Haskins LaboraUJrin Statlu /Upon on Spuch /U.earch1991, SR-107/108, 209-230

Phonological Access of the Lexicon: Evidence fromAssociative Priming with Pseudohomophones*

Georgije Lukatelat and M. T. Turveytt

Four experiments were conducted with pseudohomophones in a primed naming task. InExperiments 1 and 2, target pseudowords that sounded like real words, for example,CHARE, were preceded either by context words that related associatively to the word withwhich the target was homophonic, TABLE-CHARE, or by context words that were notassociatively related, NOVEL-CHARE. Control pairs were TABLE-THARE and NOVEL­THARE (Experiment 1) and TABLE-CHARK and NOVEL-CHARK (Experiment 2). Theprior presentation of TABLE relative to the prior presentation of NOVEL benefited thenaming of CHARE but not the naming ofTHARE or CHARK. The third experiment placedpseudohomophones· in the role of primes, that is, TAYBLE-CHAIR, with such pairscomprising only 8% of all pairs seen by a subject in order to counter guessing strategies. Ifthe prime TAYBLE activated /tablet, then Ichairl would be activated associatively and thetarget CHAIR would be named faster than if TARBLE was the prime. This result wasobtained. The fourth experiment extended the design of the third to include TABLE­CHAIR pairs and a comparison of a short (280 ms) and a long (500 ms) delay betweencontext and target onsets. The priming due to associated pseudohomophones wasunaffected by onset asynchrony and equal in magnitude to that due to associated words.The overall pattern of results suggests that lexical representations are codedphonologically and accessed phonologically.

Most research on recognizing and pronouncingwords has been motivated by the dual processtheory (Coltheart, 1978). According to this theory,two independent processes-a direct, visualprocess and a mediated, phonological process­govern the accessing of the internal lexicon, amental dictionary containing relatively permanentinformation about the identity of individual words.The two independent processes are not usedequally. The direct process which entails amapping between orthographic features ofindividual printed/written words and lexicalrepresentations is the primary access route.

This research was supported in part by National Institute ofChild Health and Human Development Grants HD-08945 andHD-01994 to the first author and Haskins Laboratoriesrespectively. The authors wish to acknowledge ArthurAbramson's help in constrocting the pseudohomophones usedin Experiment 1.

209

The secondary access route is provided by the me­diated process which involves a set of grapheme­phoneme correspondence rules turning spellingsinto phonological representations, and a subse­quent mapping from these phonological represen­tations onto lexical entries. According to the the­ory, whereas the mediated route might dominateword identification in early stages of reading, it isthe direct route which characterizes reading flu­ency. And whereas phonological mediation isneeded for reading new words and nonwords, thedirect visual route is mandatory for exceptionalspellings and is preferred for familiar words.

A recent major review of the literatureconcluded that the dual process theory isessentially wrong (Humphreys & Evett, 1985).The heart of the criticism was that the appeal to arule-governed phonological process to access thelexicon is superfluous. Lexical access is achievedfor all word forms in a word-specific manner, bythe direct visual route. Evidence from studies withSerbo-Croatian language materials, and new

210 LuJazte1a and Tum:y

evidence from studies with English languagematerials suggest, however, a different criticism ofthe dual process theory: The primary constrainton lexical access is phonological, not visual, andthis constraint arises not through explicit rulesbut through continuous statistical regularity (VanOrden, Pennington, & Stone, 1990).

Until recently empirical support for phonologicalmediation was sparse. The early demonstration byRubenstein, Lewis, and Rubenstein (1971) thatlexical decisions are slower for homophones (e.g.,YOLK) and pseudohomophones, that is, nonwordshomophonous with real words (e.g., TRATE), wasoriginally taken as evidence for speech-relatedprocesses in word identification. Subsequent re­search, however, found that the homophony effecton "yes" responses could be eliminated by includ­ing many pseudohomophones as foils (Davelaar,Coltheart, Besner, & Jonasson, 1978) suggestingthat the phonological route was optional, andprobably used only for the processing of nonwords.The prelexical phonological processing of a non­word such as TRATE produces a representationwhich activates the lexical entry for the wordTRAIT, making the rejection of TRATE difficult.Unfortunately, because this evidence for prelexicalphonology is provided by the slower rejection la­tencies, it gives rise to doubts about phonology'srole in actual word identification. Various investi­gators have argued that even if phonological cod­ing does occur, it occurs too slowly to be of use inlexical access (e.g., Henderson, 1982; McCusker,Hi11inger, & Bias, 1981). As Coltheart, Davelaar,Jonasson, and Besner (1977, p. 551) remarked:"Unequivocal evidence for this view would beobtained by demonstrating that the phonologicalcode for a word is sometimes used in making the"yes" response to that word in a lexical decision orcategorization task; such a demonstration remainsto be achieved."

Recent experiments on rapid semantic catego­rization have provided such a demonstration (VanOrden, 1987; Van Orden, Johnston, & Hale, 1988).The basic observation is that subjects producelarger false positive error rates when they respondto foils that are homophonic to category exemplars(e.g., ROWS for the category A FLOWER) thanwhen they respond to spelling control foils (e.g.,ROBS). Further, false positive errors to nonwordhomophone foils (e.g., SUTE for AN ARTICLE OFCLOTHING) exceed false positive errors to non­homophonic nonword spelling controls; thephonological characteristics of the nonword foilsare critical. Many other demonstrations of phono­logical influences on positive responses are to be

found in lexical decision and naming experimentswith Serbo-Croatian materials.

Because of the use of two, partially overlappingalphabets, and because of the one letter-onephoneme principle for both alphabets, it is possibleto construct letter strings in Serbo-Croatian thatcan be read legally in more than one way. TakeBETAP as an example. Read strictly through theletter-to-sound correspondences of the Cyrillicalphabet, this letter string is pronounced /vetar/and is a high frequency word meaning "wind.".Read strictly through the letter-to-sound corre­spondences of the Roman alphabet, BETAP ispronounced /hetap/, a nonword. Read with a mix­ture of the two sets of correspondences, Cyrillicand Roman, leads to the pronunciations /vetapland /betar/, which are also nonwords. The wordmeaning "wind" is transcribed in the Roman al­phabet as VETAR. This letter string supports onlya single reading, /vetar/. No other readings arepossible. VETAR, unlike its Cyrillic mate BETAP,is phonologically unambiguous. In the lexical deci­sion and naming tasks, the latencies for BETAP,and for letter strings like it, are considerablylonger than the latencies for VETAR, and for let­ter strings like it, even though the two letterstrings are equal in frequency, syllabic structure,number of letters, and meaning. This contrast hasbeen called the phonological ambiguity effect.

The effect of phonological ambiguity on "yes" re­sponses in lexical decision and naming impliesthat phonology mediates word identification. Thesame conclusion holds for nonwords. Rejection la­tencies are slowed for the phonologically ambigu­ous nonword BEMAP relative to its phonologicallyunambiguous nonword mate VEMAR. (Arguably,BEMAP is coded into more phonological formsthan VEMAR and, because of this, BEMAP's re­jection latencies are slowed relative to those forVEMAR.) The fact that the phonological ambigu­ity effect occurs for both words and nonwords,suggests that phonological mediation must be rou­tine. Significantly, the phonological ambiguityef­fect is generally larger for words than nonwords(e.g., Feldman & Turvey, 1983; Lukatela et a1.,1980; Lukatela, Feldman et a1., 1989). As notedabove, experiments showing that homophony in­fluences nonword processing more so than wordprocessing have been interpreted to mean thatphonological coding must be rare in ordinary wordrecognition (Coltheart et aI., 1977, Henderson,1982; McCusker et al., 1981). Applying this logicto the Serbo-Croatian results, phonological codingmust be routine in ordinary word recognition be­cause phonological ambiguity influences the pro-

Phonologiclll Access of the Lexicon: Evidencefrom Associative Priming with Pseudohomophones 211

cessing of words more than the processing of non­words.

It has often been suggested that phonologicalmediation should reveal effects dependent on thenumber of letters or syllables in a word (Forster &Chambers, 1973; Frederiksen & Kroll, 1976;Green & Shallice, 1976). Letter length effects havebeen observed regularly in naming but hardlyever in lexical decision experiments using Englishlanguage materials (Henderson, 1982). The lack ofsuch effects in lexical decision is used to arguethat, contrary to dual process theory, there is nophonological route to the lexicon. Results from re­search with Serbo-Croatian materials demon­strate "number-of-constituents effects· and lead,thereby, to the contrary conclusion. KOTBA hasone ambiguous letter (B), BETAP has two am­biguous letters (B and P), and CABAHA has three(C, B, and H). In each example the remaining un­ambiguous letters are shared letters. In the lexicaldecision task with such stimuli, the phonologicalambiguity effect is larger the greater the numberof ambiguous letters (Feldman & Turvey, 1983;Feldman, KostiC, Lukatela, & Turvey, 1983). Theimplication is that there is a prelexical processthat is phonologically analytic (Turvey, Feldman,& Lukatela, 1984).

Other evidence points in the same direction. Wewould expect a phonological route to the lexicon tobe largely indifferent to variables related towords, such as grammatical role and familiarity.The inflected nouns of Serbo-Croatian are mostfrequent and most prominent grammatically inthe nominative singular form. A common findingis that nominative singulars are responded tofaster than oblique forms (e.g., Lukatela, Carello,& Turvey, 1987). For example, VENA ("vein" innominative singular) is responded to much fasterthan VENI (same word in dative or locative singu­lar). However, if VENA and VENI are written inCyrillic, that is, BEHA and BEHH, respectively,then the less frequent and less grammaticallyprominent dativellocative form BEHH is respondedto faster. The latency difference between BEHAand VENA is large (the phonological ambiguityeffect); in contrast, the latencies to BEHH andVENI do not differ (Feldman et aI., 1983). Theprocess underlying the phonological ambiguity ef­fect is not affected by the grammatical significanceof word stimuli. Further, the effect has beenshown to be indifferent to frequency; it is of thesame magnitude for both frequent and infrequentwords (Lukatela & Turvey, 1987).

There are two kinds of error patterns that en­force the notion of phonological mediation.

