The writing system
Grapheme: a unit of written language that corresponds to a phoneme.
There is much more variability in the structure of written languages than there is in
spoken languages. Whereas all spoken languages utilize a basic distinction between
consonants and vowels, there is no such common thread to the world’s written languages.
In alphabetic scripts, the basic unit represented by a grapheme is essentially a phoneme.
In idiographic (or logographic) languages like Chinese, each symbol is equivalent to a
A preliminary model of reading
Our ability to read nonword on the one hand and irregular words on the other suggests the
possibility of a dual-route model of naming. The classic dual-route model has two routes
for turning words into sounds. There is a direct access, or lexical route, which is needed
for irregular words. This must at least in some way involve a direct link between print
and sound. There is also a grapheme-to-phoneme conversion route (GPC), or the indirect
or nonlexical or sublexcial route), which is used for reading nonwords. This route carries
what is called phonological recoding. It is useful for children learning to read by
sounding out words letter by letter.
Modern dual-route theorists see reading as a “race” between these routes. When we see a
word, both routes start processing it. For skilled readers, most of the time the direct route
is much faster, so it will usually win the race and the word will be pronounced the way
that it recommends.
Relation of the dual-route model to other models
The dual route is the simplest version of a range of possible multi-route or parallel coding
models, some of which posit more than two reading routes.
The primary goal of reading is not getting the sound of a word, but getting meaning. In
the early stages of learning to read, children get to the meaning through the sound; that is
they spell out the sound of the words, and then access meaning as they recognize those
sounds. Some researchers believe adults primarily get to meaning by going from print to
phonology and then to meaning, an idea called phonological mediation. But most
believe that in skilled adults, most of the time, there is a direct route from print to
The Processes of Normal Reading
According to the dual-route model, there are two independent routes when naming a
word and accessing the lexicon: a lexical or direct access route and a sublexical or
grapheme-phoneme conversion route.
According to the dual-route model, the pronunciation of all nonwords should be
assembled using the GPC route. This means that all pronounceable nonwords should be
alike and their similarity to words should not matter. However, pronounceable nonwords
are not all alike.
The Pseudohomophone effect
Pseudohomophones are pronounceable nonwords that sound like words when
pronounced (like brane, which sounds like brain). Pseudohomophones are more confusable with words than other types of nonwords are. Participants are faster to name
them, but slower to reject them as nonwords that control nonwords.
Is the effect caused by the phonological or visual similarity between the nonword and
word? Martin and Taft argued that it’s the visual similarity that is important;
pseudohomophones are more confusable with words than other nonwords because they
look more similar to words than non-pseudohomophones do, rather than because they
sound the same.
Glushko’s experiment: Lexical effects on nonwords reading
Glushko performed an experiment on the effect of the regularity of the word-neighbors of
a nonword on its pronunciation. For example, naming latencies to nonwords such as
“tave” were significantly slower than to ones such as “taze”, since “taze” has regulary
pronounced neighbors (maze, gaze, laze, etc.). That is, reaction times to nonwords that
have orthographically irregular spelling-to-sound correspondence word-neighbors are
slower than to other nonword controls.
In summary, Glushko found that the pronunciation of nonwords is affected by the
pronunciation of similar words, and that nonwords are not the same as each other. These
are lexical effects on nonword processing.
More on reading nonwords
Pronunciation can be affected by the pronunciation of a preceding prime word. For
example, “bead” biases a participant to pronounce “yead” to rhyme with it, whereas the
prime “head” biases participants to the alternative pronunciation. Nonwords primed by a
semantic relative of a phonologically related also affects the pronunciation. This suggests
that nonword pronunciation involves more than just GPC.
Evaluation of research on reading nonwords
These data do not fit the simple version of the dual-route model. The pronunciation of
nonwords is affected by the pronunciation of visually similar words. That is, there are
lexical effects in nonword processing: the lexical route seems to be affective the non-
According to the dual-route model, words are accessed directly by the direct route. This
means that all words should be treated the same in respect of the regularity of their
spelling-to-sound correspondences. An examination of the data reveals that this
prediction does not stand up.
One problem is that pronunciation regularity affects response times, although in a
complex way. A well-replicated finding is that of an interaction between regularity and
frequency: regularity has little effect on the pronunciation of high-frequency words, but
low-frequency regular words are named faster than low-frequency irregular words. High-
frequency words can be sensitive to regularity, but the effect of regularity is moderated
by the number and frequencies of their “friends” and “enemies” (words with similar or
conflicting pronunciations. Once we control for the generally unusual appearance of
irregular words, regularity and consistency only seem to affect naming times, not lexical
decision times. Age-of-acquisition has a similar effect to frequency, and gives rise to a
similar interaction: consistency has a much bigger impact on naming time for late-
acquired than early-acquired words. Late-acquired and low-frequency inconsistent words
stand out because they can make use of the network structure of other consistent words.
In general, regularity effects are more likely to be found when participants have to be more conservative, such as when accuracy rather than speed is emphasized. The finding
that regularity affects naming might appear problematic for the dual-route model, but
makes sense if there is a race between the direct and indirect routes.
