HOW PSYCHOLOGICAL SCIENCE INFORMS
THE TEACHING OF READING
Abstract
—This monograph discusses research, theory, and practice
relevant to how children learn to read English. After an
initial overview of writing systems, the discussion summarizes
research from developmental psychology on children’s language
competency when they enter school and on the nature of
early reading development. Subsequent sections review theories
of learning to read, the characteristics of children who do
not learn to read (i.e., who have developmental dyslexia), research
from cognitive psychology and cognitive neuroscience
on skilled reading, and connectionist models of learning to
read. The implications of the research findings for learning to
read and teaching reading are discussed. Next, the primary
methods used to teach reading (phonics and whole language)
are summarized. The final section reviews laboratory and
classroom studies on teaching reading. From these different
sources of evidence, two inescapable conclusions emerge: (a)
Mastering the alphabetic principle (that written symbols are
associated with phonemes) is essential to becoming proficient
in the skill of reading, and (b) methods that teach this principle
directly are more effective than those that do not (especially for
children who are at risk in some way for having difficulty
learning to read). Using whole-language activities to supplement
phonics instruction does help make reading fun and
meaningful for children, but ultimately, phonics instruction is
critically important because it helps beginning readers understand
the alphabetic principle and learn new words. Thus, elementary-
school teachers who make the alphabetic principle
explicit are most effective in helping their students become
skilled, independent readers.
INTRODUCTION
Learning to read presents a paradox. For an adult who is a
good reader, reading feels so simple, effortless, and automatic
that it is almost impossible to look at a word and not read it.
Reading seems so natural to the literate adult that one could
easily imagine that it must rank among the simplest skills for a
child to acquire. Yet nothing could be further from the truth.
For many children, learning to read is an extraordinarily effortful
task, a long and complicated process that can last for years.
That is the essence of the paradox. How can a skill that feels so
easy to the adult be so difficult for the child to acquire? The
paradox is interesting to the scientist because learning to read
is strikingly different from other sorts of learning.
But the significance of the paradox is more general, in ways
that touch everyone. Literacy is an essential ingredient of success
in societies like ours, where so much information is conveyed
by the written word. Furthermore, a literate population is
a key to the functioning of these societies. A significant number
of people never achieve the effortless literacy of the skilled
reader. For them, the complex process of learning to read never
came to an end. To help them, as well as children just learning
to read, it is important to understand the source of their difficulty
and how to overcome it. To achieve these goals, scientists
need to understand three aspects of the paradox:
1. The starting point: What are the preconditions for learning
to read? What must a child be able to do in order to learn to
read effectively?
2. The learning process: What is the process of learning to
read? What happens when a person goes from being a nonreader
to being a reader?
3. The end point: What does skilled reading—the end point of
the learning process—look like?
As scientists learn more about the starting point, the process,
and the end point of learning to read, they can more effectively
address the vital fourth issue:
4. Appropriate educational practices: What are the best ways
to teach reading?
These are the central topics with which we are concerned in
this monograph.
The major instructional methods traditionally used to teach
reading have been
whole-word
and
phonics
instruction. In
whole-word instruction (also called the
look-say
method), a
sight vocabulary of 50 to 100 words is taught initially. Subsequent
words are also learned as wholes, although not necessar-
ily out of context. In contrast, phonics instruction emphasizes
the relationship between
graphemes
(printed letters) and
phonemes
(their associated sounds). Unfortunately, in English the
grapheme-phoneme correspondence is complex, and critics of
this approach have argued that this lack of perfect correspondence
causes confusion for the beginning reader. More recently,
an approach to teaching reading that emphasizes meaning, called
whole-language
instruction, has been widely implemented in
school districts, and the debate on how to best teach reading
has focused on whole-language versus phonics approaches.
In this monograph, we first present some background material
on different writing systems and the alphabetic principle.
Then we discuss in more detail the starting point of learning to
read (and the learning process) and the end point (taking into
account insights from research in cognitive psychology, cognitive
neuroscience, and connectionist models), while drawing
implications for learning to read. We continue with a discussion
of the methods that are used to teach children to read and what
happens in classrooms where children are taught to read. We
conclude by discussing research related to reading instruction.
WRITING SYSTEMS AND THE
ALPHABETIC PRINCIPLE
Writing systems have developed around mixes of various
principles that map a graphic form onto some unit of language.
All modern writing systems connect to the spoken language,
and do not directly encode nonlinguistic meanings. Written
language, of course, does not directly use the sound waves produced
by a human speaker: A book obviously is not a CD
player. Instead, writing systems preserve abstract language
units that are used in spoken language.
As illustrated in Figure 1, writing systems differ as to which
language units are represented by graphic units. The elementary
writing units can correspond to elementary speech sounds,
such as phonemes (the /b/ sound in
bat
) and syllables, and to
morphemes
. Morphemes are the minimal units associated with
meaning (such as the
cook
in
cooks
and
cooking
) and grammatical
form (the
s
in
cooks
and the
ing
in
cooking
). These mapping
options give rise to the three major kinds of writing
systems—alphabetic, syllabary, and morpho-syllabic systems—
that are found among the world’s languages, often in intermixed
form (1989; 1952). Each system may
have variations in orthography—the details of the mapping between
graphic units and language units. And each system allows
each of its basic units to map onto one or more
phonological units (phonemes or syllables) or morphological
units (morphemes).
English, Italian, Russian, and Korean are examples of
alphabetic
writing systems, in which graphic units (letters) are
associated with phonemes. The letter
b
in the written word
bat
corresponds to the phoneme /b/ in the spoken word “bat.” Arabic,
Hebrew, and Persian are modified alphabetic systems, in
which vowels can be omitted. In
syllabaries
, such as Japanese
Kana, the graphic units correspond to syllables. Systems whose
units correspond to specific words or morphemes are usually
called
logographic
. The logographic type of writing system is
usually said to be represented by Chinese (along with the Japanese
Kanji adaptation of the Chinese system). However, although
its origins might be logographic, Chinese writing has
evolved into a
morpho-syllabic
system (1989) in
which the characters map onto syllable units that are also usually
morphemes. The world is without an example of a writing
system that encodes meaning in any pure form. In fact, the history
of writing exhibits movement away from representation of
meaning and toward more direct representation of sound
(1989;1952; 1981).
The beginning reader must learn how a given writing system
relates to spoken units in his or her language. Thus, critical details
of learning to read depend on the writing system. Chinese,
for example, can be read by associating each symbol with a
meaning. Learning these form-meaning associations takes considerable
effort. A vocabulary of 5,000 to 7,000 characters is
typical of literate Chinese adults; during the first 6 years of
school, children master about 3,500 characters by learning 500
to 600 per year. For the past half-century, Chinese children
have been taught to read
pin yin
in the first grade prior to learning
the character system. Pin yin is an alphabet, using letters of
the Roman alphabet to spell Chinese words and adding marks
to the spellings to indicate tones, the pitch variations that ac-
company Chinese vowels. Thus, Chinese children can be said
to learn alphabetic reading as a first step toward mastery of
their own morpheme-based system (1995).
Even with this head start, Chinese children spend more time at
both school and home learning to read Chinese characters than
American children spend in mastering an equivalent number of
words. Part of this difference in time spent practicing reading
undoubtedly reflects cultural factors (1990).
However, an additional factor is the difference in economy between
an alphabetic system and a character system. An alphabetic
system gains economy by mapping written units onto a
small set of elements—the phonemes of a language—rather
than the much larger set of morphemes a language has. This association
of letters with phonemes is referred to as the
alphabetic
principle
, and it allows alphabets to be
productive
; that is,
a small set of symbols (letters) can be used to write an indefinitely
large number of words. Productivity simplifies the learning
problem, for example, allowing the child to use the mapping
between four letters and their phonemes /t/, /p/, /s/, and /o/ to
read
top
,
pot
,
stop
,
spot
,
pots
, and
tops
.
One of the main points of this monograph is that, despite the
economy of the alphabetic principle, learning to read an alphabetic
writing system like English is not easy. There are two main
sources of difficulty: the abstract nature of phonemes (especially
consonants) and the fact that most alphabets do not code each
vowel with a unique symbol. Regarding the first issue, young
children often have an imperfect idea of what phonemes are because
they are abstractions rather than natural physical segments
of speech. This is less of a problem for vowels than for consonants.
For example, the vowel sound in
bat
is about the same as
the one in
laugh
, and because vowels tend to have a relatively
long duration, in both words the sound can be clearly heard and
isolated. Thus, a teacher can point at the
a
in
bat
and say /æ/, and
the child can hear the vowel sound clearly because it has sufficient
duration no matter which sound precedes it. The pronunciation
of a consonant, in contrast, can be highly dependent on the
vowels that precede and follow it. For example, the /d/ in
dime
is
different acoustically from the /d/ in
dome
or in
lid
. The letter
d
,
then, corresponds to a phonemic representation that subsumes the
/d/ sounds in these words but is not identical to any one of them.
In such cases, it may be difficult for a child who is learning to
read to discover and make mental representations of phonemes
without some assistance. It also follows that applying the alphabetic
principle will be difficult: A child who cannot identify an
abstract phoneme (such as /d/) will have difficulty associating it
with a specific grapheme (such as
d
).
The second obstacle to learning to read an alphabetic writing
system is that many alphabets do not code each vowel with a
unique symbol. For example, American English has more than a
dozen vowel sounds but only five standard vowel letters (see Fig.
2). That means that
a
,
e
,
i
,
o
, and
u
have to do double and triple
duty, even with some help from
y
and
w
(both of which can
change vowel sounds, as in
saw
and
say
;
y
can also substitute for
i
). For example,
cat
,
car
, and
cake
each use the letter
a
for a dif-
ferent vowel phoneme. Thus, the writing system exhibits economy
(one letter represents these vowels rather than three) at the
expense of complexity (the mapping between letter and sound is
one-to-many). The trade-off is a good one because the resulting
ambiguity is greatly reduced by other regularities in the writing
system: For example, the pronunciation of
a
in
cake
is determined
by the presence of the final
e
, and the pronunciation of
a
in
car
is determined by the presence of the
r
and the absence of
the
e
, patterns that occur in many words. Thus, English is not as
irregular as is often implied. However, it is more complex than
many other alphabetic writing systems that adhere more closely
to the principle that each letter should be associated with a single
sound (Gelb, 1952; Hung & Tzeng, 1981).
The English writing system also exhibits a trade-off between
phonological explicitness and morphological transparency.
A fully explicit system would associate the letter
a
with a
single vowel phoneme, such as the /æ/ in
fat
, and use a different
symbol for the vowel in
fate
. The cost, however, is that this
would obscure facts about morphological relationships between
words. For example, the use of
a
to represent two different
phonemes in
nature
and
natural
may be confusing as a
guide to pronunciation, but it serves to remind the reader that
the two words are morphologically related. This trade-off occurs
repeatedly in English (1968).
In summary, there are two main problems associated with
understanding the alphabetic principle. First, phonemes are
perceptual abstractions, and second, alphabets sacrifice phonological
explicitness for symbol economy and morphological
transparency, thereby complicating the orthography. Because
of these problems, teaching methods that make the alphabetic
principle explicit result in greater success among children trying
to master the reading skill than methods that do not make it
explicit. We discuss evidence for this assertion later.
THE DEVELOPMENT OF READING SKILL
How does a child come to acquire reading skill? What are
the foundational competencies that reading builds upon? What
is the course of development of these competencies? This section
examines these questions. Before proceeding, we consider
some definitional issues. Considerable confusion has been created
by the fact that people mean different things when they refer
to reading. Confounding the problem is a genuine and
useful distinction between literacy and reading.
Literacy and Reading: Definitions
Each definition of reading can be defended on practical, logical,
or programmatic grounds, and each has its own set of entailments
that affect the framing of scientific and educational
issues. According to broad definitions, reading is understood as
a number of distinct literacy activities that have specific functions
(e.g., reading bus schedules, newspaper ads, tax forms, or
road signs; 1986). Narrow definitions focus on
the conversion of written forms into spoken language forms.
The most common definition has been a midlevel one: Reading
is getting meaning from print.
To see the value of the narrower definition, it is useful to
make a distinction between literacy and reading. Literacy includes
a variety of educational outcomes—dispositions toward
learning, interests in reading and writing, and knowledge of
subject-matter domains—that go beyond reading. These dimensions
of literacy entail the achievement of a broad range of
skills embedded in cultural and technological contexts. An extended
functional definition is useful in helping to make clear
the wide range of literacy tasks a society might present to its
members. For example, literacy may be defined as including
computer literacy, historical literacy, and scientific literacy,
among others. Such a functional definition takes literacy as referring
to a level of achievement, an extension of basic skill to
reasoning and discourse in a domain (1998).
However, the starting point for literacy is reading skill. Although
many children are engaged in written language at an
early age, schooling brings about specific expectations that all
children will develop the ability to read and learn from texts.
Our focus is on this necessary foundation. In this monograph,
we use the term reading to refer to the process of gaining
meaning from print. In focusing on reading’s distinguishing
features, we define learning to read as the acquisition of knowledge
that results in the child being able to identify and understand
printed words that he or she knows on the basis of spoken
language. Because words already known to a reader are sometimes
said to be represented in a mental lexicon or dictionary,
this learning process can also be described as a modification of
the mental lexicon such that it becomes
print addressable
. Put
in other terms, learning to read is learning how to use the conventional
forms of printed language to obtain meaning from
words. This definition separates learning to read from other aspects
of cognitive development. The distinguishing features of
reading center on the conventionalized, graphic input to the
reader and his or her conversion of that input into languageencoded
messages. This view implies that the child learning how
to read needs to learn how his or her writing system works.
How reading competence is achieved cannot be completely
separated from how reading is taught. However, evidence for
details of the course of reading acquisition in different instructional
settings is sparse. That means that research that informs
reading acquisition has to be considered at least partly independently
of instruction. We examine research on reading acquisition
after first considering what kinds of cognitive and language
competencies are typically in place as a child enters school.
A Developmental Perspective on Reading
Learning to read builds on cognitive, linguistic, and social
skills that have developed from the earliest age. The most important
among these is the child’s competence in language,
which provides the basic foundation for reading.
Language development prior to school
Well before the start of school, children have acquired extensive
knowledge of many aspects of language, including phonology,
grammar, word meaning, and pragmatics (the social
and communicative use of language). Although elements of each
of these subsystems continue to be acquired over long periods
(e.g., word meanings are acquired over the life span), children
normally acquire the basics of each subsystem by age 4.
The phonological system is especially important for learning
to read because, as we have observed, writing is a means of
representing speech. What are the child’s phonological abilities?
An important part of the answer to this question is that
whereas the basics of speech perception are acquired rapidly,
mental representations of abstract phonological structure undergo
further refinement well into the period when children begin
to be exposed to writing. Newborns can discriminate all of
the sounds (phonemes) that occur in spoken languages. Exposure
to the sounds of one’s native language, however, appears
to reduce this ability; by 12 months, infants readily discriminate
only the sounds of their native language (Werker & Lalonde,
1988). Note, however, that completely reliable discrimination
between words that differ by only a single speech segment may
not develop until the beginning of year 5 (1994). The
reason for this slowdown after the rapid phonological development
in the 1st year is unclear. One general possibility is that
children develop holistic strategies in word perception, relying
on prosodic and acoustic shapes more than segments
(1994). If so, further development is postponed, being renewed
when the child begins to sharpen phonological representations
by taking account of segments (phonemes). One possible reason
for this shift to segments is that the child’s lexicon becomes
larger, forcing finer discriminations (1985, 1992). Another
possibility is that increased speech production by the
child increases the demands for the child to represent speech in
terms of ordered segments (1986).
As in the case of phonology, knowledge of grammar develops
rapidly. The basic syntactic structures of the language are
learned by age 2 ( 1994;
1973; 1984). For example, children under 2 understand
that “Big Bird is tickling Cookie Monster” means
something different from “Cookie Monster is tickling Big
Bird”
( 1985, 1994). This understanding is
permitted by knowledge about how semantic notions (agent,
recipient of action) map onto syntactic structures (first noun
phrase, second noun phrase) in English. Grammatical knowledge
gets refined during the preschool years, and the child
comes to school equipped with fairly mature productive knowledge
of his or her language.
Reading also depends on a developing knowledge of word
meanings. Unlike grammar, which includes the productive machinery
of language, word meanings are concepts that must be
learned individually. Concepts and their lexical realizations are
continuously added to what the child already knows and refined
throughout development. The process starts early, with
children typically producing their first words prior to their first
birthday ( 1973). The comprehension of word meanings
appears even earlier (1987). A
dramatic increase in word knowledge (the
naming explosion
occurs during the 2nd year, typically coinciding with the
child’s first use of multiple-word phrases
( 1988). Although word meanings may not be part of the
grammatical system itself, grammar and vocabulary appear to
develop in tandem. Five- and 6-year-old children have vocabularies
of 2,500 to 5,000 words (1991). Of
course, vocabulary continues to grow after children enter school,
and it is estimated that elementary-school children learn about
7 words per day (1987). However, individual
differences in vocabulary related to reading ability and to demographics
are readily seen ( 1991). For
example,(1987) reported that first graders
from the upper socioeconomic status (SES) had about double
the vocabulary size of first graders from the lower SES. (See
also 1990.)
Finally, children acquire some understanding of the social
uses of language (pragmatics) throughout the preschool period
(1996). They learn about basic conversational
functions (e.g., turn taking) and conventional speech acts (e.g.,
requesting) that allow participation in a broad range of communicative
situations (1996).
Beyond the basics of the phonological, grammatical, semantic,
and pragmatic language subsystems (by which the child
comes to produce and understand language) are other developments
important for literacy and for the mature use of language
forms. These developments are usually summarized by the
phrase
metalinguistic awareness
—awareness of the aspects of
language just discussed, as opposed to the ability to use them.
Children make judgments, for example, about the correctness
of sentence forms, using their knowledge of grammar
(1984). Each of the linguistic subsystems—
the morphological and phonological as well as the grammatical—
is a potential area for metalinguistic awareness. Phonological
awareness, which we discuss in more detail later, is
especially important for learning to read alphabetic writing
systems.
One type of metalinguistic awareness that is outside the productive
components of language is also relevant for reading—
the concept of a word. Although young children quickly recognize
that things have names, the knowledge of single words as
basic units of meaning develops gradually over the preschool
period (1974;
1988). Many preschool children appear to confuse
the name with the object it refers to, referring to
snake
as a
long word and
caterpillar
as a short one, for example. The relevance
of this confusion for reading was demonstrated by
(1977), who found that preschool children
could not specify which of two printed words corresponded
to each of two spoken words that differed in length (see also
1978). Thus, awareness of words as spoken
forms is helpful but not sufficient for recognizing how such
words are realized in print.
