Upload
truongthu
View
214
Download
0
Embed Size (px)
Citation preview
Haskins Laboratories Status Report on Speech Research1993, SR-113, 153-170
Prelexical and Postlexical Strategies in Reading:Evidence from a Deep and a Shallow Orthography*
Ram Frostt
The validity of the orthographic depth hypothesis (ODH) was examined in Hebrew byemploying pointed (shallow) and unpointed (deep) print. Experiments 1 and 2 revealedlarger frequency effects, and larger semantic priming effects in naming, with unpointedprint than with pointed print. In Experiments 3 and 4 subjects were presented withHebrew consonantal strings that were followed by vowel marks appearing at stimulusonset asynchronies ranging from 0 ms (simultaneous presentation) to 300 ms from theonset of consonant presentation. Subjects were inclined to wait for the vowel marks toappear even though the words could be named uneqUIvocally using lexical phonology.These results suggest that prelexical phonology is the default strategy for readers inshallow orthographies, providing strong support for the ODH.
Most early studies of visual word recognitionwere carried out in the English language. Thisstate of affairs was partly due to an underlyingbelief that reading processes (as well as othercognitive processes) are universal, and thereforestudies in English are sufficient to provide a complete account of the processes involved in recognizing printed words. In the last decade, however,studies in orthographies other than English havebecome more and more prevalent. These studieshave in common the view that reading processescannot be explained without considering the reader's linguistic environment in general, and thecharacteristics of his writing system in particular.
Various writing system have evolved over timein different cultures. These writing systems,whether logographic, syllabic, or alphabetic,typically reflect the language's unique phonologyand morphology, representing them in an effective
This work was supported in part by a grant awarded to theauthor by the Basic Research Foundation administered by theIsrael Academy of Science and Humanities, and in part by theNational Institute of Child Health and Human DevelopmentGrant HD-01994 to Haskins Laboratories. I am indebted toLen Katz for many of the ideas that were put forward in thisarticle, to Oma Moshel and Itchak Mendelbaum for their helpin conducting the experiments, and to Ken Pugh, Bruno Repp,Charles Perfetti, Albrecht InhofT, and Sandy Pollatsek for theircomments on earlier drafts of this paper.
153
way (Mattingly, 1992; Scheerer, 1986). The matchbetween writing system and language insures adegree of efficiency for the reading and the writingprocess (for a discussion, see Katz & Frost, 1992).Theories of visual recognition differ in theiraccount of how the different characteristics ofwriting systems affect reading performance. Onesuch characteristic that has been widelyinvestigated is the way the orthographyrepresents the language's surface phonology.
The transparency of the relation betweenspelling and phonology varies widely between orthographies. This variance can often be attributedto morphological factors. In some languages (e.g.,in English), morphological variations are capturedby phonologic variations. The orthography, however, was designed to preserve primarilymorphologic information. Consequently, in manycases, similar spellings denote the samemorpheme but different phonologic forms: Thesame letter can represent different phonemeswhen it is in different contexts, and the samephoneme can be represented by different letters.The words "steal" and "stealth," for example, aresimilarly spelled because they are morphologicallyrelated. Since in this case, however, a morphologicderivation resulted in a phonologic variation, thecluster "ea" represents both the sounds [i) and [E].Thus, alphabetic orthographies can be classified
154 Frost
according to the transparency of their letter tophonology correspondence. This factor is usuallyreferred to as "orthographic depth" (Katz &Feldman, 1981; Klima, 1972; Liberman,Liberman, Mattingly, & Shankweiler, 1980;Lukatela, PopadiC, Ognjenovic, & Turvey, 1980).An orthography that represents its phonologyunequivocally following grapheme-phonemesimple correspondences is considered shallow,while in a deep orthography the relation oforthography to phonology is more opaque.
The effect of orthographic depth on readingstrategies has been the focus of recent and currentcontroversies (e.g., Besner & Smith, 1992; Frost,Katz, & Bentin, 1987). In general the argumentrevolves around the question of whetherdifferences in orthographic depth lead todifferences in processing printed words. What iscalled the orthographic depth hypothesis (ODH)suggests that it does. The ODH suggests thatshallow orthographies can easily support a wordrecognition process that involves the printedword's phonology. This is because the phonologicstructure of the printed word can be easilyrecovered from the print by applying a simpleprocess of grapheme-to-phoneme conversion(GPC). For example, in the Serbo-Croatian writingsystem the letter-to-phoneme correspondence isconsistent and the spoken language itself is notphonologically complex. The correspondencebetween spelling and pronunciation is so simpleand direct that a reader of this orthography canexpect that phonological recoding will alwaysresult in an accurate representation of the wordintended by the writer. A considerable amount ofevidence now supports the claim that phonologicalrecoding is extensively used by readers of SerboCroatian (see, Carello, Turvey, & Lukatela, 1992,for a review). In contrast to shallow orthographies,deep orthographies like English or Hebrewencourage readers to process printed words byreferring to their morphology via the printedword's visual-orthographic structure. In deeporthographies lexical access is based mainly onthe word's orthographic structure, and the word'sphonology is retrieved from the mental lexicon.This is because the relation between the printedword and its phonology are more opaque andprelexical phonologic information cannot be easilygenerated (e.g., Frost et al. 1987; Frost & Bentin,1992b; Katz & Feldman, 1983).
The ODH's specific predictions mainly refer tothe way a printed word's phonology is generatedin the reading process. Because readers of shalloworthographies have simple, consistent, and
relatively complete connections betweengraphemes and sub-word pronunciation, they canrecover most of a word's phonological structureprelexically, by assembling it directly from theprinted letters. In contrast, the opaque relation ofsubwords segments and phonemes in deeporthographies prevents readers from usingprelexical conversion rules. For these readers, themore efficient process of generating the word'sphonologic structure is to rely on a fast visualaccess of the lexicon and to retrieve the word'sphonology from it. Thus, phonology in this case islexically addressed, not prelexically assembled.Thus, according to the ODH, the differencebetween deep and shallow orthographies is theamount of lexical involvement in pronunciation.This does not necessarily entail specificpredictions concerning lexical decisions and lexicalaccess, or how meaning is accessed from print.
The specific predictions of the ODH must bediscussed with reference to the tools ofinvestigation employed by the experimenter. Theissue to be clarified here is what serves as a validdemonstration that phonology is mainly prelexical(assembled) or postlexical (addressed). In general,measures of latencies and error rates for lexicaldecisions or naming are monitored. The idea isthat lexical involvement in pronunciation leavescharacteristic traces. The first question to beexamined is, therefore, whether the lexical statusof a word affects naming latencies. Lexical searchresults in frequency effects and in lexicalityeffects: Frequent words are named faster thannonfrequent words, and words are named fasterthan nonwords (but see Balota & Chumbley, 1984,for a discussion of this point). Thus, one trace ofpostlexical phonology is that naming latencies andlexical decision latencies are similarly effected bythe lexical status of the printed stimulus (e.g.,Katz & Feldman, 1983). If phonology is assembledfrom print prelexically, smaller effects related tothe word's lexical status should be expected, andconsequently lexical status should affect namingand lexical decisions differently. A second methodof investigation involves the monitoring ofsemantic priming effects in naming (see Lupker,1984; Neely, 1991, for a review). If pronunciationinvolves postlexical phonology, strong semanticpriming effects will be revealed in naming. Incontrast, if pronunciation is carried out usingmainly prelexical phonology, naming of targetwords would be less facilitated by semanticallyrelated primes.
