Memory & Cognition
I 982, Vol. 10(5), 465-474
An alternative to grapheme-phoneme
conversionrules?
MARCUS TAFT
Uniyersitt o! New South Wsles,Kentingto4 Nev South lfoles 2033,Au*rslia
[
I
I
When orthographic factors wcro tightly controlled in a lerical decision task, it wae obg€rvod
that orthog?aphic similarity rather than honophony with a word led to increased r€action timos
to nonwords. This result suggested that the pseudohomophone eff€ct is not a phonological
effect. Inst€ad, a convugion of the graphemee of a stimulus itsn into different graphemeo
via a eet of graphemegrapheme conversion rulee was eupported. When phonological factora
were tightly controlled and orthographic similarity varied, evidence for the eristance of
grapheme-grapheme rules was provided in both a lerical decision task and a tack in rhich
were required to say whether an it€m wa8 pronounced in the same way as a word.
"objecte
Even in the lattar task, in which the likelihood of phouological recoding was optimized, it
appeared that graphemephonene rules were rarely, if ever, used.
I
The finding that nonwords are classifiedin a lexical
decisiontask more slowly when they are pronouncedin
the sameway as a real word (e.g.,BRANE) than when
they are not (e.9.,BRATE), has been taken as evidence
that nonwords are convertedinto a phonologicalrepresentation via grapheme-phoneme
rule conversionwhen
they are processed(Coltheart, Davelaar,Jonasson,&
Besner, 1977; Gough & Cosky, 1977; Patterson &
Marcel, 1977; Rubenstein,Lewis, & Rubenstein,l97l).
This has been called the pseudohomophoneeffect. On
the other hand, the failure to consistently find such a
homophone effect for word items has been taken as
evidence that words arc nolmally processedon the
basis of their orthography rather than their phonology
(Coltheart, 1978;Coltheartet al., 1977). Takentogether,
the presenceof a pseudohomophoneeffect and the
general lack of a homophone effect for words implies
that there are two accessroutes to the lexicon. one
visual and one phonological, and that both are used
when processingprinted material. However,the phonological route is slower than the visual one and exerts
an influence on responsetimes only when the visual
route fails to find a lexical entry, that is, when the item
presentedis a nonword.
Following from this, the pseudohomophoneeffect
has been used as a diagnosticto test the extent to which
a readerusesphonologicalrules.For example,Patterson
and Marcel (1977) report two dyslexic patientswhose
failure to produce a pseudohomophoneeffect (along
with their inability to read nonwords) was taken to
mean that these patients had a deficit in their ability
to convert the printed word into its phonologicalform
by rule.
Related to the pseudohomophoneeffect is a task
that requires subjects to explicitly report whether a
nonword is pronounced in the sameway as a proper
word (e.9., BRANE) or not (e.9., BRATE). Again, the
ability to perform this task hasbeenusedasa diagnostic
Copyright1982Psychonomic
Society,Inc.
of a reader'sability to use grapheme-phoneme
rules.
For example, Saffran and Marin (1977) examined the
pattern of errors made on this task by a dyslexic patient
and concludedthat the patientwasunableto usephonological conversionrules. Ability to perform this task was
also used by Baron and Strawson(1976) to determine
whether a group of normal readerswas phonologically
oriented or visually oriented.
The assumptionthat is made in explainingboth the
pseudohomophoneeffect and the ability to saywhether
or not a nonword is pronounced like a word is that a
phonologicalrepresentationof the nonword is generated
by a set of grapheme-phoneme
conversionrules and that
this is then matched with a phonologically represented
lexical entry. In order to be certain that this assumption
is correct, however, it must be shown that pseudohomophonesdo not show their effects simply because
they are orthographically more similar to real words
than are nonhomophonic nonwords. For this reason,
Coltheartet al.(1977) attemptedto match their pseudohomophones and their nonhomophonic nonwords on
orthographic similarity to English. They did this by
changingonly one letter of each pseudohomophoneto
form a nonhomophone. For example, the pseudohomophone ILE was matched with the nonhomophone
IFE, and FRAZE was matched with FRUZE. Coltheart
et al. supposed that the changing of only one letter
meant that the nonhomophoneswere just as orthographicallysimilar to real words as were the pseudohomophones.But that is not really true. For example,
ILE is orthographically more similar to a word (ISLE)
than IFE is and FRAZE is orthographicallymore similar
to a word (PHRASE) than FRUZE is. To be sure that
orthographic similarity is not confounded with homophony, then, one must use pseudohomophones
and
nonhomophonesthat are matched exactly on their
similarity to real words. This can be achievedin the
following way.
465
I 0$0r .2slo
0090-s02x
| 821os046s-
466
TAFT
There are orthographic structuresin English that can
be pronounced in more than one way (e'g', O-STin
GHOST and FROST). If one constructsa pseudohomophone by changing such an orthographic structure
appropriately(e.g., GHOAST)' then one can also consiruct- a nonhomophone by making exactly the same
change (e.g., FROAST). In this way, the nonhomophoric nonword is related orthographically to a_real
word in exactly the sameway asthe pseudohomophone
is. Therefote, if the pseudohomophonefindings results
from phonological recoding rather than orthographic
similarity, then pseudohomophones like GHOAST
should i.kr lottger to classify as nonwords than should
nonhomophoneslike FROAST. If, on the other !*d'
the effect is an orthographic one, then GHOAST and
FROAST should take equally longer to classify iN non'
words than a nonword like PLOAST, which is not orthographically related to a real word. Experiment I was
carriedout to test thesepredictions.
EXPERIMENT I
Method
Materials.Threetypesofnonwordwereconsttucted:pseudo.
that
homophones (H-nonword condition), nonhomophones.
were orthographically similar to a word (o-nonword condition),
to a
and nonhoirophon"r that were not orthographically similar
word (N-nonword condition). There were 20 nonwords in each
In
condition, matched in triplets on their orthographic structureand
addition, the words from which the H-condition nonwords
theo-conditionnonwordswerederivedwerematchedoverall
items on log frequency according to Carroll, Davies, and Richman
( I 97 1). Eximples of iiem tripleti were : GHOAST (from GHOST)'
(from FROST), PLOAST; PLEED (from PLEAD),
inOaSr
(fiom GERM)'
DREED (from DREAD), CLEED; and JERM
preJIFT (from GIFT), JILK. All of the nonword items are
sented in the aPPendix.