Glushko’s experiment: Results from words
Glushko found that words behave in a similar way to nonwords, in that the naming times
of words are affected by the phonological consistency of neighbors. The naming of a
regular word is lowed down relative to that of a control word of similar frequency if the
test word has irregular neighbors.
In addition, he found true naming errors of over-regularization: for example “pint” was
sometimes given its regular pronunciation – to rhyme with “dint.”
Neighbors: friends or enemies (whether or not it is regular)
The number of both friends and enemies affects naming times.
Effects of neighborhood size were found in both the naming and the lexical decision
tasks. Responses to words with large neighborhoods were faster than words with small
neighborhoods (although this may be moderated by frequency). Parkin found more of a
continuum of ease-of-pronunciation than a simple division between regular and irregular
words. This suggests that a binary division into words with regular and irregular
pronunciations is no longer adequate.
A more satisfactory but complex categorization between regular and irregular words was
provided: (1) the regularity of the pronunciation with reference to spelling-to-sound
correspondence rules, (2) the agreement with other words that share the same body. Body
is the same as rime: the final vowel and terminal consonants.
In summary, not all words behave the same way and the regularity of pronunciation of a
word affects the ease with which we can name it.
The role of sound in accessing meaning: phonological mediation
There is some experimental evidence suggesting that a word’s sound may have some
influence on accessing the meaning. For example, in categorization of words, if a “no”
word (not belonging to the category in question) is a homophone of a “yes” word,
participants make a lot of false positive errors – responding “yes” instead of “no.” We
also take longer to respond to homophones in a lexical decision task.
There is considerable evidence that the recognition of a word can be influenced by its
phonology. The dominant view is that this influence arises through the indirect route,
although word recognition is primarily driven by the direct route – a view that has been
labeled the weak phonological perspective. The alternative, strong phonological view
that we primarily get to the meaning through sound is called phonological mediation.
There is a great deal of controversy about the status of phonological mediation. Ex) Study
found that homophones were read with longer gaze duration – that is, they were
processed as though they were lexically ambiguous, even though the orthography should
have prevented this. This result is only explicable if the phonology is in some way
interfering with the semantic access.
On the other hand, researchers showed that prior phonological access only happens with
low-frequency homophones. Poor readers are more likely to have to access phonology in
order to access semantics, whereas good readers primarily activate semantics first.
Researchers reported fixation data on homophones that suggested the meaning of a word
is accessed first whereas the phonological code is accessed later, probably post-access. They found that gaze duration times were longer on an incorrect homophone (brake
instead of break) and that the fixation times on the incorrect homophone were about the
same as on a spelling control (broke). This means that the appropriate meaning must have
been activated before the decision to move the eyes, and that the phonological code is not
activated at this time. (If the phonological code had been accessed before meaning then
the incorrect homophone would sound all right in the context, and gaze durations should
have been about the same.) The phonological code is accessed later, however, and
influences the number of regressions to the target word.
Semantic categorization task was used to examine phonological mediation. Results
suggested that the sound of a word does not need to be accessed on the route to accessing
While some researchers interpret masked phonological priming as supporting
phonological mediation (why else should early phonological activation happen so early
unless it is essential?), other researchers point out that these effects are very sensitive to
environmental conditions, and are not always reliably found.
Data suggest that the sound of a word is usually accessed at an early stage. However,
there is much evidence suggesting that phonological recoding cannot be obligatory in
order to access the word’s meaning. However, there is much evidence suggesting that
phonological recoding cannot be obligatory in order to access the word’s meaning. For
example, some dyslexics cannot pronounce nonwords, yet can still read many words.
There are a number of alternative explanations to explain the data showing phonological
mediation. First, although phonological recoding prior to accessing meaning may not be
obligatory, it might occur in some circumstances. Given there is a race between the
lexical and sublexical routes in the dual-route model, if for some reason the lexical route
is slow in producing an output, the sublexical route might have time to assemble a
conflicting phonological representation. Second, there might be feedback from the speech
production system to the semantic system, or the direct-access route causes inner speech
that interferes with processing. Third, it is possible that lexical decision is based on
Silent reading and inner speech
Although it seems unlikely that we have to access sound before meaning, we do routinely
seem to access some sort of phonological code after accessing meaning in silent reading.
Subjective evidence for this is the experience of “inner speech” while reading. However
this inner speech cannot involve exactly the same processes as overt speech because we
can read silently much faster than we can read aloud.
Inner speech seems to assist comprehension; if it is reduced, comprehension suffers for
all but the easiest material. Some argued that whichever route is used for lexical access in
reading, at least part of the phonological code of each word is automatically accessed - in
particular we access the sounds of beginnings of words.
Although there is some debate about the precise nature of the phonological code and how
much of it is activated, it does seem that silent reading necessarily generates some sort of
phonological code. This information is used to assist comprehension, primarily by
maintaining items in sequence in working memory.
The role of meaning in accessing sound
Phonological mediation means that we might access meaning via sound. Sometimes we
need to access the meaning before we can access a word’s sound (ex. Bow, tear – homographs that have two different pronunciations.