Keeping the above tools of investigation in mind,the exact predictions of the ODH can be now fully
Prelexical and Postlexical Strategies in Reading: Evidence from a Deep and a Shallow Orthography 155
described. Two versions of the hypothesis exist inthe current literature. What can be called thestrong ODH claims that in shallow orthographies,the complete phonological representations can bederived exclusively from only the prelexicaltranslation of subword spelling units (letter or letter clusters) into phonological units (phonemes,phoneme clusters, and syllables). According to thisview, readers of shallow orthographies perform aphonological analysis of the word based only on aknowledge of these correspondences. Rapidnaming, then, is a result of this analytic processonly, and does not involve any lexical information(see Katz & Frost, 1992, for a discussion). Incontrast to the strong ODH, a weaker version ofthe ODH can be proposed. According to the weakversion of the ODH, the phonology needed for thepronunciation of printed words comes both fromprelexical letter-phonology correspondences andfrom stored lexical phonology. The latter is theresult of either an orthographic addressing of thelexicon (i.e., from a whole-word or wholemorpheme spelling pattern to its storedphonology), or from a partial phonologic representation that was assembled from the print and wasunequivocal enough to allow lexical access. Thedegree to which the prelexical process is active is afunction of the depth of the orthography; prelexical analytic processes are more functional inshallow orthographies. Whether or not prelexicalprocesses actually dominate orthographic processing for any particular orthography is a question ofthe demands the two processes make on the reader's processing resources (Katz & Frost, 1992).
It is easy to show that the strong form of theODH is untenable. It is patently insufficient toaccount for pronunciation even in shalloworthographies like Spanish, Italian, or SerboCroatian. This is so because these orthographiesdo not represent syllable stress and, even thoughstress is often predictable, this is not always thecase. For example, in Serbo-Croatian, stress fortwo-syllable words always occurs on the firstsyllable, but not always for words of more thantwo syllables. These words can be pronouncedcorrectly only by reference to lexically storedinformation. The issue of stress assignment iseven more problematic in Italian, where stresspatterns are much less predictable. In Italian,many pairs of words differ only in stress whichprovides the intended semantic meaning (Colombo& Tabossi, 1992; Laudanna & Caramazza, 1992).
Several studies have argued for the obligatoryinvolvement of prelexical phonology in SerboCroatian (e.g., Turvey, Feldman, & Lukatela,
1984) or in English (Perfetti, Bell, & Delaney,1988; Perfetti, Zhang, & Berent, 1992; Van Orden,198'1). Note than the weaker version of the ODH isnot inconsistent with these claims as long as it isnot claimed that naming is achieved exclusively byprelexical analysis. All alphabetic orthographiesmay make some use of prelexically derivedphonology for word recognition. According to theweak form of the ODH, shallow orthographiesshould make more use of it than deep orthographies, because prelexical phonology is more readily available in these orthographies. Substantialprelexical phonology may be generated inevitablywhen reading Serbo-Croatian, for example. Thisphonological representation may be sufficient forlexical access, but not necessarily for pronunciation. The complete analysis of the word's phonologic and phonetic structure may involve lexicalinformation as well (Carello et aI., 1992).
Evidence concerning the validity of the ODHcomes from within- and cross-language studies.But note that single-language experiments areadequate only for testing the strongest form of theODH. The strong ODH requires lexical access andpronunciation to be accomplished entirely on thebasis of phonological information derived fromcorrespondences between subword spelling andphonology. That is, the reader's phonologicalanalysis of the printed word, based on his or herknowledge of subword letter-sound relationships,is the only kind of information that is allowed.Thus, merely showing lexical involvement in pronunciation in a shallow orthography would provide a valid falsification of the strong ODH(Seidenberg & Vidanovic, 1985; Sebastian-Galles,1991). However, the demonstration of lexical effects in shallow orthographies cannot in itself beconsidered evidence against the weak ODH, and adifferent methodology is necessary to test it.Because it refers to the relative use of prelexicalphonologic information in word recognition in different orthographies, experimental evidence bearing on the weak ODH should come mainly fromcross-language studies. More important, becausethe weak ODH claims that readers in shallow orthographies do not use the assembled routine exclusively, the evidence against it cannot comefrom single-language studies by merely showingthe use of the addressed routine in shallow orthographies (e.g., Sebastian-Galles, 1991).
The experimental evidence supporting the weakODH is abundant. Katz and Feldman (1983)compared semantic priming effects in naming andlexical decision in English and Serbo-Croatian,and demonstrated that while semantic facilitation
156 Frost
was obtained in English for both lexical decisionand naming, in Serbo-Croatian semantic primingfacilitated only lexical decision. Similarly, acomparison of semantic priming effects in namingin English and Italian showed greater effects inthe deeper English than in the shallower Italianorthography (Tabossi & Laghi, 1992). A study byFrost et al. (1987) involved a simultaneouscomparison of three languages, Hebrew, English,and Serbo-Croatian, and confirmed the hypothesisthat the use of prelexical phonology in namingvaries as a function of orthographic depth. Frostet al. showed that the lexical status of thestimulus (its being a high- or a low-frequencyword or a nonword) affected naming latencies inHebrew more than in English, and in Englishmore than in Serbo-Croatian. In a secondexperiment, Frost et al. showed a relatively strongeffect of semantic facilitation in Hebrew (21 ms), asmaller but significant effect in English (16 ms),and no facilitation (0 ms) in Serbo-Croatian.
Frost and Katz (1989) examined the effects ofvisual and auditory degradation on the ability ofsubjects to match printed to spoken words inEnglish and Serbo-Croatian. They showed thatboth visual and auditory degradation had a muchstronger effect in English than in Serbo-Croatian,regardless of word frequency. These results wereexplained by an extension of an interactive modelwhich rationalized the relationship between theorthographic and phonologic systems in terms oflateral connections between the systems at all oftheir levels. The structure of these lateral connections was determined by the relationship betweenspelling and phonology in the language: simpleisomorphic connections between graphemes andphonemes in the shallower Serbo-Croatian, butmore complex, many-to-one, connections in thedeeper English. Frost and Katz argued that thesimple isomorphic connections between the orthographic and the phonologic systems in the shallower orthography enabled subjects to restore boththe degraded phonemes from the print and thedegraded graphemes from the phonemic information, with ease. In contrast, in the deeper orthography, because the degraded information in onesystem was usually consistent with several alternatives in the other system, the buildup of sufficient information for a unique solution to thematching judgment was delayed, so the matchingbetween print and degraded speech, or betweenspeech and degraded print, was slowed.