The word items used in the experiment were also set up in
order to look at the effects of orthographic similarity to other
words. Fifteen H-condition words (i'e', homophonic words'
like PANE) were compared to 15 nonhomophonic control
and
words matched on frequency and length (e'g" fE\K)'
were
15 Ocondition words (i.e., nonhomophonic words that
orthographically similar to another word; like GREET, which is
similar io GREAT) were compared to 15 matched control
words (e.g., CREST). The H-condition words could not be
the
compared- directly to the o-condition words because of
difficultyoffindingappropriatelymatcheditems.Thegraphemic
ending of each n-ionOition word was always a less frequently
ending than that of the word with which it was homoo"",rring "Fot
-ANE is a less common ending than is
.tu*ple,
phonic.
-RtN. ltr addition, each H-condition word was always less frequent than the word with which it was homophonic (e'g''
iANO is less frequent than PAIN). The same principles were
-EET is a
followed with thJ Ocondition words. For example,
-EAT and GREET is a less frequent
less frequent ending than is
word than is GREAT.
Procedure. Subjects were presented the stimulus items in one
on a
of four different iandom orders. Stimuli wele presented
item
video display unit under computer control, with each
b e i n g p-sutjects
resentedforS00msec.Theinterstimulusintervalwas
were instructed to press a right-hand "yes"
+ seJ.
presented was a word and to press a leftitem
if
ifre
button
hand "no" button if ttre item presented was not a word' They
were also told to respond as quickly, but as accurately' as they
could.
Thirty subjects were used in the experiment, each being
given course accre{itation for their participation'
Results
Table 1 provides the mean reaction times for the
nonword and word items of the first experiment'
An analysis of the reaction times to the nonwords
revealedthat, while the H nonwords differed from the
N nonwords[min F'(1,50)= 8.67, p <.01] (the normal
pseudohomophoneeffect), there was no difference at
ull btt*t.n the H nonwords and the O nonwords(both
Fr and F2 < l). An analysisof the errors showedthe
=
sarnepatte.n [H vs. N nonwords,ffih F'(1,49) 4.17,
, in F'< 1]. The O nonp < . 0 ^ 5 ; H v s . O n o n w o r d sm
words differed from the N nonwords also,but only on
reaction time [min F'(1 ,34) = 7.08, p < .02] . The
error difference between O nonwords and N nonwords
was significantonly on the subjectanalysis[Ft(1 ,29)=
5 . 3 7 , p< . O SF; z ( 1 , 1 9=) 4 . 0 2 , P> . 0 5 1
The word items were analyzedusing aZby 2 analysis
of variance,the main effects being H and O words vs'
the control conditions (orthographic similarity), and H
conditions (experimental and control) vs. O conditions
(experimental and control). The main effect of orthogtupnir similarity for reaction times was significantonly
on- the subjects analysis [Fr (l ,29) = 8'394, p 1 '02;
F2(I,29)=1.63,p>'051'aswasthedifferencebetween
th;-H conditioni and the O conditions [Ft(l ,29)=
8 . 2 8 7 , p 1 . 0 2 ; F z ( l , 5 6 ) =2 . 1 1 5 ,p > . 0 5 1 ' T h e i n t e r action 6etween these two effects was not significant
(min F'< l). Analysis of the errorsrevealedno significant differences.
Discussion
It is apparent from the data that the pseudohomophone efiect may not be a phonological effect at all'
Fseudohomophonesdid not differ in the experiment
from those nonwords that were not homophonic with
but were orthographically similar to words. Therefore,
it appearsthat the orthographic similarity of pseudohomophones to real words is sufficient to produce a
delay in classificationtimes relativeto nonwordsthat are
not similar to real words.
Table I
Mean Reaction fimes (in Milliseconds) and standard Errors
Along with Percentage Error Rate @E) for the Nonword
and Word Items of ExPeriment I
Condition
H Nonwords
O Nonwords
N Nonwords
H
H
O
O
Words
Control Words
Words
Control Words
PE
Example
RT
Nonwords
GHOAST
FROAST
PLOAST
750
7s2
7t5
26
29
25
9
664
26
6s7
6s0
2r
24
617
20
I2
11
8
4
Words
PANE
JERK
GREET
CREST
n
J
GRAPHEME-PHONEMERULES
467
Such a conclusionhasmajor implicationsfor theories Jonasson (1978) have found that the homophone
of word recognition that cite the pseudohomophone effect is also open to strategic control, thus implying
effect as evidencefor the use of grapheme-phoneme the use of rules. Davelaaret al. obtaineda homophone
were not included
conversionrules. According to the presentresults,the for words when pseudohomophones
pseudohomophoneeffect should be seeninstead as the in the experiment, but no effect when they were.
result of confusionsresulting from visual access(that is, Possibly, then, the lack of any effects significant by
accessthat is not mediated by a phonologicalcode). min F' for the word items in the presentexperiment
Similarly, when a dyslexic patient fails to produce a was a result of there being nonwords in the experipseudohomophoneeffect (Patterson& Marcel, 1977), ment that were orthographically similar to real words.
the resultspresentedhere would imply that the patient If there had been no such nonwords, then significant
is simply not being influencedby orthographicsimilarity. effects on the word items might have been observed.
If the pseudohomophoneeffect and the homophone
This, however, seemsto be untrue. On the contrary,
Saffran and Marin (1977) have demonstratedthat such effect are under strategiccontrol and if the explanation
dyslexic patients are very much influenced by ortho- for these effects is in terms of orthographic similarity
graphic similarity when required to make a response (as suggestedby Experiment1), then that impliesthat
about the pronunciation of an item. For example, a the orthographic similarity involved is rule governed.
The purposeof the secondexperiment,therefore,is to
patient might say that MANOR is homophonic with
look more directly at the question of the existenceof
MINOR, rather than with MANNER, or that RAYNE
is pronouncedas RAYON, rather than RAIN. How can rule-governedorthographic similarity, that is, the existconversionrules.
enceof grapheme-grapheme
this fact be reconcilablewith the presentresults?
