the nature of the sound codes accessed by visual language

JOURNAL OF MEMORY AND LANGUAGE 38, 70–93 (1998)ARTICLE NO. ML972536

The Nature of the Sound Codes Accessed by Visual Language

Stacy Birch

Hampshire College

and

Alexander Pollatsek and John Kingston

University of Massachusetts

The nature of the sound codes accessed by visual words was explored in four experiments.Experiments 1 and 2 employed a homophony judgment task and Experiments 3 and 4 employeda lexical decision task. Experiments 1 and 3 compared phonetic pseudohomophones (phoneticPHs) such as nootle to various controls including phonemic PHs such as speek. Phonemic PHswere more easily judged as sounding like words than phonetic PHs, but other comparisonsindicated that both phonetic and phonemic codes were involved in homophony judgments. Incontrast, only phonetic codes were clearly implicated in the lexical decision task. Experiments2 and 4 compared PHs whose morphemic complexity mismatched the baseword (e.g., counce)to PHs whose morphemic complexity matched the baseword (e.g., discloaz). There was a cleareffect of morphemic mismatching in the homophony judgment task, but little or no effect ofmorphemic mismatching in the lexical decision task. These data suggested that morphologicalstructure is part of phonemic coding but not part of phonetic coding. q 1998 Academic Press

A view that once held center stage in cogni- tions rather than the more widely used term‘‘phonological code’’ because the latter sug-tive psychology is that identification of writtengests a commitment to a phonemic representa-words occurs solely through direct visual ac-tion that we believe is premature. We willcess. That is, the letters of a printed wordintroduce more specific terminology shortly.)access a node in a ‘‘visual lexicon’’ through

A variety of evidence has accumulated froma process that does not involve intercessionseveral paradigms, however, that indicatesfrom any other type of linguistic coding. Athat sound codes have a more central place inrelated position, the classic dual-route theoryword identification. Van Orden (1987; 1991)(Coltheart, 1978), posited that direct visual ac-showed that sound codes play a crucial rolecess is the primary route to word identificationin judgments of whether words are exemplarsbut that a secondary, back-up route operatesof semantic categories. Participants madethrough the intermediary of a sound code. (Wemany more errors on homophones of realwill use ‘‘sound code’’ to refer generically towords than on other words matched on visualspeech- or pronunciation-based representa-similarity to the real category exemplar (e.g.,falsely judged meet to be a type of food much

The first author was supported by a National Institute of more often than melt). While this effect dimin-Health Post-Doctoral Traineeship (HD07327), the second

ishes when broad categories such as ‘‘livingauthor by Grant HD26765 from the National Institute ofthing’’ are used (Jared & Seidenberg, 1991),Health, and the third author by Grant DC01708 from the

National Institute of Health. We thank Charles Clifton for interference from homophones also occurs forhis comments on an earlier draft of this manuscript. semantic relatedness judgments (e.g., bare

Address requests for reprints to Stacy Birch, School and wolf; Luo, 1996), indicating a significantof Cognitive Science, Hampshire College, Amherst, MA

role for a sound code in accessing written01002 (E-mail: [email protected]) or Alexander Pol-word meanings. Other work using semanticlatsek, University of Massachusetts, Amherst, MA 01003

(E-mail: [email protected]). judgment paradigms has assessed whether

700749-596X/98 $25.00Copyright q 1998 by Academic PressAll rights of reproduction in any form reserved.

AID JML 2536 / a010h$$121 12-17-97 08:51:24 jmla AP: JML

71TYPE OF SOUND CODES

word meanings are merely activated by stored of related words produce priming (e.g., doesbeech prime sand?). While the exact condi-sound codes that are directly accessed by the

lexical entry of the written word (i.e., ad- tions for such priming are not clear, the phe-nomenon appears to be reasonably robustdressed sound codes) or whether they are acti-

vated by sound codes that are assembled when (Lesch & Pollatsek, 1993; Lukatela & Turvey,1994). Finally, Folk and Morris (1995; 1996)the graphemic input is presented. Two pieces

of data argue that assembled sound codes con- measured fixation times on heterographic(e.g., sole/soul) and homographic (calf) homo-tribute to at least part of the above interference

effect. First, Van Orden, Johnston and Hale phones and on heterophones (e.g., tear) andconsistently found evidence for early activa-(1988) showed that pseudoword homophones

also produce the phenomenon (e.g., sute was tion of sound codes during word identificationin reading.falsely judged to be clothing). Second, Lesch

and Pollatsek (1998) showed that a smaller All of these paradigms indicate that somesort of speech-based code is assembled duringeffect can be produced from possible but not

actual homophones (e.g., beard is difficult to word identification, but they say little aboutthe nature of that code. One issue is the degreejudge as unrelated to robin because beard

could be a homophone to bird, if pronounced of specificity in the sound code. One possibil-ity is that this code is an ‘‘auditory image’’like heard), indicating that assembled phonol-

ogy is involved even when words are being (Baddeley & Lewis, 1981). McCutchen, Bell,France, and Perfetti (1991) have proposed thatprocessed.

Sound coding has been shown to be im- the sound code used in sentence-level compre-hension and memory processes is ‘‘phoneti-portant in paradigms more directly related to

word identification as well. One approach has cally specific’’ (p. 102) and includes at leastword-initial consonants and articulatory infor-employed a backward masking paradigm. Per-

fetti, Bell, and Delaney (1988; see also Per- mation. That is, the assembled code may ap-proximate an actual pronunciation. However,fetti & Bell, 1991) found that a pseudohomo-

phone mask (e.g., mayd) was less disruptive not all of the pronunciation may be availableat the same time. Recent work by Berent andthan a graphemically similar mask (e.g., mard)

during identification of a target word (e.g., Perfetti (1995) has suggested that the soundcode may be built up in stages, with conso-made), implying that a sound code for the

written word is assembled before a stored one nants being constructed first and vowels beingconstructed later. Another possibility, how-is addressed. Pollatsek, Lesch, Morris and

Rayner (1992) demonstrated that a parafoveal ever, is that the ‘‘sound code’’ is quite abstractand merely contains (phonemic) codes that arehomophone preview of a word facilitates iden-

tification of the word when it is fixated about sufficient to discriminate spoken words in thelanguage from each other. A second issue with200 ms later, relative to preview of an ortho-

graphically similar word. For instance, parafo- respect to the sound code which remainslargely unexplored is the extent to which itveal preview of bare facilitates identification

of bear more than does preview of bean. This includes information about the word’s mor-phological structure.effect was demonstrated both with isolated

words using naming time as the dependent Our goal in this study is to investigate thenature of the sound code or codes generatedvariable, and with silent reading of sentences

using fixation time on the target word as the given a graphemic input. Consider the writtenword print. There are at least three relevantdependent variable. Similar homophone ‘‘fast

priming’’ results have been obtained in read- types of information that could be representedin the sound code assembled from this graphe-ing within a single fixation (Rayner, Sereno,

Lesch, & Pollatsek, 1995; Lee, Binder, Kim, mic input, which will be referred to as pho-netic, phonemic and morphophonemic codes.Pollatsek, & Rayner, 1997). A related ap-

proach has been to test whether homophones First, if the initial consonant is specified as


72 BIRCH, POLLATSEK, AND KINGSTON

aspirated [ph], in contrast to an unaspirated [p] /t/ and /d/ is an instance of morphophonemicalternation, or, in monomorphemic wordsin the sound code for sprint, then predictable

phonetic information is present in this code. such as latter and ladder, allophonic variation.The specific distinctions which are lost inIf, on the other hand, these consonants are not

distinct, but the sound code for the word prints these two examples are perhaps of less impor-tance than the fact that both produce (near)has a /t/ which is not present in the sound

code for the word prince, then the sound code homophony. Because it is possible though rarefor speakers to pronounce facts with an audi-at least includes information about phonemic

contrasts between one word and another. A ble /t/, this word is not always homophonouswith fax. Similarly, some speakers occasion-sound code in which only phonemic informa-

tion is represented approximates an idealized ally produce vowels of modestly different du-rations (Fox & Terbeek, 1977), or slightly dif-alphabetic code (Venezky, 1970), rather than

an idealized pronunciation. Finally, does the ferent qualities (Joos, 1942) in writer vs rider,so their pronunciations may not be phoneti-sound code for prints include a morphological

boundary between the /t/ and the /s/ which cally identical. However, the high frequencyof cluster simplification and the obligatorinessis absent in the code for prince? If so, then

morphophonemic information is present in of flapping substantially reduces the phoneticdifferences between these pairs of words, onthis code.1

Two further examples illustrate how we occasion even to zero.The first question we address is whether thewill investigate the distinctions between kinds

of information in the experiments reported sound codes that are most relevant in pro-cessing written language are phonetic, phone-here. First, in the phonemic representations of

the word facts, there is a /t/ in the final cluster, mic, or both. One could make a plausible argu-ment either way. On the one hand, if the writ-between the /k/ and the plural suffix/-s/. The

/t/, however, is seldom pronounced, which ten language is supposed to produce a tokenequivalent to the spoken word, then one wouldrenders this word’s phonetic realization indis-

tinguishable from that of fax, which has no expect the sound code to contain phonetic in-formation. On the other hand, written lan-/t/ in its phonemic representation. Second, the

phonemic representations of the words writer guage may more easily access the phonemiccode because it is closer in form to it; thusand rider contrast in the final consonant of the

verb stems, [-voice] /t/ in writer vs [/voice} the sound code most relevant for a given taskwould contain phonemic information, either/d/ in rider, but these consonants are pro-

nounced as an alveolar tap obligatorily in instead of or in addition to phonetic informa-tion. By ‘‘most relevant’’ we mean most ac-American English (Zue & LaFerriere, 1978)

in nouns formed by adding the agentive suf- cessible to other cognitive systems, to re-sponse systems, or possibly to consciousness.fix -er. The omission of /t/ from the phonetic

realization of facts is an instance of the so- That is, it could be that both phonetic andphonemic codes are activated by the writtencalled ‘‘cluster simplification’’ that often oc-

curs in fast or casual speech, whereas the tap- language, but the phonemic code is only in-volved as an intermediate coding step to form-ping that neutralizes the contrast betweening a phonetic representation but is otherwiseirrelevant to the processing stream. An anal-1 The distinctions we draw here between phonetic, pho-ogy might be drawn with coding in the visualnemic, and morphophonemic information overlap with

those made in the linguistics literature between ‘‘surface’’ system, in which the retinal size (and other 2-vs ‘‘underlying,’’ ‘‘postlexical’’ vs ‘‘lexical’’, and ‘‘out- dimensional aspects of the image) are codedput’’ vs ‘‘input’’ levels of representation (for the last, see at the level of the retina and primary visualPrince & Smolensky, 1993). However, we focus narrowly

cortex, but apparently are not easily accessedon what kinds of information are present in the soundfrom other systems. We further address thiscode readers construct for written words, and not on the

other baggage associated with these linguistic distinctions. issue in the General Discussion.