The psychological reality of orthographic depthis not unanimously accepted. Although it is generally agreed that the relation between spelling and
phonology in different orthographies might affectreading processes (especially reading acquisition)to a certain extent, there is disagreement as to therelative importance of this factor. What I will callhere "the alternative view" argues that the primary factor determining whether or not theword's phonology is assembled pre- or post-lexically is not orthographic depth but word frequency. The alternative view suggests that in anyorthography, frequent words are very familiar asvisual patterns. Therefore, these words can easilybe recognized through a fast visually-based lexicalaccess which occurs before a phonologic representation has time to be generated prelexically fromthe print. For these words, phonologic informationis eventually obtained, but only postlexically, frommemory storage. According to this view, the relation of spelling to phonology should not affectrecognition of frequent words. Since the orthographic structure is not converted into a phonologic structure by use of grapheme-to-phonemeconversion rules, the depth of the orthographydoes not play a role in the processing of thesewords. Orthographic depth exerts some influence,but only on the processing of low-frequency wordsand nonwords. Since such verbal stimuli are lessfamiliar, their visual lexical access is slower, andtheir phonology has enough time to be generatedprelexically (Baluch & Besner, 1991; Tabossi &Laghi, 1992; Seidenberg, 1985).
A few studies involving cross-language researchsupport the alternative view. Seidenberg (1985)demonstrated that in both English and Chinese,naming frequent printed words was not effectedby phonologic regularity. This outcome was interpreted to mean that, in logographic as in alphabetic orthographies, the phonology of frequentwords was derived postlexically, after the wordhad been recognized on a visual basis. Moreover,in another study, Seidenberg and Vidanovic (1985)found similar semantic priming effects in namingfrequent words in English and Serbo-Croatian,suggesting again that the addressed routine playsa major role even in the shallow Serbo-Croatian.
Several studies in Japanese were interpreted asproviding support for the alternative view.Although these studies were carried out withinone language, they could in principle furnish evidence in favor or against the weak ODH becausethey examined reading performance in two different writing systems that are commonly used inJapanese- the deep logographic Kanji and theshallower syllabic Hiragana and Katakana.However, the relevance of these studies to the debate concerning the weak ODH is questionable, as
Pre/exicaI and PostlexicaI Strategies in Reading: Evidence (rom a Deep and a Shallow Orthography 157
I will point out below. Besner and Hildebrant(1987) showed that words that were normallywritten in Katakana were named faster thanwords written in Katakana that were transcribedfrom Kanji. They argued that these results suggest that readers of the shallow Katakana did notname the transcribed words using the assembledroutine, as otherwise no difference should haveemerged in naming the two types of stimuli. Inanother experiment, Besner, Patterson, Lee, andHildebrant (1992) showed that naming words thatare normally seen in Katakana Were namedslower if they were written in Hiragana and viceversa. This outcome suggests that orthographicfamiliarity plays a role even in reading the shallower syllabic Japanese orthographies. Taken together, the results from Japanese provide strongevidence against the strong ODH, but they do notcontradict the weak ODH. Although these studiescompare reading performance in two writing systems, they merely show that readers of the shallower Japanese syllabary do not use the assembled routine exclusively, but use the addressedroutine as well. These conclusions, however, areactually the basic tenets of the weak ODH.
More damaging to the weak ODH is a recentstudy by Baluch and Besner (1991), who employeda within-language between orthography design inPersian, similar to the one in Japanese. In thisstudy the authors took advantage of the fact thatsome words in Persian are phonologically transparent whereas other words are phonologicallyopaque. This is because three of the six vowels ofwritten Persian are represented in print as diacritics and three are represented as letters.Because (as in Hebrew) fluent readers do not usethe pointed script, words that contain vowels represented by letters are phonologically transparent,whereas words that contain vowels represented bydiacritics are phonologically opaque. Baluch andBesner demonstrated similar semantic primingeffects in naming phonologically transparent andphonologically opaque words of Persian, providedthat nonwords were omitted from the stimuluslist. These results were interpreted to suggest thatthe addressed routine was used in naming bothtypes of words.
Thus, when the weak ODH and the traditionalalternative view are contrasted, it appears thatthey differ in one major claim concerning thepreferred route of the cognitive system forobtaining the printed word's phonology. Thealternative view suggests that in general visualaccess of the lexicon is "direct" and requires fewercognitive resources. Moreover, the extraction of
phonological information from the mental lexiconis more or less effortless relative to the extractionof phonological information prelexically. Hence,the default of the cognitive system in reading isthe use of addressed phonology. The weak ODHdenies that the extraction of phonologicalinformation from the mental lexicon is effortless.On the contrary, based on findings showingextensive phonologic recoding in shalloworthographies, its working hypothesis is that if thereader can successfully employ prelexicalphonological information, then it will be used first;the easier it is, the more often it will be used.Thus, the "default" of the cognitive system in wordrecognition is the use of prelexical rather thanaddressed phonology. If the reader's orthographyis a shallow one with uncomplicated, direct, andconsistent correspondences between letters andphonology, then the reader will be able to use suchinformation with minimal resources for lexicalaccess. The logic for the ODH lies in a simpleargument. The so called "fast," "effortless," visualroute is available for readers in all orthographies,including the shallow ones. If in spite of that itcan be demonstrated that readers of shalloworthographies do prefer prelexical phonologicalmediation over visual access, it is because it ismore efficient and faster for these readers.
The present study addresses this controversy bylooking at a deeper orthography, namely Hebrew.In Hebrew, letters represent mostly consonants,while vowels can optionally be superimposed onthe consonants as diacritical marks. The diacritical marks, however, are omitted from most reading material, and are found only in poetry, children's literature, and religious texts. In addition,like other Semitic languages, Hebrew is based onword families derived from triconsonantal roots.Therefore, many words (whether morphologicallyrelated or not) share a similar or identical letterconfiguration. If the vowel marks are absent, asingle printed consonantal string usually represents several different spoken words. Thus, in itsunpointed form, the Hebrew orthography does notconvey to the reader the full phonemic structure ofthe printed word, and the reader is often facedwith phonological ambiguity. In contrast to theunpointed orthography, the pointed orthographyis a very shallow writing system. The vowel marksconvey the missing phonemic information, makingthe printed word phonemically unequivocal.Although the diacritical marks carry mainly vowelinformation, they also differentiate in some instances between fricative and stop variants of consonants. Thus the presentation of vowels consid-
158 Frost
Figure 1. Example of the unpointed and the pointedforms of a Hebrew printed word.
There were four experimental conditions: thestimuli could be presented pointed or unpointed,for naming or for lexical decision. Forty differentsubjects were tested in each experimental
Experiment 1 was to investigate whether a similargradual deviation of naming from lexical decisionperformance would be observed in pointed relativeto unpointed Hebrew. If the weak ODH is correct,then the reader of Hebrew should use the assembled routine to a greater extent when namingpointed print than when naming unpointed print.
