The Quarterly Journal of Experimental Psychology
Section A

ISSN: 0272-4987 (Print) 1464-0740 (Online) Journal homepage: http://www.tandfonline.com/loi/pqja20

The pseudohomophone effect: The role of visual
similarity in non-word decisions
Randi C. Martin
To cite this article: Randi C. Martin (1982) The pseudohomophone effect: The role of visual
similarity in non-word decisions, The Quarterly Journal of Experimental Psychology Section A, 34:3,
395-409, DOI: 10.1080/14640748208400851
To link to this article: http://dx.doi.org/10.1080/14640748208400851

Published online: 29 May 2007.

Submit your article to this journal

Article views: 925

View related articles

Citing articles: 59 View citing articles

Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=pqja20
Download by: [Pepperdine University]

Date: 28 September 2017, At: 06:51

Quartwly Journal of Experimental Psychology (1982)34A, 395-409

THE PSEUDOHOMOPHONE EFFECT:
THE ROLE OF VISUAL SIMILARITY IN
NON-WORD DECISIONS

Downloaded by [Pepperdine University] at 06:51 28 September 2017

RAND1 C. MARTIN
Department of Psychology, The Johns Hopkins University Baltimore,
Maryland 21218, U.S.A.
?’he pseudohomophone effect, that is, the finding that non-words that are pronounced like words (e.g. MEEN) take longer to reject in a lexical decision task than
other pronounceable non-words (e.g. NEEN), has been used to support the hypothesis of phonological recoding in lexical access. The lexical decision experiments
reported here matched pseudohomophones to control non-words that were as
visually similar to words as were the pseudohomophones. For both normal and
aphasic subjects, reaction times to reject the pseudohomophones were no longer
than those for the visual controls. However, the pseudohomophones did take
longer to reject than other pronounceable non-words which were not as visually
similar to words. The results suggest that most previous findings of the pseudohomophone effect result from the greater visual rather than phonological similarity
of the pseudohomophones to words. The absence of a phonological effect on the
non-words in the present study implies that phonological coding is an optional
rather than a slow, but obligatory process.

Introduction
Because reading is usually learned by mastering spelling-to-sound correspondences,
and because many people report hearing “inner speech” as they read (Brown,
1970), it has often been argued that accessing a lexical entry from print involves
an intermediate stage of grapheme-to-phoneme conversion (Gibson 1970;
Rubenstein et al., 1971). However, recent attempts to provide experimental
evidence for phonemic recoding have shown that only under certain experimental
conditions is evidence for phonemic recoding found, and that, in many cases,
skilled adult readers use a direct, or visual, route for accessing the lexicon (see
Coltheart, 1978. and McCusker et al.. 1981, for recent reviews).
One explanation offered for the task-dependent nature of the evidence for
phonemic recoding in lexical access is that phonemic recoding is an optional
strategy that subjects will employ depending on the advantages and disadvantages
of its use in a particular task. Several studies using a lexical decision task, that
is, in which subjects must decide if a letter string is a word, have found that
evidence for phonemic recoding depends on the type of non-words used in the

0 1982 The Experimental Psychology Society

0272-4987/82/030395+15 803.00/0

395

Downloaded by [Pepperdine University] at 06:51 28 September 2017

396

R. C . MARTIN

experiment. For example, Meyer, Schvaneveldt and Ruddy (I974), in a doubleword lexical decision task, found evidence for phonemic recoding when the nonwords consisted of legal (conforming to English spelling patterns), pronounceable
letter strings. They found that reaction times to rhyming pairs of visually similar
words (e.g. pear-bear) were faster than those to unrelated pairs, while reaction
times to non-rhyming pairs of visually similar words (e.g. dead-bead) were longer
than those for the unrelated pairs. They explain their results by suggesting that
the grapheme-to-phoneme conversion rules used in developing the phonemic
representation for the first word are applied to the second when the two words are
visually similar, resulting in faster lexica1 access for the rhyming pairs and causing
a delay for the non-rhyming pairs. Shulman et at. (I@),
in a replication of the
Meyer et al. study, found no evidence for phonemic recoding when the non-words
were consonant or random letter strings. I n their study, both the rhyming and
non-rhyming word pairs had faster reaction times than the unrelated pairs, indicating a facilitating effect for visual similarity.
A possible objection to the Shulman et al. finding is that subjects in the consonant or random letter conditions could have based their “yes)’ decisions on whether
the letter strings looked suficiently like words without lexical access ever occurring.
T o counter this objection, Shulman et al., report that they found a priming effect
for semantically related word pairs, and they conclude that lexical access had
indeed occurred in their task. Shulman et al. suggested that subjects use phonemic
recoding when the non-words are legal, pronounceable letter strings as a check to
insure that the non-words are not rare words. When the non-words are visually
distinct from real words, no such check is necessary.
Davelaar et al. (1978) compared reaction times to the less frequent member of
a homophone pair (e.g. witch of the pair which-witch) to control words matched
in frequency to the less frequent homophone. They reasoned that if subjects
use grapheme-to-phoneme conversion in accessing the lexicon, and if the first
word accessed from a phonological representation is the most frequent item with
that representation, then the more frequent member of the homophone pair
should be accessed first when reading the less frequent member of the pair. A
subsequent visual check would indicate that this was not the correct word, and a
second attempt at lexical access would have to be made. Thus, the less frequent
members of the homophone pairs should have longer reaction times than the
matched control words. When the non-words were legal, pronounceable letter
strings (e.6. slint), evidence supporting the phonemic recoding hypothesis was
found. However, when the non-words were pseudohomophones (i.e. letter strings
that sound like words when pronounced such as brane), there was no significant
difference between the reaction times for the homophones and the control words.
Davelaar and co-workers hypothesized that when the non-words were pseudohomophones, subjects discovered that using phonemic recoding would lead to
errors on the non-words, and thus abandoned this strategy.
A second possible explanation for the task dependent nature of evidence for
phonemic recoding in lexical access derives from the dual encoding hypothesis
(Mcyer et al., 1974; Coltheart, 1978). According to this hypothesis, both visual
access and phonemic access via grapheme-to-phoneme conversion occur auto-