It is possible that there are in fact two types of
EXPERJI\{ENT 2
orthographic similarity: gross graphemic similarity and
rule-governedsimilarity. The former is a matter of the
In Experiment 1, it was found that nonwords that
similarity of the overall graphemicpattern of the pre(e.g.,RAYNE were orthographically similar to words took longer to
senteditem with the word that is accessed
and RAYON are orthographically similar in this way). classify than those that were nott It is unclearfrom this
The latter refers to the interchanging of particular experiment, however, what type of orthographicsimigraphemeswith other graphemesthat can be pro- larity was involved. The O nonwords were more similar
nounced identically and that do not violate laws of to words than were the N nonwords, both on gross
basis.
orthographic cooccurrence(called orthotactic rules by graphemiccharacteristicsand on a rule-governed
In order to separatethesetwo types of orthographic
Taft, 1979a).For example,EA, and EE are interchangeable in this way, but EA and OA are not. Hence,NEAT similarity, two nonword conditions were designed so
and NEET are orthographically similar in this sense, that they were matched on their general graphemic
but NEAT and NOAT are not. Similarly, DREAD and similarity to other words (and phonemic similarity)
DREED are orthographically similar, even though they but varied in the nature of this similarity. For example,
are not homophonic with each other. G and J can be CHEECE is as graphemically(and phonemically)similar
interchangeablealso,but not in every environment.For to CHEESEas BREECEis to BREEZE.Yet C and S are
example,they can be interchangedin GIMP, but not in interchangeablein the sense that they both can be
GRIMP or LING, since JR and NJ are orthotactic pronounced in the sameway, whereasC and Z cannot.
violations. Possibly,then, normal subjectsare influenced Hence, if lexical decision items are slowed by orthoby the rule-governedtype of orthographic similarity graphic similarity of a grossgraphemictype only, then
when performing the lexical decision task, whereas CHEECE (O condition) should not differ from BREECE
dyslexic subjects are unable to apply these grapheme- (G condition). If, on the other hand, they are slowedby
grapheme conversion rules and are influenced solely the fact that the application of grapheme-grapheme
rules leadsto the erroneousaccessing
of a lexicalentry,
by grossgraphemicsimilarity.
McQuade(1981) has reported that the size of the then O-condition nonwords should take longer to
pseudohomophoneeffect is variable, depending upon classify as nonwords than G-condition nonwords. Note
rules are applied, rather
the strategiesadopted by the subjects.This impliesthat also that if grapheme-phoneme
rules, then the two conditions
the pseudohomophone effect must be explained in than grapheme-grapheme
terms of someprocedureover which subjectshavesome should not differ either, since /Ui:s/ (the pronunciation
control (either conscious or unconscious).Delay in of CHEECE) is as similar to lti'zl (the pronunciation
reaction time through gross graphemic confusions of CHEESE)as /bri:s/ (the pronunciationof BREECE)
between nonwords and words is unlikely to be under is to /bri:z/ (the pronunciationof BREEZE).
strategic control. Application of rules, on the other
hand, could easily be open to strategiccontrol: But Method
Materials. Twenty O-condition nonwords were matched with
rather than grapheme-phonemerules, as McQuade 20 G-condition nonwords on orthographic structure. An item
proposes, these rules could be grapheme-graphemewas considered to be an O<ondition nonword if its pronunciarules. Similarly Davelaar, Coltheart, Besner, and tion could be spelled differently to form a word. Examples of
468
TAFT
influence decisiontimes, but this factor was held constant in the presentexperiment.The resultsof Chambers
(1979) provide evidencethat approximategraphemic
accessdoes indeed take place in word recognition.
Similarly,this experimentdoesnot say whether phonological similarity influencesdecisiontimes.However,the
resultsof Experiment1 suggestthat it does not. In this
first experiment, orthographic similarity was held constant and phonological similarity was varied, and no
effect was observed;in the secondexperiment,phonological similarity was held constant and orthographic
similarity was varied, and here, an effect was observed.
It seemsthat grapheme-grapheme
rules can account
for effectsthat have previouslybeen assumedto involve
grapheme-phoneme
rules. If this is true, it may be the
case that grapheme-phonemerules are simply not
availableto readers.One can look at this questionby
employing a task that optimizes the likelihood that
grapheme-phoneme
rules will be used. Such a task is
one in which subjectsmust make judgmentsabout the
pronunciation of nonwords.In particular, subjectsare
Rezults and Discussion
The lexical decision results for the two nonword required to classifya letter string as being or not being
pronouncedin the sameway as a real word. This is a
conditionsare presentedin Table2.
The reactiontime differencebetweenthe O-condition homophonedecisiontask. Thus a "yes" responsewould
nonwords and the G-condition nonwords proved to be be made to BRANE and a "no" responsemade to
significant [min F'(1,30)= 4.37, p < .05] , but the BRATE.
If grapheme-phoneme
rules are availableto a reader,
error differencedid not (min F' < 1).
then it would seem that the most logical way of perThis finding points to the existenceof some sort of
grapheme-grapheme
rule system.The differencebetween forming this task would be to convert the nonword into
a phonologicalrepresentation
via theserulesand attempt
the two nonword conditions could not result from
generalgraphemic similarity, since the conditions were to match this representation
to a lexical representation
matched on this factor. Neither could the difference that is phonologically coded. Baron and Strawson
(1976) and Saffran and Marin (1977), who have used
be attributed to phonological similarity, since this too
was matched.The only differencebetweenthe O condi- this task, assumethat this is how it is performed. It is
tion and the G condition was that if each of the O- possible,however,that eventhough this task requiresa
condition nonwords was respelledto producethe same judgment about pronunciation, subjects employ
pronunciation,they created a real word, whereasthe grapheme-grapheme
rules and do not use a mediating
phonologicalrepresentation.By this account, BRANE
G-condition nonwords respelled in this way did not
create a real word. For example,CHEECE could be would be classifiedas being homophonic with a word
respelled as CHEESE while maintaining its pronuncia- becauseapplication of the rule A-E-+AI leads to the
tion (asin GEESE);BREECEcould similarlybe respelled accessingof the lexical entry BRAIN. BRATE would
as BREESE,but this is not a word. BREEZE is deriv- be classified as being nonhomophonic with a word
able from BREECE, not by systematicrespelling,but
becausenone of the rules for A-E (-AI, -+EI, -+EIGH,
-+AIGH)producesa word.
by simply exchangingone letter for another.
Subjects appear to have availableto them a set of
The method that subjectsemploy in performingthis
rules that tells them what letter combinations can be task can be examinedby comparing the O-condition
pronouncedin the sameway, ffid they appearto apply nonwords and G-condition nonwords of Experiment2
these rules when presentednovel letter strings. This is in a homophone decision task. If grapheme-grapheme
not to say that gross graphemic similarity cannot also rules were applied, subjectsshould encounterproblems
in classifying CHEECE as a nonhomophone compared
to BREECE, since only in the former casewill a word
Ttble 2
be accessed,
namely, CHEESE.On the other hand,there
Mean Reaction Times (in Milliseconds) and Standard Errors
should
be
no
differencebetweenCHEECEand BREECE
Along With Percentage Error Rate (PE) for the
if only grapheme-phoneme
rules are used,since/Ui:s/is
Nonword ltems of Experiment 2
no more similarto lti:zl than /bri:s/ is to lbri:zl.