MethodSubjects. One hundred and sixty undergraduate
students from the Hebrew University, all nativespeakers of Hebrew, participated in theexperiment for course credit or for payment.
Stimuli and Design. The stimuli were 40 highfrequency words, 40 low-frequency words, and 80nonwords. All stimuli were three to five letterslong, and contained two syllables with four to sixphonemes. The average number of letters andphonemes were similar for the three types ofstimuli. All words could be pronounced as only onemeaningful word. Nonwords were created byaltering randomly one or two letters of high- orlow-frequency real words that were not employedin the experiment. The nonwords were allpronounceable and did not violate the phonotacticrules of Hebrew. In the absence of a reliablefrequency count in Hebrew, the subjectivefrequency of each word was estimated using thefollowing procedure: A list of 250 words waspresented to 50 undergraduate students, whorated the frequency of each word on a 7-pointscale from very infrequent (1) to very frequent (7).The rated frequencies were averaged across all 50judges, and all words in the present study wereselected from this pool. The average frequency ofthe high-frequency words was 4.0, whereas theaverage frequency of the low-frequency words was2.0. Examples of pointed and unpointed Hebrewwords are presented in Figure 1.
Pointed
Piano
I psanter I
Unpointed
Hebrew print
Phonologic structure
Semantic meaning
erably reduces several aspects of phonemic ambiguity (see Frost & Bentin, 1992b, for a discussion).
Several studies have shown that lexicaldecisions to Hebrew phonologically ambiguouswords are given prior to any phonologicaldisambiguation suggesting that lexical access inunpointed Hebrew is accomplished more often viaorthographic representations than viaphonological recodings of the orthography (Bentin& Frost, 1987; Bentin, Bargai, & Katz, 1984; Frost& Bentin, 1992a; Frost, 1992; and see Frost &Bentin, 1992b, for a review). The purpose of thepresent study was to show that even the reader ofHebrew who is exposed almost exclusively tounpointed print and generally uses the lexicaladdressed routine, prefers to use the prelexicalassembled routine when possible. Note that theprobability of extending the cross-languagefindings (e.g., Frost et aI., 1987) to a withinHebrew experimental design suffers from the factthat readers in general are used to regularlyemploying strategies of reading that arise fromthe characteristics of their orthography.Obviously, if it can be shown that even in spite ofthis factor Hebrew readers are willing to altertheir reading strategy and adopt a prelexicalroutine, it will certainly provide very significantevidence in favor of the ODH.
EXPERIMENT 1In Experiment 1 lexical decision and naming
performance with pointed and unpointed printwere compared. The aim of the experiment was toassess the effects of lexical factors on naming performance relative to lexical decision in the deepand shallow forms of the Hebrew orthography.This experimental design is very similar to thatemployed by Frost et al. (1987) in the first experiment of their multilingual study. Frost et aI.found that in unpointed Hebrew the lexical statusof the stimuli affected lexical decision latencies inthe same way that it affected naming latencies. Incontrast, in English and in Serbo-Croatian thepronunciation task was much less affected by thelexical status of the stimuli. The shallower the orthography, the more naming performance deviated from lexical decision performance. This outcome confirmed that, in unpointed Hebrew,phonology was generated mainly using the addressed routine, whereas in the shallower Englishorthography the assembled routine had a moresignificant role. Finally, in the shallowest orthography, Serbo-Croatian, the assembled routinedominated the addressed routine, resulting in anonsignificant frequency effect. The purpose of
Prelexical and Postlexical Strategies in Reading: Evidence from a Deep and a Shallow Orthography 159
condition. This blocked design was identical tothat of the Frost et al. (1987) study.
Procedure and apparatus. The stimuli were presented on a Macintosh SE computer screen, and abold Hebrew font, size 24, was used. Subjects weretested individually in a dimly lighted room. Theysat 70 em from the screen so that the stimuli subtended a horizontal visual angle of 4 degrees onthe average. The subjects communicated lexicaldecisions by pressing a "yes" or a "no" key. Thedominant hand was always used for the "yes" response. In the naming task, response latencieswere monitored by a Mura-DX 118 microphoneconnected to a voice key. Each experiment startedwith 16 practice trials, which were followed by the160 experimental trials presented in one block.The trials were presented at a 2.5 sec intertrialinterval.
ResultsMeans and standard deviations of RTs for
correct responses were calculated for each subjectin each of the four experimental condition!').Within each subject/condition combination, RTsthat were outside a range of 2 SDs from therespective mean were excluded, and the mean wasrecalculated. Outliers accounted for less than 5%of all responses. This procedure was repeated inall the experiments of the present study.
Mean RTs and error rates for high-frequencywords, low-frequency words and nonwords in thedifferent experimental conditions are presented inTable 1.1 Point presentation had very little effectin the lexical decision task. In contrast, in thenaming task the effect of frequency and lexicalstatus of the stimulus were more pronounced inthe unpointed presentation than in the pointedpresentation. Moreover, naming was more similarto lexical decision performance in unpointed thanin pointed print.
The statistical significance of these differenceswas assessed by an analysis of variance (ANOVA)
across subjects (Fl) and across stimuli (F2), withthe main factors of stimulus type (high-frequencywords, low-frequency words, nonwords), pointpresentation (pointed, unpointed), and task(naming, lexical decision). The main effect ofstimulus type was significant (Fl(2,312) = 267.0,MSe =1550, p<O.OOI, F2(2,157) = 119.0, MSe =4705, p<O.OOl). The main effects of point presentation and task were not significant in the subjectanalysis (Fl(I,156) = 1.7, MSe =21,506, p<0.2;Fl(l,156) = 2.1, MSe = 21,506, p<0.2, respectively), but were significant in the stimulus analysis (F2(1,157) = 63.7, MS e = 3258, p<O.OOI;F2(1,157) = 23.1, MSe = 3258, p<O.OOI, respectively). Point presentation interacted with task(F1(1,156) = 4.3, MSe = 21,506, p<0.04; F2(1,157)= 97.9, MSe = 1877, p<O.OOl). Stimulus type interacted both with point presentation and withtask (Fl(2,312) = 3.0, M S e = 1550, p<0.05;F2(2,157) = 11.9, MSe = 832, p<O.OOI; Fl(2,312) =30.0, MSe = 1550, p<O.OOI; F2(2,157)=17.4, MSe =3258, p<O.OOl, respectively). The three-way interaction did not reach significance. This wasprobably due to a similar slowing of nonword latencies relatively to the low-frequency words inthe two prints. The difference between namingpointed and unpointed presentation was mostconspicuous while examining the effect of wordfrequency. A Tukey-A post-hoc analysis (p<0.05)revealed that while there was a significantfrequency effect in naming unpointed print (28ms), there was no significant frequency effect innaming pointed print (9 ms). Finally, thecorrelation of RTs in the naming and the lexicaldecision task was calculated for both types of printpresentation. RTs in the two tasks were highlycorrelated in the unpointed presentation (r=0.56),but much less so in the pointed presentation(r=0.28). This suggests that the lexical status ofthe stimuli affected lexical decision and namingsimilarly in unpointed print but not in pointedprint.