Downloaded by [Pepperdine University] at 06:51 28 September 2017

VISUAL SIMILARITY I N NON-WORD DECISION

397

matically in parallel, with phonemic access being a slower process (Coltheart,
1978). Support for this position comes from a study by Stanovich and Bauer
(1978), who compared reaction times to regularly spelled words (e.g. bake) and
exception words (e.g. sword) in a lexical decision task. When the usual reaction
time task instructions stressing both speed and accuracy were given, the subjects’
reaction times were significantly longer for exception words than for regular
words, indicating that phonemic recoding was involved. However, when a
response deadline technique (stressing speed) was used, no difference was found
between the two types of words. Stanovich and Bauer suggest that when instructions do not stress speed over accuracy subjects adopt the conservative strategy
of awaiting the outcome of both visual and phonological access before making a
decision. When speed is stressed, subjects use only the outcome of visual access
which is available sooner.
The dual encoding model has also been used to explain why in some experiments
no effect for phonemic recoding has been found in the “yes” decisions (i.e. the
word trials), but an effect has been found on the “no” decisions (i.e. the non-word
trials). For example, Coltheart et al. (1977) found no effect for phonemic recoding
in comparing homophones to control words in the “yes” decisions of a lexical
decision task, but did find that it took longer to make “no” decision to pseudohomophones than the control non-words. T h e lexical access model proposed by
Coltheart et al. assumes that subjects adopt deadlines for deciding that no lexical
entry will be found for a non-word. Since most words will be identified before a
deadline occurs, reaction times for non-words will tend to be longer than reaction
times for words. During this longer decision time for non-words it is possible
that phonological recoding will have occurred before a decision is made, resulting
in phonological effects for the “no” decisions. On the other hand, “yes” decisions
could all be made before this phonological access is complete, resulting in no
effect on the “yes” decisions.
‘The experiment cited previously as evidence supporting phonemic recoding as
an optional strategy could also be explained by the dual encoding theory. If it
were assumed that one could decide whether or not to wait for the outcome of
the phonological process on word trials, depending on the nature of the non-words,
then evidence for phonemic coding should only be obtained when subjects are
motivated to wait for the output of the phonological route before making a decision
on the word trials. Stanovich and Bauer’s results would suggest that subjects
can control the point at which they will make a decision. Thus, most of the
literature on lexical decision could be explained either in terms of phonemic
encoding as an optional strategy or in terms of the dual encoding hypothesis. T h e
only evidence favouring the dual encoding hypothesis is the result discussed above
concerning the evidence for phonemic recoding on “no” decisions but not on
“yes” decisions.
However, it is possible that the pseudohomophone effect as reported by
Coltheart and his colleagues, as well as by others (e.g. Rubenstein et al., 1971 ;
Patterson and Marcel, 1977) is a visual rather than a phonological effect. Pseudohomophones, in order that they sound like a word, tend to have many letters in
common with the intended word, and thus, may be more visually similar to words

Downloaded by [Pepperdine University] at 06:51 28 September 2017

398

R. C. MARTIN

than are other non-words. Coltheart (1978), however, dismisses the visual
similarity argument on the basis of results obtained in a lexical decision task in
which he attempted to control the visual similarity of the pseudohomophones and
other non-words to words (Coltheart et al., 1977). The visual controls for the
pseudohomophones were created by changing one letter of the pseudohomophone
to result in a non-word that would not be pronounced like a word, but would be
close in visual similarity to a word, for example, changing wurld to murld.
Although this results in control non-words that may be close in visual similarity to
real words, they are still not as visually similar to words as are the pseudohomophones. In the above example, wurld differs from world by one letter, while
murld differs from world by two letters. Thus, for many of the pseudohomophonecontrol non-word pairs, the control non-word differed from the word by one more
letter than the pseudohomophone. While it might be the case that the control
non-word is visually similar to some other word (e.g. murky), this was not
systematically part of the experimental design. T o check the similarity of the
pseudohomophones and control non-words of Coltheart et al. to words in general,
their N measure was used which counts the number of words which can be created
from the non-words by changing one letter. The average N measure for their
pseudohomophones was 5.7, while for the control non-words it was 3.4, a
difference that was significant, t38=4.69, P<o-oI. The reason for the lower N
value for the control non-words lies partly in the difference mentioned above in the
visual similarity of the pseudohomophones and visual controls to the intended word
for the pseudohomophone. Also, when one letter of the pseudohomophone was
changed to create the control non-word, an infrequent letter or combination of
letters often resulted, a probable consequence of trying to insure that the control
non-word not sound like a word. For example, changing the L in the pseudohomophone hele to F for the visual control hefe changed the N value from 7 to I .
It remains possible therefore, that the pseudohomophone effect is due to visual
rather than phonological similarity to words, and that phonemic recoding is an
optional strategy rather than a process proceeding in parallel with visual access,
but at a slower rate.
Another finding that might be taken as evidence in favour of a dual encoding
hypothesis comes from research on brain damaged patients with a syndrome
known as deep, or phonemic, dyslexia. These patients show in a number of
tasks an inability to do grapheme-to-phoneme conversion (Shallice and Warrington,
1975; Patterson and Marcel, 1977; Saffran and Marin, 1977). They cannot pronounce non-words, recognise rhymes or homophones, nor recognise the intended
referent of a pseudohomophone. They appear to rely solely on the visual aspects
of words in accessing the lexicon in reading tasks (Saffran and Marin, 1977).
Besides showing an inability to do grapheme-to-phoneme conversion, these
patients make a number of errors in reading aloud, the most distinctive of which
are their semantic errors (e.g. reading “accident” for “disaster”). These patients
also make a good percentage of visual errors (e.g. reading “chair” for “charm”).
Rather than assuming separate semantic and visual impairments in deep dyslexia,
some have suggested that these errors are what might be expected when a person
is unable to do grapheme-to-phoneme conversion (Shallice and Warrington, 1975;

