
Shiffrin, Richard M., 270 
Simon, Herbert A., 273 
Sindelar, Paul T., 295 
Singer, Linda, 290 
Sinha, Chris, 299 
Skinner, B. E, 261 
Slater, Wayne H.. 249 
Slavin, Robert E., 310 
Smith, Frank, xi, 9, 12, 27, 190. 
203, 209, 216. 221, 222. 
230, 233, 236, 241, 248, 
260, 266, 271, 277, 278, 
279, 288, 298, 299, 302, 
316 
Sneddon, William, 296 
Snow, Catherine E., 321 
Snowling, Margaret J., 313 
Solomon, R. L., 258 
Sperling, George. 266 
Sperry. R. W., 269 
Squire, James R.. 310 
Staller. Joshua, 312 
Stanovich. Keith E.. 242, 269, 268, 
291. 296, 297, 307, 312 
Stein, Barry S.. 245 
Stein. Dieter, 248 
Stevenson, V.. 239 
Strickland, Dorothy S., 306 
Studdert-Kennedy. M.. 252 
Studt. Alice. 279 
Suhor, Charles. 247 
Sulzby, Elizabeth. 306 
Sutherland-Smith, Wendy, 316 
T 
Tao. Liang, 290 
Taylor. Barbara M.. 249 
Taylor, Denny, 306. 315 
Teberosky, Ana. 229 
Templeton. Shane. 242 
Thomas. Charlene, 306 
Thompson. G. Brian, 296 
Thomson. Richard A., 318 
Thorndike, E., 294 
Thorndike, E. L.. 136 
Tierney. Robert J., 294. 298 
Tolchinsky, Liliana. 6. 230 
Torrance. Nancy, 297, 310 
Treiman, Rebecca, 285. 287 
Trelease. Jim. 305 
Trimble, John. 308 
Truax, Roberta R.. 304. 314 
AUTHOR INDEX 
367 
Trushell, John. 316 
Tulving. Endel, 270 
Tuman. Myron C.. 310 
Tnten. Jenny, 268 
U 
Underwood, Geoffrey, 269 
Unrah, Elizabeth, 250 
Urwln, Cathy, 238 
V 
Vaihinger, Hans, 238 
van Dongen, Richard, 244 
Vellutino, Frank R., 242, 312 
Venezky, Richard L., 141, 288 
Viale, R., 250 
Vincente, Kim J., 320 
Von Sprecken, Debra, 305 
Vye, Nancy J., 245 
Vygotsky, Lev, 241, 245, 294 
W 
Wagner, Richard K., 290 
Walker, E„ 246 
Walkerdine, Valerie, 238, 299 
Walsh. Margaret A., 293 
Walters, Gloria S., 293 
Warrington, Elizabeth K., 161 
Wasik, Barbara A., 310 
Watt, Ian, 45 
Weaver, Constance, 310, 319 
Weintraub, Sam, 321 
Weisser, Susan, 244 
Wells, Gordon, 244 
West, Richard F., 297 
Wheeler, D. D„ 277 
White, Howard, 288 
Whorf, Benjamin Lee, 20 
Wildman, Daniel, 241 
Wilkes, Kathleen V., 250 
Wilkinson, Ian A. G., 190, 321 
Willinsky, John, 308 
Willis, Arlette Ingram, 303 
Willis, Brenda J., 285, 297 
Wilson, Stephanie Gray, 273 
Winograd, Peter N., 273 
Winsor, Pamela, 300 
Wixson, Karen K., 312 
Wolf, Maryanne, 284 
Wong. Bernice Y. L., 312 
Woodward, Virginia A., 247 
Y 
Yachzel, B., 296 
Yates, Jack, 241 
Yatvin, Joanne, 310, 322 
Young, Andrew W., 269 