SE
PE
Condition
Example
RT
A further consequence
of the view that the homo51
5
O Nonwords
CHEECE
683
phone decision task should be performed via simple
4
s0
G Nonwords
BREECE
636
application of grapheme-phoneme
rules is that there
item pairs are CHEECE (from CHEESE) and BREECE (from
BREEZE), BREEST (from BREAST) and CREEST (from
CREST), and KELL (from CELL) and KEND (from SEND).
Item pairs were designed so that, over all items, the nonwords in
the two conditions differed by one letter from the same number
of real words, through substitution or addition. Thus general
graphemic similarity was controlled. Phonemic similarity was
also controlled in that the relationship between the pronunciation of an O-condition item and the word from which it was
derived was the sitme as the relationship between the pronunciation of its Gcondition pair and a real word. Thus the pronunciation of the Ocondition item CHALC (namely, /dalk/) bears
the same relationship to the pronunciation of CHALK (namely,
/do.k/), as the Gcondition HALC does to HAWK (namely,
/halk/ and /hc:k/). In addition, while CHALC is graphemically
only one letter different from CHALK, so HALC is only one
letter different from HALT (as well as HALL, HALF, and
TALC). All items are presented in the appendix.
There were also 40 words used as distractor items. These
were of similar structure to the nonwords (e.g., SLEEVE,
LEECH, MAIZE).
The procedure was identical to that of Experiment 1. There
were 10 subjects.
GRAPHEME.PHONEMERULES
should be no difference between
pseudohomophones
469
of Experiment2, the reversewas true.
The error difference observedhere implies that grapheme-grapheme
rules were applied in peiforming thetask. ihe onty
t9aso1-yhy subjectsshould be more
inclined to think
that cHEEcE is pronouncedlike cHEEsE
than to think
that BREECE is pronouncedlike BREEZE
would seem
to be the greaterinterchangeability
of c and i .o,npu..a
to C and Z. What upp.un to be happening
is that sub_
jects are using a visuai rather than
pifnotoiicJstrategy
in performing this task. Grapheme-grapheme
conversions
are made, and.an attempi is then'made
to uiruuffy
accessthe resultant item in lexical
memory. A'con_
version of C to S is worth trying, since
it i, ptrriUt, fo,
these letters to be oro.rounrra"in ttre
,u_.,uV, lu, u
conversionof C to Z is not worth trying,
sinceihey are
neverpronouncedin the samewav.
Why is it, then, that subjectsio not always
make an
error on o-condition nonwords?In the
lexical
o..irion
EXPERIMENT 3
task, there can be a visual check after
acce* to-a.rrr_
mine whether the_entry accesseddoes
indeea-*.t.r,
Method
with
the presented item. In the homophor.
Materials. The O-condition and G_condition
or.iri",
nonwords that
task, a visual check is pointless.Any check
were used in the second experiment
were arso used in the third
that is made
experiment.r Thus there were 40
to avoid errors
b. a phonologicalone. ffrui-ir,-tn.
nont on,opho"i" it"_r, unO
Tu{
these are provided in the appendix.
pronunciation of
the accessedw-ord, as listed in it,
The items for which u p*itiu" response
lexical entry, must be matchedwith it. pronun.i.tio'
was required were
either nonwords or words. Twenty
iseudohomophones were
of the presented.
matched with 20 wgrds on orthtgiaphic
The pronunciationof the-presenteditem could itrT.
frequency of the words they sounJ"t-rit". structure and on
be
determined
eitherby graphe_."-ro_
r"i
was matched with RISK, STAWL with
pronunciationrules (see Coltheart,
"r"r"pri'i.rsr
CRAWi,
tl8O,bgure 10.1,
with ZINC.
"iO'OUSC
or by analogyto reaiwords (e.g.,Glushko,
Procedure.The presentationof the items
l:l{r.t-C)
was the sameasin
1979; Marcel, r9g0). Neither of thrr.
two experiments.fhis time, though,
pt*iu'i,i*
zubjectswere necessitates
,tl:. -ft::1, press
rnstructed
a phonologicalroute to the t.*i.orr. tn
to
the "yes" button if the worl *u, pionoun""a
it ,
an English word and to pressthe .,no,,
button if
former case,a phonologicalconversionof the graphlmes
t:_.i:lt_rye
was not' Ample practicewas provided
is not used to accessthe lexicon but,
so that suuir"trio"r"
rathei **r, u,
clear what the task
fi"
questionedall
a check with phonologicalinformation
subjectsafte-rthey had1n1of9a.
derivedfrom the
finishedt. u" "*p"r*ent
*"" that they pronounced lexicon. In the
latter case,the lexicon is accessed
the nonwordsin the way that wasintendeO.ft *Jr;;;;;ry
on a
to discardany data on this basis.Therewere
visual basis and
l0 subjectsin the
lhe pronunciation is determined from
experiment.
information stored
therein.
why is there.no significant reaction time
difference
Results and Discussion
between o-condition nonwords and G-condition
nonThe reaction time and error data for
words? It is possible that, in this difficuri
the third experiiurr.,'tt,
ment are presented in Table 3.
influence of grossgraphemicsimilarity
*., grruier in*
A significant difference between o-condition
in the lexical decision task. while t-he
nont",i,,""r.omowords and G-condition nonwords
phone conditions were matched
was observed for
on gross graphemic
errors [minF'(l ,29): l2.Ol, p ( .01J,
but noi io, similarity, the huge.variability introduc"eduy"ii'in-tt i,
reactiontimes(min F'< l). ln ihe lexilal
task (and possibly by other strategies
Aecision
tast
_uV well
have washed out
"froitime
significant reaction
differ^any
ences.An error effect was nevertheless
oUrr*.a, ,in.,
subjects were more_likely to make an
MeanReaction
Times," tffiti,tlirnds)andstandardErrors
error not only
{9tg With percentage Error Rate (pE) for
when
the nonword approximately resembleJthe
u ..uf
HomophoneDecisi6nTaskof n*i"rirn"nt
word, but arsowhen its-resemblance
f
to that rearword
wasrule governed.2
Condition
Example
RT
PE
^ Turning to the positive responseconditions, it was
Not pronounced Like a Word
found that words were respondedto
far rnorc quickfy
O Nonryords
CHEECE
1470
91
29
and with fewer errorsthan were orthographica[y-siJilar
G Nonwords
BREECE
iCqZ
92
10
nonwords [min F'(l ,2g) = 74.01, p < .OOl,
pronounced Like
fo, nf , LO
a Word
min
F'(1,19)= 5.32, p ( .05, foi irrorrJ
Word
.