Downloaded by [Pepperdine University] at 06:51 28 September 2017

VISUAL SIMILARITY I N NON-WORD DECISION

399

Newcombe and Marshall, 1980). Using a modified version of Morton’s logogen
model, it has been proposed that when a letter string is presented a number of
logogens for words visually similar to the letter string, or semantically related to
the word, are activated, and the subject selects the most highly activated unit for
output. According to the model, it is possible that a more frequent but incorrect
word might have a higher activation level than the correct word since activation
level is assumed to be related to frequency (Morton, 1969). In a normal subject,
the output of the incorrect word is prevented because the output of the graphemeto-phoneme route will be available for comparison with the intended output
derived from visual access. Since the deep dyslexic cannot perform this check,
reading errors result.
Thus, according to this model of visual access, phonemic encoding is not an
optional strategy, but an important mechanism in lexical access that ensures access
to the correct word. However, as Coltheart (1980) has argued elsewhere, a large
number of studies indicate that lexical access often occurs in normals without any
evidence of phonemic recoding, and that there is no evidence that access without
phonemic recoding should lead to the kind of errors found in deep dyslexia.
The present experiments were directed towards examining the pseudohomophone effect as evidence for dual encoding in lexical access. The first experiment
examined normal subjects’ performance in a lexical decision task in which control
words were created for the pseudohomophones to be as nearly equal as possible in
visual similarity to words as were the pseudohomophones. In the second experiment, performance on the same task was studied for two patients who show an
inability to do grapheme-to-phoneme conversion.

Experiment I
To create visual controls for the pseudohomophones used in this experiment,
two words were found of approximately the same frequency such that a change
made in one resulted in a pseudohomophone, while a similar change made in the
other resulted in a pronounceable non-word, but not in a pseudohomophone. For
example, the A was changed to E in mean resulting in meen. The control nonword was created by changing the A to E in plan resulting in plen. Besides
matching the frequency of the two initial words, the pseudohomophone and control
non-words were matched in number of letters and syllables. A check was also
made to insure that the pseudohomophones were not only equally similar as the
visual controls to particular words, but also to words in general using the N
measure. For the pseudohomophones, the average N measure was 3.88, and for
the visual controls 3.64, a difference which was far from significant, t24=0.28,
P>0.20.

T o provide a comparison with the Coltheart et al. study, a second type of control
non-word (to be called the “approximate visual control”) was created by changing
one letter of the pseudohomophone (e.g. changing meen to neen). As in the
Coltheart et al. study this resulted in non-words having a significantly lower N
measure (2.24) than the pseudohomophones, t24=2.84, P<o.oI. A third type
of control non-word was used (“distant visual control”) that consisted of a

Downloaded by [Pepperdine University] at 06:51 28 September 2017

400

R. C . MARTIN

pronounceable but infrequently encountered combination of letters in order that
the resulting string would be very different visually from any real word (e.g. byph).
The N measure for the distant visual controls was 0.48, again significantly different
from the pseudohomophones, t24=5-98, P<o-oI. The distant controls were
matched to the pseudohomophones in number of letters and syllables. Lists of the
pseudohomophones and control non-words are given in the Appendix. The words
used were matched in number of letters and syllables to the non-words and had a
frequency range similar to that of the target words for the pseudohomophones and
the visual control non-words.
To insure that subjects would agree that the pseudohomophones were actually
pronounced like the intended words and that the other non-words were not pronounced like any words, a pilot study was conducted. Twenty-three subjects
were given lists of the non-words on a sheet of paper in a random order. ‘rhey
were’asked to judge if the letter string would be pronounced like a word and if so
to write the word beside the letter string.
For each pseudohomophone the number of subjects who judged that the pseudohomophone was pronounced like the intended word was computed. Averaging
across the 25 pseudohomophones, this number was 20.7 out of 23 (90%). For
each of the control non-words, the number of subjects who thought that a nonword was pronounced like any word was calculated. The average number of
subjects was 0.35 out of 23 (1.5%) for the visual controls, 0.30 out of 23 (1.3%)
for the approximate visual controls, and 0.92 out of 23 (2.4%) for the distant visual
controls. The slightly higher percentage for the distant visual controls was due
to the inclusion of remm that several subjects said was pronounced like the
acronym “rem” (i.e. Roentgen Equivalent in Man).
If the visual similarity of the non-words to words is the determining factor in
reaction times, then we would not expect a difference in “no” reaction times
between the pseudohomophones and visual controls, even though the pseudohomophones are pronounced like words while the visual controls are not. A
significant difference might be expected between the pseudohomophones and the
approximate visual controls since the pseudohomophones have a greater visual
similarity to real words. If, on the other hand, phonological recoding as well as
visual similarity affects “no” reaction times, we would expect the pseudohomophones to take longer to reject than both the visual controls and the approximate
visual controls.