Researchers showed that there is an effect of imageability on skilled reading such that
there is a three-way interaction between frequency, imageability, and spelling
consistency. People are particularly slow and make more errors when reading low-
frequency exception words with abstract meanings. At least some of the time, we need to
access a word’s semantic representation before we can access its phonology.
Does speed reading work?
According to psychological research, as you increase your reading speed above the
normal rate, comprehension declines. The speed readers performed only slightly better
than a group of people who skimmed through the passages. They did as well as the
normal readers on the general gist of the text, but were worse at details. In particular, they
could not answer questions when the answers were located in places where their eyes had
Eye movements are the key to why speed reading confers limited advantages. For a word
to be processed properly, its image has to land close to the fovea and stay there for a
sufficient length of time. Speed reading is nothing more than skimming through a piece
of writing. But if you have sufficient prior information about the material, your level of
comprehension can be quite good.
Evaluation of experiments on normal readings
There are two major problems with a simple dual-route model. First there are lexical
effects on reading nonwords, which should be read by a non-lexical rout that is
insensitive to lexical information. Second, there are effects of regularity of pronunciation
on reading words, which should be read by a direct, lexical route that is insensitive to
A race model fares better. Regularity effects arise when the direct and indirect routes an
output at about the same time, so that conflict arises between the irregular pronunciation
proposed by the lexical route and the regular pronunciation proposed by the sublexical
Skilled readers have a measure of attentional or strategic control over the lexical and
sublexical routes such that they can attend selectively to lexical or sublexical information.
People need not change the extent to which they rely on sublexical information, but
instead might be responding at different points in the processing of the stimuli.
There is further evidence for strategic effects in the choice of route when reading. Lexical
access is performed so quickly for high-frequency words that there is little scope for
sublexical involvement, but with low-frequency words or in difficult conditions people
can devote more attention to one route or the other.
The neuropsychology of adult reading disorders: acquired dyslexia
Acquired disorders: a disorder caused by brain damage is acquired if it affects an ability
that was previously intact.
Developmental disorders: a disorder where the normal development or acquisition of a
process is affected.
Dyslexia: disorder of reading
Dysgraphia: disorders of writing
Damage to the left hemisphere will generally result in dyslexia, but as the same sites are
involved in speaking, dyslexia is often accompanied by impairments to spoken language processing.
Central dyslexias involve central, high-level processes and peripheral dyslexias involve
lower level processes.
If the dual-route model of reading is correct, then we should expect to find a double
dissociation of the two reading routes.
Surface dyslexia: a type of dyslexia where the person has difficulty with exception words.
Surface dyslexics often make over-regularization errors when trying to read irregular
words aloud (“steak” as “steek”). On the other hand, their ability to read regular words
and nonwords is intact. In terms of the dual route model, the most obvious explanation of
surface dyslexia is that these patients can only read via the indirect, non-lexical route:
that is, it is an impairment of the lexical (direct access) processing route. The
comprehension of word meaning is intact in these patients.
Imageability: a semantic variable concerning how easy it is to form a mental image of
word: “rose” is more imageable than “truth.”
The more pure cases of surface dyslexia, Type I, are highly accurate at naming regular
words and pseudowords. Other patients, Type II’ also show some impairments at reading
regular words and pseudowords. The reading performance of Type II patients may be
affected by lexical variables such that they are better at reading high-frequency, high
imageability words, better at reading nouns than adjectives, etc. Type II patients must
have an additional, moderate impairment to the non-lexical route, but the dual route
model can nevertheless still explain this pattern.
Phonological dyslexia: a type of dyslexia where people can read words quite well but are
poor at reading nonwords (or pseudowords).
Phonological dyslexics find irregular words no harder to read than regular words. These
symptoms suggest that these patients can only read using the lexical route, and therefore
that phonological dyslexia is an impairment of the non-lexcial (GPC) processing route.
For those patients who can pronounce some nonwords, nonword reading is improved if
the nonwords are psedohomophones (such as “nite” for “night”). Low-frequency, low-
imageability words are also poorly read. They sometimes make derivational errors
(reading a word as a grammatical relative of the target, such as reading “performance” as
“performing”), and visual errors (in which a word is read as another with a similar visual
appearance, such as reading “perform” as “perfume”).
Phonological dyslexia can result from disruption of either orthographic or phonological
Disruption of graphemic parsing: when patients are worse at reading graphemically
compex nonwords than graphemically simple nonwords), but show no advantage for
Distruption of phonological processing: when patients are better at reading
pseudohomophones than non-pseudohomophones, but show no effect of orthographic
A three-stage model of sublexical processing has emerged: (1) a graphemic analysis stage
parses the letter string into graphemes. (2) a print-to-sound conversion stage assigns
phonemes to graphemes. (3) the phonemic blending stage assembles the sounds into a
phonological representation. There are patients whose behavior can best be explained in terms of disruption of each of these stages.
The possibility why some people with phonological dyslexia have difficulty reading
function words is because they are so abstract. (However, there is evidence to the
People with phonological dyslexia show complex phonological problems that have
nothing to do with orthography. Phonological processing may be one aspect of a general
impairment of phonological processing. An important piece of evidence in favor of this