Rejecting a BETAP-type word as a word occurs ona high proportion of the times such a word is pre­sented, for example, 20% (Lukatela et at, 1978,Experiment 1), 19% (Lukatela et aI., 1978,Experiment 2), 26% «Lukatela, Popadic,Ognjenovic, & Turvey, 1980), 22% (Lukatela,Feldman, Turvey, Carello, & Katz, 1989,Experiment i), and 21% (Lukatela, Feldman etal., 1989, Experiment 2). In these examples, aver­age false negatives on phonologically unambigu­ous words were in the range of 1% to 4%. Falsenegatives of the order of 20-25% suggest a processin which the representation activating the lexiconfrequently does not correspond to a word. This isunderstandable if (a) BETAP can give rise to anumber of phonological representations that donot correspond to a word, and (b) the lexicon is ac­cessed through phonological representations. Thesame conclusion follows from a consideration offalse positive responses.

There are two significant types of phonologicallyambiguous nonwords in Serbo-Croatian. One typecomprises letter strings that are (a) composed ofshared letters, one or more of which are ambigu­ous, (b) unreadable as a word in either Roman orCyrillic, and (c) unreadable as a word in a mixtureof both alphabets. The other type satisfies (a) and(b) but not (c). Consider the nonword HAPEB asan example of the second type. If the letter­phoneme correspondences of both alphabets areapplied to HAPEB, then one of the resulting sixphonological descriptions corresponds to a word,namely, lnapevl meaning "tune.· It comes about byassigning the phoneme In! to H by the Cyrillic al­phabet, the phoneme IpI to P by the Roman alpha­bet, and the phoneme Ivl to B by the Cyrillic al­phabet. Compare HAPEB to a nonword of the firsttype, such as BEMAP. BEMAP has all but one let­ter in common with a real word (BETAP), and allbut one phoneme in common with this real word.HAPEB similarly has all but one letter in commonwith a real word (HAnEB). However, HAPEB hasall phonemes in common with a real word, ifbothalphabets apply. If lexical access is visual, thenfalse positives to BEMAP and HAPEB should befew and equal. In contrast, if lexical access isbased on phonemic descriptions computed prelexi­cally, and if the computation is analytic, assigningas many phonemes per letter as permitted, thenfalse positives should be greater for HAPEB.Experimentation shows that BEMAP and its un­ambiguous nonword control VEMAR generate thesame small number offalse positives (about 2-3%).When preceded by a neutral context word, HAPEBgenerated 31% false positives; when preceded by

212

an associate of /napev/, namely, MELODIJA("melody"), HAPEB generated 55% false positives(Lukatela, Turvey et al., 1989). The latencies ofthe false positives in the neutral and associativecontexts were 890 msec and 779 msec, respec­tively. Further, the false positive "yes· responsetimes to HAPEB-type nonwords and correct "yes·response times to their corresponding controlwords (e.g., NAPEV) were closely similar, 824msec to 795 msec. The same pattern of resultswith respect to BETAP and VEPAR, and HAPEBand NAPEV, held for naming (Lukatela, Feldmanet al., 1989; Lukatela, Turvey, Feldman, Carello,& Katz, 1989). These results are understandable if(a) a phonologically ambiguous nonword can giverise automatically to all of the phonological repre­sentations that its letter structure permits, (b) forsome phonologically ambiguous nonwords one ofthe automatically generated phonological repre­sentations corresponds to a word, and (c) the lexi­con is accessed through phonological representa­tions.

The conclusion to be drawn from the Englishand Serbo-Croatian data summarized in the pre­ceding paragraphs is that phonological mediationplays a significant role in lexical access. It is thecase, however, that the supportive data fromEnglish language studies remains limited; the ma­jority of the evidence provided above was fromSerbo-Croatian studies, and most of it was ob­tained with tasks exploiting the peculiarities ofthe Serbo-Croatian orthography. Our goal in thepresent article is to provide a further line of sup­port within the English language for this core hy­pothesis of the dual process theory. The experi­ments reported constitute a return to the pseudo­homophony manipulation introduced byRubenstein et al. (1971); they exploit, therefore, afeature of the English orthography that has nocounterpart in the phonetically precise Serbo­Croatian orthography. We place the pseudoho­mophony manipulation in the context of the asso­ciative priming phenomenon, and we use namingrather than lexical decision as the response mea­sure. Associative priming is the improvement inword identification that follows from preceding thetarget word by an associate, for example, TABLE­CHAIR, where the control is provided by the sametarget word following a nonassociate, for example,NOVEL-CHAIR. This effect is said to arise fromthe connections within the internal lexicon. Fromthe perspective of evaluating the nature of lexicalaccess, naming has the advantage that, unlikelexical decision, it does not necessitate post-lexicalprocesses. In signal detection terminology,

whereas lexical decision involves sensitivity andbias, naming involves only sensitivity; it does notentail a decision that requires a criterion(Seidenberg, Waters, Sanders, & Langer, 1984).

Consider the task of naming the pseudohomo­phone CHARE following the "associate" TABLE. IfTABLE's phonology is computed pre-lexically, andthis representation is used to access the lexicon,then TABLE would activate fully its lexical repre­sentation /table! and activate partially, throughlexical interconnections, the associated represen­tation /chair/. The preactivation of /chair/ couldbenefit the naming of CHARE in two ways. Thefirst way is based on the reasonable assumptionthat the prelexically computed phonology is in­complete, for example, it would not specify stress.Consequently, to name a word, the full phonologymust be retrieved from the lexicon. If CHARE iscoded phonologically prior to lexical access, then itwill contact /chair/ in the lexicon and, because ofthe associative priming of /chair/, retrieve the fullphonology faster following TABLE than followinga nonassociate. The second way CHARE couldbenefit from the priming of /chair/ is based on theassumption that the prelexical computation ofphonology and the accessing of lexical phonologyordinarily interact in the determination of a letterstring's name. If so, then CHARE will benefit fromthe prior presentation of TABLE through thefeedback from the preactivated lexical item /cltair/to the process by which CHARE's phonology is as­sembled. (The notions of assembled phonology andlexical phonology are not mutually exclusive, andin interpretations of Serbo-Croatian word namingthe need for both sources of phonological informa­tion has been repeatedly underscored [Carello,Lukatela, & Turvey, 1988; Lukatela & Turvey,1987; Lukatela, Turvey et a1., 1989; Lukatela &Turvey, 1990a]. An analogous, though weakerversion, of the same idea has been advanced forEnglish by McCann and Besner [1987].)Arguments of a similar kind apply to the evenmore interesting case in which the pseudohomo­phone is in the role of the prime, for example,TAYBLE-CHAIR. According to the phonologicalmediation hypothesis, the context stimulusTAYBLE will activate the lexical entry /table! and,by lexical interconnections, the associate /chair/.As a result, naming CHAIR will benefit from theprior processing ofTAYBLE.

In sum, the phonological mediation hypothesisof dual process theory makes the following -predic­tions: (a) a target pseudohomophone will benamed faster in the context of an associate of thepseudohomophone's source word than in an unre-

PhonologicJU Access of the Lexicon: EtJidence from Associative Priming with Pseudohomophones 213

lated context, and (b) a target word will be namedfaster in the context of a pseudohomophone whosesource word is an associate of the target wordthan in the context of a pseudohomophone whosesource word is unrelated to the target word.Experiments 1 and 2 test prediction (a),Experiments 3 and 4 test prediction (b).

EXPERIMENT 1The first experiment uses stimuli constructed in

the manner of McCann and Besner (1987) in theirdemonstration that pseudohmophones are namedfaster than ordinary nonwords. The nonpseudo­homophone controls (e.g., THARE) were distin­guished from the pseudohomophones (e.g.,CHARE) by a single initial letter. The experimentcompared target naming latencies for TABLE­CHARE, NOVEL-CHARE, TABLE-THARE, andNOVEL-THARE pairs.

MethodSubjects. The participants in the experiment

were 24 students from the Department ofPsychology at the University of Connecticut. Asubject was assigned to one of four groups to givea total of six subjects per group.

Materials. A set of 60 semantically related wordpairs (e.g., TABLE-CHAIR) was generated on thebasis of a high associativity rating in Palermo andJenkins (1964), and in a tabulation of associativenorms collected at the University of Connecticut.In each context-target pair the target word wasreplaced by its pseudohomophone to produce a setof 60 word-pseudohomophone pairs (TABLE­CHARE). With the same pseudohomophonetargets, another set of 60 associatively unrelatedpairs (NOVEL-CHARE) was also generated.

A third set of 60 word-nonword pairs (e.g.,TABLE-THARE) was generated by replacing thefirst (or, in some cases, the second) grapheme ofeach pseudohomophone (CHARE) with anotherlegal grapheme (THARE). A fourth set of 60 word­nonword pairs was assembled by combining theassociatively unrelated context word (GULF) withthe non-pseudohomophone target (THARE).Thecomplete set of experimental materials ispresented in Appendix 1. In addition, a foil set of48 word-word pairs was selected from a differentword list. In one half of the foil pairs the contextand target word were associatively related and inthe other half they were unrelated.

Four counterbalanced lists of stimuli pairs wereprepared. In each group, each subject saw 15 in­stances of each stimulus pair. For example, in onegroup the subject would see 15 FOOT-HANNED

(associated pseudohomophone derived fromHAND) pairs, 15 HAMMER-LALE (associatednonpseudohomophone derived from NAIL) pairs,15 FACT-KWEEN (nonassociated pseudohomo­phone derived from QUEEN) pairs, and 15DRESS-SUDDER (nonassociated nonpseudoho­mophone derived from BUTTER) pairs.Additionally each subject saw 24 related word­word pairs (GOLF-BALL),and 24 unrelated word­word pairs (FOOL-BALL).

Design. The major constraint on the design wasthat a given subject never encountered a giventarget more than once. There were four (2 x 2)stimulus types (associative relation. x target ho­mophony). Each subject was presented with 15experimental stimulus pairs from each of the fourtypes and, in addition, with 48 foils for a total of108 stimulus pairs. The experimental sequencewas divided into quarters, with a brief rest in be­tween. Stimulus types were ordered pseudoran­domly within each quarter. Experimental se­quence was preceded by a practice sequence of 32stimulus pairs.