Method
Stimuli
The non-words consisted of 25 pseudohomophones and zg each of the visual controls,
approximate visual controls, and distant visual controls. ‘The words consisted of IOO
words with regular spellings matched in frequency, number of letters and numbers of
syllables to the target words for the pseudohomophones. The stimuli were presented in
lower case on a remote CRT controlled by a PET computer.
Procedure
Eighteen Johns Hopkins University students participated in the experiment receiving
either credit towards a course requirement or $3.00 an hour.

VISUAL SIMILARITY IN NON-WORD DECISION

401

Subjects initiated a trial by pressing a start button with the left hand after “READY”
appeared on the screen. Then a fixation point appeared for 0.5 s followed by a word or
non-word. Subjects responded by pressing buttons corresponding to “no” and “yes”,
with the forefinger and middle finger of the right hand respectively. Reaction times were
recorded. The word or non-word stayed on the screen for 2.0 s. If a subject did not
respond within 2.5 s an error was recorded. Following their responses, either “correct”,
“error”, or “too slow” would appear on the screen.
Subjects saw 32 practice trials followed by a presentation of the IOO words and IOO nonwords in a random order.

Downloaded by [Pepperdine University] at 06:51 28 September 2017

Results

An analysis of variance was performed on the ‘‘no” reaction time data with type
of non-word (T) as a fixed effect, and non-word items (N) and subjects (S) as

random effects (Clark, 1973). Because the individual non-words were matched
across type for length and number of syllables, non-word items were treated as a
repeated factor across non-word type rather than nested within non-word type.
In order to perform the analysis, missing reaction times (due to errors) were
estimated for each subject using the method outlined by Myers (1979). A quasi-F
ratio suggested by Myers was computed (F’=MST/[MSTN+MSTS]), which was
highly significant, Ff3,62=
17-59, P<o-oI. Planned comparisons of the pseudohomophones against the three types of control non-words were made, using the
error term from the quasi-F ratio. There was no significant difference between
1.3, P>0.20, with 8 of the
the pseudohomophones and the visual controls, F1,62=
18 subjects and 13 out of the 25 non-word pairs having longer reaction times for the
pseudohomophones than for the visual controls. There was, however, a significant difference between the pseudohomophones and the approximate visual controls, F1,62=13.3, P<o.oI,with 15 of the 18 subjects and 20 of 25 non-word pairs
having longer reaction times for the pseudohomophones. The pseudohomophone
-distant visual control comparison was also significant, F1,,,=67.8, P<O.OOI,
with all 18 subjects and 23 of 25 non-word pairs having longer reaction times for
the pseudohomophones. Mean reaction times and error rates for the types of
non-word, are shown in Table I. Errors were positively correlated with reaction
times, indicating no speed-accuracy tradeoff,

TABLE
I
Reaction times and error rates for normal subjects, Experiment I
Type of non-word

RT (ms)
Error rate

Pseudohomophone

Visual
control

Approximate
visual
control

Distant
visual
control

609
6.0%

619
7’6%

575
34%

539
1.6%

The mean reaction time for the words was 550 ms, and the error rate was 4.7%.

402

R. C. MARTIN

Downloaded by [Pepperdine University] at 06:51 28 September 2017

Discussion
Previous studies have shown that in a lexical decision task reaction times to
reject pseudohomophones are longer than those to reject other non-words. The
results of the present experiment suggest that such findings result from the greater
visual similarity of the pseudohomophones than the control non-words to words.
A possible exception to this conclusion is the recent work of McQuade (1981).
She controlled the visual similarity of the pseudohomophones and control nonwords by matching them on summed positional frequencies of the letters making
up the non-words. However, McQuade obtained a pseudohomophone effect only
when pseudohomophones constituted a small percentage of the non-words (13%),
but not when they constituted the majority of the non-words. Her results suggest
that subjects were able to rely solely on visual access when phonological coding
would have led to many errors. It is possible that the 25% pseudohomophones
in the present study was sufficient to dissuade subjects from using phonological
coding.
In any event, the present results, and those of McQuade, demonstrate that
phonemic effects will not necessarily be found on non-word trials in the lexical
decision task. Consequently, there is no evidence that phonological access is
always occurring, but at a slower rate than visual access. It might be objected
that phonological coding is always proceeding in parallel with visual access, but
that since subjects in the present experiment were forced to make their decisions
in 2.5 s, they had to make their decisions before phonological recoding was complete. However, there is little evidence to suggest that subjects felt pressured by
the deadline to make their decisions faster than they would have without one.
There were only a total of eight errors due to subjects exceeding the deadline
across all 18 subjects (200 responses per subject). Also, in comparing the present
results to those of Coltheart et al. who employed no deadline, the mean reaction
times to the pseudohomophones, visual controls, and approximate visual controls
(601 ms) was only slightly faster than that obtained by Coltheart et al. to the
pseudohomophones and visual controls (619 ms). (Their study included no nonwords as visually distinct from words as the distant visual controls of the present
study.) Thus, it is unlikely that very many of the responses in the Coltheart et al.
study exceeded 2500 ms, and even more unlikely that their pseudohomophone
effect depended on those reaction times exceeding 2500 ms.