This
resultis
RISK
828
4 4 4
not. intuitively surprising, yet it is
Pseudohomophone RIST
strong ovidence
1139 6t
t7
against any theory that ,uppor.s that
pfi"r"i"gi..f
and actual words. To say that RIST'i,
pronoun.iJlit.
a word involves
application of the .ui.r-n_rl|,
tr
r-lrl, s-/s/, t-ltl
*d th. accessingof the lexical
entry for WRIST, being
lnstl .In exactiy th, ,urn. w"y,
to say that RISK is pronounced like
a word would
y"9lu: the appticationbf tn, rutes R_+/rl t-lt/, Sir/,
,
K-lkl and the accessing
of the lexical entry for RIsK,
being /rrsk/. The grapheme-grapheme
conversionview,
on the other han{,.woutO preOict a difference
here,
sincerecognitionof the pseudohomophone
involvesirre
applicationof at leastone rule, whereas
the actualword
can be accessedwithout any rule application
at all.
Experiment 3 was therefore set up
to ascertain
whether grapheme-grapheme
rules are used instead of
grapheme-phonemerules even in
a task that requires
decisionsto be madeabout pronunciations.
470
TAFT
conversionis of primary importancein reading.If lexical would expect PURCE to be easierthat PERCE, but
accesswere based solely on grapheme-phonemerule VERCE to be easierthan VURCE,whereasthe graphemeapplication, then only the pronunciations of the stim- phoneme account would not expect this crossover
with orthographicstructure.Experiment4 was designed
ulus items would be important (as well as difficulty of
to test the effects of the orthographic similarity of the
Whether
rule application, but this was controlled for).
pseudohomophonesto the words with which they are
or not the orthography of an item matchedwith that of
a real word should have been irrelevant. Clearly, from homophonic.
An additional prediction of the grapheme-grapheme
the result obtained here, visual accessis the primary
was also tested in Experiment4. When a nonthat
account
that
items
means of performing the task, such
match exactly with a lexical entry can be responded word is not homophonic with any word, it should be
conversions
the casethat all possiblegrapheme-grapheme
to the most easily.
are attempted before a "no" responseis made,just in
There is nothing presentedin the data so far, how'
ever, that deniesthat accessvia phonologicalconversion caseone of them does producea correctspelling.If this
comesinto play in the homophonedecisiontask assoon is so, then subjects should take longer to classify as a
as direct accessfails. The differencebetween words and nonhomophone any nonword whose pronunciation
pseudohomophones simply says that direct visual can be spelledin a number of ways comparedto one
accessis the fust zuccessfulapproachto the lexicon; whose pronunciation can be spelled in only one way.
conversionsare
it does not say whether or not grapheme-phoneme For example, no grapheme-grapheme
conversionbecomesimportant wheneverdirect access possible with the nonhomophone TARL, except for
fails. The difference between the O-condition and G- ones that violate orthotactic rules, such as TARRL
condition nonwords says only that grapheme-grapheme (R+RR) or TAWRL (R-+WR). Thus responsetimes
rules are employedin the task at somestage;it doesnot should be faster to TARL than to RAWL, for which a
say whether or not they arebrought into play only after number of legalconversionsare possible,such asR-+WR,
R+RH, AW+AU, and AW-+AL (as in WALK). Again,
direct visual accessand phonologicalaccessfail.
conversion(and also
con- on the basisof grapheme-phoneme
In order to examinewhether grapheme'phoneme
be no differthere
should
graphemic
similarity),
gross
version is more important in the homophone decision
TARL
like
and
a nonword
a
nonword
ence
between
one
needs
conversion,
grapheme-grapheme
task than is
like RAWL.
to look at a situation in which direct accessfails to
locate a lexical entry but phonologicalaccessshould not
fail. That is, one needsto manipulatethe pseudohomo- Method
Materials. Two different types of pseudohomophones were
phone condition. This was done in the fourth expericonstructed: one in which only a single graphemegrapheme
ment.
conversion was performed on the homophonic word (e.g.,
SCREAM, SKEME, PURCE, VERCE), and one in which two
conversions were performed (e.g., SKREME, SKEAM, PERCE,
VURCE). For every item in the one-grapheme-changecondition'
there was an item of similar orthographic structure in the twoIf the homophone decision task is primarily carried grapheme-change
condition (e.g., SKREAM and SKEAM ; SKEME
out via grapheme-phonemeconversion rules (when and SKREME; PURCE and VURCE; VERCE and PERCE).
direct visual accessfails), it should then be the casethat
There were two lists of items constructed, each presented to
the orthographic similarity of a pseudohomophoneto a different Soup of subjects. Each list contained eight oneand eight two-graphemethe word with which it is homophonic should not have grapheme-change pseudohomophones way
that no one subject
change pseudohomophones in zuch a
For example,it should not saw both spellings of the one homophone. For example, one
any influence on responses.
matter whether the word /skri:m/ is presented as group received SKREAM and SKEME as onegrapheme-change
SKREAM or SKREME, eventhough the former is ortho- pseudohomophones and PERCE and VURCE as two-graphemegraphically more similar to SCREAM than is the latter. change pseudohomophones; the other group received PURCE
ruleswould be applied and VERCE and SKREME and SKEAM.
hTboth cases,grapheme-phoneme
In addition to the pseudohomophones, l0 nonhomophonic
to produce /skri:m/ and, thus, accesswould be based nonwords were constructed so that no grapheme-grapheme rules
rule could be applied to them (e.g.,TARL, GWELVE). Matched with
on the same representation.Grapheme-grapheme
application, on the other hand, would lead to SKREME these on length were 10 nonhomophonic nonwords to which
being more difficult than SKREAM, since the former grapheme-grapheme rules could be applied (e.g., RAWL,
KWIEVE).3 Both groups of subjects received the same 20 noninvolvesmore conversions(K-+C,E-E-+EA).
homophonic nonwords.
It is possible,however,that the grapheme-phoneme Procedure. The procedure was the same as for Experiment 3,
rule E-E+/i:/ takeslongerto apply than EA-+/i:/, since but this time there were two groups of subjects, with eight
it is rarer. If this were so,however,then not only would subjects in each Soup.
EXPERIMENT 4
SKREME be harderthan SKREAM, but SKEME (homophonic with SCHEME) would be harder than SKEAM. Results
Table4 provides reaction times and error rates for
conversions
Should the number of grapheme-grapheme
be important instead, SKEME should be easier than the fourth experiment. The amount of orthographic
SKEAM. Similarly, the grapheme-graphemeaccount similarity between a pseudohomophoneand the word
GRAPHEME.PHONEMERULES
Table 4
Mean Reaction Times (in Milliseconds) and standard
Errors
Along With Percentage Error Rate (pE) for the
Homophone Decision Task of Experiment 4
i
I
I
I
471
the lexicon via grapheme_phonemerule conversion.