Procedure. Subjects, run one at a time, sat infront of the CRT of an Apple IIe computer in adimly lit room. A fixation point was centered onthe screen. On each trial, there was a briefauditory warning signal after which a letter stringappeared for 500 ms above the fixation point.After a 100 ms interstimulus interval, anotherletter string appeared below the fixation point for500 ms. The subject was told that he or she wouldbe viewing word-word and word-nonword pairs,and that some of the nonwords, when pronounced,would sound like English words. Subjects wererequired to pronounce each target letter string asquickly and as distinctly as possible. In allconditions, latencies from the onset of the targetto the onset of the response were measured by avoice operated trigger relay. Naming wasconsidered to be erroneous when the pronun­ciation included a phoneme not specified byallowable grapheme-to-phoneme correspondencesin English, the pronunciation was not smooth (i.e.,the subject hesitated after beginning to name), orthe response was not loud enough to trigger thevoice key. To ensure that subjects were readingthe contexts, the subject was asked occasionally(on less than 5% of total trials) by a computermessage to report orally both words (only some ofthe foil word-word pairs were controlled) after thetarget word was named. If the naming latency waslonger than 1000 ms, a message appeared on thescreen requesting the subject to name morequickly. All latencies, including those longer than

214 LulaIte1a lind Tumry

Table 1. Mean naming latencies L (in ms) and errorrate ER (in %), with the corresponding standarddeviations by subjects and by items in Experiment 1.

1000 ms, were correctly stored in the computermemory.

Results and DiscussionFor each subject, naming latencies more than

two standard-deviation units above or below thesubject's mean in all conditions were considered aserrors. (This procedure was followed for all of thereported experiments.) The mean latencies of eachgroup of subjects, standard deviations, and errorsfor related and unrelated contexts are presentedin Table 1.

Context-target Relation

oooa 5.838lb 6.6168c 12.21

ms vs. unrelated = 621 ms) proved significant bysubjects, F<l, 23) = 17.38,p < .001, and by stimuli,F(1, 59) = 3.88, p < .05.

A 2 x 2 error analysis showed no significantmain effects. There was, however, a significant in­teraction of associativeness and homophony(associative effect for pseudohomophone errors = +6.11% vs. associative effect for non-pseudohomo­phone errors = - 1.94%) by subjects, F(l, 23) =8.23, p < .01, and by stimuli, F(l, 59) = 7.19, p <.01. A separate analysis for pseudohomophone er·ror data showed that associativeness (related =6.83% vs. unrelated = 11.94%) was significant bysubjects and stimuli, F(l, 23) =6.84, p < .01, F(l,69) = 5.07,p < .05, respectively. For the non-pseu­dohomophone data, the main effect of associative­ness (related =661 ms vs. unrelated =646 ms)was insignificant in the latency data, F(l, 23) =3.74, p > .05 by subjects, and F(l, 59) = 1.42, p >.05, by stimuli. Associativeness (related = 7.22%vs. unrelated = 5.28%) was also insignificant inthe non-pseudohomophone error data, F(l, 23) =1.62, p > .05 by subjects, and F < 1 by stimuli.

In sum, CHARE was named faster and withfewer errors (as defined above) when the contextwas TABLE than when the context was NOVEL.The naming of THARE, in contrast, showedstatistically no differential sensitivity to TABLEand NOVEL.

ER

Unrelated

L

621 11.9482 9.6380 17.92

ER

Related

L

Non-Pseudohomophone

Pseudohomophone

Target

8meanbstandard deviation by subjectsCstandard deviation by items

A 2 X 2 analysis of variance was conducted onthe latency data with factors of associativeness,and homophony (pseudohomophones vs. non­pseudohomophones). Homophony was significant(pseudohomophones = 610 ms vs. non·pseudohomophones = 653 ms) by subjects F(l, 23)=35.53, p < .001, and by stimuli, F(l,59) =34.04,p < .001. Associativeness was not significant as amain effect (related =630 ms vs. unrelated =633ms), but interacted with homophony (associativeeffect for pseudohomophones = +21 ms vs.associative effect for non-pseudohomophones =-15ms) significantly by subjects, F(l, 23) = 14.83, p <.001, and by stimuli, F(l,59) =4.33, p < .05.

Because of the associativeness with pseudoho­mophony interaction, separate analyses were per­formed on the pseudohomophone data, and thenonpseudohomophone data. Thus, associativenessfor the pseudohomophone latencies (related = 600

661 7.2294 5.1778 10.79

646 5.28109 7.3590 9.45

EXPERIMENT 2Although the outcome of the first experiment

was straightforward, questions might be raisedabout (a) the high error rates for nonrelatedpseudohomophone targets, and (b) the manner inwhich the control nonwords were created. Withrespect to (a) one may note that some of thepseudohomophones (for example, KLOWD, HAHT,SMUTHE, ... standing for CLOUD, HOT,SMOOTH, ... , respectively) in Experiment 1 (seeAppendix 1) did not constrain the reader to theexpected homophonic pronunciation and,evidently, left room for various phonologicalint.erpretations. With respect to (b) one may objectthat the graphemic structure of non·pseudohomophones (e.g., THARE) was less likethe source word CHAIR than was the graphemicstructure of pseudohomophones (e.g., CHARE).Consequently, the differences obtained inExperiment 1 could be attributed wholly or in partto the graphemic advantage of thepseudohomophones over their controls. The secondexperiment addressed this potential criticism. IfCHARE was the pseudohomophone, then the newcontrol would be CHARK. that is, a letter string

Phonological Access of the LexU:on: EfJidencefrom Associative Priming with Pseudohomophones 215

that has the same letter overlap with the sourceword CHAIR as does the correspondingpseudohomophone. More precisely speaking, each"pseudohomograph" control and the correspondingpseudohomophone would share the same numberof letters with the source word in terms of bothexact letter positions and total number of letters.Because this manipulation required an almostnew basic set of 64 source pairs (different fromthat in Experiment 1), the second experiment alsotests the generality of the preceding results.Finally, in the second experiment one set of wordsserved as both associated and nonassociatedcontexts in contrast to the first experiment inwhich different sets of words functioned in theseroles. This additional manipulation guardedagainst uncontrolled effects due to differences inaverage familiarity of the context stimuli.

MethodSubjects. The participants in the experiment

were 24 students from the Department ofPsychology at the University of Connecticut. Asubject was assigned to one of four groups to givea total of six subjects per group. The subjects didnot participate in Experiment 1.

Materials. The stimuli (see Appendix 2) weregenerated from a set of 64 associatively relatedword pairs was generated on the basis of a highassociativity rating in Palermo and Jenkins(1964), and in a tabulation of associative normscollected at the University of Connecticut. In addi­tion, a list of different words, which can be conve­niently transcribed as pseudohomophones, wascompiled in part from the source words used byPerfetti, Bell, and Delaney (1988), and in partfrom the authors' own work. The list was pre­sented to a pool of 60 undergraduate students ofPsychology at the University of Connecticut. Eachstudent in the pool was asked to read through thelist on a word by word basis and after each wordto write down on a prepared form the first threewords that may have come to his/her mind. Theappropriate associates were picked up to createnew associatively related word-word pairs thatwere paired to make 32 "related word quadruples"(e.g., BITE-TEETH; CIDER.JUICE). Within eachrelated quadruple the mutual substitution of con­text words produced a new "unrelated wordquadruple" of two associatively unrelated pairs(e.g., CIDER-TEETH, BITE-JUICE). In each con­text-target pair the target word was replaced byits pseudohomophone to produce a set of 64 asso­ciatively related word-pseudohomophone pairs or32 related pseudohomophone quadruples (BITE-

TEATH; CIDER-JOOCE). With the same pseudo­homophone targets, a second set of 32 associa­tively unrelated pseudohomophone quadruples(e.g., CIDER-TEATH; BITE-JOOCE) was alsogenerated.

A third set of 64 "associated" word-pseudohomo­graph pairs or 32 related pseudohomographquadruples (e.g., BITE-TERTH; CIDER-JORCE)was generated, whereby each pseudohomographtarget shared the same letters as the pseudoho­mophone target, in position and out of position,with the source word (e.g., TEETH, JUICE, re­spectively). A fourth set of 32 unrelated pseudo­homograph quadruples was assembled by mutu­ally substituting the two context words in eachquadruple (e.g., CIDER-TERTH; BITE-JORCE).With regard to all pseudohomophones and pseu­dohomographs used in this and in the next exper­iment a word of caution might be necessary: theymet an arbitrary criterion of lexicality for a lettersequence. Namely, the pseudohomophones andpseudohomographs should have no entry in theWord Frequency Book by Carroll, Davies, andRichman (1971) which is based on the AmericanHeritage Intermediate Corpus of approximately5,000.000 words (tokens).

In order to prevent any special responsestrategy in rapid naming, a foil set of 32associated word-word pairs (e.g., SOFT-HARD)and a corresponding (i. e., having the same targetwords) foil set of 32 unrelated word-word pairs(e.g., CHESS-HARD) were selected from adifferent word list. In addition, a third foil set of32 word-nonword pairs was also created. Thenonword targets were easily pronounceable, legalEnglish letter strings.

As before, four counterbalanced lists of stimulipairs were prepared. In each group, each subjectsaw eight quadruples with 16 instances of eachstimulus pair. For example,in one group the sub­ject would see 16 BITE-TEATH (associated pseu­dohomophone derived from TEETH) pairs, 16DEAF-MERT (associated pseudohomograph de­rived from MUTE) pairs, 16 MOAN-FYVE(unrelated pseudohomophone derived from FIVE)pairs, and 16 WEAR-BUSHER (unrelatedpseudohomograph derived from BUTTER) pairs.Additionally each subject saw 16 related word­word pairs (SOFT-HARD), 16 unrelated work­word pairs (OLIVE-STRICT), and 32 word-non­word pairs (ROSE-SHAIG). In total, anexperimental list consisted of 128 stimulus pairs.