Experiment I1
If the reaction time difference between the pseudohomophones and approximate
visual controls in Experiment I is a visual rather than phonemic effect, then one
would expect that aphasic patients who cannot do grapheme-to-phoneme conversion should show the same effect as is seen with normals. However, Patterson
and Marcel (1977) showed that for the same set of stimuli for which a pseudohomophone effect was demonstrated for normals, two deep dyslexics did not show
any difference in reaction time between the pseudohomophones and the other
non-words. Coltheart (1978) maintains that if the pseudohomophone effect were
a visual effect it should, if anything, be enhanced in deep dyslexics rather than

Downloaded by [Pepperdine University] at 06:51 28 September 2017

VISUAL SIMILARITY IN NON-WORD DECISION

403

disappearing, since they rely on visual coding for lexical access. However, there
are several possible explanations for the absence of the pseudohomophone effect
for the deep dyslexic patients. One possible explanation relates to the methodology used by Patterson and Marcel (1977). Deep dyslexics, as with any braindamaged patients, may be highly variable in their response latencies, and it may
consequently be difficult to replicate small reaction time differences found in
normals. Thus, the method employed for measuring reaction times should be
as sensitive and reliable as possible. Patterson and Marcel, rather than using the
usual method of measuring reaction times to individual words and non-words,
measured the time to read through four lists of 17 words and 17 non-words, two
lists containing pseudohomophones as non-words and two lists containing other
pronounceable non-words. Subjects were required to say “yes” or “no” to each
letter string indicating whether or not it was a word. Times to read through the
lists were measured with a stopwatch. Even assuming that the deep dyslexics
were not required to articulate “yes” and “no” (both were also Broca’s asphasics),
the comparison of interest was based on only the mean of the two times for the
lists with pseudohomophones and the mean of the two times for the lists without
pseudohomophones. Under such conditions the finding of a null effect should
not be accorded too much weight.
Assuming, however, that, in fact, deep dyslexics do not show the difference
between pseudohomophones and other non-words shown by normals, another
possible explanation for Patterson and Marcel’s result is that these patients rather
than having an enhanced ability to use visual access have an impairment in their
visual access mechanism. The results from Experiment I suggest that the usually
obtained pseudohomophone effect rests on the sometimes small differences in
visual similarity to words for pseudohomophones and control non-words. In a
recent article, Shallice and Warrington (1980) reported that for several deep
dyslexics discussed in the literature the proportion of visual errors ranged from
13 to 61% of all reading errors. If these errors result from an impairment in
visual access rather than an inability to do grapheme-to-phoneme conversion, it
is possible that the lack of a pseudohomophone effect could be due to an insensitivity to small differences in visual similarity to words.
In Experiment 11, the performance of a deep dyslexic on the task from Experiment I was compared to the performance of another aphasic patient who cannot
do grapheme-to-phoneme conversion but who shows no tendency to make visual
errors. If an insensitivity to small differences in visual similarity can account for
deep dyslexics’ performance on the pseudohomophone task, then we would expect
no difference in reaction time for the pseudohomophones, the visual controls, and
the approximate visual controls, for the deep dyslexic patient. For the patient
who cannot do grapheme-to-phoneme conversion but who does not make visual
errors in reading, the same pattern of reaction times as found for the normal
subjects would be expected if the pseudohomophone effect is a visual effect.

Patient descriptions
Patient B.L. is a right-handed, English-speaking male who suffered a left-