Even if such a route does exist, it certainly appears
to
-visual
play very much of a secondary role in
word
recognition. The two findings that have been previously
Condition
Example
RT
SE PE
taken as evidence for the existenceof a phfnorogical
pronounced Like a Word
route to the lexicon, namely, the pseudohomophone
One-Grapheme Conversion
SKREAM
1007 39
effect and the ability to perftormthe homoptrone'aeci7
Two4rapheme Conversion
SKREME
rr94
s4 30
sion tasks,have been shown here to be explainable
by a
Not Pronounced Like a Word
visual route using grapheme_grapheme
rule conversion.
-to
RulesApplicable
RAWL
tlt2
56
16
produce a pseudo_
RulesNot Appticable
_ When dyslexjc zubjects fiil
TARL
t36Z 73
8
homophone effect and, in addition, are unuut.
to
successfuilyperform the homophone decision
task, it
with which it is homophonic was found to have an
effect may be becausethey ur. u.r.ble to
apply g.uph._._
on homophonedecisiontimes.The one-grapheme-change
graphemerules,and not becausethey have'utii-f"ir.a
condition was easierthan the two-graphle-r-.h*je.onphonological route to the rexicon. ittry do, hoivever,
dition as measuredboth by reacti& ii-,
seemto have a further disability in that ih.y-u.. unable
1-in f\,2e1
=9, 43, - p ( . 01] a n d b y
e rro rs [mi n F ,(i ,2 5 1 =' 9.0r, to pronounce nonwords.
This implies thai they have
p< . 0 1 J .
an impairment of the mechanism that gets from
the
It was also found that the nonhomophone manipu_
printed word to its pronunciation,be it bigrapheme_tolation was significant. When graphem._grupt
r-e rules pronunciationrules (Coltheart, l9g0)
or by an analogy
were possibleto apply, subjectstook long.i th* when
technique(Glushko, 1979; Marcel, l9S0). This impairthere were no possible rules
[min Fit ,211= 4.g3, ment means that, when RAyoN is generated
as being
p <.051 . The error difference wai signifitant only
on
homophonic with RAYNE (on tfrc basis oi gros
th e_ s ubjec t
=
'
1
0
.3
5
,
(1
,1
5
)
p l . Ot;
[F ,
graphemic similarity), the postaccesscheck for
.lyll
coirect
F 2 ( 1 , 9=
) 3 . 6 4 ,Tp>
. 0 5 1.
pronunciation cannot be carried out and, hence,
the
error is maintained. Normal readersmay also generate
Discussion
RAYON from RAYNE but reject it when the"phono_
Looking at the pseudohomophonesfirst, it can be
logical check fails. on the othei hand, the word
RerN,
seen that a decision about th; pronunciation of
an generatedby grapheme-grapheme
conversion,would be
item is influenced by orthographiciactors. If pronunciaacceptedafter the phonologicalcheckis made.
tion were determined primarily by grapheme_phoneme
The main problem with the view expressedin this
rules (even as a_backupto failed diieci visuj access),
paper is, of course,the counterintuitivenature
of the
then. the only factor influencing performance
shourd existence of grapheme-graphemerules.
Graphemebe the difficulty of applying such-rules. This
experi_ phoneme rules have been rationarizedby
assumiigirr.t
ment holds constantthis factor of rure difficulty and yet
beginning readers make use of their already
Jxtant
tinds that the more orthographically similar ih.
non- phonologically accessiblelexicon by simply
,on*rtrng
word is to the word, the more difficurt is the response.
the written word into a phonological fbrm. The
eviIf. grapheme-phonemerules are used at all, the|cer_ dence presented here, however, suggeststhat
adult
tainly appear to play very much a secondaryrole
to
readersno longer do this, at least ., tli. primary
means
visual access.Together with the findings of'the preof
vious experiment,it can be said that, evenin a task
.word recognition. lnstead, phonoiogi.j u...r,
that
might be seenas a backup for visuil u...rr-*hen visual
optimizes the likelihood that grapheme-phoneme
rules accessfails (as also suggestedby coltheart).
But when
will be used,the applicationof graptremr_g.upt
e.e rules doesvisualaccessfail in everydayreading?
appearsto be the preferredmethod of periorming
the
One encounters nonwords in norm-al reading only
task.
when one reads new words, proper narnes,misprints,
_ The finding of a delay in reactiontimes to nonhomo_
misspellings. Application of grapheme_phoneme
phones for which grapheme-graphemerules
can be rules will be of use in understandingn.w
*o.ds o,
applied further supports this view. with items like
proper namesonly if thosewords havebeen encountered
RAWL, there is a needto try out a numberof grapheme_
before verbally and if they have a regularpronunciation.
graphemeconversionsbefore a ..no', ,.rponi,
,an U,
Even then, it is possiblefor there to havetrrr, * ortho_
made. With items like TARL, subjects can respond graphic
representationset up when the word was first
more rapidly becausethe absenceof any appropriut.
encountered verbally (cf. Tanenhaus, Flanigan,
&
grapheme-grapheme
rules means that tho item clrnot
S.eid
enberg, 1980), hencerenderingphonologi..ll onurrbe homophonicwith a word.
sion unnecessary.Grapheme-phoneme
conversionwourd
be of use in understandinga misprint or a misspelling
GENERAL DISCUSSION
only if the error wete phonologicaflyidentical toln,
intended word. If it were not, only gross graphemic
-However,
- Taken together, the four experimentsreported raise approximation would lead to recognition.
doubts about the existenceof i phonologicalroute
to
when grapheme-phoneme
conversion could be used.
472
TAFT
of orthographic
equally GlusHxo, R. J. The organizationand activation
P:y'
grapheme-graphemeconversion could be as
knowledge in readini aloud' Journal of Expe!m9nt-?!.
as
there
614'691'
5'
1919'
P-erformonce'
.is
Efiectivet/ used. It seems,therefore, that
iiotogl: numon Perciption-and
B., & cosiv, M. J' one secondof readingagain'
much of a rationale in postulating grapheme-grapheme G;;;;:i.
-i;
grapheme-phoneme
D- B. Pisoni,& G' R' Potts (Eds')' Cognitive
postulating
n.'Castelian,
of
is
conversionas there
'
N'J: Erlbaum,1977
Hillsdale,
2).
ii*ry(Vol.
conversionin normal adult word recognition'aystexiaand beginningreading:A revised
Surfa..
J.
frf^*"r",'e.