Yun-Fei. Zhang, 285 
This page intentionally left blank 
Subject Index 
A 
Abduction, 247 
Aesthetic reading, 325, see also 
Reading, ejjerent! aesthetic 
Alphabet, 5, 142, 146, 148, 238 
Ambiguity, 37, 135 
Ambiguous figure, 18 
Ampersand, 119 
Apprehension (of meaning), 156, 
see also Comprehension 
Apprentices, 209 
Approximations to English, 127, 
153 
Archeologists, 174 
Artificial intelligence, 235, 261, 325 
As-if, 238 
Authority to think, 28, 207 
Authors, 214, 231 
Awareness, 241 
B 
Basal readers, 309 
Bewilderment, 15 
Bits, 255 
Boredom, 202 
Brain, 9, 72, 90, 324 
disabilities, 5 
Brain-injured readers. 161 
C 
Camels, submarines, 23 
Capitalization, learning, 190 
Car driving analogy, 182 
Caretaker talk, 199 
Categories. 16. 242, 325 
interrelations, 19, 325 
membership. 18 
Cats and dogs, 2, 195, 325 
Channel capacity, 259, 325 
Chomsky, Noam. 224, 247 
Chunking, 103, 272 
Club, literacy, 229 
of readers, 1 90 
spoken language, 209, 230 
Clubs, membership. 190 
Cognition, 325 
Cognitive psychology, 26 
Cognitive science, 28, 235, 261, 326 
Cognitive structure, 13, 326, see 
also Knowledge 
369 
370 
UNDERSTANDING READING 
dynamics, 22 
Cohesion, 50 
Comprehension, 60. 248, 296, 326, 
see also Understanding 
definitions, 179 
and knowledge, 12 
and reduction of uncertainty, 1 62 
Computers, 261 
and instruction, 221 
Confusions, 121, 275 
Consciousness, 270 
Consonants, single/double, 140 
Constructivism, 240 
Context-dependent language, 43, 
218, 326 
Control, 192 
Conventions, 48, 326 
global and focal, 186 
Crisis, 315 
Criterial set, 116, 326 
Criterion, 65, 326 
Culture, 210 
D 
Decoding to sound, 6, 326 
Deconstructionism, 250 
Deep structure, 32, 326 
Dialect, 147 
Dictionaries, 301 
Direct instruction, ix, 9, 221, 326 
Discourse structure, 21, 46, 186, 
327 
Discourse theory, 248 
Disposition for thinking, 28 
Distinctive features, 19, 122, 242, 
327 
letters, 112 
speech, 275 
words, 128 
Dual process, 292 
Dyslexia, 312, see also Seeing 
backward 
E 
Eclecticism, 236 
Errors, 6 1 , see also Miscues 
Evaluation, 227, 314, see also Tests 
Event knowledge. 22, 243. 327 
Experience, 55. 260 
Eye-movements, 263, 267, 291 
fixations, 84, 327 
regressions, 84, 329 
saccade, 84 
Eye -voice span, 40 
F 
Faces, 3 
Facky-tious phenomenon, 171 
Federal studies, 321 
Feelings, 178, 190 
Flow, 298 
Fluent/difficult reading. 224, 249. 