Design and procedure. These were the same asin Experiment 1, except for the fact that thecontext and target presentation time both were

216 LubJtela and Tum:y

Table 2. Mean naming lalencies L (in ms) and errorrate ER (in %), with the corresponding standarddeviations by subjects and by items in Experiment 2.

reduced to 400 ms each, leaving the interstimulusinterval unchanged (100 ms).

Results and DiscussionThe mean latencies, standard deviations, and

errors for pseudohomophones and pseudohomo­graphs in associatively related and unrelatedcontexts are presented in Table 2.

Pseudohomophone

557· 3.39 578 3.9170b 4.51 79 4.4459c 8.84 54 8.56

Pseudohomograph

582 2.86 574 3.9178 4.87 78 4.4448 7.08 50 10.80

8meanbStandard deviation by subjectsCstandard deviation by items

A 2 X 2 analysis of variance was conducted onthe latency data with factors of associativeness,and homophony (pseudohomophones vs. pseudo­homographs). Homophony was significant(pseudohomophones = 567 ms vs. pseudohomo­graphs =578 ms) by subjects F(1,23) =7.31, p <.01, but not by stimuli, F(1,63) =3.57, p > .05 .Associativeness was not significant as a main ef­fect (related =569 ms vs. unrelated =575 ms), butit interacted with homophony (associative effectfor pseudohomophones = +20 ms vs. associative ef­fect for pseudohomographs = - 5 ms) significantlyboth by subjects, F(l, 23) = 19.72, p < .001, andstimuli, F(1, 63) = 5.15, p < .05.

Because of the associativeness with homophonyinteraction, separate analyses were performed onthe pseudohomophone data, and the pseudohomo­graph data. Thus, associativeness for the pseudo~

homophone latencies (related = 557 ms vs. unre­lated = 578 ms) proved significant by subjects,F(l, 23) = 10.91, p < .01, and by stimuli, F(1, 63) =5.26, p < 0.5. A 2 x 2 error analysis showed nosignificant effects and no interaction.

Experiment 3Although the results of Experiments 1 and 2 can

be discussed in terms of the automatic priming of/chair/, and of the prelexical phonologicalprocesses by which CHARE's name is assembled,other interpretations are possible. In particular,given the challenge of naming nonwords rapidly, asubject might have adopted a conscious strategy ofgenerating on each trial a word associated withthe context in order to help in the pronunciation ofthe nonword target. The fact that 25% of the pairswere of the GOLF-BALL type and 16% were of theTABLE-CHARE type might have encouraged sucha strategy.

The third experiment examined the efficacy of apseudohomophone as an associated prime.Specifically, the experiment compared naming la­tencies to CHAIR when presented subsequent toTAYBLE and TARBLE in an experimental settingin which TARBLE-CHAIR pairs accounted foronly about 8% of all pairs. This experiment, there­fore, militates against the conscious association­pronunciation strategy identified above, andshould provide thereby a purer test of the phono­logical mediation hypothesis than the precedingtwo experiments. According to the phonologicalmediation hypothesis, TAYBLE will activate thelexical entry /table!, resulting in the priming ofthat entry's associate, Ichair/. The contextTARBLE, although visually similar to TAYBLE,win not activate the lexical item /table! and, there­fore, not prime /chair/. Consequently, namingCHAIR should be faster in the context TAYBLEthan in the context TARBLE.

Considering the pseudohomograph data, themain effect of associativeness was not significantfor latencies (related = 582 ms vs. unrelated = 574ms}--by subjects, F(l, 23) = 2.06, P > .1, by stim­uli, F(1, 63) = 0.2, p > .1-nor was it significant forerrors (all F < 1). The mean naming latency forassociatively related pseudohomophones (557 ms)was reliably faster than for associatively relatedpseudohomographs (582 ms), both by subjects,F(1, 23) =18.17, p < .001, and by stimuli, F(l, 63)=7.18,p < .01.

The obtained results suggest an associativepriming of pseudohomophones, and no contextualeffect for the pseudohomographic controls. Themagnitude of the obtained contextual effect (+21ma) compared favorably with the magnitude ob­tained on the foil set of 32 word-word stimuluspairs (HARD-SOFT =525ms vs. CHESS-SOFT =540ms).

ER

Umelated

L

Context-target Relation

Related

L ER

Target

Phonological Access of the Lericon: Evidencefrom Associative Priming with Pseudohomophones 217

The design of the third experiment is significantin an additional way. IfTAYBLE-CHAIR is fasterthan TARBLE-CHAIR, then it would emphasizethe prelexicalllexical nature of pseudoassociativepriming. An effect arising from the combination ofnaming as the dependent measure and pseudo­homophones as primes would pose difficulties forany post-lexical account. The consensus is thatnaming involves few, if any, post-lexical checks,and that associative priming is purely lexical(Balota &: Chumbley, 1985; Carello et al., 1988;Seidenberg et aI., 1984)

MethodSubjects. The participants in the experiment

were 24 students from the Department ofPsychology at the University of Connecticut. Asubject was assigned to one of four groups to givea total of six subjects per group. None of the sub­jects had participated in the previous experi.ments.

Materials. The stimuli (see Appendix 3) weregenerated from the same general list which wasused in Experiment 2, though not all of the word­word pairs in this experiment were necessarilyidentical to those in Experiment 2. A basic set of64 associatively related word pairs that werepaired to make 32 "related word quadruples" (e.g.,BITE-TEETH; HATE-LOVE). Within each relatedquadruple the mutual substitution of contextwords produced a new "unrelated word quadruple"of two associatively unrelated pairs (e.g., HATE­TEETH, BITE-LOVE). In each context-target pairthe context word was replaced by its pseudohomo­phone to produce a set of 64 associatively relatedpseudohomophone-word pairs or 32 related pseu­dohomophone quadruples (BIGHT-TEETH; HAIT­LOVE). With the same word targets, a second setof stimuli which comprised 32 associatively unre­lated pseudohomophone quadruples (e.g., HAIT­TEETH; BIGHT-LOVE) was also generated.

A third set of 64 "associated" pseudohomograph­word pairs which comprised 32 related pseudoho­mograph quadruples (e.g., BIRET-TEETH;HANT-LOVE) was generated, whereby each pseu­dohomograph context shared the same letters asthe pseudohomophone context, in position and outof position, with its source word (e.g., BITE,HATE, respectively). A fourth set of 32"unrelated" pseudohomograph quadruples was as­sembled by mutually substituting the two contextwords in each quadruple (e.g., HANT-TEETH;BIRET-LOVE). Summed bigram frequency(Mayzner &: Tresselt, 1965) was computed for eachpseudohomophone and pseudohomograph; in the

majority of pseudohomographs (i.e., 49 out of 64)the summed bigram frequency was higher than inthe corresponding pseudohomophones. In addi­tion, a comparison set of 32 associated word-wordpairs (e.g., BOY-GIRL) and another set having thesame target words but unrelated context words(e.g., GULF-GIRL) were selected from a differentword list. The targets words in the comparison setwere selected to have approximately the same av­erage frequency of occurrence as the target wordsin the four experimental sets. In order to preventany special response strategy in rapid naming, afoil set of 64 nonword-nonword pairs was assem­bled. In the same vein, to discourage further asubject's guessing strategy, a foil set of 32 pseudo­homophone-nonword pairs was also created. Allnonwords were easily pronounceable, legalEnglish letter strings.

As in the preceding experiments, four counter­balanced lists of stimuli pairs were prepared. Ineach group, each subject saw eight quadrupleswith 16 instances of each stimulus pair. For ex­ample, in one group the subject would see 16BIGHT-TEETH (associated pseudohomophonecontext derived from BITE) pairs, 16 DELF­MUTE (associated pseudohomograph context de­rived from DEAF) pairs, 16 MONE·FIVE(unrelated pseudohomophone context derived fromMOAN) pairs, and 16 WAUR-BUTrER (unrelatedpseudohomograph context derived from WEAR)pairs. Additionally each subject saw 16 relatedword-word pairs (BOY-GIRL), 16 unrelated work­word pairs (SKULL-LAW), 32 pseudohomophone­nonword pairs (LOGIK-SHAIG), and 64 nonword­nonword pairs (TINFER·PROLIT). The foil setswere used to counter the development of biases,such as (a) always looking for an associative rela­tion, (b) making predictions about targets basedon the sound of the context, and (c) trying to makecontexts sound like words. In total, an experimen­tal list consisted of 192 stimulus pairs.

Design. The design was similar to that inExperiments 1 and 2, except for an increase in thenumber of foil stimuli pairs. This increase pro­duced a low ratio (l : 3) in the presentation of as­sociativelyrelated relative to associatively unre­lated stimulus pairs within each quarter. The ex­perimental sequence was preceded by a practicesequence of 32 stimulus pairs.

Procedure. The procedure was the same as thatin Experiment 2.

Results and DiscussionThe mean latencies, standard deviations, and

errors for pseudohomophones and pseu-

218 Lukatela and Tum:y

Table 3. Mean naming latencies L (in ms) and errorrate ER (in %), with the corresponding standarddeviations by subjects and by items in Experiment 3.

L ER L ER

Pseudohomophone context

522a 1.82 542 0.7845b 2.90 47 2.1141c 5.24 46 3.73

Pseudohomograph context

538 3.13 534 1.0453 4.12 49 2.3851 7.63 41 3.60

lImeanbstandard deviation by subjectsCstandard deviation by items

The associativeness by context homophonyinteraction motivated separate analyses on thepseudohomophone context pairs and thepseudohomograph context pairs. Thus, the effecton naming latency of associativeness whencontexts were pseudohomophones (related = 522ms vs. unrelated = 542 ms) proved significant bysubjects, F(1, 23) = 19.07,p < .001, and by stimuli,F(l, 63) = 6.76,p < 0.01.