In summary, the child comes to school with a well-developed
language system. Many elements undergo further development,
but the functioning language system the child already
has is sufficient to support reading.
Reading readiness and emergent literacy
Learning to read can be viewed from the context of other aspects
of development, too: There are additional skills on which
reading builds, and reading is a component of other developmental
progressions. These alternative perspectives are reflected
in two different approaches to preparing children to
read. The skills tradition has been to teach and assess
reading
readiness
skills in kindergarten (around age 5), as preparation
for reading instruction in the first grade (around age 6). The
prereading experience includes skills developed through exposure
to visual forms and oral language, as well as experiences
more directly related to reading (learning the alphabet).
A more recent alternative perspective takes a developmental
view of literacy development. It emphasizes a developmental
continuity between the cognitive tasks typical of the preschool
period and learning to read (1985). This view, known
as
emergent literacy
(1991), links the young child’s activities
around books to later opportunities for actual reading.
Young children are characterized as developing concepts about
the components of literacy, and their performance on various
literacy-like tasks is used to place them on a developmental
continuum. In this framework, the assumption is that reading
and writing are developmental phenomena. Thus, literacy is
characterized in developmental stages, with children’s ideas
about literacy being qualitatively different from those of literate
adults ( 1986). The central idea of emergent literacy
is that literacy emerges in various forms in development
before being transformed into conventional reading and writing.
This view has had considerable impact on teacher training
and classroom practice. Classrooms organized on emergent-literacy
principles emphasize a variety of communication opportunities—
oral reading by the teacher, idea sharing by children,
writing, and drawing—but with little emphasis on letter-sound
relationships.
The developmental perspective is important for bringing
into focus the accumulating knowledge that supports learning
to read and providing a reminder that children come to school
with varying amounts of knowledge about literacy activities.
Some children will have acquired some knowledge of how
written forms are mapped onto spoken language, but many will
not have been so fortunate. At some point, children who learn
to read must learn how their language is represented in the
writing system. This knowledge is not a natural end point of a
developmental progression; rather, it is usually the product of
instruction and practice.
A learning perspective on learning to read
Learning to read is one specific example of learning. According
to this perspective, the salient questions are (a) what is
it that children learn when they learn to read? and (b) how does
such learning come about? The answer to the first question is
roughly that a child comes to learn that the writing system encodes
his or her spoken language in a systematic way. The answer
to the second question is that the child must be taught or
else discover how this systematic encoding works.
Learning to Read an Alphabetic Writing System
For an alphabetic writing system, a child must learn that letters
and letter strings correspond to speech segments. The alphabetic
principle, the idea that written symbols are associated
with speech sounds, is the key design principle of alphabetic
writing and must be grasped by the child. Whether this knowledge
is acquired implicitly (through the extraction of printspeech
correspondences in text) or explicitly (through direct
instruction) varies among children.
Alphabetic writing systems differ in terms of
orthographic
depth
( 1987), or the consistency of the
mapping between letters and sounds. In a
shallow
orthography,
the mapping is highly consistent. Finnish, Italian, and Dutch
are all shallower than English because letters are more reliably
associated with particular phonemes. English is deeper because
the spelling-sound correspondences are more variable, as illustrated
earlier. In comparing different orthographies’ complexity
and ease of learning, one needs to consider not just depth,
but also many other features of orthographies and the languages
they represent, as well as trade-offs between orthographies
and languages ( 1992). For example, English
has many irregularities at the level of individual graphemes and
phonemes, but most of them occur in short, high-frequency
words, such as
have
,
was
, and
the
. Moreover, English is less inconsistent
at the level of the larger, subsyllabic unit called the
rime (a vowel plus syllable ending;
1995), and this unit is highly
salient in early reading ( 1993).
Good readers have a good grasp of spelling-sound correspondences,
as evidenced by their ability to sound out novel
words ( 1989). The extent to which learning the alphabetic
principle depends on explicit instruction is not clear.
Some evidence suggests that it may be difficult for children to
infer the principles of correspondence without instruction
(1991), whereas other research suggests that this is exactly
what successful readers do in the absence of direct instruction
(1996). Both
conclusions seem reasonable. Without direct instruction, learning
to read successfully is not guaranteed, but some children do
learn to read without such instruction. To the extent that learning
occurs, whether by direct instruction or implicit learning,
the problems posed by an inconsistent orthography are overcome
just as other inconsistent-mapping problems in human
learning are overcome—by practice. With sufficient effective
practice, children acquire a context-sensitive mapping of relations
between graphemes and phonemes (and larger units). So
-ow comes to be pronounced one way in the context n_ _ and
another way in the context l_ _. As we discuss next, initial success
in learning to read depends on the extent to which the
child has developed knowledge of individual speech sounds.
Phonological Awareness
The central problem in learning to read English is that although
discovering and applying the alphabetic principle is a
key to success, this is not an easy achievement for beginning
readers. Concerning this discovery, (1979) wrote, “If the
light were not so gradual in dawning, the relationship between
speech and print might count as one of the most remarkable
discoveries of childhood” (). As already noted, this is a
difficult learning problem because (a) phonemes are abstract
categories, (b) they have not been fully developed through the
use of speech by the time reading instruction begins, and (c)
there are inconsistencies in the mapping between spelling and
these units in deep alphabetic orthographies such as the one for
English.
The term phonological awareness (or phonemic awareness)
refers to children’s knowledge of the internal sound structure
of spoken words. There is a large literature examining how
children’s level of phonological awareness relates to their success
learning to read. Phonological awareness is assessed by
tasks such as deciding if two words rhyme or if they start or
end with the same sound. Not surprisingly, children who perform
well on such tasks do markedly better in early reading
than those who do not. And, conversely, children who score
poorly on phonological-awareness tests prior to entering school
are much more at risk for not learning to read effectively than
children who score well ( 1978, 1983).
The strong relationship between phonological awareness
and learning to read has been shown by numerous studies in
several languages ( 1988; 1989;
1976; 1980;
1984; 1988;
1987). Other studies have suggested that
instruction can bring gains in phonological awareness and, in
turn, in reading (1988;
1983; 1991;
1988; 1991;
1987; 1983; 1991).
Moreover, reading programs that emphasize phonologicalawareness
training have proved to be successful in classrooms
(1989;1999). Effective reading
instruction can help teach children what they need to know
about both the alphabetic principle and phonological awareness.
Furthermore, phonological training can remediate problems
for children who have not learned to read (
1989; 1997).
The relationship between knowledge of phonological structure
and ability to read is reciprocal. At the start of reading instruction,
children’s knowledge of phonological structure is partial:
Although they have begun to discover aspects of the internal
structure of spoken words, they typically have not converged
on explicit representations of phonemic segments. These partially
structured phonological representations are sufficient to
support the use of spoken language. Exposure to orthography
and explicit instruction in the mappings between spelling and
sound lead to further refinement of children’s phonological
representations, in the direction of more explicit representations
of segments and other units such as onsets and rimes.
Learning to spell also contributes to this process
(1992). These refinements in turn facilitate further
development of reading skill.
These observations suggest that the child’s development of
phonemic representations is more closely tied to reading than
to speech. No child ready to read has trouble hearing that bad
and pad are different forms with different meanings. Making
such distinctions does not require the use of phonemes; they
can be based on acoustic phonetic information (such as the difference
in voice onset time, the lag between the release of the
consonant and the onset of the vowel, that differentiates /b/
from /p/) to which even infants are sensitive, or on the basis of
whole syllabic representations. In fact relatively few preschool
children demonstrate an awareness of phonemes despite showing
awareness of syllables
( 1974). The alphabetic writing system both builds
upon and facilitates the development of phonemic representations.
It is in keeping with the alphabetic principle that a single
letter d, for example, is used to represent the category of
sounds called the phoneme /d/. Thus, the alphabetic principle
was a unique discovery in the evolution of writing systems
( 1952), and it is a discovery not made by all children on
their own.
Three types of evidence indicate that reading experience
plays a role in developing phonemic knowledge: (a) studies of
illiterate adults ( 1979), (b)
longitudinal studies of first graders ( 1987), and
(c) studies of Chinese readers (1986;
1986). For example, Morais et al. compared how illiterate
and recently literate Portuguese speakers from the same
community performed on phonological-awareness tasks. The
illiterate participants could not add or delete an initial consonant
from a spoken utterance, but the adults who had recently
become literate by attending adult education classes performed
the task successfully. The implication is that experience with
an alphabetic orthography may be necessary for an individual
to develop full phonological representations.
Experience with an alphabetic orthography also reduces the
impact of dialect variations on phonological awareness. Despite
regional dialects, speakers of American English share
knowledge of phonology, morphology, and semantics. For example,
some Americans no longer explicitly represent final
consonant clusters in speech (e.g., test is pronounced “tes”).
However, there is an underlying tacit awareness of these phonemes
in all speakers, whether they are literate or not. This
awareness is evidenced by their overt phonological representation
in inflected forms of a word. That is, /st/ is pronounced in
“testing” by a speaker who would drop these phonemes in
“test.” However, this tacit knowledge is not sufficient for illiterate
speakers to identify the final sound in “test” as /t/
(1979). Furthermore, for literate speakers, there is the awareness
that despite reduction of consonant clusters in speech, the
phonemes /st/ are explicitly represented orthographically in
spelling (test). Thus, experience with an alphabetic orthography
draws conscious attention to the underlying phonological
representation of words and prompts literate speakers to write
final consonant clusters even though they do not pronounce
them in speech.
Finally, we add a note about the relationship between intelligence
and phonological awareness. It is tempting for laypersons
and parents to assume that intelligence plays a critical role
in learning to read. In the extreme, of course, this must be true.
However, three lines of evidence indicate that in general it is
not. First, studies of children who learn to read early (prior to
entering school) indicate that there is not a strong relationship
between IQ and early reading (1973;
1966); although the mean IQ of early readers is above average,
the range of their IQ scores indicates that many children who
are early readers do not have unusually high IQs. Second, a
number of studies have found that IQ is only weakly and nonspecifically
related to reading achievement in the first and second
grades (1984). Finally,
children who have difficulty learning to read often have aboveaverage
IQs ( 1995). Hence, it is not surprising that
phonological awareness is the strongest predictor of early reading
skill (1990).
Learning to Read at the Beginning
Prior to the onset of direct instruction in grapheme-phoneme
correspondences, when children’s knowledge of phonological
structure is limited, their earliest attempts at reading are
revealing. (1993; 1991) demonstrated
that 5-year-old children, as a first pass at reading, associate features
of print with spoken word names, apparently without using
the orthography of the words. In one experiment, children
learned to recognize a word by use of a thumbprint placed on a
card containing the printed word. When the thumbprint was absent,
so was recognition. In another experiment, children were
found to associate selective parts of the printed word to the
spoken word. After being shown a short list of words, they
were presented with either the first letters or the end letters of
the words. Children who could identify the word based on its
first letters failed to identify it when presented with only the final
letters; children who could identify the word based on the
final letters failed to identify it when presented with its initial
letters. This suggests that attending to all the letters of a word is
not something that all children do at the beginning (see also
1988). study does not imply that 5-year-olds
cannot use letter forms and their associated speech forms. It
merely shows that in the absence of reading instruction and
knowledge of letter-sound correspondences, children can approach
a reading task by memorizing the visual images of
words, without learning how the sound-letter system works.
Moving to productive reading requires more than memorizing
printed words.
Theories of Learning to Read
Progress in learning to read has often been viewed as a series
of stages (1983; 1991; 1985;
1980; 1981).
The earliest stage can be characterized as attempts to learn associations
between visual features of graphic forms (not complete
orthographic word forms) and spoken words. A subsequent
stage of graphic-phonological decoding, in which children
learn the letter-sound associations, brings on a truly productive
capability enabling them to read words they have not seen
before. The use of letter names as a bridge to phonology is a
beginning step (1991). Alternative theoretical accounts
emphasize the incremental acquisition of individual word representations
rather than discrete stages (1992). In each
of these theories, phonology plays an important role in helping
the child establish word-specific orthographic representations,
a proposal that has come to be known as the bootstrapping hypothesis,
the idea that attempting to decode an unfamiliar word
is a form of self-teaching that allows the child to acquire an orthographic
representation for the word ( 1995).
Stage theories of reading development
A proposal by (, 1980;
1991; 1991) illustrates stage theories.
The first stage is a visual association stage, which is followed
by a second stage of decoding-based learning. In the first stage,
the child, absent any knowledge of decoding, uses any conceivable
source of information to discriminate one word from another.
In doing this, the child builds up a set of words that can
be recognized on the basis of partial visual cues (e.g., an initial
letter). Gough called this first stage selective association because
the basic learning mechanism establishes idiosyncratic
associations between some part of a printed word and the name
of the word. Under the right circumstances, including an increase
in phonological awareness and an intention to encode all
of the letters of the word, the child moves into the cipher stage
of true reading (1985, called this the alphabetic stage).
As the child reaches the limits of learning associations, there is
pressure to adopt a new procedure, one based on the alphabetic
principle. Early in reading, for example, a child can attend to
the m in mouse to distinguish mouse from cat and house. But as
moon and moose are encountered, the association cue that was
sufficient earlier becomes insufficient. This problem then fuels
progress in the child’s attending to more orthographic information.
An alternative view comes from (1980, 1991;
1985). In her theory, there is no purely visual stage (as
in account). Rather, children use letter names as cues
to word identification from their very first opportunity, as when
the letter j provides its name (“jay”) as a cue for reading the
word jail. Learning the alphabet, not necessarily the alphabetic
principle, is the key that moves a child into the first stage of
reading, called phonetic cue reading. In this stage, the child
reads by using some of the associations between the printed
letter forms and the phonetic cues of some of the letters (their
phonological associations). As in the selective association
stage identified by Gough, the child is reading primarily by using
incomplete, selective associations. But in Ehri’s account,
the associations are systematic and based on letter-sound correspondences.
The process of learning to read involves establishing
complete word representations that have both phonological
and orthographic components.
A nonstage incremental theory
The theories we have discussed so far assume that children
learn to read by progressing through a series of stages defined
by different types of decoding strategies. Other theories emphasize
the incremental nature of development
(1997). The basic idea is that
many types of knowledge are acquired gradually on the basis
of many experiences. What appear to be qualitative shifts in
strategy result from changes in the amount and complexity of
the information that has been acquired. Consider, for example,
the observation that children progress from an early logographic
stage (in which printed words are directly associated
with meanings and pronunciations) to an alphabetic stage (in
which they make use of knowledge concerning components of
words such as letters and phonemes). Research (see Using
Connectionist Models) has shown that this shift does not require
a change in strategy or in the hypothesis about the nature
of print. Rather, it can be accounted for by a process of gradual
learning based on many examples. In other words, progress to
an alphabetic stage can be viewed as a change in behavior: increased
sensitivity to the internal structure of words and the
correspondences between subword components and pronunciations
(as the alphabetic principle is discovered). Thus, whereas
stage theories provide qualitative characterizations of changes
in children’s performance, the connectionist theory attempts to
explain how these changes arise from more basic mechanisms.
In (1992) nonstage framework, learning to read
involves the acquisition of increasing numbers of word representations
that can be accessed by their spellings (quantity acquisition)
and changes in the specificity and redundancy (quality
dimensions) of individual words’ representations. As a child
learns to read, his or her representations of words increasingly
have specific letters in their correct positions (i.e., increased
specificity). Also, these representations become phonologically
redundant. The addition of specific grapheme-phoneme correspondences
for a word is redundant with the word-level pronunciation
of the word. Such redundancy assists word reading
by allowing both letter-level and word-level processes to produce
a word’s pronunciation. Together, increasing specificity
and redundancy allow high-quality word representations that
can be reliably activated by orthographic input. As individual
words become fully specified and redundant, they move from
what is called the functional lexicon, which consists of words
that can be read only with effort, to the autonomous lexicon,
which includes words that can be read with minimal effort.
This theory has been applied to explain individual differences
in reading skill (, in press).
All the theories we have discussed are compatible in many
respects and indeed share the fundamental assumption that
achieving reading skill requires use of the alphabetic principle.
This principle, effectively applied to print-sound connections
and supported by phonological sensitivity, is the critical factor
in early success in learning to read (1983;
1991; 1986;
1995; 1988).
Mechanisms of Progress in Learning to Read
Beyond these rudimentary beginnings, progress in learning
to read is fueled by several factors that center on increasingly
adaptive use of the alphabetic principle and the establishment
of orthographic patterns that are associated with pronunciations.
Successful learners apply their primitive understanding
of the alphabetic principle to their encounters with words,
which are turned into opportunities to apply and extend their
alphabetic knowledge. Studies by (1990;
1988) illustrate this point: The extent to which children
made phonological errors (i.e., misidentified a word as a similar-
sounding word) in word reading early in the first grade predicted
their end-of-year reading achievement. Nonphonological
errors, including errors that shared letters but put phonemes in
the wrong position (e.g., like for milk), were associated with
low end-of-year achievement. The point at which phonological
errors became more common than nonphonological errors coincided
with when the child attained
functional phonologicalskill
(i.e., knew at least half the alphabet and was successful inat least some tests of phonological sensitivity). Also, the level
of a child’s phonological sensitivity (awareness) corresponded
in some detail to the level of the child’s achievement in word
reading.
Truly productive reading, the ability to read novel words,
comes only from an increase in knowledge of how orthography
relates to phonology. This requires attention to letter strings
and the context-sensitive association of phoneme sequences to
these letter strings. This is where sensitivity to phonological
structure should play its most important role. Children who
have attained this productive level of reading can read pronounceable
nonwords, and their errors in word reading show a high
degree of phonological plausibility. These considerations, along
with demonstrations that success in learning is associated with
a phonological approach to reading (1988),
suggest that the main learning mechanism available to the child
is phonological recoding, recoding of spellings into pronunciations.
A model of how this mechanism works comes from
(1995), who emphasized the role of self-teaching in learning
to read words. An important focus in this model is children’s
attempts to phonologically recode words. One opportunity
to do this arises during reading aloud either to a parent or to a
teacher. The feedback from these attempts gradually builds up
the orthographic representation of specific words. The role of
phonology, in effect, is to influence the development of wordspecific
orthography. The letter-by-letter processing in sequential
decoding of words may be the main factor in producing
high-quality word representations that incorporate the letter
constituents of those words ( 1990; 1980;
1992; 1979). Several studies have
found that a few exposures to a word may be sufficient for the
child to acquire word-specific orthographic information
(1977; 1985; 1983), increasing the specificity
and redundancy of the child’s printed-word lexicon. Although
other mechanisms might promote the acquisition of printaccessible
word representations, phonological recoding is the
most effective mechanism (1995).