^'--h;;ili.ris
conbe
In
Grapheme-phonemerules, however' mght
of the pronunciationof print and its impairments'
beginning
(Eds')' Deep
sidered important in word recognition for the
tri. Cottt."rt, K. E. Patterson,& J' C' Marshall
a
iytirio. London: Routledge& KeganPaul' 1980'
iiua.t. Tlie child enters the reading situation with
information
basis'
M;6-;;;;, b. v. variable .ili"n.. bn phonological
lexicon that is accessibleonly on a phonologicalLanguogeand Speech'lg8l' 24'
recognition'
word
visual
in
phonological
a
into
word can be converted
If the printed
-on
99-109.
and the
the basis of rules, then the already extant pirrr*ron,
form
K.E., & Mlncrl, A. J. Aphasia,dyslex-ia
-';h;;ldical
of
need
used
Journal
rules
the
However,
written^I"-t9:'
Quarterlv
coding-oi
lexicon can be accessed.
I
8'
the
307-3
be
22,
gv,
7
may
127
rulesbut, rather'
Exie rimZntal PsvchoIo
:
nol U. grapheme'phoneme
u., it*ttl-S. s', & Runrxstntn' M' A' Evidence
grapheme-ptonuniiation rules that adults possibly use R;;;;t;;;",
Journol of
for phonemi. ,..oainj in visual word recognition.
this
645-657.
In betermi.tit g the pronunciation of nonwords' By
1971,10,
Behavior,
ierbat
and
verbalLearning
rule
without phono.logy:
account, ttre child overtly pronouncesthe word by
s"".*"n, n. u., b Menrn, o. s. M. Reading
Psynormal
the
Evidencefrom aphasia.QuortertyJournal of Experimentol
and then the output of this passesthrough
read
I
5-526'
can
5
29,
child
1917,
the
is,
T-hat
chologY,
speech recognition system.
L...tt via an orthographic-code: The,Basic
is T;;-Ivi.'iexiial
-'Orif,ogr"ptric
onty UV tp.utittg thi word aloud, and, in fact' that
Structure (BOSS)' Journal of Verbol
Svffalic
transi
.*u.tly w'hat belinning readersseemto do' The
VerbolBehavior,l9?9' lt' 21-39'(a)
Lrirning'ond
a
frequency
tion to silent reading-might take place either when
M."i..ognition of affixed words and the word
T;*,
(b)
when
or1979,1,26t'272'
acquired
is
Cognition,
&
route
Memoiy
.rii.t.
direct orthographic accesi
stripp'ingrevisited' J-ournalof Verbal Leornins
the graph.-t-pionunciation rulesbecomemore abstract To;;, M. Prefii
'
289-297
Behavior,1981,20,
Verbat
and
present
The
of preand become grapheme-phonemerules'
T;;i M., g Fonstnn, K' I' t-exicalstoragea1d 1et1ie11l
resultssupport the former alternative'
fixedwords.Journa'lofVerbalLearningondVerbalBehavior'
I t w o u l d b e a t t h i s t r a n s i t i o n s t a g e t h a t g r a p h e m e . 1975,14,638-&7.
S'
graphemeTlNnxuaus, M. K., Fllxrcen, H' P" & Sntonxsnnc'.M'
gruph.rn. rules develop' Knowledge of
in auditoryand visual
onologicaiactivation
profrom
orthographi.
originate
must
"na-pt
irupn.tn. correspondenie-s
a cognition, 19E0,t, 513-520.
word recognition.-Me^oly
iunciational knowledge. However, once the transition
two
to direct orthographii accessis completed' these
the
that
in
types of knowl"edgebecome independelt'
NOTES
knowledge
provides
system
grupt.-t-grapheme rule
without
for ExperiiUo"t phoiroiogicalrelationshipsbetweenwords
1. Two G-condition items were in fact changed
and BOOLwaschanged
GAUVE
to
.ftung.d
*u,
recoding'
CAUVE
3.
ment
to
,r.orrtie
Phonological
realized that CAUVE could be pro-
to HOOL, since it was
could be prontun*o like covE (as in MAWE) and BooL
REFERENCES
(as
WOOL)'
in
nouncedlike BULL
one of the
2. It is interes-tin!to note that' inadvertently'
orthographic.and-wordof
Use
C'
Stnnwsox,
&
J.,
be homophonicwith a
Benox,
-';;;ih;
to
out
turned
non*o,i'
G-condition
inowteaeBin readingwordsaloud.Journalof Experi'
homophonicwith
1976' ;ord, but an itn*t"O one: CREESTbeingdetecle!
numonF"'eption andPerformance'
this' it is
mentalPsycnot-o;yt
subjects
10
of
the
2
ontv
CREASED. Since
2,386-393.
evidencefor morphologicaldecomposi;
this
see
to
tempting
HeriAmericon
The
is'lexical
.f. B., Dnvtns,P', & RtcHMlx'B'
Citi*"i",
-'iir
ii* ti#t, 1g79b,1981;Taft & Forster,1975)'That
l97l'
of a word (e.9.,
i"ra frequencybook; Boston:Houghton-Mifflin'
stem
the
of
basis
the
on
performed
is
in lexicalaccess. u.."ri
c"""""r*r, 5. rvr.L.it.r andorderinformation
CREASE)and not on the inflectedword asa whole'
1979'lE'
comply
Journol of viOii t'eorningand VerbalBehovior'
3. Someof the items used in the experiment9t9.":t
225-Ur.
w
i
t
h
o
r
t
h
o
t
a
c
t
i
c
r
u
l
e
s
'
F
o
r
e
x
a
m
p
l
e
'
G
W
o
c
c
u
r
s
i
n
i
t
i abrand
llyonlyin
In
G'
in simplereading3s$'
Co"t"ti*t, M. Lexicalaccess
'it and KW- occurs initially onlv in
names(CWEN),
London:
processrng'
;*;
i4formotion
of
be
not
it'atesiu
Underwoodfeail
was consideredthat this would
names(KWIK-GRipj
relevant
only
Press,1978'
are
rules
Academic
orthotactic
since
recoding,and deepdys- important, tro*euer, be made about the "Englishness"of the
C"ir"r^*r, tr,t.i.aaing, phonological
& J' Ci Marshall when a decisionis io
Fatterson'
e'
f'
nonword item
i"fnt"ii
r"r.