A 2 x 2 error analysis displayed a significantmain effect of associativeness (related = 2.47% vs.unrelated = 0.91%), F(l, 23) = 5.75, p < .05, F(l,63) = 7.28, p < .01. The main effect of context ho­mophony was not significant, F(l, 23) = 1, 68, p >

dohomographs in associatively related andunrelated contexts are presented in Table 3. A 2 x2 analysis of variance was conducted on thelatency data with factors of associativeness, andthe homophony of the context (pseudohomophonecontext vs. pseudohomograph context).Associativeness as a main effect (related = 530 msvs. unrelated = 538 ms) was significant only bysubjects, F(l, 23) = 4.33, p < .05, but not by stim­uli, F(l. 63) = 1.66, p > .1. Main effect of contexthomophony was not significant, F(l, 23) = 1.43, p> .1, F(l, 63) =0.96, p > .1, by subjects-and stim­uli, respectively. However, context homophony in­teracted with associativeness (associative effectfor pseudohomophone context = +20 ms vs. asso­ciative effect for pseudohomograph context = - 4ms). The interaction was significant both by sub­jects, F(l, 23) =17.84, p < .001, and by stimuli,F(l, 63) = 4.35, p < .05.

EXPERIMENT 4In Experiment 3 the stimulus onset asynchrony

(SOA) between context and target was 500 ms.With respect to the trials with nonword contexts,the delay from context to target may have allowedsubjects the opportunity to construct the context'sphonology and then to determine consciouslywhether or not there was a word with which it

.1, F(l, 63) = 1.54, P > .1; neither was the interac­tion between context homophony and associative­ness, F < 1. A similar analysis on the contextpseudohomograph data revealed that the main ef­fect of associativeness was not significant for la­tencies (related =538 ms vs. unrelated =534 ms),either by subjects or by stimuli (F <1 in bothcases).

A further observation was that the meannaming latency for word targets preceded byassociatively related pseudohomophones (557 ms)was reliably faster than the mean naming latencyfor word targets preceded by associatively relatedpseudohomographs (582 ms), both by subjects,F<1, 23) = 18.17, p < .001, and by stimuli, F(l, 63)= 7.18,p < .01.

How efficient are pseudohomophones as asso­ciative primes compared to real word contexts? Anapproximate evaluation is provided by an ANOVA(with factors of associativeness and context lexi­cality) conducted on the target latencies of the ex­perimental set of 32 pseudohomophone-word pairs(e.g., TAYBLE·CHAIR) and the set of 32 word­word pairs (BOY-GIRL). The analysis was withinsubjects and between items. The evaluation isonly approximate because of the different targetitems and different associative strengths in thecomparison pairs. Only the main effect of associa­tiveness (related =521 ms vs. unrelated =540 ms)was significant, F(l, 23) = 38.87, p < .001, F(l, 62)= 10.24, p < .01, by subjects and by stimuli, re­spectively. The important effects of context lexical­ity (pseudohomophones =532 ms vs. words =529ms), and of associativeness by context lexicality(associative effect by pseudohomophone contexts =+ 20 ms vs. associative effect by word contexts = +19 ms) were not significant, F < 1 in both cases.

Finally, we note that given the design of the ex­periment, a strategy of looking for associationswas not encouraged (only 1/6 of all stimulus pairswere associated words), neither was the assem­bling of a name for the prime (only targets had tobe named). Despite these discouraging features ofthe experimental design, TAYBLE primed CHAIRsuggesting that neither the effect of associationnor the effect of phonology were optional.

Context-target Relation

Related UnrelatedContext

Phonological Access oj the Lexicon: Evidence from AssociDtitJe Priming with Pseudohomophones 219

was homophonic. That is to say, the source ofTAYBLE's advantage over TARBLE in the prim­ing of CHAIR may have resided in strategic pro­cesses of a post-lexical nature rather than auto­matic processes of a pre-lexical and lexical nature.The inclusion in Experiment 3 of a relativelysmall proportion of TAYBLE-CHAIR pairs (8.3%of the total number of pairs) was for the purposeof minimizing conscious guessing strategies of thepreceding kind, but the possibility remains thatthis protective measure was not sufficient. In thefourth experiment the conditions of priming by anonword homophonic with an associate of the tar­get were repeated with SOAs of 280 ms and 500ms. A much reduced delay of target relative to thecontext ought to restrain the use of consciousstrategies, or at least render them less effective.Consequently, if the priming effect of TAYBLE isstrategic, then there should be an interaction be­tween onset asynchrony and the associative rela­tion between context and target; that is, relativeto TARBLE, TAYBLE should prime CHAIR lesswith SOA = 280 ms than with SOA = 500 ms.

In addition to examining the effect of SOA, thefourth experiment was designed to provide an as­sessment of the degree to which priming a targetword by a nonword homophonic with an associateof the target is comparable to priming a targetword by an associated word. In the third experi­ment the relative efficacy of pseudohomophonicpriming had been evaluated statistically by com­paring pairs such as TAYBLE-CHAIR with pairssuch as BOY-GIRL. Neither the lexical status ofthe context nor the associativeness by context lexi­cality interaction were significant. In the fourthexperiment, the comparison is made across pairsthat have the same target and the same associa­tive strength, for example, TAYBLE-CHAIR vs.TABLE-CHAIR. Evidence that pseudohomophonicnonwords prime as well as words will be looked forin the latency differences between (a) CHAIR pre­ceded by TAYBLE and BLOO and (b) CHAIR pre­ceded by TABLE and BLUE. If pseudohomo­phones prime equally as well, then the advantageof TAYBLE over BLOO in (a) should be of thesame magnitude as the advantage of TABLE overBLUE in (b), and if the pseudohomophonic prim­ing effect witnessed in the preceding experimentsis due to non-strategic processes, then this equa­tion should hold equally over both the short (280ms) and long (500 ms) SOAs.

MethodSubjects. The participants in the experiment

were 60 students at the University of Connecticut.

Subjects received either credit for participating aspart of the course requirements in IntroductoryPsychology or were paid $5. A subject wasassigned to one of two main groups (distinguishedby SOA), and within a group to one of sixsubgroups to give a total of five subjects persubgroup. None of the subjects had participated inthe previous experiments.

Materials. There were six word sets (seeAppendix 3). The first set (also the base set) con­sisted of 96 associatively related word pairs thatwere paired to make 48 "related word quadruples"(e.g., BITE-TEETH; HATE·LOVE). From this firstset, five other sets of 96 pairs were generated inthe manner described in Experiment 3: Set 2(unrelated word) represented by BITE-LOVE, Set3 (related pseudohomophone) represented byHAlT-LOVE, Set 4 (unrelated pseudohomophone)represented by BIGHT·LOVE, Set 5 (relatedpseudohomograph) represented by HANT-LOVE,and Set 6 (unrelated pseudohomograph) repre­sented by BIRET-LOVE.

Six counterbalanced lists of stimuli pairs wereprepared. In each subgroup, each subject saw 16pairs from of each stimulus set (for a total of 96),plus a foil set of 64 nonword-nonword pairs, a foilset of 32 pseudohomophone-nonword pairs (allnonwords were easily pronounceable, legalEnglish letter strings), and a small set of eightnonrelated word-word pairs used for probingawareness of the context. The foil sets were used,as in Experiment 3, to counter the development ofbiases, such as (a) always looking for an associa­tive relation, (b) making predictions about targetsbased on the sound of the context, and (c) trying tomake contexts sound like words. In total, an ex­perimentallist consisted of2oo stimulus pairs.

Procedure. The stimuli were presented in a dif­ferent manner from the previous experiments.Contexts and targets were fixed at 200 ms nomi­nal duration. The interstimulus interval was ei­ther 300 ms or 80 ms to produce nominal SOAs of500 ms and 280 ms, respectively. In further con·trast to the preceding experiments, the contextand target were presented at the same place (atthe fIXation point). The latter change in the pre­sentation of stimuli was demanded by the factthat for SOA = 280 ms the time to shift fixationwithin the procedure used in Experiments 1-3would be detrimental to performing the main task(e.g., the subject would learn to fixate on the areabelow the fixation point where the target ap­peared consistently). A major consequence of thischange in manner of displaying the paired stimuliwas the possibility for forward and/or backward

220

Table 4. Mean naming latencies L (in ms) and errorrate ER (in 'To), with the corresponding standarddeviations by subjects and by items in Experimenl4.

masking, especially at SOA = 280 ms. In thisregard, it is important to note that pilot workrevealed that, for the nominal stimulus exposuretimes of 200 ms, nominal SOAs briefer than 280ms (e.g., 230 ms) produced an unacceptable levelof error in reporting the context of the probedpairs.

Results and DiscussionThe mean latencies, mean errors, and standard

deviations for words, pseudohomophones,and pseudohomographs in associatively relatedand unrelated contexts at both SOAs arepresented in Table 4. A 2 x 3 x 2 ANOVA wasconducted with factors of associativeness (withinSs), context (within Ss), and SOA (between Ss).

Context-target Relation

Related Unrelated

The main effect of associativeness (associated =547 ms vs. nonassociated = 557 ms) wassignificant, F(1, 58) = 33.89, p < .001 (by subjects),F(l, 190) = 13.66, p < .001 (by stimuli). Alsosignificant was the main effect of context (word =543 ms, pseudohomophone = 554 ms,pseudohomograph =559 ms), F(2, 116) =25.37, p< .001 (by subjects), F(2, 380) =16.33, p < .001 (bystimuli). The main effect of SOA was notsignificant Oong SOA = 542 Dl8 vs. short SOA =562 ms) by the subjects analysis, F(l, 58) = 2.01, p> .05 but it was significant by the stimuli analysis,F(l, 190) = 36.63, p < .001. Of all possibleinteractions, only the associativeness by contextinteraction (related-unrelated difference for words= 16 ms vs related-u "related difference forpseudohomophones =1·us vs related-unrelateddifference for pseudohoT!lographs = -3 ms) wassignificant: F(2, 116) ::;;'.84, p < .001 (by sulrjects), F(2, 380) = 5.09, p < .007 (by stimuli).