Downloaded by [Pepperdine University] at 06:51 28 September 2017

404

R. C . MARTIN

hemisphere cerebrovascular accident in August of 1974 at the age of 52. The
patient had completed school through ninth grade, and had worked as a maintenance foreman at a large industrial concern for over 35 years. He was initially
evaluated for research purposes in April of 1978, at which time he demonstrated
a dense, right-sided hemiplegia and moderately severe non-fluent aphasia.
Aphasia classification using the Boston Diagnostic Aphasia Examination (Goodglass
and Kaplan, 1972) places him in the category of “Broca’s aphasia”. In addition,
research testing has revealed a striking deficit in his ability to read aloud, and his
symptoms in this regard are consonant with further classification of this patient
as a “deep dyslexic’’ (Marshall and Newcombe, 1973). For example, in reading
words aloud, he makes both semantic and visual errors. I n a reading test of 420
words he made 63% errors overall. Of these 18% were semantic and 25% visual.
H e shows an inability to do grapheme-to-phoneme conversion as he read only two
out of 20 non-words correctly, and this allowing a pronounciation one phoneme
off as correct. Out of 149 pseudohomophones, he read 37 correctly. However,
most of these were close in visual similarity to their target words (e.g. mowth for
mouth), and it is thus likely that he was using a visual strategy in choosing his
responses. Further evidence in support of the deep dyslexic classification for
this patient is given by Schwartz et al. (1977), and Nolan and Caramazza (in
press).
C T scan reveals a large area of infarct in the peri-Sylvian region of the left
hemisphere including Broca’s area and extending posteriorly to include portions
of the superior temporal and inferior parietal lobes.
Patient J.S. is a right-handed, English-speaking male who suffered a lefthemisphere cerebrovascular accident in July of 1979 at the age of 44. T h e
patient attended four years of college and served as a sergeant in the U.S. Army
until several months prior to his CVA. H e was initially seen for research evaluation in December of 1979, at which time he exhibited a moderately severe fluent
aphasia with a severe deficit of auditory comprehension. Based on the speech
production profile of the B.D.A.E., he was classified as a Wernicke’s aphasic;
however, his reading comprehension was found to be excellent. On the auditory
comprehension section of the B.D.A.E., J.S. achieved a x-score of -I or worse
on all subtests, whereas on the reading comprehension section all scores for subtests requiring no auditory processing approached the ceiling.
Experimental testing has confirmed this striking dissociation between visual
and auditory processing (see Caramazza, Berndt and Basili, manuscript in press,
and Martin and Caramazza, 1gS2), and has further demonstrated his inability to
generate a phonemic representation from print. Despite 100% accuracy in a
picture-word matching task, and a score of 127 out of 150on a visual presentation
of the Peabody Picture Vocabulary Test, on a test of single word oral reading he
read only 38% of the words correctly. When unsuccessful at reading a word he
produced no visual or semantic errors, but instead produced neologisms, though
showing no articulatory difficulties, In attempting to read non-words, he read
only 15% of the pronounceable non-words correctly, and 20% of the pseudohomophones. Further, he appears to be unable to generate a phonological form
even when no verbal output is required, for example, showing an inability to select

VISUAL SIMILARITY IN NON-WORD DECISION

405

a pseudohomophone to match a picture (e.g. nyphe), or to judge whether two
visually dissimilar words rhyme. He is also incapable of judging whether the
names for two pictures will rhyme even though no graphic form need be processed.
C T scan reveals an area of infarction in the superior temporal lobe, including
Wernicke’s area, which extends superiorly into the parietal lobe to involve the
supramarginal gyrus and an area slightly superior to it. It appears that the
infarcted area may involve only the anterior portion of the angular gyrus.
Method

Downloaded by [Pepperdine University] at 06:51 28 September 2017

Stimuli
T h e same stimuli as used in Experiment I were used here.

Procedure
The procedure for the most part was identical to that of Experiment I. Because one
of the patients (B.L.) has a right hemiplegia, the procedure was modified so that a trial was
initiated by pressing a foot pedal with the left foot. A one button response (using the left
hand) was used for obtaining reaction times for the “no” trials, with the subjects instructed
to press the button only when a non-word appeared and to do nothing when a word
appeared. Thus, reaction times for the word trials were not obtained for these patients.
The patients saw the 232 trials (including practice trials) twice, with over a month separating
the two presentations for each subject.

Results

Reaction times for the individual non-words were averaged over the two
presentations. Means and number of errors for the four types of non-words are
shown in Table 11. If no reaction time for a particular non-word was available

TABLE
I1
Reaction times and error rates for aphasic sujbects, Experiment II
Type of non-word
Pseudohomophone

Visual
control

Approximate
visual
control

Distant
visual
control
~

J.S.
R T (ms)
Error rate
B.L.
R T (ms)
Error rate

785

815
14%

725

8Yo

0YO

899

894

22 Yo

14%

914
8%

800
4%

12%

678

(because the patient made errors both times the stimulus was presented), reaction
times for the matched controls for that non-word were also eliminated from the
computation of the mean reaction time for each set.
Analyses of variance were performed on the data for each subject separately
using individual non-word items as a random effect. Rather than estimating
missing values as was done for the normal subjects, the data eliminated from the
calculation of the means were also excluded from these analyses. Significant

Downloaded by [Pepperdine University] at 06:51 28 September 2017

406

R. C. MARTIN

effects from these analyses would, of course, only indicate that the results can be
generalized to other non-words of these types for these same patients, and not that
these results can be generalised to other patients like these patients. However, it
was felt that it would be better to be able to make this kind of generahation
rather than performing no analyses at all on the data.
For J.S., a highly significant effect for type of non-word was obtained, F3,63=
14.27, P<O.OOI.
The planned comparisons for the pseudohomophones against
the other non-word types showed no effect for the pseudohomophone-visual
control comparison, F1,,,=0-56, but significant effects for the pseudohomophone
-approximate visual control comparison F1,,,=7.96, P<o.oI,and for the pseudohomophone-distant visual control comparison, F1,63=z5.30, Pto-001. Error
rates corresponded to reaction times.
For B.L., there was no significant effect for type of non-word, F3,57=~.44,
P>o.ro. Looking at the mean reaction times in Table I1 it appears that there
was a fairly large difference between the distant visual controls and the other nonword types. However, there was large variability in B.L.’s reaction times, and
even this comparison was not significant by itself, Fl,a7=z.13, P>o.og. Across
all non-word conditions, the standard deviation for B.L.’s reaction times was
220 ms, while for J.S. it was 131 ms. (For the normal subjects, the average
standard deviation was 94 ms.)
Although there were no significant differences between the non-word types in
terms of reaction times, the pattern of errors obtained for B.L. conformed to that
found for the normals and J.S. in terms of the relative difficulty of the non-word
types. A comparison for dependent proportions showed no significant difference
between the proportions of errors for B.L. for the pseudohomophones and the
visual controls, z= I -07. However the difference in errors between the pseudohomophones and the approximate visual controls was significant, z=z. I I , P<o.og,
as was the difference between the pseudohomophones and the distant visual
controls, x= 5-21, P<o.og.
On the word trials, J.S. made only three errors summed across both days for an
error rate of 1.5%. B.L. made 24 errors on the word trials for an error rate of
12%. Thus, J.S.’s overall error rate across words and non-words was g.oyo,
while B.L.’s was 12%.
Discussion