295 
Functional equivalence, 116. 327 
G 
Garden path sentences. 107, 290 
Gaze, convergence, 199 
Genre. 21. 248, 327 
schemes, 46, 186 
Grammar, 37. 186, 249 
Grapheme units. 141. 144 
Great debate, vii. 317. see also 
Never-ending debate 
H 
Hand movements. 269 
Hearing impairment, 292 
Hemispheric specialization, 269 
Hits, 65 
Hope. hot. hook .... 142 
Horse, mule or donkey. 152 
Hypothesis testing. 1 96, 327 
I 
Icons, 6. 1 75 
Identification vs. recognition, 110 
Identifying new words (reprise). 172 
Idiom. 49, 186 
Imagery, 244. 272 
Images, in visual system. 83. 120 
Imagination, 27 
Information. 255. 328 
and experience. 55 
and uncertainty. 56 
Insight, print is meaningful. 215 
spoken and written language 
different. 218 
SUBJECT INDEX 
371 
Instruction, 307 
Instructional methods, 220 
Intelligence, 307 
Intentions, of writers. 185 
Interminable controversy, 316, see 
also Never-ending debate 
J 
Jabberwocky, 217 
John, distinguishing, 133 
Juice illustration, 198 
K 
Kid watching, 302 
Knowledge. 240 
implicit, 195 
structure of, 15 
L 
Language, 254 
complexity, 51 
creates worlds, 7 
description vs. explanation, 8 
about language, 52, see also 
Metalanguage 
learning, 43 
social nature, 51 
specialization, 41 
spoken and written, 31 
written, 1 
Learning, 167, 328, see also 
Learning to read 
categories, 200 
by comprehension, 310 
conditions, 203 
is constant, 201 
conventions of writing, 190 
demonstrations, 204, 326 
difficulties, 4 
disabilities, 312 
distinctive features, 200 
engagement, 204 
to identify words, 132, 301 
about language. 197 
from other people, 209 
risks and rewards, 203 
sensitivity, 206, 329 
social nature, 208 
styles. 310 
about world, 194, 299 
about written language, 212 
Learning to read, 222, 302 
beginning, 225 
by hearing impaired, 292 
by reading, 213 
Letter identification, 110, 118 
clusters, 122, 275 
feature analysis, 114, 327, see 
also Distinctive features 
template matching, 113 
theories, 274 
Lexicon, 174, 293, 328 
Libraries, 303 
Lion street, 1 8 
Literacy, overrated, 298 
and schooling, 306 
Long-term memory, 100, 328 
Luggage/footwear illustration, 216 
M 
Making sense, 2, 12 
Masking, 82, 266 
Meaning, 32, 136, 151, 218, 246, 328 
beyond words. 165 
inferring, 174 
and prediction, 167 
priority in reading, 160 
use in reading, 157 
Meaning identification, 156, 289 
learning, 167 
mediated, 170, 328 
Meanings, multiplicity of, 36 
Memory, 3, 13, 47, 95 
alternative representation, 103 
aspects, 96 
bottlenecks. 95 
buffer, 97 
children's, 273 
control. 106 
input/capacity/persistence/ 
retrieval, 96, 109 
from inside, 106 
long-term, 100, 328 
overcoming limitations, 102 
persistence, 271 
recognition/reproduction, 270 
short-term. 97, 330 
theories, 269 
typical representation, 97 
working, 97 
372 
UNDERSTANDING READING 
Metacognition, 26, 29, 250, 328 
Metalanguage, 52, 328 
Metalinguistic awareness, 311, 328 
Miscues, 160, 166 
Misses, 65 
Missing letters, 164 
Moliere, 53 
Mother-child exchange. 168 
Motherese, 199 
Motivation, 302 
Moving lights, 192 
N 
National Reading Panel, 322 
Neuroscience. 9 
Never-ending debate, 221, 316 
Noise, 61, 328 
Nonvisual information, 9, 73, 88, 
328 
use of, 164 
Normalizing, 114 
O 
Onions, 19, 37 
Orthography, 131, 293, 328, see 
also Spelling 
P 
Paragraphing, learning. 190 
Partial recall, 266 
Pattern recognition, theories, 111, 
274 
equivalence, 112 
Phonemes, 251, 329 
Phonemic awareness, 146, see also 
Phonological awareness 
training, 286 
Phonics, 128, 138, 280, 329, see 
also Sound- spelling 
relationships 
complexity, 140, 281 
efficiency, 143 
result of reading, 152 