The importance of a phonological-recoding mechanism,
therefore, goes beyond its role in learning decoding rules. In
addition, the application, even the imperfect application, of this
mechanism helps the child learn specific word forms. Models
such as (1995) self-teaching model emphasize the child’s
acquisition of individual word representations, rather than stages
of development. Such models (1999;
1992; 1995) ask “which words can a child read?”
rather than “what stage is the child in?” The rapid buildup of
the child’s lexicon through reading promotes many words to a
functionally high-frequency status (i.e., they become familiar).
Texts that contain a high proportion of familiar words will be
read well, and the occasional low-frequency word provides an
opportunity for phonological self-teaching. Because the child
will face many low-frequency words over time, the phonological-
recoding mechanism is a very powerful, indeed essential,
mechanism throughout reading development, not merely for
beginners. Research has shown, for example, that third-grade
children who are skilled in reading can quickly and accurately
read a novel word that they have previously only heard; less
skilled readers tend to reach the same level of accuracy and fluency
reading these words only when they have previously actually
seen the words (1978).
An important fact about the acquisition of reading skill is
that it improves with practice. What is it that is improved?
Practice improves many components, but central among them
is knowledge of individual words. Experience in reading allows
the increasingly accurate representation of a word’s spelling
(its specificity), as well as a strengthening of the connection
between the phonological form and the spelling, and this specificity
increases the speed of word identification. Practice in
reading brings about an increasing facility with words because
it increases the quality of lexical representations. It turns lowfrequency
words into high-frequency words. The result is what
is commonly known as fluency in reading. Fluency entails developing
rapid and perhaps automatic word-identification processes
( 1974). The main mechanism for
gains in automaticity is, in some form or another, practice at
consistent input-output mappings (1977).
In reading, automaticity entails practice at retrieving word
forms and meanings (the output) from printed words (the input).
Automaticity is a characteristic of specific words, not
readers. Words move from the functional lexicon to the autonomous
lexicon as a result of practice reading text.
Experience not only builds automaticity, it also establishes
an important lexical-orthographic source of knowledge for
reading (1989). This lexical-orthographic
knowledge centers on increasing familiarity with the letters
that form the printed word. It is reflected in performance on
tasks that assess spelling knowledge, as opposed to those that
assess mainly phonological knowledge, and is indexed by the
amount of reading a person has done. Phonological and lexicalorthographic
abilities are correlated, but each makes a unique
contribution to reading achievement. The result is two complementary
but overlapping kinds of knowledge that support the
reading of words.
One benefit of reading practice is that it supports comprehension
ability, vocabulary growth, and spelling skill. Stanovich
( 1992; 1989)
measured college students’ reading experience (or print exposure)
and correlated it with measures of cognitive and reading
abilities. On the Author Recognition Test, a print-exposure
measure, readers are given a list of 80 names, 40 names of real
authors and 40 other names, and are asked to indicate which
are the names of authors. Correctly identifying the real authors
on this test is presumably an indicator of reading experience,
and greater reading experience, measured in this way, correlated
with better comprehension, spelling, and vocabulary
skills. Furthermore, print exposure accounts for variance in
word recognition and spelling that is not accounted for by phonological
processing in adults
( 1990; 1989) and children
(1991).
Notice that the Author Recognition Test allowed differentiation
of print exposure within a relatively homogeneous population
of college-age readers. Print exposure appears to be the
literacy equivalent of practice in skills like chess; just as practice
in chess separates grand masters from excellent tournament
players, practice at reading separates skilled readers from
less skilled readers in the college population. Thus, this research
is important in establishing that the amount of reading
makes an independent contribution to reading skill. This contribution
appears to be mediated not by phonological processes,
which readers must acquire anyway, but rather by the
more general facilitation that arises from accessing words repeatedly.
It is the dilemma of the less able reader that he or she
will not get as much practice as the more able reader. The gap
between more and less able readers thus increases with time
( 1986). Furthermore,
(1997) found that 1st-grade reading ability was a strong
predictor of 11th-grade reading ability and suggested that the
rapid acquisition of reading ability helps develop the lifetime
habit of reading. Thus, although it might be tempting to believe
that initial differences in reading ability wash out over time, the
data suggest just the opposite.
Spelling
Children’s initial expression of the alphabetic principle appears
more often in spelling than in reading (1985). Indeed,
children’s attempts at spelling prior to formal reading
instruction typically reveal an understanding of basic lettersound
associations (1970; 1971). In producing
spellings, preliterate children often use the sounds associated
with the names of the letters, spelling car as “KR,” for
example. In effect, this spelling uses the name of the letter R (/ar/)
to capture both vowel and consonant. As
(1997) pointed out, the tendency to use letter names in spelling
is affected by phonological properties of the letter names. R is
more likely to be used for /ar/ than T is for /ti/ or L for /l/ because
it is harder to segment the /r/ from its preceding vowel
than to segment the /t/ and /l/ from the vowels following and
preceding them. These observations have two important implications:
findings about R, L, and T reflect
the difficulty of reliably segmenting syllables into phonemes
and reinforce the conclusion that full awareness of phonemes is
difficult to achieve prior to literacy. But the broader implication
is that one underestimates the child’s potential grasp of the alphabetic
principle—or at least the idea that speech sounds are
associated with letters—if one considers only decoding. Spelling
is the primary early indicator of this potential and can form
the basis for later expression of the alphabetic principle in decoding.
It is also clear, however, that early spellings are guided by
more than the child’s attempts to map sounds onto letters.
(1993;1997) has shown that at an
early age children become sensitive to both the orthographic
structures and the morphological structures that are present in
spellings. Even in first grade, children’s classroom spellings reflect
a number of orthographic conventions, including, for example,
the fact that ck never occurs as a spelling of /k/ at the
beginning of a word, but does occur following a vowel. First
graders show their sensitivity to morphology when they refuse
to spell dirty as “dirdy,” even though the /d/ sound is what is
heard. The conventional spelling honors the fact that dirty contains
the lexical morpheme dirt plus a suffix. Children are more
likely to substitute a d for a t in a single-morpheme word such
as duty (1994). At this age, children’s
use of morphology is very incomplete, and many of their
errors reflect a preference for spelling a sound over spelling a
morpheme. But clearly, at least as soon as literacy is under
way, children begin to show an awareness that conventional
spellings honor both phonological and morphological structure,
as well as conventional orthographic constraints.
Eventually, in many languages, the learner confronts an important
fact about spelling: Typically, the mapping from pronunciation
to spelling is less consistent than the mapping from
spelling to pronunciation. Reading is more reliable than spelling.
An important idea about the relation between spelling and
reading comes from recent research on word identification. The
more ways a sequence of phonemes can be spelled, the longer
it takes to read a word containing that sequence. For example,
shelf is more efficiently read than sneer because its rime unit
/lf/ is always spelled elf, whereas the rime unit /ir/ is spelled
variously as eer, ear, ier, or ere. Notice that this is not a question
of consistency in the direction of orthography to phonology:
eer is always pronounced /ir/.
(1997) reported the first demonstration of this backward consistency
effect (i.e., effect of phonology on orthography). Although
the reliability of this phenomenon is a focus of research
(1998), it is interesting to note
the obvious implication it would have for reading and spelling.
If reading words really includes a feedback mechanism from
phonology to orthography, then the reader would not merely
convert a written input into a phonological representation but
would also, in effect, verify (rapidly and unconsciously, to be
sure) that the phonological representation could be spelled in
the way presented. More generally, the hypothesized feedback
mechanism illustrates one way that spelling and reading may
be intimately related.
The relationship between reading and spelling is represented
in Figure 3, which presents a system in which lexical
representations include information needed for both spelling
and reading (both letter and phoneme strings). Some identities
(letters or phonemes) can be missing or variable or incorrectly
specified, especially for children. Reading can be successful
with an underspecified representation, but spelling cannot.
It has been observed that children can sometimes spell
words that they cannot read (1980). However,
this phenomenon is no more frequent than that of failing
to read a word after having read it successfully once before, or
failing to spell a word after successfully spelling it once before
(1992). Thus, variability in performance,
which is presumably due to an unreliably specified
mental representation, characterizes both reading and spelling
and is consistent with the assumption that reading and spelling
have a common representation system.
It is the case, of course, that spelling is more difficult than
reading, and indeed many skilled readers identify themselves
as “terrible spellers.” Spelling is typically more difficult than
reading not because they use different representations, but because
spelling requires additional processes that benefit from
specific practice at spelling. Because spelling requires production,
whereas reading requires recognition, success in reading
does not lead automatically to success in spelling. Interestingly,
this is true even among children who are learning to read
a shallow orthography through phonics-based teaching, as is
the case in the Netherlands (1997). Children’s
spelling ability benefits from instruction that specifically
targets the production of conventional spellings.
Comprehension
It can be reasonably argued that learning to read enables a
person to comprehend written language to the same level that
he or she comprehends spoken language. Thus, reading comprehension
is not so much an issue of reading as it is an issue of
general language comprehension. However, the reality is that
assessments of reading skill are concerned with reading comprehension
and that this concern reflects the expectation that
children should understand what they read. Thus, although we
focus more in this monograph on word reading than on comprehension,
it is important to emphasize the importance of
comprehension as a part of a more complete picture of reading
skill.
There are several important points to emphasize in considering
comprehension. First, as we just pointed out, comprehension
is a matter of language understanding, not a unique feature
of reading. Thus, the acquisition of reading comprehension
skill includes two highly general components: the application
of nonlinguistic (conceptual) knowledge and the application of
general language comprehension skills to written texts. An important
question for instruction is the extent to which either of
these applications needs to be targeted in the classroom. The
research literature is clear in showing profound effects of specific
conceptual knowledge on the comprehension of texts
(1983). Knowledge is a matter of general education,
inside and outside the classroom, and has little specific
claim on reading, however. Its contribution, as important as it is
to every comprehension event, is not an intrinsic component of
reading, which includes mechanisms that can compensate to
some extent for limited knowledge (1985).
Two conclusions about the role of comprehension in learning
to read seem warranted. First, comprehension is critical as
part of the acquisition of reading skill. Second, much of what is
important about comprehension is highly general to language,
and not unique to reading. Evidence for this comes from the
high correlations (in the range of r .9) observed between
written and spoken language comprehension among adults
( 1994; 1990;
1990). These correlations are lower for children
and increase with age ( 1980; 1984), as
would be expected if general language-comprehension skills
show their importance as basic literacy skills are mastered.
If written and spoken language comprehension go together,
what about children who can read words but whose reading
comprehension is not as good as their spoken language comprehension?
The frequency of such cases may be exaggerated
by the anecdotal impressions of teachers who have not had the
luxury of assessing carefully both the comprehension (spoken
and written) and word-identification skills of such children
( 1985). There is surprisingly little convincing documentation
of pure reading comprehension deficits accompanied
by high levels of both word-identification and listening
comprehension skill. Some research ( 1996;
1991) suggests that some children have better
decoding than reading comprehension skills. However, these
children appear not to have specific reading problems, but
rather to have general comprehension problems associated with
both spoken and written language ( 1996).
It is important to note that spoken language skills, acquired
in conversation and play, may not transfer to reading comprehension
for typical written texts. Indeed, there are differences
between spoken and written language that should lead to processing
differences (D1977). This would mean that
skills developed in oral settings would not transfer to spoken
versions of written texts either. The high correlations between
spoken and written language comprehension have been obtained
in studies that use a single set of materials that are pre-
sented in two different modalities. These studies have not
correlated spoken comprehension of typical spoken language
with written comprehension of typical written language. Accordingly,
the high correlations partly reflect the effects of specific
text material that is more typical of written than spoken
language. The conclusions that reading comprehension depends
on spoken comprehension may need to be qualified a bit.
It may be more correct to say that the potential for comprehending
a written text is set by the ability to comprehend that
same text when it is spoken. The typical differences between
the uses of spoken and written language suggest that special efforts
directed toward the distinctive aspects of written text
comprehension may be warranted as part of reading instruction,
especially as written texts become more complex later in
elementary school.
Implications for Teaching Reading
The instructional goals for alphabetic systems seem clear:
Children need to learn that the letters of their alphabet map
onto speech segments of their language. However, controversies
over how to teach reading abound. Rather than focusing on
letter-sound correspondences, the dominant instructional approaches
for the past 20 years were meaning focused, built
around story reading, exposure to print, and enhanced language
environments. Nevertheless, for more than 30 years, research
has supported the effectiveness of methods that are based on
direct instruction in phonics or decoding. In a major review of
research on reading acquisition titled
Learning to Read: TheGreat Debate
, (1967) concluded that the majority of theresearch tended to favor phonics instruction, and her comprehensive
study has been confirmed by more recent studies
(1990;
1998). However, the practice of reading instruction has
remained out of touch with research, emphasizing a variety of
language activities but minimizing the teaching of graphemephoneme
relationships. Much of the debate has been fueled
by philosophical stances and professional advocacies that
have little to do with the research basis for effective teaching
( 2000). However, there are recent indications that
a stronger consensus is emerging in favor of research-guided
practice.
Summary
Most children come to reading instruction with normally
developed language abilities, which can provide the foundation
for learning to read. Those children with less developed language
abilities still benefit from phonological-awareness instruction.
The central achievement of early reading is learning
to read words, which requires knowledge of the phonological
structures of language and how the written units connect with
the spoken units. Phonological sensitivity at the subword level
is important in this achievement but is not provided to all children
by the spoken language in their environment. Phonological
training helps the acquisition of reading. Very early, children
who will turn out to be successful in learning to read use phonological
connections to letters, establishing decoding as a mechanism
for productive reading, the ability to read previously
unencountered words. An important mechanism for this is phonological
recoding, which helps the child acquire high-quality
word representations. Gains in fluency (automaticity) come
with increased experience, as does increased lexical knowledge,
which supports word identification.
DEVELOPMENTAL DYSLEXIA
Individual differences in reading achievement are often due
to differences in the ability to read words, and, indeed, children
and adults display a wide range of ability to read words. When
reading skill is sufficiently low relative to certain standards, the
individual has a disability called dyslexia. In discussing dyslexia,
we need to distinguish between acquired dyslexia and
developmental dyslexia. Individuals who have acquired dyslexia
were previously able to read quite fluently, but because of
some type of brain injury (resulting from head injury, stroke, or
degenerative neuropathology, such as Alzheimer’s disease),
they can no longer read efficiently. There are several patterns of
acquired dyslexia (2000), and we make reference
to some of them here to support certain arguments.
Our focus, however, is on developmental dyslexia, which
varies along a continuum from mild to severe. The term has traditionally
been reserved to apply specifically to children who
have normal intelligence and do not exhibit frank sensory or
neurological impairments (1970). More generally,
the dyslexia label is typically restricted to individuals whose
reading levels are discrepant from the potential implied by their
IQs. However, children who have reading disability appear to
present the same reading problems whether or not they meet
the discrepancy criterion (1994). In either
case, these children have not mastered the task of learning to
read efficiently, and intervention by the end of the second grade
is often critical for their successful reading development
(1995). Their reading difficulties must be understood in
terms of underlying cognitive processes that can go wrong. A
signature problem for all struggling readers is their poor skill at
reading words and pseudowords (nonwords, like mard, that are
pronounceable and could be words because they conform to
English spelling rules). These problems, as we discuss later in
this section, are often traceable to problems in phonological
processing.
Contemporary research suggests that developmental dyslexia
is caused largely by language-related deficits, whose neurological
bases are beginning to be identified
(2000). Although these deficits sometimes affect other aspects
of behavior, they have a particularly large impact on reading.
We now describe some of the suspected causes of developmental
dyslexia.
The Phonological-Deficit Hypothesis
Given the extensive evidence indicating the importance of
phonological information in reading acquisition, it is not surprising
that the leading hypothesis about the cause of developmental
dyslexia is that it is due to deficits in the representation
and use of this information ( 2000;
1994). Phonological information plays important roles in
reading familiar words and sounding out new ones; phonologically
based working memory capacities are used in integrating
and comprehending words in sentences
(1989 1983;
1992). There is extensive evidence that these functions
are impaired in many dyslexics ( 1999;2000)
Although there is broad agreement that dyslexia is typically
associated with impaired use of phonology, the basis of the
deficit is less clear. One hypothesis is that the deficit is secondary
to subtle impairments in the processing of auditory information
(1980); another is that the deficit is limited to the
processing of speech (1997).
The basic idea is that speech perception involves processing a
complex, rapidly changing auditory signal. Impairments in the
capacity to process this information (because of either a
speech-specific or a general auditory deficit) would be expected
to interfere with the development of the phonological
representations that are critical to reading. The auditory-processing
hypothesis gains support from studies that have identified
this type of deficit in some dyslexics (1989); the
speech-processing hypothesis gains support from studies in
which dyslexics demonstrated impaired processing of speech
but not other auditory information (1997).
It is important to note, however, that many developmental
dyslexics perform normally on tests of both auditory and speech
processing. For example, there have been numerous studies of
the perception of consonants in nondyslexic and dyslexic individuals
( 1990; 1999;, 1987).
These studies involved assessing listeners’ capacity to perceive
the differences between stimuli such as /pæt/ and /bæt/. The
studies have found that dyslexic children whose performance
on measures of phonological knowledge is impaired exhibit
normal categorical perception of phonemes
( 2000). These data point to an information
processing deficit that has a significant, debilitating effect
on learning to read but little impact on speech perception or
other uses of language. As we have noted, segmental phonological
information is critical for reading in alphabetic orthographies;
it develops in part through exposure to print, and may be
less crucial to the perception of spoken language. Thus, the effects
of a mild information processing deficit might be manifested
only in the reading context (H1999).
In contrast, deficits in speech perception and auditory processing
have been reliably observed in a subset of dyslexics
who exhibit impairments in the use of spoken language; these
children are often said to have developmental language impairment,
or developmental dysphasia (2000;
1980; 1987). Whether dyslexia that involves a
phonological deficit and developmental language impairment
derive from a common underlying impairment varying in severity
is also the focus of ongoing research (1997).
Other Possible Causes
Reading is a complex task involving several cognitive capacities.