The results from Experiment I1 again provide evidence that the commonly
obtained pseudohomophone effect is a visual effect. A patient (J.S.) who has a
number of difficulties related to phonological processing including an inability to
do grapheme-to-phoneme conversion, showed the same pattern in his reaction
times and errors as that obtained for the normal subjects. Assuming that the
pseudohomophone-approximate visual control comparison is similar to the
pseudohomophone-non-word comparison made in other studies, J. S. could be
said to show the pseudohomphone effect. However, the lack of a significant
difference between the pseudohomophones and the visual controls for J.S. and the
normals indicates that this effect is visual rather than phonological.

Downloaded by [Pepperdine University] at 06:51 28 September 2017

VISUAL SIMILARITY IN NON-WORD DECISION

407

For B.L., the deep dyslexic patient, the results were not as clear cut. In terms
of his reaction times, there was no significant difference between any of the nonword types. However, his reaction times were highly variable, with even a 99 ms
difference between the pseudohomophones and distant visual controls not reaching
significance. Thus, more weight should be given to his error rates (which were
high relative to the other subjects) than to his reaction times. The pattern of his
error rates conformed to that found for J.S. and the normal subjects for both
their reaction times and error rates. I n terms of error rates, B.L. also showed the
“pseudohomophone effect”.
Thus, it appears plausible that the previous failure to find the pseudohomophone
effect for deep dyslexic patients may have resulted from the methodology employed
and/or highly variable performance on the part of the patients tested. At any
rate, in the present experiment where a more controlled procedure was used, both
patients, neither of whom can do grapheme-to-phoneme conversion, showed the
“pseudohomophone effect”. T h e other explanation offered for the previous null
result was not confirmed since B.L. did show a sensitivity to differences in similarity of the non-word types to words in his error rates. Although his overall
higher error rate may have resulted from impairments related to his deep dyslexic
syndrome, the differences in education and intelligence between B.L. and J.S.
most likely also contributed to his lower level of performance.

Conclusion
The results of the present study have both theoretical and practical importance.
Of practical concern is the use of a test for the pseudohomophone effect as a means
of assessing an aphasic patient’s ability to generate a phonological representation
from print. Unless care is taken to insure that a pseudohomophone effect found
in normals is not due to the greater visual similarity of the pseudohomophones
than control non-words to words, little can be concluded concerning the graphemeto-phoneme conversion abilities of an aphasic patient who either does or does not
show the effect.
Of theoretical interest is the implication that most reports of the pseudohomophone effect confound visual with phonological similarity to words. I n the present
study where control non-words were more carefully matched in visual similarity
to words, no evidence for phonological coding was found on the non-word trials.
Thus, the proposal that phonological access always occurs in parallel with visual
access but at a slower rate has lost its major empirical support. It might be
argued, however, that phonological coding, though obligatory, can be slowed down
under the control of the subject to the extent that it has no effect on either the
word or non-word trials. However, this would be to assert that some process is
always occurring, but that it will sometimes have no discernible effect on any
measure of performance. It seems simpler to assume that phonological coding
is an optional strategy that subjects will employ when conditions favour its use.
The author would like to thank Betty Ward for her help in testing the subjects.
This research was supported by National Institutes of Health (NINCDS) grant N S 14099
to The Johns Hopkins University.

R. C. MARTIN

APPENDIX
Non-words
-

Downloaded by [Pepperdine University] at 06:51 28 September 2017

Pseudohomop hone
werd
(Word)
sune
(Soon)
groe
(Grow)
gerl
(Girl)
werk
(Work)
meen
(Mean)
turm
(Term)
dert
(Dirt)
wite
(White)
whife
(Wife)
rufe
(Roof)
whall
(Wall)
ded
(Dead)
gard
(Guard)
consept (Concept)
shure
(Sure)
reech
(Reach)
sertain (Certain)
breth
(Breath)
hert
(Hurt)
sircle
(Circle)
merder (Murder)
munny (Money)
lern
(Learn)
snoe
(Snow)

Visual control
cest
buke
dree
feve
leng
plen
nuck
vece
tink
sheem
fute
thalk
ner
gand
resords
the11
plent
sountry
tught
pesh
sloser
jenior
tuddy
trin
gree

(Cost)
(Book)
(Drew)
(Five)
(Long)
(Plan)
(Neck)
(Vice)
(Think)
(Seem)
(Foot)
(Talk)
(Near)
(Grand)
(Records)
(Tell)
(Plant)
(Country)
(Taught)
(Push)
(Closer)
(Junior)
(Today)
(Train)
(Grew)

Approximate
visual control
serd
wune
derl
ferk
neen
lurm
derf
wute
shife
lufe
thall
ved
gark
monsept
shurb
seech
mertain
preth
zert
hircle
perder
murny
lorn
skoe

Distant
visual control
kyse
gher
phou
remm
twup
bYPh
naff
unck
puhm
mulst
vuls
sckop
irm
kenj
icktion
zhamp
aungs
irdkasp
knuth
haan
oquars
ihrmot
wrulo
WZP
ooms

*Target words for the pseudohomophones and visual controls are shown in parentheses.