Table 1. RTs and percent errors in the lexical decision and naming tasks the for high-frequency words. low-frequencywords. and nonwords in unpainted and pointed print.
UNPOINTED PRINT POINTED PRINT
High-freg. Low-freg. Nonwords High-freg. Low-freg. Nonwords
LEXICALDECISION 529 617 659 545 627 664
4% 2% 5% 5% 5% 5%
NAMING 569 597 664 541 550 6040%. 6% 2% 0% 2% 9%
160 Frost
Discussion
The results of Experiment 1 replicate to a certain extent the findings of Frost et al. (1987), witha within-language, between-orthography design.The pattern of naming and lexical dec~sion ~aten
cies in unpointed Hebrew was almost IdentIcal tothe pattern obtained in the Frost et a1. study.However the relations of lexical decision to naming late~cies in pointed Hebrew were similar tothose obtained by Frost et al. in the shallower orthographies of English and Serbo-Croatian. ~hepatterns of response times in naming and le~cal
decisions were fairy similar in unpointed prmt.This suggests a strong reliance on the addressedroutine in naming when the vowel marks were notpresent. In contrast, naming deviated from thepattern obtained in the lexical decision task whenthe vowel marks were presented. The frequencyeffect almost disappeared, and the overall difference between naming high-frequency words andnonwords was considerably reduced. This outcomesuggests that the presentation of vowel marks encouraged the readers to adopt a strategy of .assembling the printed word phonology by usmgprelexical conversion rules. Note, however, thatalthough the overall difference between high-frequency words and nonwords was smaller inpointed print than in unpointed print, the difference between low-frequency words and nonwordswere fairly similar in the two conditions (54 inpointed print vs. 67 in unpointed print). This outcome is not fully consistent with a prelexical computation of printed stimuli in the shallow pointedprint. A possible explanation for the unexpect~d
slow naming latencies for pointed nonwords IS
that although phonology could be easily computedprelexically in pointed print, subjects were alsosensitive to the familiarity of the printed stimuli.Readers in Hebrew read mostly unpointed printbut have a large experience in reading pointedprint as well. Because they are obviously morefamiliar with reading words than nonwords, theslower latencies for nonwords could be attributedto a response factors rather than factor related tothe computation of phonology.
EXPERIMENT 2The aim of Experiment 2 was to examine the
effects of semantic facilitation in the naming taskwhen pointed and unpointed words are presented.Semantic priming effects in the naming task havebeen used in several studies to monitor the extentof lexical involvement in pronunciation (e.g.,Baluch & Besner, 1991; Frost et a1. 1987; Tabossi
& Laghi, 1992). In general, it has been shown thatin languages with deep orthographies such asEnglish, semantic facilitation in naming caneasily be obtained, although the effects areusually smaller than the effect obtained in thelexical decision task (Lupker, 1984; Neely, 1991).In contrast, in shallow orthographies such asSerbo-Croatian, semantic priming effects were notobtained in some studies (e.g., Katz & Feldman,1983; Frost et a1., 1987), but have been shown inother studies (e.g., Carello, Lukatela, & Turvey,1988; Lukatela, Feldman, Turvey, Carello, &Katz, 1989).
An examination of the weak version of the ODHentails, however, a comparison of semantic priming effects in a deep and a shallow orthography.Katz and Feldman (1983) showed greater semantic facilitation in English than in Serbo-Croatian.Similarly, Frost et a1. (1987) demonstrated agradual decrease in semantic facilitation fromHebrew (deepest orthography) to English(shallower) to Serbo-Croatian (shallowest).Similar results were suggested by Tabossi andLaghi (1992) who compared English to Italian.Recently, Baluch, and Besner (1991) have challenged these findings, suggesting that all thosecross-language differences were obtained becausenonwords were included in the stimulus lists.According to their proposal the inclusion of nonwords encouraged the use of a prelexical namingstrategy in the shallower orthographies. Indeed,when nonwords were not included in the stimuluslists, no differences in semantic facilitation werefound in naming phonologically opaque andphonologically transparent words in Persian.
Experiment 2 was designed to examine theweak version of the ODH using a semanticpriming paradigm, while considering thehypothesis that the effects of orthographic depthare caused by the mere inclusion of nonwords inthe stimuli lists. For this purpose semanticfacilitation in naming target words was examinedin pointed and unpointed Hebrew orthography,when only words were employed.
Methods
Subjects. Ninety-six undergraduate studentsfrom the Hebrew University, all native speakers ofHebrew, participated in the experiment for coursecredit or for payment.
Stimuli and design. The stimuli were 48 targetwords that were paired with semantically relatedand semantically unrelated primes. Relatedtargets and primes were two instances of asemantic category. In order to avoid repetition
Prelexical and Postlexical Strategies in Reading: Evidence [rom a Deep and a Shallow Orthography 161
effects, two lists of stimuli were constructed.Targets presented with semantically relatedprimes in one list were unrelated in the other list,and vice versa. Each subject was tested in one listonly. There were two experimental conditions: inone condition all stimuli were pointed, and in theother they were unpointed. Forty-eight subjectwere tested in each condition, 24 on each list. Thetargets were three to five letter words and hadtwo or three syllables. Both primes and targetswere unambiguous, and each could be read as ameaningful word in only one way.
Procedure and apparatus. An experimentalsession consisted of 16 practice trials followed bythe 48 test trials. Each trial consisted of apresentation of the prime for 750 ms followed bythe presentation of the target. Subject wereinstructed to read the primes silently and to namethe targets aloud. The exposure of the target wasterminated by the subject's vocal response, andthe intertrial interval was 2 seconds. Theapparatus was identical to the one used in thenaming task of Experiment 1.
Results
RTs in the different experimental conditions arepresented in Table 2. Naming of related targetswas faster than naming of unrelated targets inboth pointed and unpointed print. However,semantic facilitation was twice as large withunpointed print than with pointed print.
Table 2. Reaction times and percent errors to relatedand unrelated targets with pointed and unpainted print.
Unpainted PointedPrint Print
Unrelated 53\ 5\2(1%) (2%)
Related 509 503(4%) (1%)
PrimingEffect 22 9
The statistical significance of these effects wasassessed by an ANOVA across subjects (Fl) andacross stimuli (F2), with the main factors of semantic relatedness (related, unrelated), and printtype (pointed, unpointed). The effect of semanticrelatedness was significant (Fl(1,47) = 20.9, MSe= 656, p< 0.001; F2(1,94) = 33.2, MS e = 363,p<O.OOl). The effect of print type was significant
in the subject analysis (Fl(1,47) = 30.8, MSe =232, p<O.OOl), but not in the stimulus analysis(F2(1,94) = 1.0). More important to our hypothesis,the two-way interaction was significant in bothanalyses (F1(1,47) = 6.6, MSe = 207,p< 0.01; F2(1,94) = 4.7, MSe = 363,p<0.03). A Tukey-A posthoc analysis of the interaction (p<0.05) revealedthat the semantic facilitation was significant onlywith unpointed print, but not with pointed print.