The second question addressed in the cur- mation may not be. If that is the case, morphe-mic information should be relevant to lan-rent experiments is the role of morphophone-

mic information in sound coding. There is guage processing tasks only when phonemicinformation is shown to be relevant. An alter-very little empirical evidence that bears on the

question. Even though the role of morphemes native possibility is that morphological con-stituents are available to all other componentsin visual word identification has been studied

extensively, the major focus has been on of the system that identifies written words.The third question addressed in the currentwhether individual morphemes are accessed

as part of the normal process of word identifi- experiments is whether the nature of the soundcode used varies with the task. There is somecation. One possibility is that the morphemes

of polymorphemic words are not a component evidence that the degree to which readers usesound coding varies according to the taskof the process of word identification, but are

accessed only after the lexical entry has been (Davelaar, Coltheart, Besner, & Jonasson1978; Hawkins, Reicher, Rogers, & Peterson,accessed. Another possibility is that polymor-

phemic words are identified morpheme by 1976), raising the possibility that the natureof the sound code varies according to contextmorpheme. The work investigating the role of

morphemes in lexical access has employed a or task demands. Crowder and Wagner (1992)speculated that for access of individual words,variety of paradigms, including priming stud-

ies (e.g., Deutsch, Frost & Forster, 1996; the relevant sound code is likely to be a moreabstract representation (i.e., a phonemic code),Fowler, Napps, & Feldman, 1985; Stanners,

Neiser, Hernon, & Hall, 1979); comparing ef- whereas for storage in short-term memorywhile sentence comprehension processes arefects of the frequency of an entire polymor-

phemic word to the frequency of component being carried out, the relevant sound code islikely to be a type of verbatim memory (i.e.,morphemes (e.g., Burani & Caramazza, 1987;

Taft, 1979); and recording eye movement pat- a phonetic code). In the present experiments,we compare not word and sentence-levelterns (Beauvillain, 1996; Hyona & Pollatsek,

1997). Results from such studies indicate that tasks, but two word-level tasks. One task, ho-mophone judgment, specifically queries read-morphemic analysis does play a part in lexical

access. ers’ intuitions regarding the pronunciations ofletter strings. For the other task, lexical deci-In the present study, the focus is on the

relation between morphological structure and sion, the pronunciation of a letter string islogically irrelevant and potentially interfering.the sound code. In one sense, this question is

simply another way of probing the representa- Thus, readers may tap a different type ofsound code for the two tasks; plausibly, theytion of the sound code. It seems likely that

phonemic coding often relies on morphemic may generate a richer sound code in the homo-phone judgment task than in the lexical deci-decomposition (e.g., the pronunciation in

broad transcription of Coltheart would likely sion task.be [koluart] without morphemic analysis) and

Plan of Researchtherefore that morphological codes are eitherpart of the phonemic representation or that the We investigated the preceding questions in

four experiments, all of which employed non-two coding systems are in close communica-tion with each other. In contrast, there seems words that are homophonic to real words (i.e.,

pseudohomophones). The rationale for thisto be no a priori reason to expect that morpho-phonemic coding plays any direct role in pho- strategy was twofold. First, by using nonword

homophones, we ensured that judgments ofnetic coding (though it is likely to shape pho-netic coding through phonemic coding). Ac- homophony were based on a sound look-up

at the time of the experiment. With real words,cording to this hypothesis, morphological andphonemic information may be represented to- judgments of homophony could logically be

based on ‘‘fact retrieval’’ (e.g., a previous ob-gether, but morphological and phonetic infor-



servation that their and there are homo- word (grew); and (2) phonetic pseudohomo-phones (henceforth, phonetic PHs), nonwordsphones). The use of nonwords also ensured

that some kind of assembled or prelexical pro- like sticts or drifded whose phonetic represen-tations are identical or nearly identical to thosecess must be used in looking up the sound

code. The use of nonwords has some draw- of real words (sticks and drifted), but whosephonemic representations differ from all realbacks, the chief one being that the issue re-

mains whether the processes used for encod- words. (Note that our phonemic PHs are pho-netically identical to a real word as well asing nonwords are involved when real words

are encountered in text. However, we believed phonemically identical to it.) In both cases,we will refer to the real word homophone asit was a reasonable research strategy to start

with a process in which a sound code must the baseword.The phonetic PHs were used in two com-be accessed by an assembly process and then

move from there to tasks closer to actual word parisons. In Experiment 1a, they were con-trasted with deletion controls (i.e., pseu-identification.

In each of four experiments we compared dowords constructed from the phonetic PHswith the final letters removed). Removing theresponse times and error rates among pseudo-

homophones that varied according to their ends of the words renders the phonetic codefor the deletion controls distinct from that ofcorrespondence to real words, where the cor-

respondences tapped phonetic, phonemic and their basewords: stict and drifd are not ho-mophonous with stick and drift, respectivelymorphophonemic coding. In Experiments 1

and 2 we employed a processing task, ho- (though there is a large degree of sound over-lap). Because the phonetic PHs and deletionmophony judgment, that explicitly required

consultation of a sound code. In Experiments controls do not differ in their phonemic ororthographic similarity to real words, if pho-3 and 4 we employed the lexical decision task,

which requires processes more similar to those netic PHs are more often or more quicklyjudged to sound like a word than deletion con-used during normal lexical access. In Experi-

ments 1 and 3, we assessed whether the sound trols, that would suggest that the phonetic sim-ilarity to real words is being used in makingcode relevant to the task contained phonetic

information, phonemic information, or both. these judgments. On the other hand, if thelikelihood and/or speed of homophone judg-In Experiments 2 and 4, we assessed whether

codes relevant to the task specified morpho- ments are the same for the phonetic PHs andthe deletion controls, it would suggest thatlogical structure.phonetic similarity is irrelevant, and that pho-

EXPERIMENT 1 nemic representations are used in judgingwhether these pseudowords sound like words.In Experiments 1 and 2, we probed the na-

ture of the sound codes employed when skilled In Experiment 1b, the phonetic PHs werecontrasted with phonemic PHs. If the soundreaders judged whether a pseudoword sounds

like a real word. In fact, both experiments code used in these judgments corresponds tothe phonetic representation (i.e., to somethingwere run on the same set of participants in the

same experimental session, but exposition is like the actual pronunciation of the words),then participants should judge both phonemiceasier if they are treated as two separate exper-

iments. The focus in Experiment 1 was on and phonetic PHs to sound like words withequal frequency and speed. In contrast, if thewhether a phonetic level, a phonemic level,

or both, are involved when people judge the phonemic representation is activated (possiblyin addition to the phonetic representation),sound of a printed nonword. Two kinds of

pseudohomophones were used: (1) phonemic participants should judge phonemic PHs assounding like real words more often and/orpseudohomophones (henceforth, phonemic

PHs), nonwords like grue whose phonemic faster than they do phonetic PHs, because theformer, but not the latter, are phonemicallyrepresentations are identical to those of a real



identical to a real word. That is, this outcome the homophony of the items. The participantswere told to pronounce them as quickly aswould show that identity in phonemic speci-

fication to a real word is a more salient feature possible, as best they could. Only those itemsthat were pronounced correctly (i.e., as ho-of the sound code than identity of pronuncia-

tion. mophonous to their baseword) by 7 or moreof the 10 norming participants were used inFor both types of pseudohomophones (and

the deletion controls), yoked orthographic the experiment.In Experiment 1a, the phonetic PHs werecontrols were constructed, in which the first

letter or two of the pseudohomophone was all inflected. Examples are attens (for attends)and sticts (for sticks). The deletion controlchanged so that the control was no longer a

pseudohomophone. In Experiment 1a, these condition consisted of the phonetic PHs withthe inflections removed so that these itemsorthographic controls allowed an assessment

of whether there was any difference in diffi- no longer sounded identical to a possiblebaseword. Examples are atten and stict. Thereculty of converting the spelling patterns of the

phonetic PHs and their deletion controls to was often substantial phonological overlap be-tween the deletion controls and the basewordsound codes, due to the added length and com-

plexity of the PHs. In Experiment 1b, these for the phonetic PHs. The remaining materialsfor Experiment 1a were the orthographic con-orthographic controls allowed an assessment

of whether the spelling patterns of the two trols for the phonetic PHs and for the deletioncontrols. These items were derived by chang-types of pseudowords were equally easy to

convert to sound. In addition, the orthographic ing the first letter of the corresponding PH ordeletion control, with the constraint that thecontrols provided a control for whether the

spelling patterns of the two pseudowords be- resulting nonword be pronounceable yet notsound like any English word. In a few cases,ing compared looked equally ‘‘wordlike’’.it was necessary to change more than one let-

Method ter and/or a non-initial letter from the corre-sponding item, to ensure that the yoked ortho-Participants. The participants were 32 stu-

dents from the University of Massachusetts graphic control was pronounceable but did notsound like a real word. Examples are ittenswho participated for course credit. All were

native speakers of American English. and blicts and otten and crict.In Experiment 1b we compared responsesMaterials and design. There were 36 sets

of items, 18 sets each in Experiments 1a and to phonetic PHs to those to phonemic PHs,and made a similar comparison between their1b (see Appendix A). Each set of items con-

sisted of four nonwords, one for each of corresponding orthographic controls. Theitems in the phonemic PH condition were thefour experimental conditions. Both experi-

ments employed phonetic PHs–nonwords that same as their basewords at both the phonemicand phonetic level (e.g., speek is the same assounded like their basewords when pro-

nounced–but which contained a letter or let- speak at both levels). These items werematched with the phonetic PHs in terms ofters whose phonemic representation was in-

consistent with that of the baseword. For orthographic similarity to their basewords(i.e., they differed by the same number of let-instance, in American English, the pronuncia-

tion of nootle is virtually the same as that of ters and in roughly the same location). Forinstance, the phonemic PH speek differs fromnoodle, but the phonemic representations of

the two orthographic strings differ in having its baseword speak visually in that its next-to-last letter is changed, as is the case for thea /t/ or /d/ in the representation, respectively.