Impairments in any of these capacities could in principle
interfere with reading. Moreover, there could be variability among
dyslexics with respect to etiology; although phonological deficits
are very prominent in dyslexics, there could be subgroups
with other deficits (or multiple deficits). Two possibilities have
been the focus of considerable research.
Visual impairments
One hypothesis is that some manifestations of dyslexia are
secondary to impairments in the processing of visual information.
Indeed, visual processing was the central feature of dyslexia
described by (1925), who hypothesized that failure
to achieve normal hemispheric dominance causes dyslexics to
confuse letters such as b and d. Explaining dyslexia as a visual
deficit has been the standard lay theory of dyslexia, despite a
relative lack of evidence (1979).
Interest in a visual explanation has been revived by studies
more closely tied to current theories of vision. Several studies
have reported that some dyslexics exhibit impairments in visual
processing that are functions of the dorsal visual pathway
in the brain, the where system
(1982). This system is dominated by input from magnocellular
cells in the lateral geniculate nucleus, and plays an important
role in the detection of transient visual information (e.g., rapid
changes in illumination, motion detection). (1996)
reported a brain-imaging study in which a small sample of dyslexics
was found to be impaired in judging the relative velocities
of visual stimuli; they also exhibited abnormal activation in
area V5/MT, part of the magnocellular system. However, the
overall evidence for a fundamental deficit in the visual system
remains in doubt. Other careful studies have failed to observe
the hypothesized impairment
(1996). Nor is it known how the visual deficit relates to the phonological
impairment identified in many other studies. The
dyslexics in the study by were also impaired on a
nonword reading task, an indication of an impairment in phonological
processing. (1999) reviewed other
evidence for the magnocellular deficit hypothesis, which is the
focus of considerable controversy (see chapters in Willows,
1993). It is fair to conclude that although further
research may confirm the existence of a type of dyslexia
that arises from magnocellular processing deficits, the current
evidence suggests that visual impairments by themselves are a
very small part of dyslexia (1999)
Developmental delay
Children with phonological dyslexia exhibit a characteristic
atypical developmental pattern: They read significantly below
grade level, their knowledge of phonological structure is poor,
and they have difficulty sounding out unfamiliar words. Some
research has identified a smaller group of dyslexics who do not
fit this pattern (1993; 1996;1994;1997)
They also read significantly
below grade level, and their phonological knowledge
is also below grade level, but their levels of reading achievement
and phonological knowledge are more closely matched
than among children in the larger group. Unlike phonological
dyslexics, their performance is very similar to that of younger
children learning to read normally. Thus, they exhibit a general
developmental delay in reading acquisition, rather than an
overtly aberrant developmental trajectory. This delay may arise
from several factors, including environmental ones (e.g., insufficient
reading experience, ineffective teaching methods). However,
in cases in which these environmental factors can be ruled
out, several endogenous factors may produce this pattern. Such
factors may include a general learning impairment, a limitation
on the brain resources that are recruited for reading, or an impairment
that affects the representation of orthographic rather
than phonological structure ( 1999;
1996;2000). The possible contributions of
these factors to what is likely to be a heterogeneous subgroup of
dyslexics are not as yet well understood, however.
Summary
A considerable body of evidence supports the phonologicaldeficit
hypothesis, allowing research to focus on specific questions
about the nature of this deficit and ways it can be remediated.
At the same time, there is growing evidence that dyslexia
may have multiple causes, which may require different types of
intervention, and that the causes may have different effects in the
case of different writing systems. Although research has identified
a narrow range of possible causes, basic questions about
their frequencies of occurrence and co-occurrence, how they affect
reading, and whether they give rise to different patterns of
behavioral impairment remain to be determined. Answering
these questions is essential to developing effective procedures
for identifying dyslexic children and remediating their problems.
Given the research, it appears that three main elements are necessary
for reading intervention with dyslexics: (a) phonologicalawareness
training, (b) systematic phonics instruction that is
linked to spelling, and (c) oral reading practice with decodable
texts (i.e., texts that include only words using the accumulating
set of letter-sound correspondences that have been taught).
SKILLED READING
Although our focus is on learning to read, we now examine
the end point of learning to read, skilled reading. We begin by
discussing research from cognitive psychology and then move
to recent findings from cognitive neuroscience that are relevant
for understanding reading. One emergent trend in cognitive
psychology is the development of formal models implemented
as computer programs that simulate behavior. In the present
section, we discuss some such models that address various aspects
of reading.
The View From Cognitive Psychology
Within cognitive psychology, there has been widespread interest
in the processes associated with skilled reading. Cognitive
psychologists have been extremely interested in how
words are recognized, how eye movements are controlled in
reading, how text is comprehended, and a variety of other issues.
In 1879, established the first laboratory
of experimental psychology, in Germany, and this event resulted
in considerable interest in language processing
(1970). One of the topics that received particular
attention in the early days of experimental psychology was
reading. This early interest reached its peak with the publication
of (1908) classic work
The Psychology and Pedagogyof Reading
. The kinds of questions asked by Huey and hiscontemporaries, and covered in his book, parallel the kinds of
questions addressed in modern textbooks on the psychology of
reading ( 1982; 1992;
1987; 1985; 1989).
was keenly interested in the mental activities involved in
reading. However, with the onset of the behaviorist revolution
in 1913, research on mental processes in reading ceased. A
number of psychologists who had previously been interested in
cognitive processes during reading turned to the burgeoning
field of test development, and worked to develop tests to measure
reading ability.
The reemergence of cognitive psychology in the mid-1960s
resulted in widespread interest in reading once again. In the
early 1970s, reading was viewed as a psycholinguistic guessing
game (1970) or a hypothesis-testing activity
( 1970), and tenets such as “reading is only incidentally
visual” were espoused. According to this view, readers engaged
in a cycle of activity in which they generated a
hypothesis about what the next word would be, moved their
eyes to that word, quickly confirmed their hypothesis, and then
generated a hypothesis about what the next word would be.
Obviously, processing was viewed as being largely contextually
driven. Also, this approach suggested that there was a bottleneck
during skilled reading at the stage of getting visual
information into the processing system. A large amount of research
on skilled reading has led to the replacement of this hypothesis-
testing view by one in which the processing activities
involved in reading occur very rapidly, so that the information
needed for reading gets into the processing system very
quickly. Thus, there is no bottleneck at the visual input stage.
Furthermore, although context has an important effect on inter
PSYCHOLOGICALpreting meaning, skilled readers identify words quickly with
little help from context. It is readers of lower skill who rely on
context to support word identification
(1979;1980). Two lines of research have
been very influential in shaping current views about skilled
reading: research on eye movements during reading and on
word identification. We discuss these two topics, and then issues
related to comprehension.
Eye movements in reading
Although people have the phenomenological impression
that when they read their eyes glide smoothly across the page,
this impression is an illusion. In fact, the eyes make rapid
movements separated by periods of time when they are relatively
still (called fixations). These fixations typically last about
200 to 250 ms, and it is during these periods that people acquire
information from the text. During the actual movements
of the eyes (called saccades, which typically cover about eight
to nine letter spaces and last about 20–40 ms), vision is suppressed
so that no new information is acquired. Reading is thus
like a slide show in which the text is on for about a quarter of a
second and then off for a brief period of time while the eyes
move. In addition to making forward-moving saccades, skilled
readers move their eyes backward in the text to reread material
about 10 to 15% of the time; these regressions are often driven
by breakdowns in the comprehension process.
The reason readers move their eyes so frequently has to do
with acuity limitations in the visual system. Because acuity is
best in the center of vision (the fovea), people move their eyes
so as to place the text they want to process on the fovea. Outside
the fovea, acuity drops off markedly in parafoveal and peripheral
vision, where the anatomical receptors are not able to
discriminate the fine details of the letters making up the words.
Some classic research ( 1975;
1975; 1979) using an
eye-contingent display-change
paradigm, in which readers’ eye movements weremonitored by a highly accurate eye-tracking system and changes
in the text were made contingent upon where the reader was
looking (see Fig. 4), demonstrated that the perceptual span (or
region from which useful information is acquired) is restricted
to an area extending from 3 or 4 letter spaces to the left of fixation
to 14 or 15 letter spaces to the right of fixation for readers
of English. Information used to identify words is typically restricted
to no more than 7 to 8 letter spaces to the right of fixation,
with more gross information (such as the length of upcoming
words) acquired out to 15 letter spaces.
The consequence of these acuity and perceptual-span limitations
is that skilled readers fixate on about two thirds of the
words in text; the words that are typically skipped are short
words and words that are highly predictable from the preceding
context. But even though words are not directly fixated, there is
evidence that they have been processed (as fixations preceding
and following skips have inflated durations). Two other facts
are highly relevant for the current discussion
(1998). First, the information needed for reading
gets into the processing system very quickly. As long as the
text is available for 50 to 60 ms before a masking pattern appears
to obliterate it, reading proceeds quite normally. Second,
although readers are not consciously aware of their eye movements,
how long the eyes remain fixated on a word is very
much influenced by the ease or difficulty of understanding that
word. Thus, for example, low-frequency words are fixated
longer than high-frequency words. The conclusion that follows
from all of this evidence is that readers are not engaging in all
sorts of guessing activities, but rather are efficiently and
quickly (at an unconscious level) processing the text. Indeed,
all the letters in a word are being processed during word identification,
as we discuss in the next section.
We conclude this section with three final observations. First,
beginning readers’ eye movements are quite different from
those of skilled readers (1986). Beginning readers fixate
virtually every word (and make more than one fixation on
many words). Thus, their saccades are much shorter (around
three letter spaces) than skilled readers’. Furthermore, their average
fixation durations are much longer (between 300 and 400
ms) than skilled readers’, and they regress much more frequently
(so that up to 50% of their eye movements are regressions).
Their perceptual span is also smaller than that of skilled
readers (see Fig. 5). Basically, their eye movements reflect the
difficulty they have encoding the words in text. Second, research
on skilled readers shows quite clearly that phonological
codes are activated for words very early in eye fixations
( 1992;1995).
Finally, there are now computer simulation models
(1998) that do a very good job of
predicting where readers fixate and how long they fixate.
Word identification
An important issue with respect to how words are read deals
with whether they are processed as wholes (in parallel) or letter
by letter (serially). More than 100 years ago, (1886) addressed
this issue by asking people to report what they saw
when words and letters were briefly exposed. In fact, they were
better able to report words than letters. These results were used
by educational reformers to advocate whole-word teaching
methods. However, when (1969) and (1970)
replicated this finding with an improved experimental design,
the results did not support whole-word instruction. The characteristics
of their paradigm are shown in Figure 6. Basically, a
word, single letter, or nonword letter string was presented very
briefly (about 25–40 ms) and followed immediately by a masking
pattern that would interfere with any extended processing
of the stimulus after its offset. In addition, two letter choices
were presented: One was the correct letter (in the word and
nonword conditions, the letter that had been previously shown
in the position indicated), and the other was a letter that had not
been presented. Notice, however, that in the word condition,
this incorrect alternative, if substituted in the word in the position
indicated, always resulted in a different word (e.g., in Fig.
6, the incorrect alternative, k, also made a word if substituted
for d). Reicher and Wheeler found that responses were more
often correct in the word condition than in the other conditions.
Thus, a letter is better identified when it is embedded in a word
than when it is presented in isolation or in a nonword. This result,
called the word-superiority effect, suggests that
phenomenon is real: Letters in words are identified more accurately
than letters in isolation.
The phenomenon forces two conclusions. First, the serial,
letter-by-letter view of word recognition cannot be correct: If
this view were correct, readers would have been more accurate
with single letters than words given the limited processing
time. Second, all the letters in the word must have been processed,
as readers were able to identify correct letters at all positions
with equal accuracy. This latter point has implications
for reading instruction because it indicates that all the letters in
a word need to be processed in order for the reader to build a
specific representation of the word that can be accessed quickly
and accurately. Although memorizing the shapes of words at
the beginning of reading may get children started reading, it is
not enough for reading development to continue(1983)
In general, the word-superiority effect has been taken as evidence
that all of the letters in short words are processed in parallel.
The effect ultimately led to the development of powerful
computer simulation models designed to account for the results.
These initial models, called the interactive activation
model ( 1981; 1982)
and the verification model
( 1982), were the forerunners of even
more elegant and powerful connectionist models
How Psychological Science Informs the Teaching of Reading
( 1996; 1989)
that we discuss in Using Connectionist Models
to Understand Reading and Dyslexia.
A second general issue regarding word identification is the
extent to which phonological (or speech-based) codes are involved
in identifying words. In principle, the meanings of
words could be identified in two ways: either directly from
print (traditionally termed direct access) or by computing a letter
string’s phonological code and using that information to access
meaning (phonologically mediated access). There has
been a long-standing debate about which of these is more efficient.
(1973) forcefully argued that direct access is necessarily
more efficient because it does not require the extra
phonological-recoding step and because English has numerous
irregularly pronounced words (have, pint, etc.). His observations
had considerable impact on educational practice, lending
support to whole-word (and later, whole-language) approaches.
The problem with direct access, however, is that, unlike the
mapping between spelling and sound, the mapping between
spelling and meaning is largely arbitrary
(1990) and thus difficult to learn. Moreover, consider
what happens when a beginning reader encounters an
unfamiliar word. The direct-access mechanism cannot operate
because the letter string has not been encountered before and
thus the association between form and meaning has not been
established. However, if the child can phonologically recode
the letter string (“sound it out”), it can be matched to knowledge
of the word derived from spoken language. Thus, phonological
recoding provides a basis for generalization, as well as
an important self-teaching mechanism ( 1983).
The debate about the extent to which word reading is phonologically
mediated clearly parallels (and indeed contributed to)
the debate concerning the extent to which reading instruction
should emphasize phonics.
Most contemporary theories of word reading assume that
both direct and phonologically mediated mechanisms are available
to skilled readers ), but they make different assumptions
about the division of labor between them
(1998; 1995). Some theories assume the primacy of
the direct pathway (, 1978), others the primacy of
phonological recoding 1995;
1990). In (1991) influential model, processing
is attempted in the two pathways in parallel, with a race
between them. Which pathway “wins” for any given word depends
on word factors (such as familiarity and spelling-sound
consistency), reading skill, and the nature of the writing system
(how directly and consistently it represents phonological information).
In s (2001) recent model, both
pathways jointly determine the activation of meaning (see Using
Connectionist Models).
There is now a large body of evidence that phonological information
plays an important role in word reading, even among
highly skilled readers ( 1998) and for nonalphabetic writing
systems such as Chinese (, 1995). We have
already noted that eye movement experiments have demonstrated
that phonological codes are activated very early in an
eye fixation. Other compelling evidence comes from studies by
(1987; 1988), who
devised a clever way to diagnose the activation of phonology.
Subjects were presented with a question (e.g., “Is it a flower?”)
and then had to read a target word (e.g., rose) and decide if it is
a member of the designated category. On critical trials, the target
was a homophone of a category exemplar (e.g., rows). On a
significant number of such trials, subjects incorrectly identified
the word as a member of the category. Moreover, such false
positive responses also occurred for nonword targets that
sounded like words (e.g., article of clothing: sute). These responses
would not have occurred unless subjects had phonologically
recoded the letter strings. Similar results have been
obtained in studies of several other writing systems
(1998). Thus, phonological recoding plays a much more prominent
role in skilled reading than (1973) asserted. This is
among the most important findings in contemporary research
on reading, and it strongly suggests the achievement of reading
skill depends in part on learning to use phonological information
efficiently.
Reading comprehension
Whether reading uses just those processes that serve spoken
language or requires something more turns out to be a difficult
question. Reading comprehension clearly depends on spoken
language comprehension. Correlations between spoken language
and reading comprehension are modest early in learning
to read (1980) but, as noted earlier, increase with the
development of reading skill (1984), reaching
.90 for adult samples ( 1990). Thus, reading comprehension
has little variance unique to reading for adults, but
shows both variance shared with listening comprehension and
variance unique to reading among children. This pattern of correlations
indicates that reading comprehension skill approaches
listening comprehension skill as printed word identification is
mastered.
Reading comprehension involves a number of different interacting
processes that have been the target of much research.
Some of this research has emphasized the guidance given to
comprehension by higher-level sources of knowledge. For example,
readers are sensitive to the causal relations among characters’
actions as they read a story
(1985). They may also use a wide range of knowledge outside
the text to guide comprehension in other ways, including by
making inferences ( 1994). It is
clear that readers do more than merely comprehend sentences
when they read. However, some general models of comprehension
describe in detail how the comprehension of text can be
built up over the reading of successive words and sentences,
from the bottom up (1988, 1998;
1992; 1998). Thus, comprehension includes
guidance from knowledge outside the text, but even this influence
can be understood in part by how the meanings of words
actually read in the sentence trigger such knowledge. Even
reading comprehension begins with the word.
Accordingly, accessing word meaning in context, parsing
sentences, and drawing inferences are all part of the overall
process of comprehension. In addition, the skilled reader monitors
his or her comprehension to verify that the text is making
sense. When readers have problems in comprehension, the
source might be any of these processes. Indeed, the interdependence
among the components of comprehension means that
multiple problems are likely to be observed in such cases and
that finding a single cause of comprehension failure, as a general
case, is unlikely. For example, because the higher levels of
processing rely on output from lower levels, an observed problem
in text comprehension can also result from lower-level processes,
including word identification, basic language processes,
and processing limitations.
The interplay among these lower-level factors can be complex.
For example, differences in basic language skill lead to
individual differences in comprehension, and less skilled readers
show a wide range of problems with syntax. The question is
whether such problems arise from a syntactic-processing deficit
or from some other source that affects performance on syntactic
tasks (such as working memory limitations, lack of
practice, or lexical processing limitations). Research with children
(1988) and adults
(1994) suggests that syntactic parsing problems can
arise from processing limitations rather than from problems
with syntax itself. Comprehension difficulties may be localized
at points of high processing demands—whether from syntax or
something else. If this analysis is correct, then the problem is
not intrinsic deficits in syntax, but the processing capacity to
handle complexity. More generally, the hypothesis that working
memory factors produce individual differences in comprehension
has received wide support over years of research
(1979; 1992;
1977; 1986). It is clear that among children
as well as adults, working memory factors constrain comprehension
of both spoken and written language.
An interesting view on individual differences is the
structure-building framework
(, 1990), which framescomprehension skill around the assumption that readers, in
constructing a coherent framework for a text, activate and enhance
relevant concepts while suppressing irrelevant concepts.