References
H.and WILLIAMS,J. P. (Eds),
BROWN,R. (1970). Psychology and reading. I n LEWIS,
Basic Studies in Reading. New York: Basic Books.
CLARK,H. H. (1973). The language-as-fixed-effect fallacy: A critique of language statistics
in psychological research. Journal of Verbal Learning and Verbal Behaviuur, 12,335-59.
G. (Ed.),
COLTHEART,
M. (1978). Lexical access in simple reading tasks. I n UNDERWOOD,
Strategies in Information Processing. London : Academic Press.
COLTHEART,M. (1980). Reading, phonological recoding, and deep dyslexia. In
COLTHEART,
M., PATTERSON,
K. and MARSHALL,
J. (Eds), Deep Dyslexia. London:
Routledge and Kegan Paul.
COLTHEART,
M., DAVELAAR,
C., JONASSON, J. and BESNER,
D. (1977). Access to the internal
lexicon, in DORNIC,S. (Ed.), Attention and Performance, 11. New York: Academic
Press.
DAVELAAR,
C., COLTHEART,
M., BESNER,D. and JONASSON,J. (1978). Phonological
recoding and lexical access. Memory and Cognition, 6 , 391-402.
GIBSON,E. (1970). The ontogeny of reading. American Psychologist, 25, 136-43.
GOODGLASS,
H. and KAPLAN,C. (1972). The Assessment of Aphasia and Related Disorders.
Philadelphia: Lea & Ferbinger.

Downloaded by [Pepperdine University] at 06:51 28 September 2017

VISUAL SIMILARITY IN NON-WORD DECISION

409

MARSHALL,
J. C. and NEWCOMBE,
F. (1973). Patterns of paralexia. Journal of Psycholinguistic Research, 2 , 175-99.
MARTIN,R. C. and CARAMAZZA,
A. (1982). Short term memory performance in the
absence of phonological coding. Brain and Cognition, I, 5 0 7 0 .
L., HILLINGER,
M. and BIAS,R. (1981). Phonological recording and reading.
MCCUSKER,
Psychological Bulletin, 89, 217-45.
MCQUADE,
D. V. (1980). Variable reliance on phonological information. Language and
Speech, 24, 99-100.
MEYER,D. E., SCHVANEVELDT,
R. and RUDDY,M. (1974). Function of graphemic and
phonemic codes in visual word recognition. Memory and Cognition, 2, 309-21.
MYERS,J. (1979). Fundamentals of Experimental Design, third edition. Boston: Allyn
and Bacon, Inc.
MORTON,J. (1969). Interaction of information in word recognition. Psychological
Review, 76, 165-78.
NEWCOMBE,
F. and MARSHALL,
J. C. (1980). Transcoding and lexical stabilization in deep
dyslexia. In COLTHEART,
M., PATTERSON,
K. and MARSHALL,
J. C. (Eds), Deep
Dyslexia. London: Routledge and Kegan Paul.
A. (1982). Modality-independent impairments in word
NOLAN,K. and CARAMAZZA,
processing in a deep dyslexic patient. Brain and Language, in press.
PATTERSON,
K. and MARCEL,
A. J. (1977). Aphasia, dyslexia, and the phonological coding
of written words. Quarterly Journal of Experimental Psychology, 29, 307-1 8.
RUBENSTEIN,
H., LEWIS,S. and RUBENSTEIN,
M. (1971). Evidence for phonemic recoding
in visual word recognition. Journal of Verbal Learning and Verbal Belaaviour, 10,
647-57.
SAFFRAN,
E. and MARIN,0. S. M. (1979). Reading without phonology: evidence from
aphasia. Quarterly Journal of Experimental Psychology, 29, 5 I 5-25.
SCHWARTZ,
M., SAFFRAN,
E. and MARTIN,0. S . M. (1977). An analysis of agrammatic
reading in aphasia. Paper presented at the International Neuropsychological Society
Meeting.
SHALLICE,
T. and WARRINGTON,
E. (1975). Word recognition in a phonemic dyslexic
patient. Quarterly Journal of Experimental Psychology, 27, 187-99.
SHALLICE,
T. and WARRINGTON,
E. (1980). Single and multiple component control
M., PATTERSON,
K. and MARSHALL,
J. (Eds),
dyslexic syndromes. In COLTHEART,
Deep Dyslexia. London : Routledge and Kegan Paul.
SHULMAN,
H., HORNAR,
R. and SANDERS,
C. (1978). The effects of graphemic, phonetic,
and semantic relationships on access to lexical structures. Memory and Cognition, 6,
115-23.
YTANOVICH, K. and BAUER,D. (1978). Experiments on the spelling-to-sound regularity
effect in word recognition. Memory and Cognition, 6, 410-5.
Revised version received 21 September 1981