These phonetic PHs (and those in the re- phonetic PH greety and its baseword greedy.Similarly, traening differs from training to themaining experiments) were selected from a

set of phonetic PHs that were presented to 10 same degree, and in roughly the same locationas drifding and drifting. The remaining condi-participants to pronounce, in order to pre-test



tions consisted of yoked orthographic controls each trial a nonword appeared in the middleof the CRT and disappeared as soon as thefor the two PH conditions. Examples of ortho-

graphic controls are sootle, brifding, and erter participant responded to it.Participants were instructed to pull the right(for nootle, drifding, and orter) for the pho-

netic PHs and treek and fraening (for speek lever of the response box if the letter stringsounded like a real English word, and to pulland traening) for the phonemic PHs.

To summarize, there were 18 sets of items the left lever otherwise. They were told torespond as quickly as possible while re-in Experiment 1a and 18 sets of items in Ex-

periment 1b. The phonemic PHs (and possibly maining accurate. To prevent participantsfrom making YES responses to all nonwordsthe phonetic PHs) were expected to receive a

response of YES in the sounds-like-a-word that could hypothetically be real words in En-glish (i.e., that were orthographically legal),task, whereas the orthographic controls were

expected to receive a response of NO. Another further instructions were given. Participantswere told to pull the YES lever only if the60 sets of nonword items (240 nonwords)

were included that served as materials in Ex- letter string sounded like a real English word,not just if it could be a word. They were toldperiment 2; half of them were intended to

sound like real words, and half were not. that about half of the nonwords were intendedto sound like real words, and half were not.There were thus 96 sets of nonwords in all,

for a total of 384 in the experimental session. Also, during the 20 practice trials, error feed-back was given: If participants respondedAnother 20 nonwords served as practice items,

half of which were phonemic PHs, and half YES to a nonword that was not a PH, or re-sponded NO to a nonword that was, they re-of which were nonhomophonic pseudowords.

The 96 sets of nonwords (from both Experi- ceived an error message. Feedback was notgiven during the experiment itself because aments 1 and 2) were presented during the ex-

periment in four blocks. Each block consisted major question was how they would judgephonetic PHs and the deletion controls.of 96 nonwords, i.e., one of the four members

of each of the 96 sets of items. There were The participants were told to keep theirhands on the response box at all times duringthus 24 items in each of the 4 conditions pre-

sented in each block. Within these constraints, a block of trials, so as to respond as quicklyas possible. They were told (and a messagethe order of presentation of the 384 items was

randomized for each participant. That is, every on the screen was presented) to take a breakafter each block of trials, and not during aparticipant saw the items in a different random

order, but all participants saw four blocks of block. The 20 practice trials and 384 experi-mental trials took about 15 min.trials, each of which consisted of 24 members

of each of the four experimental conditions.ResultsThus, participants saw all four members of

each set of stimuli in the experiment. Experiment 1a. The primary question waswhether phonetic information influenced theProcedure. Participants were seated in a

sound-attenuated chamber in front of a CRT judgment of homophony. This was exam-ined by comparing the phonetic PHs withand response box. They were told that they

would judge whether letter strings sounded their deletion controls (e.g., attens vs atten).Table 1 shows the percentage of YES re-like real words. They were shown examples

of nonwords that did (trane) and did not sponses (i.e., ‘‘sounds like a word’’) to thephonetic PHs and their deletion and ortho-(trake) sound like real words. Each block of

96 trials began with the push of a lever on graphic controls, along with the responsetimes (RTs) for YES responses to the pho-a response box. Following the initiation of

the block of trials, each of the nonwords netic PHs and their deletion controls (leftcolumn), and for NO responses to the ortho-appeared automatically 500 ms after the par-

ticipant responded to the preceding item. On graphic controls (right column).



TABLE 1

Percentage Yes Rates and RTs to Phonetic PHs and Deletion Controls, along with Their Respective OrthographicControls, in Judgments of Whether the String Sounds like a Word, Experiment 1a

Pseudohomophone vsorthographic control

Pseudohomophone Responsecondition measure Pseudohomophone Control

attens ittensPhonetic % Yes 78.2 17.5Pseudohomophone RT 1330 1959

atten ottenDeletion % Yes 49.6 15.7Control RT 1373 1799

Difference % Yes 28.6 1.8RT 043 160

Note. Response times are calculated from Yes responses for the pseudohomophones and the deletion controls andfrom No responses for the orthographic controls.

As seen in Table 1, there was a sizable trols for the phonetic PHs, F1(1,31) Å 4.9,MSe Å 409440, p õ .03; F2(1,17) Å 1.5, MSedifference in the YES rates between the pho-

netic PHs and their deletion controls. Partici- Å 127806, p Å .23, suggesting that the ortho-graphic controls, for at least some of the PHs,pants classified the phonetic PHs as sounding

like words 78% of the time, whereas they clas- may have sounded more like words than thosefor the deletion controls. In the examplessified the deletion controls as sounding like

words only about 50% of the time, F1(1,31) listed in Table 1, for instance, the orthographiccontrol differs from the PH only by an un-Å 56.91, MSe Å 1.33, p õ .01; F2(1,17) Å

20.03, MSe Å .75, p õ .01. The 43 ms differ- stressed syllable, making it quite similar insound to the PH. Another possibility is thatence in the latency with which these stimuli

were classified as words was not significant, the inflections on the orthographic controls forthe PHs led to longer RTs. In any case, theF1 õ 1; F2(1,17) Å 1.54, MSe Å 130923, p Å

.22. On those occasions when the participants difference in frequency of YES judgments be-tween the PHs and deletion controls wasjudged the phonetic PHs and their deletion

controls as not sounding like words, they were larger than that between the two orthographiccontrols, F1(1,31) Å 78.63, MSe Å .75, p õ94 ms faster on the deletion controls than on

the PHs (1965 ms vs 1871 ms); however again .01; F2(1,17) Å 27.14, MSe Å .42, p õ .01 forthe interaction. However for RT, the differ-this difference was not significant, F1 õ 1;

F2(1,17) Å 2.26, MSe Å 678976, p õ .15. ence between the orthographic controls waslarger, F1(1,31) Å 9.94, MSe Å 330180, p õThus, the phonetic PHs were judged as sound-

ing like words more often than the deletion .01; F2(1,17) Å 8.05, MSe Å 258720, p õ .01.The finding that participants judged the de-controls. There was also a suggestion that the

PHs were judged as words more rapidly, and letion controls as sounding like words 50%of the time was somewhat unexpected. Theseas nonwords less rapidly.

The orthographic controls for the phonetic items were constructed not to sound likewords at either a phonemic or phonetic level.PHs were judged as words about 2% more

than the deletion controls (however, F1 and F2 Some post-hoc item analyses indicated thatwe may not have completely succeeded in thisõ 1). In addition, the RT for NO responses

was 160 ms slower for the orthographic con- endeavor. The YES rates for six of the dele-



tion controls (contack, deteck, strick, conflick, Experiment 1b. The primary question waswhether phonemic coding is involved in theexack, and affeck) were between .66 and .94,

for a mean YES rate of .80 (.97 for the corre- judgment of homophony. This was examinedby comparing the phonetic and phonemicsponding PHs), vs .34 for the remaining 12

deletion controls (.69 for the corresponding pseudohomophones. Table 2 displays the RTsand YES rates for Experiment 1b. RTs are forPHs). Apparently our participants did not re-

quire an inflectional ending (e.g., -s) to judge YES responses to the phonetic and phonemicPHs (left-hand column) and for the NO re-that an item missing a /t/ sounded like a word.

It seems that the large degree of overlap be- sponses for both sets of orthographic controls(right-hand column). There were large andtween many deletion controls and their base-

words, even in the absence of homophony, significant differences in both RTs and YESjudgments between the phonemic and pho-prompted participants to judge them as sound-

ing like real words. Thus, the difference in the netic PHs. Participants were 18% less likelyto classify the phonetic PHs as sounding likepercent of YES responses between the pho-

netic PHs and the deletion controls can be words, F1(1,31) Å 38.12, MSe Å .5, p õ .01;F2(1,17) Å 14.96, MSe Å .28, p õ .01, andseen as a reflection of the difference between

the (near) identity of phonetic codes and par- took over 200 ms longer to do so, F1(1,31) Å11.32, MSe Å 827645, p õ .01; F2(1,17) Åtial phonetic overlap.2

13.12, MSe Å 750533, p õ .01.2 It is difficult to control all possible confounding vari- The differences among the control condi-

ables and have a useable number of stimuli. As indicated tions were small and generally unreliable. Thein the Method section, the phonetic and phonemic PHs

controls for the phonetic PHs were about 5%were equated on the number and approximate location ofless likely to be judged as sounding like realletters by which they differed from their basewords. There

are two other orthographic variables that could influence words than the controls for the phonemic PHs,how wordlike these stimuli look and thus influence deci- F1(1,31) Å 2.87, MSe Å .038, p õ .1; F2(1,17)sions in a way unrelated to phonology: (a) the frequency Å 1.12, pú .1, and were judged as not sound-of small orthographic units such as bigrams and trigrams;

ing like real words about 82 ms faster,(b) the number of words that were orthographically simi-F1(1,31) Å 2.7, MSe Å 198916, p Å .102;lar to each of the pseudowords in each condition (Colt-

heart, Davelaar, Jonasson, and Besner’s, 1977, N mea- F2(1,17) Å 5.5, MSe Å 315282, p õ .03. Be-sure). The orthographic controls were reasonable controls cause the differences between the experimen-for the former problem. The second variable was less well tal conditions were much larger than thosecontrolled, as the PHs were controlled on their similarity

between the control conditions, F1(1,31) Åto the baseword, but not on their similarity to all other28.55, MSe Å 919029, p õ .01; F2(1,17) Åwords. Accordingly, we computed N for each PH and

deletion control and used the difference in each matched 15.45, MSe Å 10119354, p õ .01, for the RTpair as an independent variable in a regression analysis interaction, F1(1,31) Å 58.15, MSe Å .41, pto predict the difference between the PHs and deletion õ .01; F2(1,17) Å 5.94, MSe Å .23, p õ .03,controls (conducting separate analyses for percent of YES

for the YES rate interaction, the factors caus-responses and RTs). Because the intercept can be interpre-ing the differences between the control condi-ted as the size of the effect when the differences in N are

controlled, of primary interest were (a) what the intercept tions were unlikely to explain the entire differ-of the regression line in an analysis was and (b) whether ence between the PH conditions.it was significantly different from zero. The intercept val- The above differences between the phone-ues for the difference in the percent of YES responses

mic and phonetic PHs indicate that partici-and RTs between the phonetic PHs and the deletion con-pants were not merely relying on the phonetictrols were 24.1% (põ .001) and 59 ms (põ .05), respec-

tively. As these differences are very close to those calcu- code in making their judgments. However, thelated in Table 1, where N is not controlled, we may con-clude that differences in N had little influence on ourparticipants’ responses. Another plausible variable con- (e.g., attend for atten). This variable had no predictive

power, and the intercept values and significance levelsfounding the difference between these conditions was thedifference in frequency between the basewords. The were virtually identical to those in the ANOVAs in the

main body of the text.‘‘baseword’’ for the deletion control was the root word



TABLE 2

Percentage Yes Rates and RTs to Phonetic PHs and Phonemic PHs and Their Respective Controls in Judgments ofWhether the String Sounds Like a Word, Experiment 1b