Tuming to the partial interactions, both that be­tween words and pseudohomographs and that be­tween pseudohomophones and pseudohomographswere significant: F(l, 58) = 22.02, p < .001(bysubjects), F(l, 190) = 7.70, p < .006 (by stimuli)and F(l, 58) = 28.14, p < .001 (by subjects),F(1,190) = 7.77,p < .006 (by stimuli), respectively.The partial interaction between words and pseu­dohomophones was not significant-F < 1 in boththe subjects and stimuli analyses. Importantly,the partial interaction between words and pseu­dohomophones was not significant at either SOA,all Fs < 1, suggesting that the associativeness bycontext interaction was due solely to the behaviorof the TARBLE-CHAIRIBLOM-CHAIR pairs rela­tive to the word and pseudohomophone pairs.

This fourth experiment was designed todetermine whether or not associative priming ofword targets through pseudohomophones wasdependent on the magnitude of time elapsingbetween context and target. Experiment 3 hadused an SOA of 500 ms. The present experimentused an SOA of the same magnitude and one thatwas considerably shorter, viz., 280 ms. Theanalyses show that the efficacy ofpseudohomophone associative priming was (a)indifferent to the SOA difference, and (b) of thesame magnitude as word associative priming. ForSOA = 500 ms, TABLE-CHAIR differed from itsunrelated control by 18 ms (this plannedcomparison was significant by both subjects andstimuli, F(l, 29) = 14.51, p < .001 and F(l, 95) =8.46, p < .005, respectively), TAYBLE-CHAIRdiffered from its unrelated control by 15 ms(significant by both subjects and stimuli, F(l, 29)

ERLER

0.83 556 1.462.16 61 3.554.02 48 5.23

1.46 573 1.673.91 59 3.265.23 50 6.27

0.42 571 0.831.59 59 2.162.87 47 4.02

2.29 545 1.253.84 54 2.547.61 45 4.87

0.21 552 1.041.14 47 2372.04 43 4.47

1.25 544 0.833.03 54 2.165.67 43 4.02

Context

L

SOA = 500 ms

Word527a

53b

47c

Pseudohomophone537

4744

Pseudohomograph549

5046

SOA = 280ms

Word543

6050

Pseudohomophone555

5443

Pseudohomograph572

5749

8meanbstandard deviation by subjectscstandard deviation by items

PhonologiaU Acass of the Lexicon: Evidence from Associative Priming with Pseudohomophones 221

=27.26, p < .001 and F(1, 95) =5.82, p < .02,respectively), and TARBLE-CHAIR differed fromits unrelated control by -5 ms (insignificant byboth subjects and stimuli, F(1, 29) = 3.12, p > .05and F < 1, respectively). For SOA = 280 ms,TABLE-CHAIR, TAYBLE-CHAIR, and TARBLE­CHAIR differed from their unrelated controls by13 ms (F(1, 29) =24.29, p < .001 and F(1, 95) =4.22, p < .05), by 18 ms (F(1, 29) = 17.98, p < .001and F(1, 95) = 5.23, p < .03), and by -1 ms (F < 1for both subjects and stimuli), respectively. Theimplication is that one and the same mechanismunderlies both pseudohomophone and wordassociative priming, and that pseudohomophoneassociative priming, therefore, is not due to aspecial (post-lexical) strategy.

There was a main effect of context lexicality inthe sense that naming CHAIR was faster to thesame degree following each of the two wordcontexts (TABLE and BLUE) than following eachof the two pseudohomophone contexts (TAYBLEand BLOO). Clearly, this faster naming after wordcontexts has nothing to do with the mechanism ofassociative priming. It might be due to a post­lexical verification process in which the subjectcompares the assembled name with the initialrepresentation of the stimulus prior to executingthe response (see Paap, Newsome, McDonald, &Schvaneveldt, 1982). Suppose that the verification(match-mismatch) process is modulated over theshort term by the kind of letter string justprocessed. Then the possibility can be entertainedthat processing a pseudohomophone raisestemporarily the criterion for verification; as aconsequence, the time to name an after-comingword is slowed. That pseudohomophone contextsmay have exerted an inhibitory effect on wordtargets is suggested by the slower response toCHAIR following BLOO than following BLOM inthe 500 ms SOA condition. The absence of asimilar inhibitory effect in the 280 ms SOAcondition would suggest that the raising of theverification criterion is not instantaneous.

GENERAL DISCUSSIONA large body of evidence favoring the idea that

phonologic coding is the key process in printedword identification is available for Serbo-Croatian(see above and Lukatela & Turvey [1990a] for areview). The results of the experiments reportedin the present article, especially Experiments 3and 4, might be interpreted as adding to a grow­ing body of evidence that phonological coding maybe similarly dominant in printed word identifica­tion in English (Perfetti et al., 1988; Van Orden,

1987; Van Orden et al., 1988). In Experiments 1and 2 we showed that a word prime can facilitatethe naming of a nonword target if the nonwordhad the same phonology as an associate of theprime. In Experiments 3 and 4 we showed that anonword prime can facilitate naming of a wordtarget if its phonology is the same as that of anassociate of the target. The general consensus isthat associative priming-whether the mechanismbe spreading activation, expectancy set, or reso­nant matching (Stone & Van Orden, 1989}-is ale:.rical process. In order for the observed associa­tive effects to have occurred, there must have beenlexical access. And given the homophony condi­tions under which these effects occurred, it musthave been the case that the lexical representa­tions accessed were phonological, and that the ac­cess was through phonology. It seems unlikelythat an account of the data can be given in termsof a model that does not grant a central role tophonology in lexical access. Consider analogymodels (e.g., Glushko, 1979), for example.TAYBLE would be assigned a phonology by anal­ogy with its orthographic neighbors. To enhancethe naming of CHAIR, this assembled phonologi­cal representation /tablel would have to affect thelexical representation of CHAIR. The minimalimplication is that /tablel must function as a lexi­cal access code, triggering lexical processes benefi­cial to the eventual naming of CHAIR.

In the light of the evidence presented here andelsewhere for phonological mediation, we aretempted to ask whether or not there isexperimental support for another process separatefrom phonological mediation. A basic prediction ofdual process theory is that irregular/exceptionwords can only be processed by the direct accessroute because they cannot be coded by grapheme­phoneme correspondence rules. Van Orden et 81.(1990) have argued that the logic by whichirregular words constitute a priori evidence fordirect processing depends on the validity of theclaim that explicit, discrete rules map spellings tosounds. If the mapping is achieved in analternative fashion, for example, by a continuousversion of regularity (captured by the degree ofcovariation between orthographic and phonologicfeatures across neighborhoods of words that arespelled similarly), then the prediction is not valid.These same authors also question critically thearguments for an independent direct processbased on the phenomena of the word-superiorityeffect (McCusker et aI., 1981) and successfulreading by the deaf (McCusker et aI., 1981;Seidenberg, 1985). As they note, there is

222 Lukatel4 and Tumey

considerable experimental evidence contrary toboth arguments; phonological variables affect boththe word-superiority effect (e.g., Chastain, 1981,1984) and word identification by deaf readers (e.g.,Hanson & Fowler, 1987).

One experimental result offered in favor of aseparate direct access is the demonstration that,in a masking paradigm, phonologic priming failsto benefit target identification over and aboveorthographic priming (Evett & Humphreys, 1981).Thus, TILE primes FILE (phonologicallysimilar/graphemically similar) no more thanTOUCH primes COUCH (phonologically dissimi­lar/graphemically similar). The logic here is that ifone can demonstrate orthographic priming with­out phonological priming, then one has evidencefor an independent direct visual access. VanOrden et al. (1990) have suggested that the Evettand Humphrey's application of this logic is flawedin that the difference between the critical stimu­lus pairs was too small to produce a noticeable dif­ference in identification. Moreover, there are othersources of data that suggest that TOUCH can af­fect the processing of COUCH detrimentally be­cause of the phonological inconsistency betweenthe items (Meyer, Schvanaveldt, & Ruddy, 1974).Although Hillinger (1980) provided evidence forstrictly phonological priming in the lexical deci­sion task, the observation has failed to replicate infurther experiments with lexical decision (Martin& Jensen, 1988) and in experiments using naminglatency (Peter, Turvey, & Lukatela, 1990). Thesesame experiments, however, also failed to demon­strate any benefits to processing of a graphemi­cally similar preceding stimulus. This confusingstate of affairs with respect to forward primingexperiments with English language materials con­trasts sharply with the results from (a) similarforward priming experiments in Serbo-Croatian,and (b) backward priming experiments withEnglish and Serbo-Croatian, which are in contra­diction of an independent visual direct route.

With respect to (a), a large number of experi­ments demonstrate unequivocally that there is noorthographic priming in lexical decision or namingover and above phonological priming (Lukatela &Turvey, 1990a; Lukatela, Carello, & Turvey,1990). In these experiments, the primes and tar­gets consist of phonologically unique letter stringswritten either in different alphabets (e.g., prime isin Cyrillic, target is in Roman) and different cases,or in the same alphabet (both are in Cyrillic) andthe same case. When they are in the same casethe graphemically dissimilar/phonologically simi­lar pairs never share more than one letter in

common, and often they share none; the graphem­ically similar/phonologically similar pairs shareall but one letter in common. In different casesand alphabets, there are no visually identical let­ter forms. Especially significant aspects of the re­sults are that phonological similarity effects inboth lexical decision and naming are independentof graphemic similarity; that phonological simi­larity expedites the naming of words and non­words, and to the same degree; and that phonolog­ical similarity effect occurs even when the contextis a masked nonword. The latter result inparticular has special bearing on the issue of anindependent direct visual route.

By hypothesis, once a word unit has been acti­vated directly on the basis of its orthographicproperties, phonological information about theword is made available. Suppose that the directaccess is mandatory (automatic) and that thephonological mediation way is optional(nonautomatic), as has been frequently proposed.Then under masking conditions that minimizeoptional strategies, only the lexical way shouldwork. On the basis of the preceding argumentHumphreys, Evett, and Taylor (1982, p. 581) hy­pothesized that: "If phonological information is ac­tivated via a nonlexical route, a pseudohomophonepriming effect should occur." Finding that thepseudohomophone condition (tial-TILE) in theirmasking experiment did not differ from itsgraphemic control condition (tirl-TILE),Humphreys et al. (1982) concluded that the lexical(direct visual access) way of deriving phonology isthe only one of the two that is automatic and isprobably the only way to activate phonological in­formation. To the contrary, the successful demon­stration of "pseudohomophone priming" undermasking in Serbo-Croatian confirms theHumphreys et a1. (1982) hypothesis and suggests,therefore, that it is the phonological access routethat is automatic and (continuing their logic) isthe only one available.