According to this approach, people with poor reading skills
have deficient suppression mechanisms. Consider, for example,
the sentence “He dug with the spade.” The final word in this
sentence has two meanings, but only one fits the context of the
sentence. However, when adults are asked to decide whether a
word is related to the meaning of the sentence, their decisions
are initially slow for ace (related to the inappropriate meaning
of spade). In other words, both appropriate and inappropriate
meanings may be activated at first. If the amount of time between
reading the sentence and seeing ace is increased, however,
skilled readers show no delay in rejecting it (i.e., they
suppress the irrelevant meaning), but less skilled readers continue
to react slowly to ace, as if they have not completely suppressed
the irrelevant meaning of spade. A failure to use
context is not what is involved here. Research with children
found that less skilled readers use context in word identification
at least as much as and perhaps more than do skilled readers
(1985; 1980). However, either a failure
to suppress irrelevant information, an insufficient level of specific
and reliable knowledge about word forms and meanings
(s), or both can lead to comprehension
failures.
The complexity of text comprehension implies several possibilities
for processing failure beyond those already discussed.
There are also higher-level skill differences in text comprehension.
For example, problems in making inferences have been
the target of much research. (1988) summarized
evidence suggesting that less skilled readers fail to
make a range of inferences. When skill differences in inferences
are observed, their uniqueness—that is, whether they occur
in the absence of differences in lexical, working memory,
or general language processes—is seldom clearly demonstrated.
However, there has been some success in identifying a
small percentage of children whose problems can be considered
comprehension-specific, although highly general across
reading and spoken language ( 1996).
Another example of a high-level contributor to comprehension
problems is comprehension monitoring (a reader’s implicit
attempt to ensure a consistent, meaningful understanding). Skilled
readers can use the detection of a comprehension breakdown
(an apparent inconsistency) as a signal for rereading and repair.
Less skilled readers fail to engage this monitoring process
(1984;1980). However, such differences may not be independent
of the ability to construct a simple understanding of
the text ( 1992). Because comprehension monitoring,
like inference making, both contributes to and results
from the reader’s text representation, it is difficult to attribute
comprehension problems uniquely to failures to monitor comprehension,
as opposed to more basic comprehension failures.
For comprehension to succeed, readers must import knowledge
from outside the text. Thus, a powerful source of individual
differences in comprehension skill is access to knowledge
needed for a given text
(1977). However, readers of high skill compensate for lack of
knowledge to some extent
(1995). It is the reader who lacks both knowledge and reading
skill who is assured failure. Moreover, the deleterious effect of
low reading skill (and its motivational consequences) on learning
through reading creates readers who lack knowledge of all
sorts.
Perhaps nothing is so important to successful reading comprehension
as practice, by which we mean repeated engagements
with reading texts of various types. Reading itself
increases familiarity not only with words but also with text
structures and written syntax, which are not identical to the
typical structures and syntax of spoken language. Thus, continuing
development of reading skill as a result of initial success
at reading—and the parallel increasing failure as a result
of initial failure—is undoubtedly a major contributor to individual
differences in reading comprehension.
Implications for learning to read
As we noted at the outset of this section, the view of skilled
reading has changed dramatically over the past 30 years.
Whereas skilled reading was once considered a guessing game
because of bottlenecks in the processing system, it has become
quite clear that skilled readers are able to quickly process the
visual information in the text, and that it is more efficient to
process what is there than to make guesses about what may be
there. In addition, a considerable amount of research suggests
that phonological codes are intimately involved in skilled reading.
Whereas it was once thought that phonological processing
in reading was simply a carryover from the way children are
taught to read (orally before silently), it now appears that phonological
codes are activated very early in processing words
and that this information is used in both accessing the meanings
of words and remembering information in the text. What
all of this suggests is that instruction that helps children efficiently
and quickly understand words, and that provides an analytic
strategy for learning new words, should provide the most
effective way to become a skilled reader.
The research on comprehension has less precise implications
for learning to read. It does suggest that a combination of
(a) simple comprehension procedures closely linked to word
processing and syntax and (b) more complex knowledgedriven
processes is characteristic of skilled comprehension.
Various ideas about comprehension instruction in the context
of reading have been proposed and have been the target of active
research. However, there is little in the basic research on
comprehension nor in instructional research to give strong
guidance on comprehension instruction.
The View From Cognitive Neuroscience
Although modern brain-imaging work is still in its infancy,
methods such as functional magnetic resonance imaging (fMRI)
and positron emission tomography (PET) have provided new
and converging information about the functional neuroanatomy
of reading. These imaging methods provide spatial information
on a centimeter or even millimeter scale, but with limited temporal
information. Questions about time course are better addressed
by event-related potential (ERP) methods. Although the
standard brain-imaging question may appear to be “What lights
up in the brain when words are read?” the potential of brainimaging
research goes well beyond this. For example, do studies
yield evidence for the phonological processes of word identification
that have proved so important for understanding
reading? What do imaging studies have to say about the reading
problems identified in dyslexia research?
Brain imaging
Neuroimaging studies of reading generally require comparisons
between images made during the performance of a reading
task and those made during baseline conditions. A simple
comparison that has provided basic data on reading is one between
reading aloud a single word and looking at a small fixation
cross. This comparison has identified regions that show
greater activation during the reading task. In many cases, the
regions are also activated during other kinds of tasks. For example,
primary motor cortex is activated during reading because
it is involved with movements of the mouth required by
oral reading. Identifying areas that play a more distinctive role
in reading itself requires other comparisons.
Research following this general approach has identified
brain regions that play some role in word reading
. Imaging studies have sought to link specific
orthographic, phonological, and semantic components of
word identification ( 1999;
1997) to specific locations in inferior
frontal cortex, the left temporoparietal cortex, and the left
basal temporal cortex (near the occipital-temporal boundary).
For example, (1999) found
that a left frontal region responded differentially to words with
consistent spelling-to-pronunciation mappings and words with
inconsistent mappings. Thus, these three areas, illustrated in
roanatomy underlying the knowledge components (orthographic,
phonological, and semantic) that are needed in word reading.
We now consider briefly what has been learned about each
source.
The visual areas in the occipital cortex that are used for object
recognition are an obvious candidate for supporting orthography
in the brain. In examining this hypothesis, researchers
have focused comparisons on activations associated with reading
words versus looking at pictures and on activations associated
with reading different kinds of letter strings (e.g., words,
pseudowords, and strings of letters). The search for an area that
is dedicated to printed words (a word-form area) has led to
some strong candidates, especially in an area near the occipitotemporal
border, the left middle fusiform gyrus. This area responds
differently to nonwords, pseudowords, and words
( 1998). In patients with lesions in occipitotemporal
areas, severe disturbances in ability to read words as wholes,
with a reliance on letter-by-letter reading (pure alexia), has
been reported ( 1999).
As we have emphasized, the conversion of an orthographic
form into a phonological form is a central part of reading, and
how the brain carries out this task is of great interest. Furthermore,
the inability to read pseudowords—a process that relies
on this conversion process without the aid of word meaning—
has become a marker for phonological dyslexia
(1993). Studies of patients have identified
two brain regions where lesions lead to deficits in phonological
decoding—the left inferior frontal lobe and the temporoparietal
cortex. Lesions in one or both of these areas are associated with
difficulty in reading pseudowords (1998;
1999). Patients with these lesions tend to
read words relatively well (compared with pseudowords), as if
they are able to use a stored lexicon that remains intact with these
lesions. Recent evidence from direct electrical stimulation of the
temporoparietal region produces an interesting convergence
with the data from patients. The ability of normal readers to
name pseudowords, a signature task for sublexical processing,
is disrupted by stimulation in this region, but their ability to
name real words is not ( 2000). Also, frontal regions
have shown greater activation for pseudowords than for
real words in some studies ( 1998). Thus, both
left frontal and temporoparietal regions are active in reading in
tasks that require or encourage phonological processing
(1996). However,
whether sublexical and lexical processes can be neatly separated
remains uncertain.
In discussing dyslexia, we noted the pervasive extent of behavioral
evidence for a phonological-processing deficit, and
imaging studies provide a convergent picture
1999; 1997; 1999; Sh
1998; 1998). In particular, dyslexics
show lower levels of activation in both left frontal and temporoparietal
regions compared with skilled readers
(1999; 1998). Recent evidence adds an intriguing
possibility that the processing problems of dyslexics
may depend on the writing system. For example, a phonological
deficit may have more of an impact for a reader of a deep
orthography (e.g., English) than for a reader of a shallow orthography
(e.g., Italian), in which spelling-sound correspondences
are highly consistent. (2001) reported a
brain-imaging study of Italian, English, and French dyslexics.
All three groups were impaired on tests of reading and phonology
and showed reduced activity in left-hemisphere regions
implicated in reading. However, the Italian dyslexics performed
better on tasks involving the pronunciation of words
and nonwords. The phonological deficit common to all dyslexics
in all three languages appears to have had less of an impact
in Italian because it is a shallow orthography.
Finally, both left frontal (1997) and basal temporal regions
( 1998) have been identified as candidates for semantic
processing (which has been studied using tasks that
require the retrieval of word names and concepts). Different
kinds of dyslexia have been linked to these two regions: developmental
phonological dyslexia to left frontal and temporoparietal
regions, and acquired surface dyslexia to basal temporal
lesions. Surface dyslexics, who experience reading problems as
a result of brain damage, have problems with reading words
lexically, as whole words (as opposed to reading sublexical
units). Thus, their problem is manifest on words that contain
inconsistently pronounced spelling patterns, or so-called irregular
words (e.g., choir). Patients with basal temporal lesions
tend to show the same problem when reading words, as well as
a more general deficit in picture naming
(1999). Although overlapping regions in the basal tem-
poral area are activated during picture and word naming, there
also appears to be differential activation (across frontal vs. posterior
areas of the fusiform gyrus) for naming pictures versus
naming words ( 1999). Interestingly, frontal
and temporal regions found to be activated in studies of reading
alphabetic systems overlap with those found to be activated in
reading Chinese ( 1999; 2000).
ERPs
Complementary to the spatial information from fMRI and
PET are the time-sensitive recordings obtained from ERPs,
measures of electrical activity from electrodes placed on the
surface of the head. The voltages associated with brain activity
vary in both polarity and magnitude over time, resulting in a
series of electrical peaks and valleys that can reflect the dynamics
of reading over a few milliseconds. ERP evidence has provided
information about the time course of events associated
with reading, confirming ideas about the rapidity of orthographic
and phonological processing. Word-level processes are
observed within 130 to 200 ms after the onset of a word
(1999;
1998). Orthographic processes are observed within 200 ms in
occipital and occipitotemporal regions, and mainly in the left
hemisphere
(1999;
1999), consistent with the idea that learned forms of the
writing system acquire a functionally distinct status within the
visual processing system. For orthographic inputs that allow
phonological processing, a second stage of processing is reached
in temporal areas within 350 ms. For inputs that are processed
to the semantic level, anterior temporal and frontal regions show
their roles within 400 to 450 ms.
Thus, various methods, including eye movement, brain-imaging,
and ERP studies, converge on a coherent picture of word
reading as a rapid process from graphic input to phonological
and meaning outputs distributed across functional brain regions.
Implications for learning to read
The knowledge gained from neuroscience methods is becoming
informative with respect to questions about skilled
reading and reading problems. The big question for learning to
read, how the brain supports the acquisition of reading skill, remains
to be addressed. However, there are studies that illustrate
the promise of neuroscience methods for showing how the brain
responds to training. For example, repeated practice affects ERP
components, presumably reflecting gains in efficiency through
specific learning ( 1997). Moreover, such learning
can actually influence brain development and neuronal connectivity.
A recent fMRI study found a thicker band of callosal
connective fibers between parietal lobes for literate than for illiterate
adults ( 1999). A study designed to
shift an acquired dyslexic from a whole-word reading strategy
to a phonological strategy found that activation patterns in the
brain changed following the intervention ( 1998).
In both studies, learning apparently produced an alteration in
brain circuitry, and there is the implication that neuronal connectivity
remains plastic into adulthood.
We can illustrate the effect of learning on brain circuitry for
an ability assumed to be central to learning to read—phonological
processing. (1999) examined oral language
processing in illiterate adults in Portugal (this group was
fully functional socially; illiteracy was widespread in rural portions
of Portugal until recently). The key comparisons concerned
performance and brain activations for the literate versus
illiterate participants when they repeated words or pseudowords
spoken to them. With real words, the literate and illiterate participants
performed comparably and showed similar brain-activation
patterns. But with pseudowords, the illiterate participants
performed more poorly than the literate participants, and their
brain-activation patterns were not the same. Castro-Caldas et
al. suggested that literacy acquired during childhood affects the
functional organization of the brain. Cognitive neuroscience
methods promise to be useful for investigating many topics in
the study of reading—the components of word identification,
cross-language comparisons, the acquisition of skill, and even
comprehension. There remains much to learn about the specific
brain regions that support these specific processes. And there is
much more to learn about how the brain reorganizes itself during
learning to read.
USING CONNECTIONIST MODELS TO
UNDERSTAND READING AND DYSLEXIA
The tools available to study reading have expanded in recent
years to include computational models, which are computer
programs that simulate detailed aspects of how children learn
to read, of skilled reading, and of dyslexia. So, for example, a
model might be taught to recognize letter strings and compute
their meanings or pronunciations using the same principles
thought to govern children’s performance. Such models complement
behavioral and neuroimaging studies that have provided
the main empirical evidence concerning reading. The
models typically focus on components of the reading system
such as eye movements (1998), pronunciation
(1993;
2001), and orthographic processing (1994)
In focusing on connectionist models in this section, our goal
is not to assess whether they provide better explanations of
reading than other models, such as the dual-route model developed
by Coltheart and colleagues ( 1993, 2001).
The connectionist and dual-route models account for many of
the same phenomena, and the differences between them are not
highly relevant to the issues discussed here. We use connectionist
models to illustrate how important phenomena in word
reading, learning to read, and dyslexia can be understood
within relatively simple systems that learn to translate between
orthographic, phonological, and semantic codes.
Connectionist models are useful for several reasons. First,
they incorporate ways of thinking about how knowledge is represented,
acquired, and used that may deviate from intuitive accounts
of cognitive phenomena. For example, words are not
represented as entries in a mental lexicon but rather as patterns
of activation over units encoding orthographic, phonological,
and semantic information. This notion invites reconsidering the
data previously taken as evidence for other types of mechanisms.
Second, connectionist models attempt to achieve theoretical
generality by explaining reading in terms of basic
principles of learning, knowledge representation, and information
processing that govern many aspects of language and cognition.
This approach is consistent with the observation that
reading, a relatively recent development in human culture, relies
on capacities that evolved for other purposes.
Third, the models provide a strong method for testing the
adequacy of theoretical proposals. With the more informal approach
to theorizing characteristic of much reading research, it
is not always clear if the proposed mechanisms work in the intended
ways. Connectionist models provide a unique way to
test causal hypotheses about the bases of normal and disordered
reading. Consider the long-standing debate about methods
for teaching reading. Phonics methods assume that learning
the relationship between the written and spoken forms of language
is an important step in becoming a skilled reader. Wholelanguage
methods assume that using phonological information
distracts the beginning reader from the goal of learning to read
for meaning; hence, the approach emphasizes instructional practices
that promote the development of literacy rather than phonological
decoding. There are plausible intuitive arguments for
both approaches. One way to assess these competing claims is
by conducting studies in which a model is trained in different
ways. Modeling complements field studies by providing insight
about why particular outcomes are obtained, and it allows
close control over potentially confounding factors.
Basic Elements of Connectionist Models of Reading
Most connectionist models use distributed representations,
in which spelling, sound, and meaning are represented by small
sets of units that participate in many words
(1986). For example, a model might include
units that correspond to phonetic features, and each such unit is
activated for all the words that contain that sound. Like other
aspects of the theoretical framework, the use of distributed representations
is motivated by the desire to use mechanisms that
are consistent with evidence about brain function; in this case,
the use of distributed representations is consistent with the fact
that large networks of neurons encode information in the brain.
Connectionist models also include hidden units that allow the
network to encode more complex mappings between orthographic,
phonological, and semantic codes.
In a connectionist model, units are linked to one another to
form a network; when a letter string is presented as an input,
the corresponding units are activated, and this activation
spreads to other units (representing phonological or semantic
information). The connections between units carry weights that
determine how much activation is passed along. The goal is to
find weights that allow the model to perform these computations
proficiently; learning means adjusting the weights on the
basis of experience. The word-reading models have used a
learning procedure (backpropagation) in which the output that
the model produces for a word is compared with the correct,
target pattern. Small adjustments to the weights are made on
the basis of discrepancy between the two. Performance improves
gradually as the weights assume values that minimize
this discrepancy. Backpropagation is a kind of learning algorithm
in which the output the model computes is compared
with a target pattern ( 1989).
Whether the brain uses this algorithm or others that have very
similar effects is a topic of current research. Although the
learning procedure used in the reading models is undoubtedly
simplified, it captures basic elements of how learning occurs in
this and other domains.
Behavior in connectionist networks (as in people) is determined
by both constitution and experience. The reading models
(see Fig. 9), for example, were constructed with the capacity
to represent different lexical codes (orthography, phonology,
semantics) and the capacity to learn. These representations are
themselves thought to have resulted from the joint influence of
biological endowment and experience. Like children, the models
learn through exposure to many examples (e.g., letter patterns
paired with pronunciations or meanings). The fact that
words are internally structured emerges through exposure to a
large ensemble of words that overlap in different ways and degrees.
The models gradually pick up on this structure because it
allows phonological and semantic codes to be computed more
efficiently.
Insights From Connectionist Models
Connectionist models provide a new perspective on many of
the issues discussed in this monograph. Here we focus on three
issues that have been the focus of considerable debate among
reading researchers.
How do children learn the correspondences between spelling
and sound?
We noted earlier that learning about the correspondences
between the written and spoken forms of words is a critical
step in becoming a skilled reader. However, English is notorious
for having many exception words that have irregular pronunciations
(e.g., give, said, was, have, does). Many of these
are high-frequency words that are among the first that beginning
readers must master. As noted earlier, (1973) argued
that because of these irregularities, reading that involves
phonological recoding (i.e., translating from spelling to sound
to meaning) cannot be efficient. Connectionist models suggest
that this reasoning is flawed.
(1989) model learned the pronunciations
of both regular words such as gave and exceptions
such as have with relatively little training (at most 250 trials,
for very low-frequency exceptions such as torque; most words
were learned with many fewer trials). Acquisition is rapid because
what is learned about one word (e.g., gave) also applies
to overlapping words (save and gate). The same thing is true
for exceptions: The model’s performance on have benefits from
exposure to overlapping words such as had, has, and hive.