Pseudohomophone vsorthographic control

Pseudohomophone Responsecondition measure Pseudohomophone Control

nootle sootlePhonetic % Yes 78.4 12.6Pseudohomophone RT 1417 1830

speek treekPhonemic % Yes 96.1 17.5Pseudohomophone RT 1189 1912


Note. Response times are calculated from Yes responses for the pseudohomophones and from No responses forthe orthographic controls.

fact that they judged the phonetic PHs to judged phonemic PHs as sounding like realwords more often than the phonetic PHs. Wesound like words nearly 80% of the time indi-

cates that the phonetic representation strongly defer a discussion of other possible interpreta-tions to the General Discussion section.influenced their judgments.3

Summary EXPERIMENT 2The results of Experiments 1a and 1b sug- In Experiment 1, we examined whether the

gest that both phonemic and phonetic repre- sound code used to judge the homophony ofsentations are involved when people judge pseudowords with real words represented pho-whether pseudowords sound like real words. nemic or phonetic information, or both. In Ex-We infer that phonetic levels were implicated periment 2, we examined whether morpho-because participants judged phonetic PHs to phonemic codes are used in homophony judg-sound like real words more often than their ments. There were two comparisons. Theydeletion controls, and we infer that phonemic both involved comparing PHs in which therelevels were implicated because participants was a morphemic mismatch between the non-

word and its baseword with PHs in whichthere was a morphemic match between the3 To control for confoundings of differences in N and

in baseword frequency, we conducted regression analyses nonword and baseword. For one set of mor-analogous to those in Experiment 1a. When the difference phemic mismatches, which we will refer to asin N was used as the predictor, the intercept values for bare-PHs, the morphemic PH has an unin-the difference between the percent YES responses and

flected (i.e., bare) structure and its basewordRTs for the phonemic and phonetic PHs were 15.8% (phas an inflected structure (e.g., counce andõ .001) and 252 ms (p õ .01), respectively. Again, these

differences were similar to those when N was not con- counts). For the other set, suffixed-PHs, thetrolled. The difference in the frequencies of the basewords morphemic PH had an inflected (i.e., suffixed)between the phonetic and phonemic PHs had virtually no structure, while the baseword had an unin-predictive power and the intercept values for the differ-

flected structure (e.g., amays and amaze).ences in percent YES and RTs and their significance levelsEach morphemic match PH was matched withwere virtually identical to those in the ANOVA in the

main body of the text. a morphemic mismatch PH on the number of



letters that needed to be changed, deleted, or The experimental materials consisted of 60sets of items, 30 bare-PH sets in which theadded to get from PH to homophone (e.g.,

discloaz and disclose are both uninflected, and mismatch PH was uninflected (derived froman inflected baseword, e.g., drize); and 30 suf-differ to the same degree and in roughly the

same location as counce and counts; grue and fixed-PH sets in which the mismatch PH wasinflected (derived from an uninflectedgrew are uninflected, and differ from each

other in the same way as amays and amaze). baseword, e.g., amays). Each set of items con-sisted of four nonwords: a morphemic mis-If the sound code includes morphemic

structure (i.e., a parse into morphological con- match PH, a morphemic match PH (e.g., dis-cloaz and grue), and yoked orthographic con-stituents–a bare stem in bare-PHs and a stem

/ suffix in suffixed-PHs), then the mis- trols for each PH (e.g., lounce and prue). Theitems in the morphemic match PH conditionmatches between both kinds of PHs and their

basewords should reduce the number and were the same as their basewords in termsof morphemic complexity, and were matchedspeed of ‘‘sounds-like-a-word’’ responses,

compared to the morphemic match controls. with the morphemic mismatch PHs in termsof orthographic similarity to their basewordsBecause this kind of morphemic information

may be combined with phonemic information (i.e., differed to the same degree and inroughly the same location). Yoked ortho-in a single representation, such a result would

also be consistent with evidence that phone- graphic controls for the two PH conditionswere constructed as in Experiment 1. The mis-mic information is salient in the sound code

generated in this task. If the sound code does match and the match PHs were expected toreceive a response of YES in the sounds-like-not include morphemic structure, then differ-

ences between morphemic mismatch and mor- a-word task and the orthographic controlswere expected to receive a response of NO.phemic match PHs may not occur in ‘‘sounds-

like-a-word’’ judgments. As indicated in Experiment 1, there were 96sets of nonwords (36 in Experiment 1 and 60

Method in Experiment 2), yielding 384 experimentaltrials. These were preceded by 20 practice tri-Participants and procedure. The partici-

pants and procedure were the same as de- als.scribed in Experiment 1.

ResultsMaterials and design. As indicated above,the key comparison is between morphemic The %YES homophony judgments and RTs

are shown in Table 3. The pattern of data formismatch PHs and morphemic match PHs.(There were orthographic controls for both.) the bare-PH and suffixed-PH mismatches was

quite similar. Overall, responses to the bare-The mismatch PHs were nonwords thatsounded like their basewords when pro- PH materials were about 3% more accurate

and about 60 ms faster than for the suffixed-nounced, but which contained a morphemicmismatch with the baseword from which they PH materials, F1(1,31) Å 24.9, MSe Å .003, p

õ .001, and F1(1,31) Å 15.7, MSe Å 15191,were derived. For instance, the pronunciationof counce is the same as that of counts, but p õ .001, respectively, but neither F2 was

significant. However, type of mismatch failedthe morphological representation of counce issimple, whereas that of counts is complex. to interact with any of the other variables (all

but one of the interaction terms had an F lessSimilarly, amays is morphemically complex,whereas amaze is simple, even though the pro- than 1). Accordingly, we will only consider

the averaged data in subsequent discussion.nunciation of these two items is the same.Other examples of morphemic mismatch PHs There were significant differences between

the morphemic mismatch and the morphemicare: bare-PHs, drize (for dries), cleerd(cleared); suffixed-PHs, focks (fox), and giffed match pseudohomophones: participants were

about 15% less likely to judge the morphemic(gift) (see Appendix B).



TABLE 3

Percentage Yes Responses and Response Times to Morphemic Mismatch Pseudohomophones and MorphemicMatch Pseudohomophones (and Their Respective Controls) in Judgments of Whether the String Sounds Like a Word,Experiment 2

Pseudohomophone vs controlPseudohomophone Response

condition measure Pseudohomophone Control

counce amays lounce omaysMorphemic mismatch % Yes 68.1 20.2Pseudohomophone RT 1465 1841

discloaz grue niscloaz prueMorphemic match % Yes 83.2 16.3Pseudohomophone RT 1292 1834


Note. Response times are calculated from Yes responses for the pseudohomophones and from No responses forthe orthographic controls.

mismatch pseudohomophones as sounding like Summary. Experiment 2 indicates that amismatch in morphemic structure and/or com-words, F1(1,31) Å 133.8, MSe Å .006, p õ

.001; F2(1,58) Å 23.1, MSe Å .001, p õ .001, plexity between a PH and its homophonicbaseword makes it harder to judge that the PHand were 173 ms slower in making the judg-

ment, F1(1,31)Å 47.7, MSeÅ 20130, põ .001; is homophonic to the baseword. This indicatesthat morphemic structure is involved in theF2(1,58) Å 21.9, MSe Å 58530, p õ .001. The

differences between the two orthographic con- homophony judgment. One explanation forthe result is that morphological structure istrol conditions were small and generally not

significant. The 7 ms difference in response part of the sound code used in the judgment.This is not the only possible conclusion, how-times was unreliable (both Fs õ 1). The 3.9%

difference in YES rate was significant by sub- ever, as a difference in morphemic structurecould be registered in a separate system, butjects, F1(1,31) Å 10.7, MSe Å .005, p õ .01,

but not by items, F2(1,58) Å 1.29, MSe Å .013, participants are unable to ignore this differ-ence in morphology while making YES (i.e.,p ú .20. Moreover, the error difference for the

controls was in the opposite direction from the ‘‘homophone’’) decisions, just as they are un-able to ignore irrelevant aspects of stimuli inPHs: the controls for the mismatch PHs were

judged to sound like words more often than Stroop-like tasks. We will consider alternativethose for the match PHs. Critically, the differ-ence between the pseudohomophone condi- morphemic mismatch and morphemic match PHs was

used as the predictor, the differences in percent YEStions was significantly larger than the differ-responses and RTs between the mismatch and matchence between the controls, F1(1,31) Å 29.94,PHs were 14.7% (p õ .01) and 180 ms (p õ .001).MSe Å .007, p õ .001; F2(1,58) Å 5.37, MSe Again, the effects were similar to when N was not con-Å .003, p õ .02, for the YES rate interaction; trolled. The difference in the frequencies of the base-

F1(1,31) Å 29.3, MSe Å 14998, p õ .001; words between the phonetic and phonemic PHs wasvery small (38 vs 39 per million) and the interceptF2(1,58) Å 10.49, MSe Å 53635, p õ .005, forvalues for the differences in percent YES and RTs andthe RT interaction.4their significance levels were virtually identical to thosein the ANOVA in the main body of the text. The differ-

4 Regression analyses were done analogous to those ence in frequency, however, did have some predictivepower for RTs (p Å .06).in Experiment 1. When the difference in N between the



explanations of these results in the General due to interference from the activation of thebaseword by some aspect of the sound codeDiscussion section. It is also worth noting that

the results were quite similar when the mor- of the PH.phemic mismatch pseudohomophone was The pseudohomophone effect has been rep-morphemically complex (suffixed) and when licated in a number of studies even when or-it was morphemically simple (bare). This sug- thographic factors are controlled (e.g.,gests that morphological complexity does not Besner & Davelaar, 1983; Coltheart et al.,have much influence on the difficulty of con- 1977; Seidenberg, Petersen, Plaut, & Mac-structing the sound code(s) used in the ho- Donald, 1996). Moreover, deep dyslexics,mophony judgment task. who seem to have an impaired phonological

rule system, do not show a pseudohomophoneEXPERIMENT 3 effect (Patterson & Marcel, 1977). It is possi-