rules and exceptions, 140 
for secondary students, 320 
Phonological awareness, 225, 282, 
329, see also Phonemic 
awareness 
Piaget. Jean, 247 
Pigeons, 274, 278 
Power, 192, 303 
Prediction, 23, 181, 241, 291. 329 
and comprehension, 25. 38 
defined. 25 
global and focal, 182 
need for, 24 
Prepositions, 36 
Prior knowledge, 13, 28. 329 
Problem readers, 40 
Processes, 8, 12 
Pronunciation, 147, 171, 176 
Prototype theory. 242 
Psycholinguistics, 233. 329 
Punctuation, learning, 190 
Puns, 34 
Q 
Questions and answers, 180 
R 
Radar, 66 
Readability. 223 
Reading, aloud/silently, 81, 166. 175 
benefits, 189, 297 
to children, 305 
definitions, 179 
efferent/aesthetic, 260, 327 
free voluntary, 304 
fluent/difficult. 188 
and information, 68 
instruction. 217 
jargon, 224, 261 
for learners, 231 
is natural, 1 
"practicing,'' 284 
readiness. 226 
recreational, 297 
right to left. 1 43 
social activity. 308. 310 
speed. 86. 164, 295 
teaching, 222, 309 
theories, 233 
and thinking, 191, 294 
trade-off, 73 
in workplace. 308 
the world, 1 
and writing, 298 
see also Learning to read 
Recognition/identification. 274 
SUBJECT INDEX 
373 
Redundancy, 63, 130, 256, 289, 329 
use by children, 279 
of letters, 117 
among words, 279 
Register, 50, 186 
Rehearsal, in memory, 99 
Reinforcement, 214 
Reminding, 107 
Research 220. 236, 250, 302, 310 
“Scientific,” 4 
Rhyme, 283, 288 
Right to ignore, 169 
Rules and exceptions, 144, 282 
S 
-s, -ed pronunciation 147, 175 
Scenarios, 20, 329 
Scenes, 243 
Schemes, 20, 243, 329 
School language, 219 
Scripts, 20, 243 
Seeing backward, 89, 269 
Semantics, 186, 329 
Semiodcs, 48, 247 
Sense, search for, 158 
expectation of, 168 
Sensory store, 81, 97, 330 
Sight vocabulary, 154, 196 
Signal detection theory, 66 
Significant differences, 112, 252, 
330 
Situation-dependent language, 43, 
218, 330 
Skills. 8, 12, 330 
Skinner, B. E, 247 
Skipping, 149 
Sound-spelling relationships, 142, 
214, 321, see also Phonics 
table, 145 
Sounding out, 150 
Specification of a text, 181, 187 
Speech, 285 
distinctive features, 275 
sounds, 145, 251 
Spelling, 186, 258, 287, 298 
instruction, 288 
learning, 190 
and meaning, 147 
reform, 146 
scrambled. 278 
units, 141 
Standards, 227, 314 
Stories, 20, 243 
Story grammar, 47 
Strategies, 249 
Style, learning, 190 
Sub vocalization, 166, 245 
Surface/deep structure, 32, 245, 
330 
Syllables, 150 
Syntax, 37 
T 
Tachistoscope, 77, 121, 330 
Teachers, 212 
Tests, 227, 314 
Text, 330 
backward, 159 
English and French, 160 
organization. 46, 248 
Theories, 238, 330 
of the world, 14, 330 
constructing, 195 
Thinking. 26, 178, 191, 245 
constraints, 28 
modes of, 244 
narrative basis, 244 
Tip of the tongue. 101 
Transformational grammar, 224. 
330 
Tunnel vision, 77, 95 
U 
Uncertainty, 56, 255. 330 
Understanding, 12, see also 
Comprehension 
familiar words, 172 
unfamiliar words, 173 
V 
Vision, limits, 72, 262 
speed. 265 
Visual information. 32, 73, 120. 331 
Vocabulary, 170, 297, 299 
W 
Whole language, 309. 317, 331 
Word identification. 125 
by analogy, 150, 287 
374 
UNDERSTANDING READING 
feature analysis, 128 
immediate, 126 
immediate/mediated compared, 
157 
learning, 132 
letter clusters, 128, 277 
letter by letter, 127 
mediated, 138, 149, 280. 328 
by analogy. 150 
learning, 153 
strategies, 149 
template theory, 126 
theories, 125 
whole word view, 126 
Words, 34, 135, 254 
ambiguity of, 1 35 
and images, 131, 294 
and meanings, 35 
and pictures, 131 
rare, 136 
Writing, 178, 185 
Chinese, 7, 161. 238, 290 
discrimination, 4 
Greek, 141 
Japanese, 290 
and reading, 190, 298 
systems, 239 
Written/spoken language differences. 
41