Thus, the same network and learning procedure are used for all
words; regular and exception words differ only in the degree to
which their spelling-sound correspondences overlap with those
of other words. Learning the exceptions is less of a problem
than (1973) asserted because they are not arbitrary and
the learning procedure is efficient.
How does phonological awareness develop?
Earlier we noted that there is a reciprocal relationship between
the development of segmental phonemic representations
and learning to read. Prereaders’ knowledge of phonological
structure is causally related to success in learning to read; at the
same time, learning to read changes the nature of phonological
representations, making them more segmental. Connectionist
models provide additional insight about the mechanisms underlying
the interactions between phonological knowledge and
reading. Children come to the task of reading with extensive
knowledge of spoken language, but their knowledge of phonological
structure is relatively coarse; being able to tell that bag
and bat are different spoken words can be done on the basis of
nonsegmental information. Phonological representations are
further shaped by children’s participation in reading.
To make this point concrete, (1999)
developed a model of the orthography-phonology mapping that
was trained in two phases. The first phase, which involved exposing
the model to the phonological forms of words, was
meant to capture the kinds of phonological knowledge children
acquire from speech. The second phase involved training the
model to map from orthography to the pretrained phonological
system. Learning this mapping changed the organization of
phonology, promoting clearer representations of subword units
such as phonemes, onsets, and rimes. The development of
these representations in turn facilitated learning to read additional
words.
What was crucial in the model was not having full phonemic
representations prior to reading (though that would be useful)
but rather having the capacity to develop such representations
with reading experience. Reducing this capacity interfered with
the development of subword units; thus, when the model was
exposed to a word such as gave, there was less carryover to
partially overlapping words such as save. Two consequences of
this limitation were observed. First, the model took much
longer to learn the phonological codes for words. Second, it
generalized poorly; lacking knowledge of subword structures,
it could not piece together the pronunciations of novel words
very well. The model did not address many aspects of phonological
development, and further research is needed. However,
the initial results provide additional causal evidence linking the
development of phonological representations and the acquisition
of reading skill. Moreover, the modeling suggests that constitutional
differences in the capacity to encode phonological
structure may contribute to variable outcomes in learning to
read. Finally, it suggests that segmental representations are
closely tied to knowledge of orthography rather than speech.
Is word reading direct or phonologically mediated?
As noted earlier, there has been a long-standing debate
about whether words are recognized visually (by a direct mapping
between orthographic patterns and meanings) or via phonology.
(2001) attempted to break this
impasse by treating the issue as a computational one: How
does a connectionist model learn to compute meanings quickly
and accurately? That is, what division of labor does the model
converge on, given the availability of both pathways? The an-
swer is that Harm and Seidenberg’s model pooled the input
from both sources for almost all words. Given orthographic input,
the semantic pattern that was computed reflected the joint
effects of both pathways. This property contrasts with the independence
of the orthography-semantics and orthography-phonology-
semantics pathways in race models
(1991), in which meaning is accessed by the process that finishes
first. The connectionist model performs more efficiently,
using both pathways rather than either one in isolation; thus, it
is a question not of which pathway wins the race, but rather of
how the pathways cooperatively solve the problem.
Early in the model’s training, semantic activation is largely
driven by input from the orthography-phonology-semantics
pathway. The phonology-semantics component is trained prior
to the introduction of orthography on the view that prereaders
possess this knowledge from their use of spoken language. The
orthography-phonology mapping is easy to learn because the
codes are highly correlated; the orthography-semantics pathway
takes longer to become established because the mapping is
more arbitrary. Over time, however, the orthography-semantics
pathway begins to exert its influence, particularly for high-frequency
words. Note, however, that what changes is the relative
division of labor between the two pathways; there is some input
from both pathways for almost all words.
In summary, the division-of-labor model explains why the
phonological pathway predominates in early reading. However,
it also contradicts the intuition that the orthography-semantics
association is too arbitrary to play a useful role. Given the cooperation
between the pathways, the orthography-semantics
pathway only has to be good enough to clean up the pattern activated
via phonology. When the model is trained to a high
level of proficiency, both pathways contribute significantly.
Summary
Connectionist models have provided insights about many
aspects of normal reading and reading impairments. The models
have brought new ideas about learning and information processing
into discussions of reading, lending support to some
claims (e.g., about the role of phonology in reading ability and
disability) while challenging others (e.g., that rule-governed forms
and exceptions are processed by distinct subsystems). Although
considerable progress has been made, the models raise
many questions that need to be addressed in future research.
The models do not address all aspects of reading; their implications
concerning instruction and remediation have not been explored
in depth; and it will be necessary to link the models more
closely to the evidence concerning the brain bases of lexical
processing that is emerging from neuroimaging (see the previous
discussion in The View From Cognitive Neuroscience).
The goal of developing an integrated account of reading behavior
and its brain bases, with computational models providing
the interface between the two, nonetheless seems a realistic one
and is likely to be the focus of considerable research.
METHODS OF TEACHING READING
Much of the history of reading instruction in the United States
has involved two general methods: whole-word instruction and
phonics instruction. However, meaning-emphasis instruction,
and especially whole-language instruction, has dominated the
philosophy of training of reading teachers over the past 20
years. Originally, the time-honored ABC method was used to
teach reading for about 200 years. This method was a basic
type of phonics instruction in which children were taught letter
names, then simple syllables, then words. The child would
spell the syllable and then pronounce it: “double-you-ay-ell-ell—
wall.” Later, more syllables and words were mixed in, usually
with the same spelling requirement prior to pronunciation. The
New England Primer, Webster’s Spelling Books, and
Readers
were the major sources of reading programs from the1700s until the 1900s. Pictures were introduced into these programs,
but for the most part the emphasis was on phonics drill.
Although there is little evidence indicating how successful the
ABC method was in teaching reading, there apparently was
some dissatisfaction with it because educational reforms in the
late 19th century led to the whole-word method becoming the
predominant method of teaching reading. More recently, the
whole-word method has been supplanted by whole-language
instruction. In this section, we describe each method of instruction.
Whole-Word Instruction
To some extent, ideas about whole-word instruction are congruent
in a general way with facts about spoken language. Spoken
language is an almost continuous stream of sound with
little or no silence separating the individual words, a fact that is
especially noticeable when one tries to pick out word boundaries
in a foreign language. However, compared with phonemes,
words are more readily detected as units, and children
come to recognize the spoken word as a meaningful unit. The
relative invisibility of phonemes encouraged the view that
whole words are the appropriate units for instruction. Typically
in this kind of instruction, the child is shown a flash card with a
word on it, and the teacher pronounces it and asks the child to
say it. Generally, the teacher starts with a small set of words
and gradually expands the set.
Another argument that has been used to support the wholeword
approach is the low reliability of letter-to-phoneme mapping,
a complex issue that we discussed in previous sections.
The irregularity of English spellings, which occurs mainly in
the vowel system, allows the word pint a different pronunciation
than most words that end in int, which are pronounced as
in hint. Similarly, have is irregular because most words that
end in ave are pronounced as in gave. Irregular words such as
pint and have, in any approach to teaching reading, must be
learned at least partly in terms of their distinctive properties.
The whole-word approach generalizes this approach to learning
to read all or most words.
Advocates of the whole-word approach have also argued
that it promotes reading for meaning at an early stage of reading.
Words have meanings; speech sounds do not. When a child
has developed a small sight vocabulary, this vocabulary is deployed
in various combinations to construct meaningful sentences,
and new words are introduced so that the context clarifies
their meaning. The pronunciation is given by the teacher,
who indicates, wherever possible, the similarity in spelling between
the word to be read and a word already in the sight vocabulary.
This makes it possible (after an initial sight vocabulary
is established) to emphasize that the letter symbols represent
sounds.
Phonics Instruction
Phonics instruction, in its purest form, starts with a limited
set of correspondences between letters and speech sounds.
These letters are used immediately to build many different kinds
of words. In this way, phonics instruction takes advantage of the
productive aspects of alphabetic writing systems. Gradually,
more letters are added, and then consonant digraphs (th, ch) and
eventually consonant clusters (st, tr) are introduced. As simple
words are presented over and over, the child also naturally develops
a sight vocabulary during these early stages, but the development
of a sight vocabulary is largely incidental in phonics
instruction (much as knowledge of the alphabetic principle is
incidental in whole-word instruction). The individual letters are
taught by the sounds they make, and then children are induced
to blend the sounds of novel letter combinations.
The main rationale behind a phonics approach is that it explicitly
teaches children both the alphabetic principle and the
specific letter-phoneme correspondences that generalize across
many English words. An additional benefit of phonics is that it
promotes an analytic approach to words that can serve the child
in encounters with unknown words. One criticism of phonics,
repeated for more than 100 years ( 1990), is that it
is boring for the child. The reason for this criticism is the emphasis
phonics places on letter-sound correspondences at the
expense of reading for meaning. However, this complaint is
more often about the practice of phonics lessons, which are often
derided as “rote drill,” than about the essence of the approach.
The teaching of letter-phoneme correspondences is not
the same as “phonic drills,” as consumers of computer-based
phonics programs can attest. Certainly, some forms of practice
at producing phonemes and blending them together into a word
may at times be boring for children (just as learning basic math
facts may be boring). But learning to read new words independently
can be very rewarding for children.
Meaning-Emphasis Instruction
In general, meaning-emphasis programs focus on language
experiences of the child. Thus, the child dictates short stories
and is taught to read the words he or she has dictated. Instruction
in learning individual words usually emphasizes memorizing
whole words, though some phonics drill may be incorporated
into the program at later stages.
Within the class of meaning-emphasis programs are wholelanguage
instruction and an earlier related approach, the psycholinguistic
approach, based on the work of
(1970, 1986; 1971, 1973;
1971). Goodman suggested that reading is a “psycholinguistic
guessing game” in which readers try to figure out
the meaning of a text by using a variety of partly redundant
cuingsystems
. There are three types of cues in this guessinggame: semantic, syntactic, and graphophonic. The graphophonic
cues represent general knowledge of spelling-sound relations;
the syntactic cues represent knowledge of syntactic
patterns and the markers that cue these patterns (such as function
words and suffixes); and the semantic cues represent
knowledge about word meanings and the topic. More recently,
Goodman’s ideas have been incorporated into the larger wholelanguage
instruction movement. This type of instruction, like
other meaning-emphasis approaches, relies heavily on the
child’s experience with language. Children are encouraged to
guess words that are presented in the context of short stories,
and the primary motivation of the method is to make reading
fun for the child. Whole-language teaching typically includes
frequent oral reading by the teacher and the use of authentic literature,
rather than decodable text.
In an approach that otherwise avoids any specification of
what should be taught, provided one clear
suggestion: Phonics should not be taught. Furthermore, they
argued that children should not be corrected when they make
errors reading words. Neither the development of phonological
awareness in general nor the development of specific knowledge
about letter-sound correspondences is a priority in wholelanguage
instruction. Whole-language advocates do not deny
that phonological awareness and phonics knowledge are components
of reading (though they are not considered as central to
the process as the research findings indicate). But they do deny
that the explicit teaching of phonics is necessary. Their basic
argument is that explicit teaching of phonics does more harm
than good to beginning readers because many of them find
phonics somewhat difficult and boring.
Whole-language proponents suggest that the knowledge
necessary for skilled reading—including phonics knowledge—
can develop in the same natural way that spoken language develops.
(1994), for example, in a book aimed at elementary-
school teachers, wrote: “Just as they learn the patterns
of oral language, so most children will unconsciously learn
common phonics patterns, given ample opportunity to read environmental
print and predictable and enjoyable materials, and
ample opportunity to write with invented (constructive) spelling”
The various tenets of whole-language teaching are a set of
interrelated ideas that suggest a coherent perspective. The logic
is as follows: (a) Reading is a natural extension of language;
(b) explicit teaching of phonics treats reading as a technical ex
ercise rather than a natural extension of learning and thus has
the potential to do harm by boring and frustrating the child; (c)
explicit teaching of phonics is unnecessary for learning; (d)
therefore, explicit teaching of phonics should be avoided; and
(e) because phonics represents just one of several redundant
cuing systems, if a child fails to learn some piece of phonics
knowledge, other cuing systems will fill in the gaps when the
child actually reads.
The problem with this method lies not in the logical connections
among its tenets, but in the extent to which they are true.
Goodman’s suggestion that skilled reading is a psycholinguistics
guessing game, for example, has largely been refuted by
research on skilled reading, which demonstrates that skilled
reading is not a guessing game and that phonological information
is critically important in word identification. In fact, the
three cuing systems are not equivalent in determining what
word is actually read; the graphophonic mechanism plays a
highly prominent role, particularly in reading acquisition. Furthermore,
the view that learning to read, like learning to speak,
is a natural act that the child teaches him- or herself how to do
stands in marked contrast to the view more common among researchers—
that learning to read is not a natural act
( 1980; 1999), and is very different from
learning to speak, which is effortless and automatic for almost
all children brought up in normal circumstances. No child
needs a teacher to show him or her how to speak. It is sufficient
to be a normally developing human being surrounded by other
human beings speaking their language.
Learning to read presents an entirely different picture. All
schools of thought agree that some amount of teaching is often
(or even always) necessary. In particular, learning to read often
requires some explicit instruction in the alphabetic principle.
Contrast this with learning to speak. No child needs to be
taught the phonemes of his or her language, but every child
needs to be taught the symbols that make up his or her writing
system. That is why there is an alphabet song, but not a phoneme
song. Furthermore, though all human societies have language,
many do not have reading and writing.
It should be clear that we have some fundamental disagreements
with some of the claims made by whole-language advocates.
However, to this point, we have presented only arguments
regarding the possible efficacy of whole-language instruction.
In a later section, we review research findings more directly.
Here we do want to note some positive contributions of the
whole-language movement (more recently called literaturebased
instruction). First, and foremost, whole-language advocates
have focused attention on the need to ensure that children
are enthusiastic about books and eager to learn to read. They
may go too far in their reliance on enthusiasm and eagerness as
components of the process of learning to read, but no one can
doubt the importance of these components, and whole-language
proponents have been largely responsible for the growing
trend to make reading instruction more meaningful. Furthermore,
they have replaced an emphasis on the teacher as an agent
of instruction with an emphasis on the child as an agent of his
or her own learning. Certainly, appreciating the mutual roles of
learner and teacher is an important step in establishing effective
reading instruction, even if learning to read is not natural
and spontaneous.
WHAT HAPPENS IN CLASSROOMS DURING
READING INSTRUCTION?
The way we have discussed different methods used to teach
reading may have implied that the method adopted determines
what actually goes on in the classroom. However, many good
teachers are adaptive rather than rigid in their approach to
teaching children and only loosely base their instruction on a
given method. They know by instinct that when learning is
made meaningful and exciting, children learn more. It should
also be noted that most schools are populated by a couple of
generations of teachers who were taught that whole-language
instruction is good and phonics is evil. So teachers may be
asked to teach phonics and not know how, which also affects
what happens in classrooms.
These days there are a large number of commercially available
reading programs in schools. Because these programs can
be rather expensive and because bulk buying is generally
cheaper than buying different programs for different schools,
school districts often adopt a single program for districtwide
use. These programs, called basal reading series, consist of
teacher guides, student materials (e.g., minibooks, anthologies,
and workbooks), and ancillary materials (e.g., letter cards,
posters, and CDs). Although each reading program generally
adheres to one particular method for teaching reading, within
individual classrooms, teachers have some flexibility in what
they actually do if they are trained in a variety of methods.
The key word currently used to describe classroom reading
instruction is balance. Reading instruction that balances instruction
in phonics with exposure to good literature and opportunities
to write is found more and more in classrooms. However,
what balanced reading instruction involves in practice varies
with respect to the explicitness with which skills are taught, the
kinds of materials used to practice these skills, the size of the instructional
group, and the extent to which assessment informs instruction.
Truly balanced instruction should integrate skills
instruction with reading for meaning and opportunities to learn
(1998). However, what often happens in
primary-grade classrooms is a fragmentation of the literacy curriculum
into activities based on the latest teacher training workshop.
More generally, variations in actual classroom practice
roughly fall under two approaches to teaching that we call prescriptive
(direct) versus responsive teaching.
Prescriptive Teaching
Reading instruction that explicitly or directly teaches skills
such as letter-sound correspondences typically consists of a
curriculum with a prescribed set of activities, which together
are called a scope and sequence. Across kindergarten and first
grade, the curriculum systematically introduces phonological
awareness and phonic skills with practice in decodable texts
(containing letter-sound correspondences taught by the teacher).
Beyond second grade, there is still some phonics in spelling instruction
but not in reading instruction.
Instruction based on prescriptive teaching may vary in the
amount of whole-class versus small-group instruction and in
the amount of assessment. For example, the Open Court (2000)
basal reading series emphasizes whole-class instruction and
additional independent practice, with little emphasis on assessment.
In contrast, in Reading Mastery
(1995) reading instruction takes place in groups of six based on
placement tests. The school-reform model Success for All
( 1996) has multiple grouping
formats—whole-class instruction according to reading level determined
by 8-week assessments, partner reading, independent
reading, and collaborative group work.
First-grade teachers adhering to prescriptive techniques tend
to plan their lessons around the following activities: (a) review
of letter sounds previously taught, (b) introduction of new letter
sounds, (c) practice blending sounds into words, (d) practice
reading in decodable texts, (e) teacher read-alouds from
children’s literature to teach vocabulary and comprehension
strategies, and (f) language arts (spelling, writing, grammar,
and mechanics). For example, in the Open Court reading program,
the teacher introduces /e/ (which corresponds to the linguistic
symbol /ε/) spelled e by hanging the sound-spelling
card hen on the wall (along with the other sound-spelling cards
already introduced). The capital and lowercase printing of the
letter (E e) is at the top, the picture of the keyword—hen—appears
below, and in a green field at the bottom is the printed letter
e (see Fig. 10). The children have been taught that the green
field denotes short vowel sounds.
At the start of the lesson, the teacher reads the decodable
story Jen’s Hen—“Jen’s pet hen likes to peck, peck, peck. She
pecks at a speck on the new red deck. This is how her pecking
sounds: /e/ /e/ /e/ /e/ /e/.” Then, the teacher asks the children to
listen to words and to signal thumbs-up when they hear a word
that has the /e/ sound at the beginning or at the middle of the
word. So, the children give a thumbs-up to ever, etch, every,
and echo, but not hand or flavor, for /e/ sounds at the beginning,
and a thumbs-up for hen, pest, wet, desk, next, bed, and
feather, but not for tape or bike, for /e/ sounds at the middle of
the word. In the next step, the teacher has the children blend
words (both in isolation and in sentences) that contain short e.