In Experiments 3 and 4 we probed the na- ble, as Coltheart et al. (1977) argued, thatture of the sound codes employed when skilled while the pseudohomophone effect indicatesreaders made lexical decisions. As with Ex- the involvement of (assembled) phonology inperiments 1 and 2, the experiments were run lexical judgments of nonwords, it does nottogether but are described separately for ease necessarily indicate such involvement forof exposition. Experiment 1 indicated the word processing. Nonetheless, the pseudoho-involvement of both phonetic and phonemic mophone effects found with the lexical deci-codes when participants judged the sounds of sion task are consistent with results from othervisually presented nonwords. Experiments 3a tasks such as categorization (e.g., Van Orden,and 3b assessed whether either or both codes 1987) and priming (Perfetti et al., 1988) andare involved in the lexical decision task. In thus provide converging evidence for soundthese experiments we used the same materials coding. Thus, we believe that the pseudoho-as in Experiments 1a and 1b and asked partici- mophone effect is a reasonable tool to assesspants, in essence, to judge whether the letter the nature of sound coding in lexical access.5strings ‘‘look like’’ words. Of course, real The test for the involvement of the phoneticwords were added as well, which had charac- code is the comparison of the pseudohomo-teristics similar to the pseudohomophones and phone effect for the phonetic PHs with that oftheir controls. the deletion controls. A greater effect for the

The design of Experiments 3a and 3b was phonetic PHs would indicate an involvementsimilar to that of Experiments 1a and 1b, and of phonetic coding. The test of whether therethe key comparisons were similar. However, is involvement of phonemic information is thethe effects could be examined a bit differently comparison of the pseudohomophone effectsbecause the correct response to all the stimuli for the phonetic PHs and the phonemic PHs.was now NO. This allowed a direct compari- If the effects are equal, then there is no evi-son of the responses to the PHs with their dence for the involvement of phonemic codes,orthographic controls (unlike in Experiment but if the effect is greater for the phonemic1, where the correct response to the PH wasYES and to the control was NO). This differ-ence between a pseudohomophone and its or-

5 As a convenient locution, we will typically say thatthographic control has been termed the pseu- a pseudohomophone effect is evidence that the sounddohomophone effect in lexical decision tasks code generated by the pseudohomophone is accessing the

lexical entry of the baseword. This is not necessarily the(Rubenstein, Lewis & Rubenstein, 1971) andcase, as the response in the lexical decision task is com-should provide an indication of sound coding.plex, and the effect could instead, for example, reflect anIf sound coding is involved in making lexicalincrease in the total activity in the lexicon. However, we

decisions, responding NO to PHs should be do not think that this distinction is particularly importantslower and less accurate than to the non-ho- for our present purposes: in either case, they are tapping

the input of sound codes to the lexicon.mophonic orthographic controls, presumably



PHs, then there is evidence for such involve- although some were phonologically similar towords (e.g., disrept); all were orthographicallyment.legal.

Method The 98 sets of items were presented duringthe experiment in 7 blocks. Each block con-Participants. The 42 University of Massa-

chusetts students who participated in the ex- sisted of 98 items, i.e., one of the 7 membersof each of the 98 sets. Within these con-periment for course credit were all native

speakers of American English. None were in straints, the order in which each participantviewed the 686 items was randomized.Experiments 1 and 2.

Materials and design. There were 38 sets Procedure. The procedure was identical tothat of Experiment 1, except for the changeof items, 20 in Experiment 3a and 18 in Exper-

iment 3b. Each set consisted of four nonwords in task, i.e., all nonwords were to receive aresponse of NO, and all words were to receiveand three words. The nonword sets in Experi-

ment 3a were the same as those in Experiment a response of YES. Also, because the taskchanged to lexical decision, in which the cor-1a (with two added) and the nonwords in Ex-

periment 3b were the same as in Experiment rect classification of the stimuli is clear, parti-cipants were given feedback on trials on which1b (see Appendix A). Thus, there were pho-

netic PHs and deletion controls (along with they made errors. They began with the 21practice trials, and then completed the 686their yoked orthographic controls) in Experi-

ment 3a and phonetic and phonemic PHs experimental trials in about 22 min.(along with their yoked orthographic controls)in Experiment 3b. In both experiments, for Results and Discussioneach set of four nonwords there were three

Experiment 3a. The RTs (for correct NOreal words added to the set, chosen to be asresponses) and error rates for the four typesorthographically similar to the nonword stim-of nonwords in Experiment 3a, along with theuli as possible. The orthographic similaritypseudohomophone effects (third column) arewas based both on the beginnings and the end-shown in Table 4. There was an 82 ms pseudo-ings of the PHs. For instance, for the set inhomophone effect in the RTs for the phoneticExperiment 3a containing the phonetic PH at-PHs, F1(1,41) Å 23.58, MSe Å 142025, p õtens and the deletion control atten (ortho-.01, F2(1,19) Å 20.15, MSe Å 127577, p õgraphic controls ittens and otten), the real.01. However, there was also a 96 ms effectwords were intends, eaten, and attest. For thefor the deletion controls, F1(1,41) Å 32.37,set of items in Experiment 3b containing theMSe Å 194978, p õ .01, F2(1,19) Å 17.09,phonetic PH nootle and the phonemic PHMSe Å 108160, p õ .01. There was no differ-speek (and orthographic controls, sootle andence in the sizes of the two effects (F1 and F2treek), the words were settle, noose, and creek.õ 1 for the interaction). In contrast, there wasTo summarize, there were 38 sets of 7 ex-a substantially bigger pseudohomophone ef-perimental items in Experiment 3 (a total offect for the phonetic PHs in the error rates.266 items). The experimental session had aThe phonetic PHs were judged to be wordstotal of 98 sets of items (38 from Experiment12.7% more often than their orthographic con-3 and 60 from Experiment 4), so that theretrols, F1(1,41) Å 94.88, MSe Å .34, p õ .01;were a total of 686 experimental trials. AsF2(1,19) Å 17.0, MSe Å .16, p õ .01, whereaseach set had 3 words and 4 nonwords, 57%the deletion controls were erroneously judgedof the trials were nonwords (i.e., were to re-to be words only 5.7% more often than theirceive a response of NO during the lexical deci-orthographic controls, F1(1,41) Å 19.1, MSesion task), and 43% were words (i.e., YESÅ .07, p õ .01; F2(1,19) Å 3.43, MSe Å .03,trials). The practice stimuli consisted of 12p õ .07. The 7% difference in the twononwords and 9 words. Of the practice non-

words, none were homophonic to real words, effects was significant, F1(1,41) Å 19.52, MSe



TABLE 4

Error Rates and Response Times to Phonetic PHs and Deletion Controls Along With Their RespectiveOrthographic Controls in Lexical Decision Judgments, Experiment 3a

Pseudohomophone vs orthographic controlPseudo-

homophone Response Differencecondition measure Pseudo-homophone Control (Pseudohomophone effect)

attens ittensPhonetic %Errors 16.0 3.3 12.7Pseudo-Homophone RT 874 792 82

atten ottenDeletion %Errors 8.4 2.7 5.7Control RT 821 725 96

Difference %Errors 7.6 0.6 7.0RT 53 67 014

Note. Response times are for correct responses.

Å .05, p õ .01; F2(1,19) Å 4.8, MSe Å .03, p Experiment 3b. Table 5 shows the RTs anderror rates for the lexical decisions in the fourõ .05.

Although the RT results provide no evi- nonword conditions, along with the pseudoho-mophone effects (third column). For the RTs,dence that a phonetic homophonic relationship

between a nonword and a real word is any there was a 61 ms pseudohomophone effectfor the phonetic PHs, F1(1,41) Å 17.74, MSemore important than mere phonetic similarity,

the error results support a role for phonetic Å 79242, p õ .01; F2(1,17) Å 6.5, MSe Å49136, p õ .01, and an 89 ms pseudohomo-homophony. Phonetic PHs were much more

likely to be erroneously accepted as words phone effect for the phonemic PHs, F1(1,41)Å 37.38, MSe Å 166694, põ .01; F2 Å 12.15,than were the deletion controls, indicating that

phonetic codes significantly influenced the MSe Å 91809, p õ .01. The 28 ms differencebetween the two was not significant, F1(1,41)lexical decision task.6

Å 2.14, MSe Å 8078, p õ .15; F2 õ 1. Therewere also large pseudohomophone effects in6 The regression analyses (using N as the predictor) inthe error rates for both types of PHs. Therethe lexical decision task used the pseudohomophone ef-

fect as the dependent variable and the difference in N was a 10.3% effect for the phonetic PHs,between the phonetic PH and its orthographic control (or F(1,41) Å 55.7, MSe Å .22, p õ .01; F2(1,17)the deletion control and its orthographic control) as the Å 3.3, MSe Å .1, p õ .08, and a 14.6% effectpredictor. For the RTs, the intercept (i.e., the predicted

for the phonemic PHs, F1(1,41)Å 110.96, MSepseudohomophone effect when difference in N is re-Å .44, p õ .01; F2(1,17) Å 6.55, MSe Å .19,moved) was 59 ms for both the phonetic PHs (p õ .01)

and the deletion controls (p õ .05). The pseudohomo- põ .02. The 4.3% difference between the twophone effects for the errors were 7.4% for the phonetic effects was significant, but by subjects only,PHs (p Å .06) and 0.0% for the deletion controls, so that F1(1,41) Å 4.82, MSe Å .02, p õ .05; F2 õthe difference between these two intercepts was virtually

1. Note that the differences between the pseu-identical to that in the main analysis. To control for differ-dohomophone effects are largely driven byences in frequency of the basewords, differences between

the frequency of the basewords were used as a predictor differences in the orthographic controls, not(as in Experiments 1 and 2), but the difference in pseudo-homophone effect between phonetic PHs and the deletioncontrols was used as the dependent variable. The interceptwas 8.3% (p õ .05) when the difference in frequency both the difference in frequency and the differences in N

were both used as predictors.was used as the sole predictor and 7.8% (p õ .05) when