DISCUSSIONSimilar to experiment 1, naming in pointed
print was found to be faster than naming inunpointed print, as revealed by the difference inRTs in the unrelated condition. However, whereassemantic relatedness accelerated naming in therelated condition in unpointed print, it had amuch smaller effect on accelerating naminglatencies in pointed print. Thus the results ofExperiment 2 suggest that semantic facilitation isstronger in the deeper than in the shallowerHebrew orthography. This outcome is in completeagreement with the findings of Frost et al. (1987),and Tabossi and Laghi (1992). Note, however, thatthe greater effects of semantic facilitation inpointed print than in unpointed print wereobtained even though nonwords were not includedin the stimulus set. These results conflict with thefindings of Baluch and Besner (1991). We willrefer to this in the general discussion.
EXPERIMENT 3The major claim of the weak ODH is that in all
orthographies both prelexical and lexicalphonology are involved in naming. Thus, the useof lexical or prelexical phonology is not an all-ornone process but a quantitative continuum. Thedegree to which a prelexical process of assemblingphonology predominates over the lexical routinedepends on the costs involved in assemblingphonology directly from the print. The ODHsuggests that unless the costs are too high interms of processing resources (as is usually thecase in deep orthographies), the default strategyof the cognitive system is to assemble a phonologiccode for lexical access, not to retrieve it from thelexicon following visual access. The following twoexperiments examined this aspect of thehypothesis by manipulating the processing costsfor generating a prelexical phonologicalrepresentation.
The manipulation of cost consisted of delayingthe presentation of the vowel marks relative to thepresentation of the consonant letters. InExperiment 3 subjects were presented with
162 Frost
unambiguous letter strings that could be read asone meaningful word only (i.e., only one vowelcombination created a meaningful word). Theletters were followed by the vowel marks, whichwere superimposed on the consonants, but atdifferent time intervals ranging from 0 ms (in fact,a regular pointed presentation) to 300 ms from theonset of consonant presentation. Subjects wereinstructed to make lexical decisions or to name thewords as soon as possible.
The vowel marks allow the easy prelexical assembly of the word's phonology using spelling-tophonology conversion rules. However, since theletter strings were unambiguous and could beread as a meaningful word in only one way, theycould be easily named using the addressedroutine, by accessing the lexicon visually. In fact,as we have shown in numerous studies, thisnaming strategy is characteristic of readingunpointed Hebrew (see Frost & Bentin, 1992b, fora review). The question was, therefore, whether'subjects are inclined to delay their response andwait for the vowel marks that are notindispensable for either correct lexical decisions orfor unequivocal pronunciation. If they are willingto wait for the vowel marks to appear then it mustbe because they prefer the option of prelexicalassembly of phonology, just as the ODH wouldpredict. The relative use of the assembled andaddressed routines was also verified by using bothhigh-frequency and low-frequency words in thestimulus lists. It was assumed that the moresubjects rely on the vowel marks for naming usingGPC conversion rules, the smaller would be thefrequency effect in this task. Nonwords wereintroduced as well to allow a baseline assessmentof the lagging costs. Note that nonwords cannot beunequivocally pronounced before the vowel marksare presented. In order to correctly pronouncethem, subjects have to wait the full lag period.However, since at least some of the articulationprogram can be launched immediately after theconsonants are presented, the absolute differencein RTs between the presentation at lag 0 and atlag 300 of nonwords would reflect the actual costin response time for lagging the vowel marks by300 ms. Any lag effect smaller than this differencewould suggest that subject did not wait the fulllag period but combined both prelexical andlexical routines to formulate their response.
Method
Subjects. Ninety-six undergraduate studentsfrom the Hebrew University, all native speakers ofHebrew, participated in the experiments for
course credit or for payment. Forty-eightparticipated in the lexical decision task and 48 inthe naming task.
Stimuli and Design. The stimuli were 40 highfrequency words (mean frequency 5.0 on thepreviously described 1-7 scale), 40 low-frequencywords (mean frequency 3.6), and 80 nonwords. Allwords were unambiguous; their pronunciation wasunequivocal, that is, they could be read as ameaningful word in only one way. Nonwords werecreated by altering one letter of a real word, andcould not be read as a meaningful word with anyvowel configuration. All stimuli were three to fiveletters long, and contained two syllables with fourto six phonemes. The average number of lettersand phonemes was similar for the three types ofstimuli.
The words were presented to the subjects forlexical decision or naming. Forty-eight differentsubjects were tested in each task with the samestimuli. The stimuli were presented in four lagconditions 0, 100, 200, and 300 ms. Each lag wasdefined by the SOA (stimulus onset asynchrony)between the presentation of the consonants andthe vowel marks. Thus, at lag 0 the consonantsand the vowel marks were presentedsimultaneously, at lag 100 the consonants werepresented first and the vowel marks weresuperimposed 100 ms later, etc. Four lists ofwords were formed: Each list contained 160stimuli that were composed of 10 high-frequencywords, 10 low-frequency words, and 20 nonwordsin each of the four lag conditions. The stimuliwere rotated across lists by a Latin Square designso that words that appeared in one lag in one list,appeared in another lag in another list, etc. Thepurpose of this rotation was to test each subject inall lagging conditions while avoiding repetitionswithin a list. The subjects were randomly assignedto each list and to each experimental condition(lexical decision or naming).
Procedure and apparatus. The procedure andapparatus were identical to those used in theprevious experiments. The only difference wasthat the letters and the vowel marks appearedwith the different SOAs. The clock for measuringresponse times was initiated on each trial with thepresentation of the letters, regardless of vowelmarks. The subjects were informed of the SOAmanipulation but were requested to communicatetheir lexical decisions or vocal responses as soonas possible, that is, 'Yithout necessarily waiting forthe vowel marks to appear. Each session startedwith 16 practice trials. The 160 test trials werepresented in one block.
Prelexical and Postlexical Strategies in Reading: Evidence from a Deep and a Shallow Orthography 163
ResultsMean RTs for high-frequency words, low
frequency words and nonwords in the different lagconditions for both the lexical decision and thenaming tasks are presented in Table 3. Althoughthe response pattern at each of the different lagsis important, for the purpose of simplicity the "lageffect" specified in Table 3 reflects the differencebetween the simultaneous presentation (lag 0) andthe longest SOA (lag 300).
The lagging of the vowel information hadrelatively little effect in the lexical decision task.The presentation of vowels marks 300 ms afterthe letters delayed subjects' responses by only 25ms for high-frequency words, and 37 ms for lowfrequency words. The lagging of vowels had asomewhat greater effect on the nonwords. Acrossall lags, the frequency effect was stable, about 85ms on the average. In contrast to lexical decisions,the effect of lagging the vowel information had amuch greater influence on naming. However, thefrequency effect was very small at lag 0 and 100.