A specific procedure is outlined for teaching blending. For example,
to blend fed, the child is taught to isolate the initial
sound (/f/) and the medial sound (/e/), then to combine them (/fe/)
before adding the final sound (/d/) to produce the entire word
(/fed/). Finally, the phonics lesson ends by practicing the accumulating
letter sounds in Jen’s Pen (see Fig. 11). Before reading
the story, the teacher reviews the high-frequency words
would, my, did, laugh, out, her, of, and move and teaches two
new nondecodable words, darts and feeds. The teacher notes
that one word with a variant spelling for “short e” (i.e., __ea_
in bread) is included in the story. This emphasis on decoding
instruction is complemented in the language arts section of the
lesson by teaching children to encode the sound-spelling /e/
through dictation practice.
Responsive Teaching
In contrast to prescriptive teaching, responsive teaching is
loosely based on the constructivist notion of scaffolding
1997, for
further discussion). Rather than working from a scope and sequence,
the responsive teacher responds to what the child is
perceived to need at the moment in the context of reading real
books. The teacher provides a scaffold against which the child
can construct knowledge of reading. Like whole-language instruction,
responsive teaching is steeped in the belief that children
inherit three cuing systems (syntactic, semantic, and
graphophonic knowledge) from their oral language abilities.
For example, in the tutorial program Reading Recovery
(1993), the classroom teacher not only provides feedback on
oral reading errors, but also responds to an error by extending
the child’s knowledge of the alphabetic system. Furthermore,
the responsive teacher keeps a running record of reading miscues
to inform the next day’s alphabetic activities of making
words and breaking words into constituent elements. Because
instruction in alphabetic coding is conducted incidentally in the
context of reading books, responsive teaching methods inherently
lack a systematic approach to phonics instruction. The
absence of sequential instruction and practice makes it difficult
for many children to acquire and transfer decoding skills.
Responsive teaching can be highly effective when knowledgeable
teachers work with individual children. But responsive
teaching in the hands of a teacher who does not have the
knowledge to seize the moment productively or who is teaching
a large group of heterogeneous readers may be ineffective
with the lower-achieving students. A popular system for responsive
teaching at the classroom level is Guided Reading
( 1996). Guided reading starts with wholeclass
discussion of a reading selection to elicit prior knowledge
and introduce difficult vocabulary. Then the teacher scaffolds
children’s reading of the passage in whole-group, small-group,
or partner-reading formats. If a small-group format is adopted,
children work in centers or independently while the teacher
works with one reading group after the other.
The current basal reading series have accommodated guided
reading by providing leveled text (i.e., texts that are ordered by
difficulty, according to a number of factors such as number of
words, predictability of syntactic patterns, word frequency, and
picture clues). These texts stand in contrast to the decodable
texts of the prescriptive approach because they are selected for
their sense of story and predictable syntactic patterns, and
words are selected for frequency, not for sound-spelling patterns.
But the issue of what makes text decodable for which
readers at which phase of reading development is a largely unanswered
empirical question (1985;
1999). An example of leveled text is shown in Figure
12. The pages in this figure are from the story The Bus Ride
( 1976), which is appropriate
for mid-fall of first grade. In contrast to Jen’s Pen (),
which shows tight control on vocabulary with an emphasis on
the /e/ sound from the lesson and previously taught lettersounds,
The Bus Ride emphasizes the predictable pattern “got
off/on the bus.”
Because of the new emphasis on phonological awareness
and phonics, guided reading now includes a separate emphasis
on words (1995; 1998).
This means that in addition to guided reading, independent
reading, and process writing blocks of an integrated readinglanguage
period, many primary-grade classrooms now include
a block of time devoted to word-level activities. During this
daily block, the teacher teaches children how to read and spell
high-frequency words and discusses strategies for decoding
and spelling. (1999) developed a curriculum
of activities for word work (e.g., organizing words on a wall
chart according to letter-sound patterns, building words with
magnetic letters, using words you know, guessing the missing
word in a sentence). These activities all emphasize analogical
reasoning around onsets and orthographic rimes, in contrast to
the synthetic phonics approach taken by the majority of prescriptive
teaching approaches. That is, a responsive teacher will
have children sort printed words by initial sounds and then by
word families (such as the ould in could). The danger is that
these strategies can result in a look-at-first-letter-then-guess decoding
strategy such as pronouncing shot as ship or categorizing
only by rimes (make, bake, rake) rather than also by vowel
spellings (make, rain, play, eight). Full mastery of the alphabetic
system requires understanding how spellings represent
speech sounds, and the inconsistency of this mapping for vowels
provides a challenge for beginning readers of English (see
Fig. 2). However, as we discuss in the section on research findings,
there is an adequate empirical base to support phonics
over nonphonics instruction, but not an adequate base to support
one type of phonics instruction over another (e.g., synthetic
vs. analogical).
Summary: Balanced Reading Instruction
In the first years of the new millennium, the language of balanced
reading instruction has swept America’s classrooms and
basal textbook market. However, underlying the rhetoric lurks
the reality of the debate over reading methods. Prescriptive
teaching follows a scope and sequence of phonic elements,
with texts based on the accumulating set of letter-sound correspondences
taught. Responsive teaching eschews a scope and
sequence in favor of strategies that enable the child to construct
meaning in texts leveled by difficulty according to a number of
factors (such as word frequency). Recently, responsive teachers
have begun to attend more to word-level work that emphasizes
an analogical approach to reading and spelling words and are
often turning to phonics kits to supplement their letter-sound
instruction. Unfortunately, the continued dichotomy of reading
philosophies produces fragmented instruction in classrooms
rather than the integrated balance of skills and meaningful applications
that research suggests are needed to produce successful
readers.
READING INSTRUCTION IS A POLITICALLY
CHARGED ISSUE
During the 1990s, concerns about the effectiveness of reading
instruction led to what are often called reading wars in a
number of states where state officials became very involved in
debates about reading instruction. In this section, we focus on
Texas, California, and Massachusetts as examples of how politically
charged the issue is. We also note that concerns among
many parents have led to increased enrollments in many private
schools. In general, this trend has occurred because many private
schools have relied on phonics instruction, whereas many
public school systems have used whole-language approaches to
teaching reading.
Because of their large populations, Texas and California
have had special status since the 1960s in decisions regarding
the adoption of basal readers. The readers adopted by school
districts in Texas and California often become the textbooks for
the entire nation. Both states have grass-roots parents organizations
and business coalitions that push educational agendas
which affect national educational policy (such as California’s
Proposition 227 ending bilingual education in 1998). In Texas,
the classic example of a politician involved in educational issues
is Ex-Governor George W. Bush, who made education
(and, in particular, reading) the focus of election campaigns,
with a rallying call of “All children shall read at or above grade
level by third grade!”
In the 1980s, the book
A Nation at Risk: The Imperative forEducational Reform
(National Commission on Excellence inEducation, 1983) inspired a number of educational reforms. In
Texas, the legislature passed reforms establishing minimum
passing scores for courses and tests, minimum competency
tests for teachers, and a “no pass—no play” rule that prohibited
students who failed from participating in extracurricular activities.
Legislation established an accountability system tied to
students’ performance on statewide tests and the development
in the early 1990s of the Texas Assessment of Academic Skills
(TAAS). Much has been written (2000) about
Texas’s experience with the TAAS as the epitome of the negative
consequences of high-stakes testing (e.g., bonuses to administrators
and bribes to students led to test anxiety, cheating,
and a TAAS-oriented curriculum). However, the public display
of TAAS results school by school in local newspapers and
across the state in the magazine Texas Monthly resulted in a
groundswell of parent and business involvement in school reform
at the local and state level. Much of this groundswell of
parents was a cry of “We want more phonics!” directed at local
school boards and the state school board. In fact, phonics advocates
joined forces with social conservatives on the state board
of education to try to mandate a phonics-oriented scope-andsequence
set of standards. Although this attempt failed, the social
conservatives and phonics advocates were successful in
passing a mandate that first-grade texts selected during the
basal-reader adoption process be 80% decodable.
Texas also institutionalized its reading initiative at the localdistrict
level. This was accomplished through several steps.
First, a reading czar, who reported directly to the commissioner
of education, was appointed at the Texas Education Agency,
and this model was followed by several of the large school districts
across the state. Second, the governor’s office worked
with the legislature to fund reading-related initiatives, such as
early-reading assessments, early-reading interventions, and 4
days of professional development for the state’s 40,000 kindergarten
through second-grade teachers. Third, the state’s curriculum
standards were approved by a consensus committee, and
textbooks that were subsequently adopted were aligned with
the state standards. Fourth, coalitions of business leaders, university
researchers, and professional educators used TAAS data
and research findings to spearhead school reform in low-performing
schools. At each step, districts are expected to make
their own decisions about how state directives will be achieved.
For example, districts can use “accelerated learning” funds for
whatever early-reading intervention programs they want, but
are held accountable for student outcomes. Additionally, dis
tricts choose from the commissioner’s list of approved instruments
which early-reading assessment will be administered.
In contrast to Texas’s local-control approach to reading reform,
California implemented a top-down approach that was a
result of the state’s plummeting scores on national assessment
tests. From the early 1980s to 1994, California went from near
the top to near the bottom in national comparisons. In 1995, the
state superintendent of public schools put together a reading
task force that prepared guidelines for reading reform. In that
same year, the legislature passed the ABC Bill, which mandated
explicit teaching of phonics, spelling, and math. This was
followed by legislation providing money for staff development
for teachers and administrators in effective early-reading techniques,
including techniques for systematic, explicit instruction
in phonics. In 1996, the California Department of Education
released a program advisory on reading and language arts instruction,
which was followed by the Reading/Language Arts
Framework in 1999. Both documents were written by outside
experts. While all of this was going on, California was also reducing
class size, setting up an accountability system, calling
for another textbook adoption, and eliminating bilingual education.
With so much change occurring so fast, it is not surprising
that California’s reading-reform movement was severely criticized
by university faculty and teachers ( 1999;2000)
In Massachusetts, the situation was slightly different. Public
education in Massachusetts has traditionally been much more
decentralized than in Texas or California. There has typically
been no statewide textbook adoption, and little history of state
involvement in curriculum choice. These matters have been
largely left up to local school systems. The 1993 passage of the
Massachusetts Education Reform Act was thus a notable departure
from tradition. Education reform promised major increases
in state funding for public education. In exchange,
local school systems were required to meet new state curricular
standards, which were to be drawn up by committees appointed
by the Massachusetts Department of Education and approved
by the state Board of Education. The standards were to be enforced
through a program of statewide testing.
Despite the traditional lack of centralized control, the curriculum
in Massachusetts public schools looked rather uniform
across the state. This is understandable: As everywhere, teachers
and administrators took similar courses at the same handful
of universities, attended the same workshops, and bought the
same textbooks, responding to national trends and fashions.
Consequently, reading instruction in Massachusetts was strongly
influenced by the whole-language movement in the early 1990s.
It is thus not surprising that the committee of educators charged
with drawing up the curriculum framework on English language
arts produced a document heavily influenced by the whole-language
approach. The document that they produced contained
no mention whatsoever of the scientific literature on reading,
yet claimed support from research. It highlighted the idea that
children could learn to read the same way they learned to
talk—and presented a vision of language acquisition that attributed
the process to curiosity and enthusiasm alone. Crucially,
the document claimed support from research on language:
By analogy, the same shift was endorsed in reading instruction.
As it happens, Massachusetts is home to well-known centers
of research in linguistics and the psychology of reading at
the Massachusetts Institute of Technology and the University
of Massachusetts at Amherst. Once the content of the proposed
curriculum document became known to some of these researchers,
a letter signed by 40 researchers in linguistics and
psychology was sent to the Commissioner of Education on July
12, 1995. The letter objected to the document’s claims about
language and reading:
By chance, the letter came along just as the reading wars
were heating up in other states, and attracted considerable attention.
Massachusetts responded quite positively, first revising
its curriculum framework and finally rewriting it completely.
The commonwealth’s curriculum document (though controversial
in other ways) has won praise for being a lucid and scientifically
accurate description of the reading process that at least
points to acceptable standards for early-reading instruction. In
a sense, the Massachusetts story represents a victory for sound
educational practice. Nonetheless, it is unclear how much has
really changed for Massachusetts schoolchildren now that
Massachusetts has a scientifically sound document underlying
its standards for reading instruction. The most recent round of
state testing continues to reveal harsh differences between rich
and poor communities in reading achievement, and there is little
sign that teachers are being educated about the research lit
PSYCHOLOGICALerature that underlies the commonwealth’s current standards
document.
Massachusetts, Texas, and California each reveal part of the
way out of the reading wars. It is helpful when the research
community is mobilized and gets itself involved, as happened
in Massachusetts. But researchers do not make policy and do
not teach young children to read. For true progress, there must
be guidance from above, as in California, supporting and fostering
strong local interest and enthusiasm from both parents
and teachers, as in Texas.
Many obstacles stand in the way of this sort of progress.
One is the peculiar alignment (particularly in North America)
of phonics with conservative politics. Considerable publicity
for the Massachusetts researchers’ letter was generated by conservative
newsletters and Web sites (a bit of an irony for a letter
signed by 40 Massachusetts professors, including several wellknown
leftist activists). In like fashion, writings from the
whole-language community proclaim that the essence of this
approach lies not in its particular ideas about reading, but in its
“belief in the empowerment of learners and teachers” and the
“acceptance of all learners and the languages, cultures and experiences
they bring to their education”
(2000)—as if a pedagogy based on the alphabetic
principle was inevitably inconsistent with empowerment and
acceptance. In reality, of course, the pedagogy that empowers
most is the pedagogy that teaches best.
RESEARCH FINDINGS ON TEACHING READING
We turn now to research findings related to how to best
teach reading skills. We begin by reviewing two well-known
summaries of research on teaching reading and then providing
overviews of the findings of two recent reports on the issue: the
National Reading Panel (NRP; 2000) report and a report of the
National Research Council (NRC) called
Preventing ReadingDifficulties in Young Children
(S 1998). We then turnto a discussion of laboratory and classroom studies.
Meta-Analyses of Teaching Reading
The questions surrounding how reading is most effectively
taught have been the object of several comprehensive reports
over the years, including two major books (1990;
1967). The question at the center of Chall’s “Great Debate”
review was, what does evidence have to say about the effectiveness
of direct instruction—explicit phonics—compared
with whole-word instruction or implicit phonics? Should beginning
instruction focus on directly teaching the correspondences
between letters and sounds (phonemes)? The logical answer to
this question appears to be that these correspondences, and the
alphabetic principle they instantiate, should be the central initial
focus of instruction. However, the tendencies of actual practice
have been otherwise. As noted earlier, a variety of alternative
pedagogies have emphasized instead meaning-focused instruction
built around story reading, exposure to print, and enhanced
language environments. These alternatives are too varied to capture
with a single characterization. For example, when Chall
coined the term Great Debate in 1967, the alternative to direct
instruction was whole-word teaching, in which basal readers
and limited (and later) phonics instruction were typical components.
In the past 20 years, the dominant alternative has been
whole-language instruction. Chall’s conclusion, based on a
careful analysis of some 22 programs, classroom observations,
and reviews of published studies, was that children who received
direct code-based instruction (emphasis on decoding or
phonics) tended to have higher achievement in the first three
grades than did children in whole-word classrooms. Although
initially, for beginning readers, whole-word classrooms performed
better on measures of comprehension and reading rate,
in later grades the advantage of decoding-based instruction became
highly general, encompassing spelling, word recognition,
and comprehension. This conclusion, in its general form, was
confirmed in later less comprehensive reports.
(1990) provided a thorough treatment of these
research reports and, more generally, an evaluation of teaching
methods in the context of research findings. Furthermore, she put
the Great Debate in its historical context, and explained why there
has been so much resistance to the direct teaching of decoding.
An emphasis on meaning and comprehension not only coincides
with the main goal of reading, but also appeals to beliefs that the
child’s experience in school should reflect purposeful learning in
authentic contexts. In that spirit, the exclusive use of commercially
published children’s literature (which is often not decodable)
has become characteristic of whole-language classrooms.
Modern phonics advocates point out that there is nothing incompatible
between these meaning values and good phonics instruction,
which aims to quickly provide the child with the basics of
the letter-sound system and practice with decodable texts while
at the same time introducing children’s literature. Like Chall,
Adams argued that phonics approaches were more successful
than nonphonics approaches in teaching children to read.
The NRC report
The NRC (the research arm of the National Academy of
Sciences) revisited this issue in its report
Preventing ReadingDifficulties in Young Children
(1998). Unlike reportsthat have focused on the question of how to teach reading,
the NRC report asked how available research findings can
inform recommendations directed at reducing children’s reading
difficulties. A distinctive feature of the NRC report was that
it considered early-childhood factors. It reviewed research on
early childhood, including research on parental influences on
cognitive and social development, family literacy, and the role
of preschools as language and literacy environments. It also reviewed
the problems of teacher preparation and made specific
recommendations about what teachers need to know about
reading. The report reviewed studies that reported low levels of
teacher preparation in the foundations of reading. Adding to
the growing call for stronger teaching preparation, the report
recognized the need to improve both the college education
(preservice training) and the in-service training of teachers.
Although the NRC report steered clear of specific curriculum
recommendations, it emphasized the importance of promoting
knowledge and practice in decoding. For example, it
recommended that kindergarten instruction “be designed to
provide practice with the sound structure of words, the recognition
and production of letters, knowledge about print concepts,
and familiarity with the basic purposes and mechanisms of
reading and writing” (). It concluded that research shows
that beginning reading “depends critically on mapping the letters
and the spellings of words onto the sounds and speech
units that they represent” (). Furthermore, counter to the
idea that somehow comprehension can proceed on its own, the
report added that “failure to master word recognition impedes
text comprehension” ().
The report’s focus on language and literacy experiences
prior to school and on the importance of decoding knowledge
as a goal of beginning reading instruction achieves a meaningful
balance. It is clear that coming to school with certain relevant
skills (some degree of phonological awareness) and dispositions
(an interest in books) eases the burden of school instruction.
It is equally clear that schooling can organize its efforts
along the lines supported by research, making sure that children
acquire the ability to decode words and have sufficient
reading practice to gain fluency and increase comprehension.