TABLE 5

Error Rates and Response Times to Phonetic PHs and Phonemic PHs and Their Respective Controls in LexicalDecision Judgments, Experiment 3b

Pseudohomophone vs orthographic control

Pseudo-homophone Response Differencecondition measure Pseudo-homophone Control (Pseudohomophone effect)

nootle sootlePhonetic %Errors 18.1 7.8 10.3Pseudo-homophone RTs 884 823 61

speek treekPhonemic %Errors 16.9 2.3 14.6Pseudo-homophone RTs 858 769 89

Difference %Errors 1.2 5.6 04.3RTs 26 54 028


the PHs, whose RTs and error rates are quite reject as the phonemic PHs, which are homo-phones of words at both the phonemic andsimilar.7

Summary. Experiment 3 indicated that phonetic levels. Because the difference be-tween the two pseudohomophone effects wassound codes had a clear effect on lexical deci-

sion times. Moreover, phonetic codes were at small and not significant, there is minimal sup-port for involvement of phonemic codes inleast part of the effect. In Experiment 3a, the

phonetic PHs produced a substantially larger the lexical decision task. This pattern differedfrom that in the ‘‘sounds like’’ task, wherepseudohomophone effect than the deletion

controls in the error data, and in Experiment there was clear evidence for the involvementof both kinds of codes.3b, there was a significant pseudohomophone

effect for the phonetic PHs in both RT andEXPERIMENT 4error measures. Moreover, in Experiment 3b,

the phonetic PHs were almost as difficult to In this experiment we used the pseudoho-mophones from Experiment 2 to explore theinvolvement of morphological structure in the7 When the difference in N between the PH and its

orthographic control was used as a predictor, the inter- lexical decision task. A greater pseudohomo-cepts for pseudohomophone effects for the phonetic PHs phone effect for the morphemic match PHswere 80 ms (p õ .01) and 10.7% (p õ .05), and the than for the morphemic mismatch PHs wouldintercepts of the regression for the pseudohomophone ef-

indicate involvement of morphemic coding infects for the phonemic PHs were 94 ms (p õ .01) andthe lexical decision task.12.5% (p õ .10). The fact that the pseudohomophone

effects were even more similar than in the main analysesfurther casts doubt that there is any real pseudohomo- Methodphone effect difference between the phonetic and phone-

Participants and procedure. The partici-mic PHs. When the difference in frequency of the base-words between the phonetic and phonemic PHs was used pants and procedure were as in Experiment 3.to predict the difference in pseudohomophone effect be- Materials and design. The nonword stim-tween them, the intercept of the regression for the error uli in Experiment 4 were the same as in Ex-probability was about 3% either when the difference in

periment 2. In addition, for each set of fourfrequency was the sole predictor or when the differencenonwords (morphemic mismatch PH, mor-in N was added as a predictor (both ts õ 1) and about

20 ms for either regression (both ts õ 1). phemic match PH, and their respective con-



TABLE 6

Error Rates and Response Times to Morphemic Mismatch PHs and Morphemic Match PHs(and Their Respective Controls) in Lexical Decision Judgments, Experiment 4

Pseudohomophone vs orthographic control

Response Pseudo- DifferenceType of homophone measure homophone Control (Pseudohomophone effect)

counce amays lounce omaysMorphemic Mismatch %Errors 12.6 4.9 7.7Pseudohomophone RT 855 771 84

discloaz grue niscloaz prueMorphemic Match %Errors 16.4 4.2 12.3Pseudohomophone RT 813 763 50

Difference %Errors 03.9 0.7 04.6RT 42 8 34


trols), three words were added according to F2(1,58) Å 33.3, MSe Å 7832, p õ .001, anda 7.7% pseudohomophone effect for errors,the same criteria as described in Experiment

3a. For instance, for the set of items con- F1(1,41) Å 74.28, MSe Å .003, p õ .001;F2(1,58) Å 16.93, MSe Å .026, p õ .001.taining the morphemic mismatch PH counce

and the morphemic match PH discloaz (and For morphemic match PHs, there was a 50-ms pseudohomophone effect for RTs,orthographic controls, lounce and niscloaz),

the words were bounce, quartz, and pizzaaz F1(1,41) Å 21.87, MSe Å 5112, p õ .001;F2(1,58) Å 16.99, MSe Å 6780, MSe Å 6780,(see Appendix B). There were thus 60 sets

of 7 stimuli in Experiment 4 (four nonwords p õ .001, and a 12.3% pseudohomophoneand three words), plus the 38 sets of items effect for errors, F1(1,41) Å 103.24, MSe Åfrom Experiment 3. The practice stimuli .003, p õ .001; F2(1,58) Å 18.44, MSe Åwere described in Experiment 3. .05, p õ .001.

The pattern of differences between theResults morphemic match and morphemic mismatch

PHs was similar to that of a speed-accuracyRTs and error rates for the morphemictradeoff effect. That is, using error rate asmismatch and match PHs and their controlsthe measure, the pseudohomophone effectare shown in Table 6, along with the differ-for the morphemic match PHs was 4.6% big-ence between PHs and controls (the pseudo-ger than for the morphemic mismatch PHs,homophone effect) for each morphemicbut using the RT measure, the pseudohomo-type. With one exception to be discussedphone effect was 34 ms smaller. However,below, the overall pattern of data was simi-neither interaction effect was reliable overlar for the bare-PH and suffixed-PH pseudo-items: for errors, F1(1,41) Å 11.86, MSe Åhomophones; hence, we will focus on the.004, põ .005; F2(1,58) Å 1.86, MSe Å .002,combined analysis. First, both morphemicp õ .18, and for RTs, F1(1,41) Å 10.93, MSematch and morphemic mismatch PHsÅ 2215, p õ .005, F2(1,58) Å 1.98, MSe Åshowed clear pseudohomophone effects. For7418, p õ .17. These results suggest thatmorphemic mismatch PHs, there was an 84-morphemic mismatching affects some sortms pseudohomophone effect for RTs,

F1(1,41) Å 61.92, MSe Å 4820, p õ .001; of decision criterion in the lexical decision



task: the presence of a morphemic mismatch cipants a bit more cautious in making YESdecisions, leading to somewhat lower errorbetween pseudohomophone and baseword

causes subjects to respond a bit slower, but rates but somewhat longer RTs. We will returnto a fuller discussion of these data below.be a bit more accurate (perhaps an occa-

sional double checking mechanism). How-GENERAL DISCUSSIONever, it does not appear that morphemic mis-

matches affect the basic pseudohomophone The current experiments examined the na-ture of the sound code that is assembled fromeffect and hence there is no evidence that

morphemic mismatches directly influence orthographic information. Our results indicatethat the code may be task dependent. Whenphonological coding in this task. There also

appeared to be some differences between the participants were required to use the soundcode to decide whether a pseudoword soundstwo types of morphemic mismatches, most

notably that the bigger RT effect for mor- like a real word, there was clear evidence thatphonemic, phonetic and morphophonemicphemic mismatches only occurred for suf-

fixed-PH items; however the interaction was codes were all involved in the decision. Incontrast, in deciding whether an orthographicalso not reliable over items, F1(1,41) Å 5.92,

MSe Å 5749, p õ .05; F2(1,58) Å 1.99, MSe string is a visual word, where the sound codecould be incidental to the task, there was clearÅ 7418, p õ .17.8

evidence that phonetic codes were involved,Summary but only hints that either phonemic or morpho-

phonemic codes contribute on their own to theThe data in Experiment 4 are different fromthose in Experiment 2. In the sounds-like-a- decision. Moreover, the homophony judg-

ments were much slower than the lexical deci-word task, there were clear effects of morphe-mic mismatch, with people finding it signifi- sion judgments.

One possible interpretation of the abovecantly harder to judge that a pseudowordsounded like a word when there was a mor- pattern of data is that phonemic and morphe-

mic coding are not obligatory in the encodingphemic mismatch with the baseword. Thecomparable effect in Experiment 4 would of orthographic information and only come

into play when (a) people are required to usehave been for a morphemic mismatch to makethe PH less like the baseword and hence easier the phonological information and/or (b) pro-

cessing times are slowed down. If this inter-to judge as a nonword. There was a suggestionof a morphemic mismatching effect in the lex- pretation is correct, it suggests that the sound

code involved in identifying words in readingical decision task, but it was different from thepattern expected from Experiment 2; instead a (Pollatsek et al., 1992) and/or accessing the

meaning of words (e.g., Van Orden, 1987) ismorphemic mismatch appeared to make parti-largely or exclusively phonetic. Moreover, itsuggests that a morphemic match or mismatch8 The difference in N had virtually no effect on thebetween the orthographic string and the targetpseudohomophone effects. The intercepts of the regres-

sion for the pseudohomophone effects were 87 ms (p lexical item is not particularly important inõ .001) and 5.9% (p õ .05) for the morphemic mis- the activation of the lexical entry of the target.match PHs, and 52 ms (p õ .01) and 14.0% (p õ The non-involvement of either phonemic or.001) for the morphemic match PHs. The difference in

morphophonemic codes in the processing offrequency of the basewords had virtually no predictiveprinted English (except in specific tasks), ifpower over the difference in pseudohomophone effect

between the morphemic match and mismatch effects. true, is a bit surprising and counter-intuitive,When both difference in N and difference in frequency because the creation of the phonetic codewere used as predictors, there was a bigger pseudoho- would appear to depend on both phonemicmophone effect for the morphemic match PHs (9.2%)

and morphemic analysis. That is, if Englishin the error data (p Å .08), but a smaller pseudohomo-orthography corresponds more closely to thephone effect (29 ms) in the RT data (tõ 1). This pattern

mirrored that in Table 6. phonemic or morphophonemic representation



of lexical items than to their pronunciations at present, only attempted to explain how mo-nomorphemic words are pronounced, it re-(e.g., Chomsky & Halle, 1968; Venezky,

1970), then it seems reasonable that visual mains an open question whether an adequatemodel of pronunciation can be constructedword recognition would initially involve acti-

vation of a phonemic or morphophonemic that doesn’t have intermediate levels of codingthat are distributed equivalents of morphemicsound code. Venezky’s (1970) analysis ac-

cords with the conclusion that phonetic forms and phonemic codes.We think the second alternative is the moreof words are derived from the phonemic or

morphophonemic representation. plausible. That is, the phonetic representationsthat are apparently activating the lexical en-One could argue that such prior stages are

not necessary for encoding words because the tries of the pseudohomophones and, as a re-sult, slowing lexical decision times, have beenphonetic form of the word could be activated

directly from the lexical entry. However, such produced by means of intermediate phonemicand morphemic codes. However, these inter-‘‘direct lookup’’ of the sound code would not

be possible for pseudowords, requiring assem- mediaries are encapsulated to the extent thatthey do not directly communicate with otherbly of the sound code (this was a major reason

for using pseudowords in our experiments). coding systems (or do not communicate verystrongly with these other systems). In the caseThe sense of ‘‘assembly’’ that we intend is

that getting to the sound code is something of phonemic codes, the logic of this interpreta-tion seems relatively straightforward. If pho-other than going to a ‘‘look-up table’’, rather

than that the sound code itself needs to be nemic codes had strong direct connections tolexical activation, then there should have beenbuilt out of smaller units.