In
lexia.
oJftogiuptty of an item' In this experiment'the
&
R-outledge
London:
[eganPaul'1980' itself was no, ift" subject of such a decision'On the other
(Eds.),O"rp ayirti'
rtr.1-o"ut*n*, E" Jownssox'J' T" & Bnsxnn' hand, the decisionwhether or not a letter string createdby the
c;;;;;;t'
(Ed')' Attenis a word mav well
D. Access,o itti internallexicon'In S' Dornic 1977'
;;ph;;elrapheme.rules
apptcation
Erlbaum'
N'J:
Hillsdale'
//'
perfor:mance
"f Ui tie Engiistrness
of that letter string,and thus
and
tion
influenceO
be
T'
J'
E.,d;;;;;enr, M" BnsNon'D" & Jorussox'
Dnvnr,AAn,
aracter"isticsoftheletterstringscreatedby
Memory& Cognition' t h e o r t h o t a c t i c , c hrules
andlexicalaccess'
recoaing
Phonologicaf
grapheme-grapheme are important'
39142.
197E,6,
473
GRAPHEME-PHONEMERULES
APPENDIX
kxical
Table A
Decision Times CI) and Errors (E) for the Nonword ltems Used in Experiment I
BROE
HATVE
DWE
MOOTH
PUTCH
LOAZE
1
750
763
740
726
707
744
737
741
726
756
73s
76r
816
7s9
772
744
GLOE
SAIVE
MUVE
YOOTH
TUTCH
HOAZE
TOOM
HOAM
GHOAST
LEACE
BRETH
LEEF
PLEED
WOSP
WOOND
JERM
WIRK
DERTH
SWARE
FLUD
N Nonwords
O Nonwords
H Nonwords
J
0
4
2
2
5
I
0
2
cooM
J
4
10
0
2
0
0
5
5
2
7tr
742
833
796
SOAM
FROAST
TEACE
WRETH
DEEF
DREED
GOSP
ROOND
JIFT
FIRK
HERTH
CLARE
BRUD
1
0
0
2
0
2
2
1
0
3
1
0
0
0
I
0
0
1
I
2
710
701
704
729
718
732
743
722
69r
769
674
757
743
703
661
667
7t4
688
716
683
SPOE
FAIVE
BUVE
NOOTH
RUTCH
MOAZE
YOOM
GOAM
PLOAST
HEACE
FRETH
SEEF
CLEED
DOSP
TOOND
JILK
SIRK
LERTH
PLARE
GRUD
0
4
I
7
5
0
I
2
0
2
I
0
2
I
5
0
0
0
l0
1
7s8
803
755
793
831
683
727
718
73r
716
790
772
787
740
693
849
682
772
79s
672
Table B
I-exical Decision Times (LT) and Errors (LE) for Nonword ltems in Experiment 2, Along With
Homophone Decision Times (HT) and Errors (HE) for Those Items in Experiment 3
G Nonwords
O Nonwords
LT
STEEK
TOWCH
SWARD
CHEECE
SWEER
BRODE
HALPH
HIERCE
YOWTH
KELL
COAPY
WICE
CHALC
SOOL
BREEST
HATVE
MUSKLE
WUDE
GLOAVE
SOOGAR
Note-llords
(steak)
(touch)
(sword)
(cheese)
(swear)
(broad)
(hal0
(hearse)
(youth)
(cell)
(copy)
(wise)
(chalk)
(soul)
(breast)
(have)
(muscle)
(wood)
Glove)
(sugar)
6s5
683
649
735
807
830
563
690
605
72s
6t6
7rs
648
734
725
632
630
639
682
668
HE
LE
0
2
0
0
I
0
0
0
2
I
0
0
0
2
0
1
0
0
1
0
1s40
r402
1430
1683
r549
1506
r349
1486
1 53 3
t37 |
1 4 81
t253
l 303
r320
1498
1695
1445
1s84
13 8 0
t652
4
I
4
4
2
2
3
I
4
1
1
I
6
1
2
2
6
5
3
4
FLEEK
MOWCH
SWOFT
BREECE
BLEER
FRODE
STALPH
VIERCE
TOWTH
KEND
PoAPY
SICE
HALC
BOOL
CREEST
CAUVE
RUSKLE
SHUDE
PROTVE
SOOVEL
(fleck, flake)
(much)
(swift, soft)
(breeze)
(bleed,blare)
(froze,fraud)
(stalk, staf0
(verse)
(tooth)
(send)
(poppy)
(size)
(halt,hawk)
(bowl)
(crest)
(cave)
(rustle)
(shade,should)
(prove)
(shovel)
LT
HT
HE
595
s90
706
734
619
s94
611
632
631
607
601
616
682
610
662
680
627
637
652
624
t2r4
t444
1400
1616
t43t
t293
1348
1339
1s46
1358
t673
1470
1415
1326*
1601
1 3 5l *
1378
1692
1380
t723
0
0
0
0
2
I
0
I
3
0
2
I
0
0
2
1
I
3
3
0
0
0
J
2
1
0
0
0
0
0
0
0
0
0
0
0
I
0
0
0
in parentheses are words that ore graphemically andfor phonemically similar to the nonword item.
*See Footnote l-
Table C
Homophone Decision Times (T) and Enors (E) for the Homophonic Items in Experiment 4
Two4rapheme
One$rapheme Change
SKEME
SKREAM
PURCE
937
862
9s8
0
0
0
SKEAM
SKREME
PERCE
Change
1008
1278
1344
I
0
3
474
TAFT
Table C Continued
TwoCrapheme Change
OneCrapheme Change
VERCE
SKAIT
SCAIR
ROAP
WREQUE
ROOLE
SOOP
PHEIGN
KRANE
GNOAM
KNEAL
SKWIRM
PHERN
947
843
890
r026
IIO2
1080
1084
11 8 8
8s8
11 5 6
820
11 8 7
1259
2
0
0
0
0
0
2
1
0
a
J
0
0
1
VURCE
SCAIT
SKAIR
WROAP
REQUE
ROOL
SOOPE
PHANE
KREIGN
KNOAM
GNEAL
SKWERM
PHIRN
16 2 8
r246
1003
t192
11 5 0
969
11 8 5
11 9 8
r02l
tr44
998
t37 3
1,394
J
1
0
2
4
2
1
,7
5
2
J
0
5
Table D
(E) for the Nonhomophonic ltems in Experiment 4
Homophone Decision Times (T) and Errors
Rules Not APPlicable
Rules Applicable
RAWL
SKACE
PHLESK
CLEED
WRAD
JIRCE
KWIEVE
WOONE
KRAIST
WHOSC
rs26
13 5 4
13 8 s
1406
t472
t434
1 s8 2
1465
t644
t67l
2
1
6
0
I
0
4
8
2
2
TARL
SPANT
THRISP
BLOIN
PRUB
LINGE
GWELVE
VOOTH
GLANCH
DWUSP
(Received
for publicationNovember24,l98li
MaY I I, 1982.)
revisionaccePted
I 159
t265
tzl6
t437
r228
1518
r422
1488
r233
r436
I
0
2
0
J
0
1
2
2
1