The statistical significance of these differenceswas assessed in a three-way ANOVA with thefactors of task (lexical decision, naming), stimulustype (high-, low- frequency, nonwords), and lag (0,100, 200, 300), across subjects and across stimuli.·The main effect of task was significant (F1(l,94) =4.2, MSe = 70,811, p<0.04; F2(l,157) = 35, MSe =8320, p<O.OOl), as was the main effect of stimulustype (Fl(2,188) = 252, MSe = 4914, p<O.OOl;F2(2,157) = 106, MSe = 13,354, p<O.OOl), and themain effect of lag (Fl(3,282)=98, MSe = 2495,p<O.OOl; F2(3,471) = 94, MSe = 2587,p<0.00l).Task interacted with stimulus type (F1(2,188) =23.1, MSe =4914, p<O.OOl; F2(2,157) = 13,MSe =8320, p<O.OOl) and with lag (Fl(3,282) = 13.4,MSe = 2495,p<0.001, F2(3,471) = 22, MSe = 1387,p<O.OOl). Lag interacted with stimulus type(F1(6,564) = 9.3, MSe = 2076,p<0.00l; F2(6,471) =9, M Se = 2587, P <0.001). The three-wayinteraction did not reach significance in thesubject analysis(F1=1.3) but was marginallysignificant in the stimulus analysis (F2(6,471) =2.1, MSe = 1387;p<0.05). A Tukey-A post-hoc testrevealed that the frequency effect in naming wassignificant only at the longer lags (200, 300 msSOA), and not in the shorter lags (0,100 ms SOA),whereas in lexical decisions the frequency effectwas significant at all lags (p<0.05).
DiscussionThe results of Experiment 3 suggest that the
effective cost of delaying vowel marks in lexicaldecisions in Hebrew is relatively low. A lag of 300
ms in vowel marks presentation resulted in 25 msdifference in response time for high-frequencywords and 37 ms for low-frequency words. Thisoutcome suggests that subjects were not inclinedto wait for the vowel marks to formulate theirlexical decisions; rather, their responses werebased On the recognition of the letter cluster. Thisinterpretation converges with previous studiesthat showed that lexical decisions in Hebrew arenot based on a detailed phonological analysis ofthe printed word, but rely on a fast judgment ofthe printed word's orthographic familiarity basedon visual access to the lexicon (Bentin & Frost,1987; Frost & Bentin, 1992b). The slightly highercost of lagging the vowel marks with lowfrequency words and the even higher one fornonwords proposes that stimulus familiarityplayed a role in the decision strategy. Thissuggests that for words that were less familiar, orfor nonwords, subjects were more conservative intheir decisions and were inclined to wait longer forthe vowel marks, for the possible purpose ofobtaining a prelexical phonological code as well. Inaddition to the lag effect, a strong frequency effectwas obtained at all lags. This provides furtherconfirmation of lexical involvement in the taskregardless of the vowel mark presentation. '
A very different pattern of results emerged inthe naming task. The effective cost of delaying thevowel marks in naming was much higher. The lageffect obtained in naming words was about 70 mson the average, twice as large as the effect foundfor lexical decisions. Thus, although the phonologic structure of the unambiguous words could beunequivocally retrieved from the lexicon followingvisual access (addressed phonology), subjects weremore inclined to wait for the vowels to appearin the naming task than in the lexical decisiontask, presumably in order to generate a prelexicalphonologic code. The lag effect for words suggeststhat subjects combined both prelexical and lexicalroutines to name words, but the relative use of theprelexical and lexical routines varied as a functionof the delay in vowel mark presentation. Thislag effect should be first compared to the effect obtained for naming nonwords. Because they cannotbe named correctly without the vowel marksthe lag effect for nonwords reflects the overall costin response time due to a 300 ms delay of vowelmark presentation. This cost is smaller than thelag itself because the articulation program canbe initiated as the letters appear, previous tothe vowel mark presentation. The results suggestthat the cost of lagging the vowel marks by300 ms is about 150 ms in response time.
164 Frost
Table 3. Lexical decision and naming RTs and percent errors for high-frequency words (HF), low-frequency words(LF), and nonwords (NW) when vowel marks appear at different lags after letter presentation. Words areunambiguous.
Lexical Decision Naming
Lag Lag
LAG 0 100 200 300 Effect 0 100 200 300 Effect
HF 546 551 560 571 25 587 608 610 648 62(6%) (6%) (7%) (6%) (4%) (3%) (4%) (3%)
LF 626 642 644 663 37 598 624 648 678 80(12%) (10%) (11%) (10%) (5%) (3%) (4%) (5%)
Frequencyeffect 80 91 84 92 11 16 38 30
NW 641 664 684 710 69 655 700 736 799 144(9%) (8%) (8%) (9%) (5%) (7%) (5%) (7%)
Thus, the adoption of a pure prelexical strategy ofnaming words should have resulted in a similarlag effect. The results suggest that this was notthe strategy employed in naming words: The lageffect for words was only half the effect obtainedfor nonwords. However, the lag effect for namingwords should be also compared to the lag effectobtained in the lexical decision task. Note that, asin the lexical decision task, subjects did not needthe vowel marks (and a prelexical phonologic code)to name the words correctly. Nevertheless, in con·trast to lexical decision, they preferred to waitlonger for the vowel information. It could be ar·gued that the introduction of nonwords introducedthe prelexical strategy in naming. However, sincethe lag effect for words was much smaller than thelag effect for nonwords, it suggests that a combined use of the assembled and addressed routinewas used in pronunciation.
Another possible interpretation could besuggested, however, to account for the lag effect innaming. Because in the naming task subjects arerequired to produce the correct pronunciation ofthe printed word, they might have adopted astrategy of waiting for the vowels in order toverifY the phonologic structure of the printed wordwhich they generated lexically. By thisinterpretation, phonology was lexically addressedbut verified at a second stage against the delayedvowel information to confirm the lexicallyretrieved pronunciation. The verificationinterpretation would regard the difference innaming words and nonwords in this paradigm asthe difference between having words verified bythe subsequently presented vowels, which is short,and having to construct a pronunciation from thevowel information, which is longer.