The question thus remains of how a pho- a substantially weaker pseudohomophone ef-fect for the phonetic PHs than for the phone-netic representation could be assembled and

involved in the processing of print without mic PHs in Experiment 3. There was a small,nonsignificant, difference between the twoapparent involvement of either phonemic or

morphemic coding. We think there are two pseudohomophone effects; however, it wouldseem most plausibly explained by the fact thatalternatives. The first is that the latter codes

are not involved in normal word identification, some of the phonetic PHs may not have beenexact homophones at the phonetic level.and that the phonetic code can be constructed

from lower-level codes such as bigrams or Another possible explanation for our datain Experiment 3 is that the sound code is non-trigrams. The second is that phonemic and

morphemic codes are involved in the assem- exact, and that only a partial match is neededto get a strong pseudohomophone effect. Onbly of the phonetic code but are encapsulated

to the degree that these codes only communi- some level, this explanation is similar to thephonetic/phonemic difference we intended tocate with the phonetic coding system and not

with lexical access or with the response sys- establish: the phonetic code fails to distinguishbetween words that have different phonemictem making lexical decisions.

The first possibility would be consistent and orthographic representations. We thinkthat a non-specific ‘‘imprecision’’ explana-with some parallel distributed processing

models of visual word encoding in which the tion, however, is less satisfying than the pho-nemic/phonetic distinction that we have beenactivation of a sound code is not necessarily

mediated by linguistically inspired word units, making. Note that the robust PH effect thatwe observed is caused by the difference in abut rather, by patterns of activity over ‘‘hidden

units’’ that learn the regularities of English single grapheme and phoneme (e.g., amays vsomays), with the phoneme that they differ byorthography and the regularities of grapheme

to sound conversion. At present, most of these often in an unstressed syllable. That is, wehave obtained reliable and fairly large differ-models presuppose something like an interme-

diary phonemic code. Moreover, as most have, ences in the lexical decision task between



stimuli which differ from the base words by which forces the participant to introspect onthe sound code and which has substantiallytwo phonemes and those that differ from them

by one phoneme. In contrast, the differences longer RTs than the lexical decision task, re-cruits systems that are largely unavailable dur-were small and unreliable between stimuli that

differ from the base word by one phoneme ing lexical decision judgments. That is, mak-ing a homophony judgment invites the partici-(the phonetic PHs) and those that differ from

it by no phonemes (the phonemic PHs). If the pant to consciously introspect on the soundrepresentation of the pseudohomophone (anddegree of activation of the base word were

merely predicted by the degree of phonemic those of candidate base words). It is likely thatthis introspective process is colored by thesimilarity to the baseword, the most natural

pattern to expect would be a larger difference orthographic representation together with theparticipant’s intuitive theories of grapheme toin activation between a perfect match and a

one phoneme difference than between a one sound code conversion. The lexical decisiontask, on the other hand, does not require (in-phoneme difference and a two phoneme dif-

ference. Hence, we think the most plausible deed should discourage) a conscious assess-ment of how a word sounds, and requires onlyexplanation for the lack of a relibable differ-

ence between the phonemic and phonetic a match with the orthographic string. As aresult, although some sort of sound code getspseudohomophone effects in Experiment 3 is

that the phonemic level of coding rarely, if computed in the lexical decision task whichactivates a lexical item (hence the pseudoho-ever, activates the lexical entries of the base-

words of the PHs, and instead activates them mophone effect), it is unlikely to be the resultof such an introspective process.through phonetic representations.

The case of morphemic coding is somewhat It thus appears that deliberation regardingthe speech code results in consultation of moremore complex, as there appeared to be a mor-

phemic effect in Experiment 4; however, it representations than when deciding on lexi-cality. Whether this is due to active suppres-appeared in the form of a speed-accuracy

tradeoff, rather than as a simple interference sion of the more abstract (phonemic and mor-phemic) codes in the lexical decision task oreffect. If morphemic constituents were di-

rectly represented in the sound code that was to an active recruitment of them in the ho-mophony judgment task is an open question. Ifproducing the pseudohomophone effect in Ex-

periment 4, then one would most likely expect phonemic and morphemic codes are recruitedspecifically for the homophony judgment task,that morphemic mismatches would produce a

smaller pseudohomophone (i.e., interference) it could either be that participants recruit thesecodes because they think that they need to useeffect than morphemic matches. Instead, mor-

phemic mismatches appeared to slow down them, or that the conscious computation andassessment of the speech code automaticallyresponses while decreasing the number of

false positive responses to PHs. There obvi- recruits more abstract codes.Our data indicate that the sound code thatously could be many explanations for such an

effect, but the most likely is that the morphe- produces the pseudohomophone effect in thelexical decision task is likely to be a phoneticmic mismatch between PH and baseword was

detected, but this did not directly influence the rather than a phonemic code, and one in whichmorphemic structure is not directly repre-degree to which the sound code activated the

lexical entry of the baseword. Instead, it set sented. This, of course, leaves open the ques-tion of the nature of the sound code activatedup a ‘‘flag’’ for the participant to be more

cautious in making a YES decision. when printed words are encountered in naturaltasks such as silent reading. Our results indi-We now return to the question of why pho-

nemic and morphemic codes were involved in cate that the phonetic code is involved in boththe homophone judgment task and the lexicalthe homophony judgments of Experiments 1

and 2. The most likely answer is that this task, decision task, and thus appears to contact the



lexicon in a relatively ‘‘automatic’’ way. In quictly quict buictly kuict exactly quickenquilt (quickly)contrast, the phonemic code only appears to

be more directly involved in the slower and *bricts brict dricts krict edicts brinks evicts(bricks)more introspective homophone judgment task.

We thus propose as a working hypothesis that pedding ped ledding ged wedding bed feared(petting)the sound code that is directly involved with

lexical access and activation of meaning in dirdy dird mirdy lird dirge tardy birdie (dirty)cludder clud pludder plud cluster bladdermost natural reading tasks is the phonetic

code. If this is true, Crowder and Wagner clubbed (clutter)affecks affeck offecks iffeck affection effects(1992) were wrong when they assumed that

the code used in the activation of the meaning checks (affects)sticts stict blicts crict sticky depicts clicksof words is phonemic. We suspect, however,

that the other part of their conjecture is true, (sticks)conflicks conflick ponflicks bonflick con-and that the code used in processing the mean-

ing of written discourse (as well as that of flicted lipsticks convicts (conflicts)exackly exack oxackly axack exalted cracklyindividual words) is phonetic.

abstractly (exactly)APPENDIX A *Not used in Experiment 1a.

Items from Experiments 1 and 3 Items from Experiments 1b and 3bFor each line, the order of items is: 1: pho- nootle speek sootle treek settle noose creek

netic PH; 2: deletion control (Experiments 1a (noodle)and 3a) or phonemic PH (Experiments 1b and greety streek dreety atreek greens sleety sleek3b) ; 3–4: orthographic controls for 1 and 2; (greedy)5–7: matched words used in Experiment 3 sattle awaik vattle swaik cattle await flakeonly. The baseword is in parentheses. (saddle)

grittle paraid prittle garaid brittle afraid unpaidItems from Experiments 1a and 3a (griddle)

drifding traening brifding fraening dreadingattens atten ittens otten intends eaten attest(attends) drifters trailing (drifting)

shouded reeched chouded veeched cloudedgrouns groun prouns troun nouns grouchlaugh (grounds) leeched perched (shouted)

chapder freazer shapder dreazer slanderremines remine temines kemine rewinds re-mains refine (reminds) chapel freaked (chapter)

brudal feable frudal beable bridal brutes use-frens fren brens cren wrens fret endues(friends) able (brutal)

empdy creeky umpdy breeky candy empirebarnds barnd larnds jarnd brands blond barks(barns) creepy (empty)

cludder mistaik pludder sistaik shudder mis-*gownds gownd lownds cownd gower gougehounds (gowns) took fresher (clutter)

baddle afrade faddle ofrade paddle afreshfokes jote lokes fote jokes folds note (folks)contacks contack gontacks rontack haystack badly (battle)

swedder despare kwedder sespare shreddercontains caution (contacts)detecks deteck fetecks geteck wrecks detest prepare sweeper (sweater)

spondser roadent apondser soadent spongedesk (detects)strickly strick utrickly atrick quickly unpick cleanser student (sponser)

randsom baykery bandsom raykery handsomestringy (strictly)clocts cloct glocts ploct tracts cloths concoct blossom mockery (ransom)

spolk skait apolk okait yolk sport trait (spoke)(clocks)



loating applawd poating epplawd boating dride twede gride dwede pride impede stam-pede (dried)dawdle apply (loading)

orter prowd erter trowd outer crowd orbit (or- worreed cheared correed theared misdeedcleared smeared (worried)der)

murter clawse lurter glawse porter browse coyld coyle joyld noyle scold style toyed(coiled)clawed (murder)

fayld fayle layld layle payer loyal yieldAPPENDIX B (failed)

addrest bryter eddrest kryter fairest writer har-Items from Experiments 2 and 4vest (addressed)

For each line, the order of the items is: kist cawt nist vawt list cart fist (kissed)1: morphemic mismatch PH; 2: morphemic danst forren banst corren blast barren dandymatch PH; 3–4 orthographic controls for 1 (danced)and 2; 5–7: matched words used in Experi- forst enuff lorst unuff worst forts bluffment 4 only. (forced)

playst skweek flayst akweek priest cheekBare-PH Items sketch (placed)

drize scrue brize icrue prize issue scrub (dries)counce discloaz lounce niscloaz bounce quartzpizazz (counts) flozz cartune plozz martune fuzz topaz com-

mune (flaws)plance rintz daince kintz glance chintz chance(plants) attrax fyting ittrax gyting climax biting citing