The NRP report
In 1997, the U.S. Congress asked the National Institute for
Child Health and Human Development and the Department of
Education to convene a committee to examine applying reading
research to classroom practice. Topics studied by the NRP were
alphabetics (phonological awareness and phonics), fluency, comprehension,
how teachers can be taught to teach reading better in
certification and professional development programs, and the
use of computer technology in reading instruction. Meta-analyses
based on available data on these topics were undertaken.
The NRP (2000) study is valuable for what it found in the
alphabetics area and what it did not find in the other areas (the
committee decided that there was generally not enough goodquality
research to make valid conclusions in some areas). The
report noted the validity of the research we discussed previously
in the section on phonological awareness. With respect to
phonics instruction, meta-analyses revealed that (a) systematic
phonics instruction produces significant benefits for students in
kindergarten through sixth grade and for students with reading
disabilities (regardless of SES), (b) the impact of phonics is
strongest in kindergarten and first grade, and (c) phonics must
be integrated with instruction in phonological awareness, fluency,
and comprehension. The report noted that a strong empirical
base supports the importance of instruction in phonological
awareness, in conjunction with phonics instruction, for the beginning
stages of reading instruction. However, the report also
noted that there are not enough data to draw conclusions about
the best way to teach vocabulary, fluency, and comprehension,
or the best way to prepare teachers to teach reading.
Laboratory Studies
The results of some important experimental studies suggest
two interrelated conclusions. First, learning correspondences
between letters and sounds is more productive (so there is more
transfer to new words) than learning whole words, even though
learning whole words may be faster at first. Second, providing
instruction that lets children infer these correspondences may
not be as effective as directly teaching them. The first conclusion
was demonstrated by (1964), who trained
two groups of adult subjects to respond to novel visual stimuli.
One group learned to make phoneme responses to individual
Arabic letters, whereas the other group learned to make word
responses to strings of Arabic letters. In each case there was a
1:1 correspondence between the graphic stimulus and the pronunciation.
In the case of single letters, the correspondence
was between the letter and the phoneme; in the case of the
words, the correspondence was between the printed word and
its pronunciation. Although training was faster for the wholeword
group than the letter-phoneme group, transfer showed the
opposite result: The letter-phoneme group could read many
more new words than the whole-word group.
(1967) carried out a similar study using
kindergarten children and a set of specially constructed letters.
They found that children who had learned the sounds of individual
letters could correctly read many more new words than
could children whose training required them to learn whole
words. Although both groups learned the intended pronunciation
of the new words, the word group needed twice as many
trials as the letter group to reach this level of performance.
Thus, laboratory research has long established the value of
learning letter-sound correspondences for productive transfer of
reading skill. Other laboratory studies with children have shown
how difficult acquiring these correspondences can be in the absence
of instruction ( 1984, 1996). (1991) taught
young children to read one-syllable words by pairing the words
with their meanings; for example, fat was associated with a picture
of a fat boy and bat was associated with a picture of a bat.
Then, with the pictures withdrawn, the children demonstrated
that they could read the words alone. One might think that the
children had inferred that the f made the sound /f/, because the f
was the only letter that distinguished fat from bat and the phoneme
/f/ was the only sound that distinguished the spoken word
“fat” from “bat.” But instead, the children were unable to demonstrate
that they had learned this association. When they were
asked to judge whether the printed word fun said “fun” or “bun,”
their responses were incorrect about as often as they were correct.
Thus, in at least some conditions, children do not spontaneously
infer letter-sound correspondences on the basis of being
able to read whole words. This finding reinforces the importance
of teaching children directly what they need to learn.
Classroom Studies
Classroom studies of teaching reading typically have compared
phonics instruction with some form of nonphonics
(whole-word or whole-language) instruction. As noted, there
have been many reviews of such research (1990;
1967) in addition to the NRC (1998) and
NRP (2000) reports. All of these reviews (1983,
1996; 1988) concluded that systematic phonics instruction
produces somewhat higher reading achievement for
beginning readers compared with the nonphonics alternative.
Results are most impressive for students at risk for reading failure,
such as children in Title I programs and those with learning
difficulties.
If reviews are unanimous in their support for phonics, why
does the debate continue? It continues because the debate is not
just about whether the unit of instruction should be graphemephoneme
mappings or whole words; rather, the debate is enmeshed
in philosophical differences between traditional versus
progressive education that divided American educators throughout
the 20th century and continue to this day. The progressives,
drawing upon the writings of (1938), champion student-
centered learning over traditional education’s emphasis on
the uniformity of method and curriculum. The whole-language
movement’s emphasis on teacher empowerment, child-centered
learning, and authentic literature is an outgrowth of progressive
education. Likewise, its emphasis on constructivist
psychology and on ethnographic and case study methodologies
challenges the credibility of the empirical data-based approaches
that dominate every field of scientific research
(1995). Hence, conclusions based on classroom studies
are contested on the basis of broad philosophical stances that
are inconsistent with standard assumptions about research.
It is important to note that most of the whole-language
movement’s educational values are not necessarily inconsistent
with teaching phonics. Indeed, schools of education could include
a course in the alphabetic principle and phonics instruction
for teachers in training. Then these teachers could enter
first-grade classrooms empowered with the knowledge necessary
to teach phonics without following scripted programs or
relying on worksheets ( 1994). In fact, few schools of
education offer such a course.
First-grade studies
The U.S. Office of Education conducted the Cooperative
Research Program in First Grade Reading Instruction between
1964 and 1967. These studies are commonly referred to as “the
first-grade studies.” (1967) concluded from
these studies that classroom approaches that emphasized (a)
systematic phonics, (b) reading for meaning in vocabularycontrolled
text, and (c) writing produced superior achievement
compared with approaches that relied on mainstream basal
readers that did not include phonics (only recently have systematic
phonics instruction and decodable text been incorporated
into mainstream basal reading series). They found a
definite advantage for code-emphasis approaches but concluded
that no single method worked for all teachers or all children.
Phonics proponents emphasize the first part of the
conclusion; whole-language proponents emphasize the latter
part of the conclusion. Only recently have the multilevel modeling
and statistical techniques become available to test for the
separate and interactive effects of characteristics of students,
teachers, and programs.
(1985) evaluated two programs in 20 firstgrade
classrooms. Half of these were traditional teacher-directed
classrooms in which instruction involved basal readers with
phonics drills and applications. The other half were studentcentered
classrooms in which instruction by the teacher constituted
only 35% of the day’s activity. In the latter classrooms,
reading was taught primarily by an individualized language-experience
method in which students produced their own workbooks
of stories and banks of words to be recognized (a wholelanguage
approach). Evans and Carr characterized these two
groups as decoding oriented and language oriented. Despite
some differences in emphasis regarding how teaching should
be conducted, the two groups did not differ in the amount of
time spent on reading tasks. The two groups were also matched
on relevant socioeconomic variables, and they were virtually
identical on measures of intelligence and language maturity.
The clear result, however, was that the decoding group scored
higher on year-end reading achievement tests, including comprehension
tests. Additionally, the language-oriented group did
not show higher achievement in oral language measures based
on a storytelling task. The results were consistent with the
Pittsburgh Longitudinal study (1981;
1982), which also showed quite clearly that instruction
that emphasizes the alphabetic principle does not produce
word callers who are insensitive to contextual meaning.
Effective-schools research
In the late 1970s and the 1980s, several syntheses of research
on effective teaching were written
(1984; 1986). Effectiveness was defined
in terms of correlations between classroom processes and student
outcomes. The strongest correlates of achievement were
instructional time engaged in academic tasks, classroom management,
and certain patterns of teacher-student interactions
( 1973; 1986). For
disadvantaged students, the link between explicit instruction
and achievement was notable ( 1986), a finding
supported in other classroom-observation research
(1978; 1975). Current reading research
builds upon this research by employing longitudinal,
multilevel designs that nest time within student, student within
classroom, and classroom within school. By so doing, re
searchers are able to examine the impact of teachers and
schools on the growth and outcomes of individual students
over time. Thus, although the effective-schools research provides
correlations between students’ time-on-task and achievement
outcomes, new methodologies in the 1990s made it
possible to model how teachers’ effectiveness mediates the impact
of students’ initial skill levels on skill development and
achievement. And by modeling these effects within schools, researchers
can begin to understand why students in some
schools perform so much better than students in other schools.
Best practices
During the heyday of the whole-language movement (the
mid-1980s to the end of the 1990s), educational researchers
turned away from large-scale studies of classroom instruction
and instead engaged in case studies of exemplary teachers and
culturally different students. Case studies of culturally responsive
instruction emphasized the small-group, collaborative approaches
( 1980; , 1972) and the importance of
relating classroom instruction to the home and community experiences
of children ( 1995;1991;1983)
In response to the assumption that best practices occurred in
literature-based classrooms and not in skills-based classrooms,
some recent research contrasted these two approaches
( 2000). The literature-based perspective is grounded
in reader response theory (1978), according to which
readers play a central role in the construction of meaning, and
in social-constructivist theory ( 1987), according to
which literacy is acquired in a book-rich context of purposeful
communication. Literature-based instruction emphasizes sustained
use of authentic literature for independent reading, readalouds,
and collaborative discussions. Skills-based programs,
in contrast, are typically defined as traditional programs that
use a commercially available basal reading program and follow
a sequence of skills ordered according to their difficulty. Systematic
phonics instruction falls under this definition of skillsbased
programs, whereas literature-based instruction is a more
recent term for the whole-language approach. Literature-based
instruction was found to benefit literacy acquisition in kindergarten
( 1994; 1989)
and at the elementary level (F 1991; 1995;1990)
In sum, studies of “best practices” provided ethnographic and
case studies of a small number of exemplary teachers, in contrast
to the effective-schools research, which examined process-
product correlations in a large number of classrooms in
schools of varying SES and achievement levels.
Recently, the combination of literature-based instruction
with traditional basal reading instruction has been found to be
more powerful than traditional instruction alone
( 1999; 1992; 1997)
. In fact, balanced reading instruction
seems to be replacing literature-based reading instruction
(2000; 1998), as the pendulum
of reading rhetoric swings away from whole-language approaches
toward phonics.
Evidence-based practices
Rather than simply describing teaching strategies of teachers
nominated by peers for their best practices (or correlating
processes and products as in the effective-schools research),
current research is drawing on longitudinal and multilevel designs
to examine the impact of student-level and teacher-level
variables on skill development and achievement.
While whole-language proponents were advocating the virtues
of literature-based instruction and condemning phonics
and skills-based instruction in the 1980s and 1990s, researchers
continued to examine how children’s reading development
was affected by the interaction of their characteristics with instructional
factors. These researchers (1990;
1998; 1994; H 1999; 1992)
addressed the complex mappings of phonology to orthography
that are required when learning to read English; they
also appreciated that phonics is an ad hoc system of 90 or so
rules for teaching reading that provides only a beginning focus
on grapheme-phoneme relations when, in fact, there are as
many as 500 spelling-sound connections that must be learned
( 1992). Because of the sheer number of these
connections, self-teaching is hypothesized as the mechanism
by which children continue their reading development beyond
basic levels (1995; 1995). Selfteaching
assumes a foundation of phonological awareness and
decoding skill upon which to bootstrap new orthographic information.
Several researchers have investigated how this knowledge
interacts with instruction in classroom settings. Juel and
(1985) found that if the dominant instructional
strategy in the classroom was decoding unknown words
letter by letter, children learned the strategy quicker and went
on to infer untaught letter-sound relations faster if their beginning
reading textbooks contained decodable text. This was particularly
true of children with low initial levels of skill.
(1991) found that
students in three first-grade classrooms with more letter-sound
instruction improved at a faster rate in reading and spelling
than students in three first-grade classrooms with less lettersound
instruction. Initial scores on phonemic segmentation
tasks predicted reading and spelling outcomes for all children.
Exploratory data analysis revealed that children who were slow
to improve in phonemic segmentation were also slow to spell
and read phonetically, especially among children receiving less
letter-sound instruction ( 1994).
In a subsequent study, (1998) examined the
reading development of 285 first and second graders in 66
classrooms in eight Title I schools to determine how the nature
of letter-sound instruction interacted with entering skill in phonological
awareness. These students scored in the bottom 18%
on the district’s early literacy assessment. Some teachers par
ticipated in one of three kinds of experimental classroom reading
programs, and some participated in an unseen control
group involving the district’s standard curriculum. Instruction
in all four groups included a language arts emphasis on writing
and read-alouds from good-quality literature. The three types
of experimental programs were differentiated by the kind of
phonics instruction: (a) direct instruction in letter-sound correspondences
practiced in decodable text (direct code), (b) less
direct instruction in systematic sound-spelling patterns embedded
in authentic literature (embedded code), and (c) implicit
instruction in the alphabetic code while reading authentic text
(implicit code). The 53 teachers for these three groups participated
in ongoing generic staff development as well as training
specific to their program. The remaining 13 teachers participated
in the district’s whole-language staff development, and
their students formed a control group for the implicit-code approach.
Children receiving direct-code instruction improved in word
reading at a faster rate and had higher word recognition skills
than those receiving implicit-code instruction. The improvement
was particularly impressive for students who began the
year with low phonological awareness. Children receiving direct-
code instruction, in contrast to children in the other
groups, also had word-reading and reading comprehension
skills that approximated national averages at the end of the
year. Despite the direct-code group’s generally good outcomes,
35% of them remained below the 30th percentile in reading
achievement. (2000) multiplied the percentage of students
remaining below the 30th percentile (35% in this case)
by the percentage of the distribution of reading scores represented
by the students at the beginning of the year (18% in this
case) to derive a population-based failure rate. Accordingly,
Torgesen computed the population-based failure rate for the
(1998) study as 6% (35% 18%).
(1998) pointed out that a failure rate of 6% represents a
substantial reduction in the 15% to 20% of students with reading
difficulty in the United States. The finding that explicit instruction
in letter-sounds can prevent reading difficulties for
children at risk for reading failure because of poor phonological
awareness or lack of home literacy has been demonstrated a
number of times
(1985; 1998; 2000; 1999;1996;1980)
The other side of this apparent ability-by-treatment interaction
is that first graders who enter with middle-range literacy
skills benefit from classrooms with ample opportunities to read
trade books (2000). In an investigation of 4,872 kindergarten
children in 114 classrooms where reading curricula (informed
by ongoing professional development) varied in the
degree of teacher choice and in the degree to which phonological
awareness was incorporated, less teacher choice and more
explicit incorporation of phonological awareness was associated
with less variability across teachers in letter knowledge
and phonological awareness at the end of kindergarten and in
reading achievement at the end of first grade
(2001). More teacher choice and a moderate number of phonological-
awareness activities (mostly in the form of letter-sound
instruction) were associated with more outliers—high-scoring
children—at the end of kindergarten and first grade.
The effects of instruction can persist beyond the first grade,
and they can be manifest in spelling as well as reading.
(1998) compared
spelling in third-grade children who had whole-language instruction
throughout school and third graders who instead had
received phonics instruction. The phonics-instructed children
were better spellers, and their spelling of pseudowords included
more conventional, phonologically accurate patterns.
In general, it appears that the clarity and organization of research-
based components in the curriculum make a difference
to reading outcomes. However, out-of-the-box implementations
of basal reading programs are not likely to be effective. Again,
ongoing professional development that provides the rationale
for each component of reading (and spelling) instruction and
provides classroom coaching to deal with the pacing of instruction,
classroom management, and grouping of students is what
helps teachers develop successful readers. Expecting teachers
to put aside their basal readers and create their own curricula is
not realistic given the lack of resources and of the knowledge
base to do so ( 1994).
Summary
Since the 1960s, classroom studies of reading methods have
consistently shown better results for early phonics instruction
compared with instruction emphasizing meaning at the level of
words and sentences. This effect is particularly strong for children
at risk for reading failure because of lack of home literacy
or weak phonological-awareness skills (children who have attention
problems, chronic ear infections, articulation problems,
or a history of dyslexia in their families). This interaction between
children’s characteristics and curricular focus is moderated
by instructional factors such as teachers’ knowledge and
competency. Thus, the kinds of materials (curriculum) and instructional
strategies used interact with a child’s stage of reading
development in determining the child’s success in learning
to read. This fact has important policy implications for improving
literacy levels nationwide. Yet in the national arena, reading
methods have become highly politicized, and the Great Debate
has turned into the reading wars. Proponents of literature-based
instruction ( 2000;
2000; 1998) have attacked research supporting
skills-based instruction, despite the fact that this research investigates
processes fundamental to learning to read rather than
skills-based instruction per se. In return, skills-based researchers
have pointed out how these attacks have misrepresented the
research and are based primarily on philosophical objections
(2000;
2000). Despite the controversy, there is no question
that continued scientific study of what constitutes effective
reading instruction will benefit children and teachers by improving
understanding of how particular children best learn to read.
SUMMARY AND CONCLUSIONS
In this monograph, we discussed a wide range of topics relevant
to how children learn to read. We discussed evidence
from developmental psychology concerning both the characteristics
of children’s language competency when they enter
school and the nature of early reading development. We also
discussed research on skilled reading (from cognitive psychology,
cognitive neuroscience, and connectionist modeling) and
the implications of this work for learning to read and teaching
methods. Included in our discussion were arguments based on
linguistic analyses, data from studies of skilled reading and
brain activity during reading, and implications that follow from
the implementation of connectionist models. Finally, we presented
evidence from laboratory and classroom studies regarding
the most effective methods for teaching reading.
From all these different perspectives, two inescapable conclusions
emerge. The first is that mastering the alphabetic principle
is essential to becoming proficient in the skill of reading,
and the second is that instructional techniques (namely, phonics)
that teach this principle directly are more effective than
those that do not. This seems to be especially the case for children
who are at risk in some way for having difficulty learning
to read. It is also the case that the absence of instruction in
phonics may increase the number of children at risk for becoming
poor spellers (particularly because whole-language instruction
often tolerates incorrect spellings). We do not deny the
value inherent in various principles of whole-language teaching
methods. As we noted many times throughout this monograph,
instructional techniques that move beyond phonics
practice to ensure the application of alphabetic principles to
reading clearly support the process of learning to read. We also
emphasized that the child’s learning is every bit as important as
the teacher’s instruction. Obviously, using whole-language activities
to supplement phonics instruction helps make reading
fun and meaningful for children. Such activities may be most
beneficial to children from environments where reading is not
highly valued. But, at the end of the day, phonics instruction is
critically important because it does help the beginning reader
understand the alphabetic principle and apply it to reading and
writing. Thus, the empirical data clearly indicate that elementary
teachers who make the alphabetic principle explicit are
most effective in helping their students become skilled, independent
readers.
0 comments:
Post a Comment