Let us now consider (b), the backward primingexperiments. These entail the identification ofbriefly exposed target words under backwardmasking conditions with the following key fea­tures: The masks are nonwords, phonologically re­lated or unrelated to the targets, and themselvesfollowed by patterned stimuli to reduce theiridentification and, thereby, guessing strategiesabout target/mask relationships. The figuralstructure of the masks, and comparable intensi­ties of the targets and masks, confine the maskingeffects on the targets to primarily central pro­cesses (Michaels & Turvey, 1979; Turvey, 1973).

PhonologiaU Access of the Lexiron: Evidencefrom Associlltive Priming with Pseudohomophones 223

Perfetti et al. (1988) argued that if phonology iscomputed automatically, then phonological simi­larity between the mask and target will reducethe interruption of central processing normallyinduced by the mask. They reasoned that a phono­logically similar mask will reinforce the phonolog­ical information activated partially by the target.In contrast, a phonologically dissimilar mask willactivate partially other phonological information.If it is the case that lexical activation follows fromphonological information, then a target precedinga phonologically similar mask will be identifiedbetter than a target preceding a phonologicallydissimilar mask. The idea is that lexical entriespartially activated by a target will be activatedfurther by a subsequent mask with commonphonological properties. The outcomes of experi­ments by Naish (1980) and by Perfetti et a1.(1988), both using native speakers/readers ofEnglish and English language materials, were inagreement with this prediction, as were the exper­iments of Lukatela and Turvey (1990b) using na­tive speakers/readers of Serbo-Croatian andSerbo-Croatian materials. All of these experi­ments showed significantly higher levels of targetidentification for homophonous masking than fornonhomophonous masking. Put differently, noneshowed graphemic backward priming effects overand above phonological priming effects. With re­spect to Evett and Humphrey's (1981) logic, thisoutcome is contrary to that expected from an in­dependent direct visual access.

Van Orden et at (1990) have remarked thattraditionally the best evidence cited in favor of theindependent direct access hypothesis comes fromexperiments that did not yield predicted effects ofstimulus phonology-a strategy of arguing formother-than-positive effects. With respect to thesemantic category task, the direct accesshypothesis predicts that the effects of homophonyshould be reduced as a function of the familiarityof the homophones (direct access is the exclusiveroute for familiar words). Contrary to thisprediction, the effect of homophone familiaritywas null (Van Orden, 1987; Van Orden et aI.,1988). In the spirit of the other-than-positiveeffects strategy, this outcome should beinterpreted in favor of phonological mediation andagainst direct visual access (Van Orden, in press;Van Orden et al., 1990).

In sum, despite the ubiquity of the claim fordirect visual access the evidence is not strong,raising the possibility that the overarchinghypothesis of Coltheart's (1978) original theory oftwo independent processes could well be false. In

contrast, there is substantial and accumulatingevidence, including that reported in the presentarticle, for phonological mediation. We concludeby noting that, although discussions of differencesamong lexical access codes have usually beenexpressed in terms of activation times (e.g.,orthographic features are earlier sources ofconstraint on lexical coding than phonologicfeatures, or vice versa), other ways of expressingthe differences are possible. One interesting,contemporary idea from dynamical interpretationsof lexical access is that all codes are activated inparallel, but the phonological code is initially morecoherent (less noisy, closer to its "attractor") thanthe others and acts as a dynamic frame thatconstrains other linguistic encodings that areinitially less coherent (Van Orden et al., 1990). Itremains to be seen whether or not this newversion of the phonological mediation hypothesis,in conjunction with the new sources of data, candeflect the criticisms levied at earlier versions(e.g., Humphreys & Evett, 1985).

REFERENCESSalota, D. A., &r Chumbley, J. I. (1985). The locus of word­

frequency effect in the pronunciation task: Lexical accessand/or production7/ountill ofMemory Imd lAnguage, 24, 89-106.

Carello, C., Lukatela, G., &r Turvey, M. T. (1988). Rapid naming isaffected by association but not by syntax. Memory iii Cognition,16, 187-195.

Carrol, J. B., Davies, P., &r Richman, B. (1971). Word fmluem:y b<d.Boston: Houghton Mifflin.

Chastain, G. (1981). Phonological and orthographic factors in theword-superiority effect. Memory iii Cognition, 9, 389-397.

Chastain, G. (1984). Phonological access to the mental lexicon in atarget discrimination task. /ourntll ofGeneral Psychology, 110, S3­60.

Coltheart, M. (1978). Lexical access in simple reading task. In G.Underwood (Ed.), Strlltegies of informlltion processing. London:Academic Press.

Coltheart, M., Davelaar, E., Jonasson, J. T., &r Besner, D. (1977).Access to internal lexicon. In S. Darnic (Ed.), Attention lindperformslnce VI (pp. 535-555). New York: Academic Press.

Davelaar, E., Coltheart, M., Besner, D., &r Jonassen, J. T. (1978).Phonological receding and lexical access. Memory iii Cognition,6,391-402.

Evett, L. J., &r Humphreys, G. W. (1981). The use of abstractgraphemic information in lexical access. Quslrterly /ountill ofExperimentlll Psychology, 33, 325-350.

Feldman, L. B., Kostic, A., Lukatela, G., &r Turvey, M. T. (1983).An evaluation of the "Basic orthographic syllabic structure" ina phonologically shallow orthography. PsychologiCill Resetzrch,45,55 -72..

FeldInan, L. B., &r Turvey, M.T. (1983). Word recognition in Serb<>­Croatian is phonologically analytic. /01lf7lll1 of Experiments"Psychology: Hunum Perception lind Perfornumce, 9, 288-298.

Forster, K. I. &r Chambers, S. M. (1973). Lexical access and namingtime. /DUmIIl ofVerbill Lellming lind Verbill Behtltlior, 12, 627-635.

Fredriksen, J. R. &r Kroll, J. F. (1976). Spelling and sound:approaches to the internal lexicon. /oumlll of ExperimentlllPsychology: Hunum Perception lind Perftmnsmce, 2, 361-379.

224 Lu1aItela and Turrey

Glushko, R. (1979). 1lte organization and activation of ortho­graphic knowledge in reading aloud. JOllnllll of ExperimentalPsychology: Hllmlln Perception and Perftmrulnce, 5, 674-691.

Green, D. W. ck Shallice, T. (1976). Direct visual access in readingfor meaning. Memory and Cognition, 4, 753-758.

Henderson, L (1982). Orthognlplry and word m:ognition in muling.New York: Academic Press.

Hillinger, M. L. (1980). Priming effects with phonemica1ly similarwords: the encoding-bias hypothesis reconsidered. Memory iiiCognition, B, 115-123.

Humphreys, G. F., Evett. L. J. ck Taylor, D. Eo (1982). Automaticphonological priming in visual word recognition. Memory iiiCognition, 10, 57&-590-

Humphreys, G. W., ck Evett, L. J. (1985). Are there independentlexical and nonlexical routes in word processing? Anevaluation of the dual-route theory of reading. The Behaf1iorallind B,.in Sciences, B,689-74O.

Lukate1a, G., Popadit, D., Ognjenovil!, P., ck Turvey, M. T. (1980).Lexical decision in a phonologically shallow orthography.Memory iii Cognition, B, 124 - 132.

LukateIa. G., Savit, M., Gligorij~, B., Ognjenovit, P., ck Turvey,M. T. (1978). Bi- alphabetical lexical decision. ulngwtge andSpeech, 21, 142-163.

Lukatela, G., Careno, C., ck Turvey, M. T. (1987). Lexicalrepresentation of regular and irregular inflected nouns.Languageand Cogniti~ Proasses, 2, 1-17.

Lukatela, G., Careno, C., ck Turvey, M. T. (1990). Phonemicpriming by words and pseudowords. Ellropean JOIII7IIII ofCogniti~Psychology, 2, 375-394.

Lukate1a, G., ck Turvey, M. T. (1987). Loci of phonological effectsin recognizing words written in a shallow orthography.PsychologiCill Research, 49, 139-146-

LukateIa. G., ck Turvey, M T. (19908). Automatic and prelexicalcomputation of phonology in visual word identification.Ell1'Opt#n JouI7III1 ofCogniti~ Psychology, 2,325-344.

Lukatela. G., ck Turvey, M. T. (1~). Phonemic similarity effectsand prelexical phonology. Memory iii Cognition., IB, 128-152.

Lukatela, G., Feldman, L.B., Turvey, M. T., Care11o, C., ck Katz, L(1989). Context effects in bi-alphabetical word perception.JOUI7IIIl ofMemory and Language, 2B, 214-236.

LukateIa. G., Turvey, M. T., Feldman, L. B., CareUo, C., ck Katz, L.(1989). Alphabet priming in bi-alphabetical word perception.Journal ofMemory and Language, 2B, 237-254.

Martin, R. C. ck Jensen. C. R. (1988). Phonological priming in thelexical decision task: A failure to replicate. Memory iii Cognition,16, 505-521.

Maymer, M.S., ck Tresselt, M. Eo (1965). Tables of single letter anddigram'" frequency counts for various word length and letterposition combinations. Psychcmomic Monograph SlIpplements, 1,13-32

McCann, R. 5., ck Besner, D. (1987). Reading Pseudohomophones:Implication for models of pronunciation assembly and the lo­cus of word-frequency effects in naming. JOllTTIQI ofExperimentRlPsychology: Hllmlln Perception and PerformIInce, 13, 14-24.

McCusker, L. X., Hillinger, M. L., ck Bias, R. G. (1981).Phonological recoding and reading. Psychological BlIlldin,89,217-245.

Meyer, D. E., Schvaneveldt, R. W. ck Ruddy, M. G. (1974).Functions of graphemic and phonemic codes in visual wordrecognition. Memory iii Cognition, 2, 309-321.