(attracts)attact flite ittact plite intact spite attract (at-tacked) correx drowt morrex browt convex stout

drown (corrects)chect picturesk shect ficturesk sect asteriskelect (checked)

drild friten prild kriten gild whiten build Suffixed-PH Items(drilled)

amays grue omays prue plays okays truepland caut sland vaut bland taut gland(amaze)(planned)

excues seveer axcues teveer ensues veneer ar-adord getto odord fetto accord ditto mottogues (excuse)(adored)

compoes procede dompoes trocede cargoescleerd surch pleerd turch shepherd churchforgoes concede (compose)cheers (cleared)

revies excede sevies oxcede envies advise re-bluft plak kluft glak tuft flak craft (bluffed)cede (revise)chopt blite thopt clite adopt trite crept

choes shete thoes thete oboes choke delete(chopped)(chose)golft slite holft glite aloft white graft (golfed)

slied sereen blied yereen spied plied careenbeept nautey meept wautey swept hockey(slide)tautly (beeped)

inclued redeme anclued sedeme subduedthaud skeme shaud okeme fraud theme kitingpursed scheme (include)(thawed)

crued brede drued drede glued suede briefscrude jule sprude wule delude yule prude(crude)(screwed)

fents chue nents thue vents clue tents (fence)delade frate nelade drate parade grate remadebounts squair dounts aquair mounts bounds(delayed)

repair (bounce)glode coffing clode boffing erode puffing codedants whele fants chele pants steel allele(glowed)

(dance)applide adheer epplide odheer provide careerconfide (applied) noys foyle foys noyle boys boyle foyer (noise)



fects in visual word recognition: Evidence for phono-focks paugh gocks vaugh docks bough rockslogical processing. Canadian Journal of Psychology,(fox)37, 300–305.

relacks desque selacks fesque attacks unpacks Bradley, D. (1979). Lexical representations of deriva-mosque (relax) tional relations. In M. Aronoff and M. Kean (Eds.),

abducked brisque ibducked crisque untucked Juncture. Cambridge, MA: MIT press.Burani, C., & Caramazza, A. (1987). Representation andunlocked plaque (abduct)

processing of derived words. Language and Cogni-facked junque gacked kunque packed uniquetive Processes, 2, 217–227.sacked (fact)

Chomsky, N., & Halle, M. (1968). The sound pattern ofelecked thaugh olecked whaugh wrecked English. New York: Harper & Row.

thawed thatch (elect) Coltheart, M. (1978). Lexical access in simple readingtasks. In G. Underwood (Ed.), Strategies of informa-helled flasque lelled glasque yelled jelledtion processing. London: Academic Press.brusque (held)

Coltheart, M., Davelaar, E., Jonasson, J. T., & Besner, D.kepped saugh mepped waugh ripped sauce(1977). Access to the internal lexicon. In S. Dornictopped (kept) (Ed.), Attention and Performance VI. London: Aca-

giffed straugh hiffed spraugh buffed giggle demic Press.straws (gift) Crowder, R. G., & Wagner, R. K. (1992). The Psychology

of reading. New York: Oxford.yarred kightes zarred pightes marred kindDavelaar, E., Coltheart, M., Besner, D., & Jonasson, J. T.spites (yard)

(1978). Phonological recoding and lexical access,spenned invighted ipenned anvighted scannedMemory and Cognition, 6, 391–402.

delighted specked (spend) Deutsch, A., Frost, R., & Forster, K. I. (1996, November).bined inseckt gined unseckt dined fined in- Morphological Processing of Phonological Word

Patterns in Hebrew. Paper presented at the 37th An-spect (bind)nual Meeting of the Psychonomic Society. Chicago,miled scowel siled acowel piled trowel scoutsIll.(mild)

Folk, J. R., & Morris, R. K. (1995). Multiple lexical codesmoled cayble woled nayble holed soled stable in reading: Evidence from eye movements, naming(mold) time, and oral reading. Journal of Experimental Psy-

blayed taybel klayed baybel played label taste chology: Learning, Memory, and Cognition, 21,1412–1429.(blade)

Folk, J. R., & Morris, R. K. (1996, November). Earlyshayed moddle thayed soddle toddle swayedPhonological Mediation During Reading. Presentedstayed (shade)at the meeting of the Psychonomic Society, Chicago,

wighed naybor nighed waybor sighed harbor IL.nation (wide) Fowler, C. A., Napps, S. E., & Feldman, L. (1985). Rela-

tions among regular and irregular morphologicallydivied preceed livied treceed defied proceedrelated words in the lexicon as revealed by repetitiondivine (divide)priming. Memory and Cognition, 13, 241–255.hokes saque lokes kaque pokes pique sadly

Fox, R. A. & Terbeek, D. (1977). Dental flaps, vowel(hoax) duration and rule ordering. Journal of Phonetics, 5,

27–34.REFERENCES Francis, W. N., & Kucera, H. (1982). Frequency analysis

of English usage: Lexicon and grammar. Boston:Baddeley, A. D., & Lewis, V. J. (1981). Inner active pro-Houghton Mifflin.cesses in reading: The inner ear, inner voice, and

Hawkins, H. L., Reicher, G. M., Rogers, M., & Peterson,inner eye. In A.M. Lesgold & C.A. Perfetti (Eds.),L. (1976). Flexible coding in word recognition. Jour-Interactive processes in reading (pp. 107–129).nal of Experimental Psychology: Human PerceptionHillsdale, NJ: Erlbaum.and Performance, 2, 380–385.Beauvillain, C. (1996). The integration of morphological

Hyona, J., & Pollatsek, A. (1997). The role of componentand whole-word form information during eye fixa-morphemes in processing Finnish compound wordstions on prefixed and suffixed words. Journal ofin reading text. (submitted).Memory and Language, 35, 801–820.

Berent, I., & Perfetti, C. A. (1995). A rose is a reez: The Jared, D., & Seidenberg, M. S. (1991). Does word identi-fication proceed from spelling to sound to meaning?two-cycles model of phonology assembly in reading

English. Psychological Review, 102, 146–184. Journal of Experimental Psychology: General, 120,358–394.Besner, D., & Davelaar, E. (1983). Seudohomofoan ef-



Joos, M. (1942). A phonological dilemma in Canadian information across saccades in word identificationand reading. Journal of Experimental Psychology:English. Language, 19, 141–144.

Lee, Y-A, Binder, K. S., Kim, J-O, Pollatsek, A., & Human Perception and Performance, 18, 148–162.Prince, A., & Smolensky, P. (1993). Optimality theory.Rayner, K. (1997). Activation of phonological codes

during eye fixations in reading. Submitted. Cambridge, MA: MIT Press. [in press]Rayner, K., Sereno, S., Lesch, M., & Pollatsek, A. (1995).Lesch, M., & Pollatsek, A. (1993). Automatic access of

semantic information by phonological codes in visual Phonological codes are automatically activated dur-ing reading: Evidence from an eye movement para-word recognition. Journal of Experimental Psychol-

ogy: Learning, Memory and Cognition, 19, 285–294. digm. Psychological Science, 6, 26–32.Rubenstein, H., Lewis, S. S., & Rubenstein, M. H. (1971).Lesch, M., & Pollatsek, A. (1998). Evidence for the use

of assembled phonology in accessing the meaning Evidence for phonemic recoding in visual word rec-ognition. Journal of Verbal Learning and Verbal Be-of printed words, in press. Journal of Experimental

Psychology: Human Perception and Performance. havior, 10, 645–647.Seidenberg, M. S., Petersen, A., Plaut, D. C., & MacDon-Lukatela, G., & Turvey, M. T. (1991). Phonological ac-

cess of the lexicon: Evidence from associative prim- ald, M. C. (1996). Pseudohomophone effects andmodels of word recognition. Journal of Experimentaling with pseudohomophones. Journal of Experimen-

tal Psychology: Human Perception and Perfor- Psychology: Learning, Memory, and Cognition, 22,48–62.mance, 17, 951–966.

Lukatela, G., & Turvey, M. T. (1994). Visual lexical ac- Stanners, R. F., Neiser, J. J., Hernon, W. P., & Hall, R.(1979). Memory representation for morphologicallycess is initially phonological: 1. Evidence from asso-

ciative priming by words, homophones, and pseudo- related words. Journal of Verbal Learning and Ver-bal Behavior, 18, 399–412.homophones. Journal of Experimental Psychology:

General, 123, 107–128. Taft, M. (1979). Recognition of affixed words and theword frequency effect. Memory and Cognition, 7,Luo, C. R. (1996). How is word meaning accessed in

reading? Evidence from the phonologically mediated 263–272.Van Orden, G. C. (1987). A ROWS is a ROSE: Spelling,interference effect. Journal of Experimental Psychol-

ogy: Learning, Memory, and Cognition, 22, 883– sound and reading. Memory & Cognition, 15, 181–198.895.

McCutchen, D., Bell, L. C., France, I. M., & Perfetti, Van Orden, G. C. (1991). Phonologic mediation is funda-mental to reading. In D. Besner & G. HumphreysC. A. (1991). Phoneme-specific interference in read-

ing: The tongue-twister effect revisited. Reading Re- (Eds.), Basic processing in reading: Visual word rec-ognition (pp. 77–103). Hillsdale, NJ: Erlbaum.search Quarterly, 26, 87–103.

Patterson, K. E., & Marcel, A. (1977). Aphasia, dyslexia, Van Orden, G. C., Johnston, J. C., & Hale, B. L. (1988).Word identification in reading proceeds from spell-and the phonological code of written words. Quar-

terly Journal of Experimental Psychology, 29, 307– ing to sound to meaning. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 14,318.

Perfetti, C. A., Bell, L. C., & Delaney, S. M. (1988). Au- 371–385.Venezky, R. L. (1970). The structure of English orthogra-tomatic (prelexical) phonetic activation in silent

word reading: Evidence from backward masking. phy. Janua Linguarum, series minor, no. 82. Mouton:The Hague.Journal of Memory and Language, 27, 59–70.

Perfetti, C. A., & Bell, L. (1991). Phonemic activation Zue, V., & LaFerriere, M. (1978). Acoustic study of me-dial /t,d/ in American English. Journal of the Acousti-during the first 40 ms of word identification: Evi-

dence from backward masking. Journal of Memory cal Society of America, 66, 1039–1050.and Language, 30, 473–485.

Pollatsek, A., Lesch, M., Morris, R. K., & Rayner, K. (Received March 12, 1997)(Revision received August 25, 1997)(1992). Phonological codes are used in integrating


the nature of the sound codes accessed by visual language

Documents