Michaels, C. F., ck Turvey, M. T. (1979). Central sourc:es of visualmasking: Indexing the structures supporting seeing at a singlebrief glance. PsychologiCIII Research, 41, 1-61.

Naish, P. (1980). The effects of graphemic and phonemicsimilarity between targets and maslcs in a backward visualmaslQng paradigm. Quarterly joumII1 ofExperimentlll Psychology,32,57-68.

Paap, Ie. R., Newsome, S. L., McDonald. J. E., ck Schvaneveldt. R.W. (1982). An activation-verification model fro letter and wordrecognition: 1lte word superiority effec:t. PsychologiCill Reuiew,89, S73-S94.

Perfetti, C. A., Bell, L. C., ck Delaney, S. (1988). Automatic(prelexical) phonetic activation in silent word reading:Evidence from backward masking. Jovnurl·of Memcry lindLanguage, 2:7, 59-70.

Peter, M., Turvey, M. T., ck Lukatela. G. (1990). Phonologicalpriming: Failure to replicate in the rapid naming task. BlIllninof the Psychonomic Society, 28, 389-392.

Rubenstein, H., Lewis, S. 5., ck Rubenstein, M. A. (1971). Anevidence for phonemic reading in visual word recognition.JOUI7IIIl ofVerba1 Learning and Veri:¥ll BehlltIior, 10, 645-654.

Seidenberg, MS., Waters, G. S., Sanders, M. A., ck Langer, P.(19M). Pre- and posUexicalloci of contextual effects on wordrecognition. Memory iii Cognition, 12,315-388.

Stone, G. 0., ck Van Orden, G. C. (1989). Are words representedby nodes? Memory iii Cognition, 17,511-524-

Turvey, M. T. (1973). On peripheral and central processes invision: Inferences from masking with patterned stimuli.Psychological Review,80, I-52

Turvey, M. T., Feldman, L., ck Lukatela. G. (19M). 1lte Serbo­Croatian orthography constrains the reader to a phonologicallyanalytic strategy. In L. Henderson (Ed.), Orthognlphies andmuling (pp.81-89). Hillsdale, NJ: Erlbaum.

Van Orden. G. C. (1987). A ROWS is a ROSE: Spelling, sound andreading. Memory iii Cognition, 15, 181-198.

Van Orden, G. C., Johnston, J. C., ck Hale, B. L. (1988). Wordidentification in reading proceeds from spelling to sound tomeaning. JOIII7IIIl of Experimental Psychology: Learning, Memory,and Cognition, 14,371-385.

Van Orden, G. C. (in press). Phonologic mediation is fundamentalto reading. In D. Besner ck G. Humphreys (Eds.), Basicprocesses in reading: VislUll word recognition. Hillsdllle, NJ:Erlbaum.

Van Orden, G. C., Pennington. B. F., ck Stone, G. O. (1990). Wordidentification in reading and the promise of subsymbolicpsycholinguistics. Psychological Rerriew, 97, 488-522.

FOOTNOTES·Jo1I17III1 of Experimental Psychology: HllmAn Perception and

PerformIInce, in press.1"tJniversity of Belgrade, Belgrade, Yugoslavia.

ttAlso University of Connecticut, Storrs.

226 Lulaztela and Tum:y

SAD SADD YADD HAPPYSAIL SAYL THAYL BOATSAND SANDE MANDE BEACHSLEEP SLEAP CLEAP BEDSMOOTH SMUTHE SLUTHE ROUGHSONG SAWNG NAWNG MUSICSOUTH SOUTHE POUTHE NORTHSTAR STAHR SLAHR MOONSTORE STOAR SLOAR MARKETTAKE TAICK NAICK GIVETALL TAWL NAWL SHORTUGLY UGLIE IGLIE BEAUTIFULWEAK WIEK FIEK STRONGWOOL WULL TULL SHEEPWRONG RONG FONG RIGHT

228 Lulattela and Turvey

SERVE SURVE SARVE OBEYSHEEP SHEAP SHEMP HERDSHOE SHOOH SHORN SOLESLEEP SLEAP SLEMP DREAMSOAP SOPE SORP SUDSSOUP SUPE SUPL EATTEETH TEATH TERTH BITETILE TIAL TIRL FLOORTOMB TOOM TORM GRAVETRACKS TRAX TRAW TRAINTREE TREA TREM PINETWO TWOO TWOK ONEWEB WHEB WERB SPIDERWEIGHT WATE WRAT TONWHEAT WHEET WHEST FLOURWHITE WHYTE WHOTE CHALKWRITE WRYTE WRUTE READYOUNG YUNG YING OLDZERO ZEERO ZEBRO NONE

Phonologico1 Access of the laimn: Evidence from Associative Priming with Pseudohomophones 229

APPENDIX 3Context source words (SW), pseudohomophones (PH) with their summed bigram frequencies,pseudohomographs (PG) with their summed bigram frequencies, and the corresponding target wordsused in Experiments 3 and 4. Stimuli marked t did not appear in Experiment 3.

TargetContext

SW PH PG

ACHE AlKE,204 ARRE,232 HEADBAKE BAlK, 192 . BAWl{, 78 * CAKEBIBLE BYBLE,64 BOBLE,83 GODBITE BIGHT, 303 BIRET,147* TEETHBLADE BLAID,129 BLARD,199 KNIFEBLUE BLOO,24 BLOM,185 SKYBOAT BOTE, 129 BOTS,115 * SAILBRAVE BRAIV,139 BRARV,201 COWARDBREAD BREDD,127 BREND,191 BUTl'ERBREAK BRAIK, 139 BRACK, 187 GLASSBRIGHT BRITE, 179 BRITH,195 DULLtCHEAT CHEET,364 CHENT,367 HOAXCIDER SIDER, 332 MIDER,339 APPLECODE COAD,114 COlD, 128 SECRETCRATE CRAIT,120 CRAST,195 BOXDATE DAlT,166 DAST,347 FRIENDDEAF DEFF,51 DELF,170 MUTEDEAL DEEL,226 DERL, 193 * BARGAlNtDOOR DORE,429 DORN, 227 * KNOBDREAM DREEM,129 DRERM,244 SLEEPEARLY URLY,63 ORLY,63 LATEEAST EEST,255 ERST, 216 * WESTEAT EET,96 ERT,6 * DRINKtEIGHT AIT,55 DIT,77 NINEFAME FAIM,175 FADM, 59 * FORTUNE tFATE FAIT, 181 FANT,207 DESTINYtFLOOR FLORE, 311 FLOTR,141 * TILEFOUR FOAR,127 FOGR, 45 * FIVEFRUIT FRUTE,102 FRUST,129 PEARGLOBE GLOAB,93 GLOOB,121 WORLDtGLUE GWO,21 GLAM;37 STICKGONE GAWN,88 GERN,174 AWAYGOAT GOTE,121 GOOT,233 MILKGOLD GOALD,236 GOOLD, 233 JEWEL tGRADE GRAID,146 GRALD,178 SCHOOLGRAVE GRAIV,131 GRARV,193 TOMBGRIEF GREEF,151 GROEF, 114 * SORROWGROUP GRUPE, 57 GRUSP,77 PEOPLEHATE HAlT, 321 RANT, 347 LOVEHOPE HOAP,67 HORP,l77 DISPAIRHOUSE HOWSE, 217 HOLSE,234 HOMEJAIL JALE,92 JALL,232 PRISONtLAKE LAIK, 182 LASK, 160 PONDtLIGHT LITE, 400 LITH, 596 DARKMAIL MAYL,168 MAAL, 140 * LETTERt

230 Lukatela and Turvey

MAINMEALMOANMONTHMOSTNAVYNEARNEATNONEOATSOBEYONEPAINPIECEPHONERAINREADROADROOMSAFESAMESHADESIXSMOKESNOWSOAPSOLESOUPSTATESTEAKSTONESTOVETABLETAKETAMETAXTEACHTEASETHIEFTIGHTTOADTONTONETRAINTROOPTRUTHWARWADEWAITWHITEYOUR

MAYN,168MEEL,213MONE,254MUNTH,l08MOAST,268NAIVY,53NEER,298NEET,238NUNN,15OTES,46OBAY,30WUN,23PAYN,87PEECE,122FOAN,129RANE,l84WREED,215ROED,117RUME,242SAlF,235SAIM,237SHAID,150SIKS,107SMOAK, 55SNOE,64SOPE,145SOAL,137SUPE,73STAIT,223STAlK, 213STOAN,216STOAV,176TAYBLE,162TAlK, 165TAIM, 167TAKS,l06TEECH,190TEEZE,58THEEF,414TITE,323TODE,130TUNN,22TOAN,152TRANE,160TRUPE,61TROOTH,94WOAR,123WAlD,256WATE,138WHYTE, 237YURE,248

MARN,212MERL,l80*MOND,271MINTH,151MOYST,204*NARVY,75NEBR,37*NERT, 211 *NANO,113ONTS,l77OBLY,73WEN, 34PARN,131PRECE,164FORN,l88RAWN,90*WREND,l48*ROND,243REME,245SARF,189SARM, 199 *SHALD,282SILE,140SMONK, 131SNOP,74 .SOPH, 118 *SORL,240SURP,74STANT, 255STREK, 187 *STORN,214STORV,208TARBLE,201TARK, 140 *TARM, 129 *TAMS, 124TERCH, 159 *TERLE,133THREF, 391 *TITH,519TORD,225TANN, 106TOON,233TRANK, 167TRAPE,105TREETH,61 *WAYR, 112 *WAAD, 105 *WATH,334WHOTE,326YURM, 38 *

STREETFOODtGROANYEARtLEASTMARINE tFARTIDytZERORYESERVE

.TWOHURTPIENUMBERtMUDtWRITEHIGHWAYBEDtSECUREtOTHERtTREESEVENtFIREWINTERtBATHtSHOESALADtPOLICE tMEATtROCKtPIPECHAIRGIVEtWILDtMONEYLEARNANNOYSTEAL tLOOSEFROGWEIGHTSOUNDtTRACKSARMYLIESPEACEWATERtSTOPtBLACKMINE

*The summed bigram frequency of the pseudohomograph is lower than that of the correspondingpseudohomophone.