Although this interpretation is plausible, it isnot entirely supported by the effect of wordfrequency on response latencies. The suggestionthat the combined use of prelexical and lexicalphonology varied as a function of the cost ofgenerating a prelexical phonological code isreinforced by examining the frequency effect innaming. In contrast to lexical decision, thefrequency effect in naming was overall muchsmaller, especially at the shorter SOAs. At the 0ms SOA the frequency effect decreased from 80 msin lexical decision to 11 in naming. This supportsthe conclusion that naming with the simultaneouspresentation of vowels was mainly prelexical andscarcely involved the mental lexicon. The longerthe SOA, the larger the cost of generating aprelexical phonologic code and, consequently, thegreater the use of addressed lexical phonology.This is reflected in the increase in the frequencyeffect at the longer SOAs (38 and 30 ms for 200and 300 ms SOA, respectively). This patternstands in sharp contrast to the lexical decisiontask, which yielded very similar frequency effectsat all lags.2
EXPERIMENT 4The aim of Experiment 4 was to examine the
effect of lagging the vowel marks whenphonologically ambiguous words (heterophonichomographs) are presented for lexical decision ornaming. In contrast to unambiguous words, thecorrect phonological structure of Hebrewheterophonic homographs can be determinedunequivocally only by referring to the vowelmarks, which specify the correct phonologicalalternative and consequently the correct meaning.Thus, if the analysis provided in the previous
Prelexical and Postlexical Strategies in Reading: Evidence from a Deep and a Shallow Orthographl/ 165
experiment concerning the cost of lagging thevowel marks is correct, the effect of presentingambiguous words on naming should be similar tothe effect of presenting nonwords. In both cases,subjects would have to rely on the vowel marks forgenerating a phonological representationnecessary for pronunciation. In contrast, laggingthe vowel marks of ambiguous words should affectlexical decisions to a much lesser extent. This isbecause lexical decisions for ambiguous wordshave been previously shown to be based on theabstract orthographic structure, and occur prior tothe process of phonological disambiguation(Bentin & Frost, 1987; Frost & Bentin, 1992a).
In Experiment 4 subjects were presented withletter strings that could be read as twomeaningful words, depending on the vowelconfiguration assigned to the letters. Nonwordswere presented as well. The vowel marks weresuperimposed on the letters at different lags, thesame as those employed in Experiment 3. Therelative use of orthographic and phonologic codingin the two tasks was again assessed by measuringthe effect of lagging the vowel information onlexical decision and naming.
MethodSubjects. One hundred and sixty undergraduate
students from the Hebrew University, all nativespeakers of Hebrew, participated in theexperiment for course credit or for payment.Eighty participated in the lexical decision taskand 80 in the naming task. None of the subjectshad participated in the previous experiment.
Stimuli and design. The words were 40 ambiguous consonant strings each of whichrepresented both a high-frequency and a lowfrequency word. The two phonological alternativeswere mostly nouns or adjectives that were notsemantically or morphologically related. Theprocedure for assessing the subjective frequenciesof the words was similar to the one employed inthe previous experiments: Fifty undergraduatestudents were presented with lists that containedthe pointed disambiguated words related to the 40ambiguous letter strings, and rated the frequencyof each word on a 7-point scale. The ratedfrequencies were averaged across all 50 judges.Each of the 40 homographs that were selected forthis study represented two words that differed intheir rated frequency by at least 1 point on thatscale. As in the previous experiments, thenonwords were constructed by altering one or twoletters of meaningful words. The design wassimilar to that of Experiment 3. However, in orderto avoid repetition of the same letter string with
different vowel marks, eight lists of words werepresented to the subjects instead of four (hencethe larger number of subjects). Each list containedonly one form (the dominant or the subordinate) ofeach homograph, at one of the possible four lags,so that each subject was presented with 40 wordsand 40 nonwords in a list. The stimuli wererotated across lists by a Latin Square design, andconsequently each letter string was presentedwith both vowel mark configurations at allpossible lags.
Procedure and apparatus. The procedure andapparatus were identical to those of Experiment 3.Each session started with 16 practice trials,followed by the 80 test trials, which werepresented in one block.
ResultsMean RTs for the dominant alternatives, the
subordinate alternatives and the nonwords in thedifferent lag conditions for both the lexical decision and the naming tasks are presented in Table4. As in Experiment 3, the effect of lagging thevowel information had little influence on lexicaldecision latencies. In fact, the use of ambiguouswords reduced the difference between the simultaneous presentation of vowels and their presentation 300 ms after the letters to 21 ms for wordson the average, and to only 6 ms for nonwords. Adifferent pattern emerged in the naming task. Theeffects of lagging the vowel marks on RTs weretwice as large as the effects found for unambiguous words in Experiment 3, where there was onlyone meaningful pronunciation. In fact, the lag effect on ambiguous words was virtually identicalwith the lag effect on nonwords.
The statistical significance of these differenceswas assessed in a three-way ANOVA with thefactors of task (lexical decision, naming), stimulustype (high-, low- frequency, nonwords), and lag (0,100,200, 300), across subjects and across stimuli.The main effect of task was significant (F1(l,158)= 39, MSe =114,2811, p<O.OOl; F2(1,117) = 314,MSe = 7379, p<O.OOl), as was the main effect ofstimulus type (F1(2,316) = 42, M Se = 7541,p<O.OOl; F2(2,1l7) = 10.5, MSe = 1l,737,p<0.001),and the main effect oflag (Fl(3,474)=69, MSe =8459, p<O.OOl; F2(3,351) = 113, MS e = 2528,p<O.OOl). Task interacted with stimulustype (F1(2,316) = 6.6, M S e = 7541, p<O.OOl;F2(2,117) = 5.8, MSe = 7,379,p<0.001) and withlag (F1(3,474) = 46, M S e = 8459, p<O.OOl,F2(3,351)=76, MSe = 2616, p<O.OOl). Lag did notinteract with stimulus type (F1, F2 <1.0). Thethree-way interaction did not reach significance(F1 = 1.3, F2 = 1.4).
166 Frost
Table 4. Lexical decision and naming RTs and percent errors for high-frequency words (HF), low-frequency words(LF), and nonwords (NW) when vowel marks appear at different lags after letter presentation. Words are ambiguous.
LAG
Dominant
Subordinate
NW
Lexical Decision Naming
Lag Lag
0 100 200 300 Effect 0 100 200 300 Effect
585 608 608 611 26 619 650 692 763 142(4%) (5%) (3%) (3%) (3%) (2%) (4%) (3%)
611 620 626 627 16 641 699 739 790 150
(7%) (5%) (4%) (2%) (5%) (4%) (6%) (4%)
624 629 635 630 6 677 713 755 828 151
9%) (9%) (10%) (10%) (5%) (4%) (8%) (7%)
DiscussionThe results of Experiment 4 suggest that the
cost of delaying the vowel marks forphonologically ambiguous letter strings in thelexical decision task was very low. This outcomesupports the conclusions put forward by Frost andBentin (1992a), suggesting that lexical decisionsin Hebrew are based on the recognition of theabstract root or orthographic cluster and do notinvolve access to a specific word in the phonologiclexicon.
The longest delays of response times due to thelagging of vowel information occurred in thenaming task. This was indeed expected. Becausethe correct pronunciation of phonologicallyambiguous words was unequivocally determinedonly after the presentation of the vowel marks,subjects had to wait for the vowels to appear inorder to name those words correctly. Thus, thesestimuli provide a baseline for assessing the effectof lagging the vowel marks on naming latencies.This baseline confirms the previous assessment,which was based on the responses to nonwords inExperiment 3, and suggests that, overall, 300 msin delaying the vowel marks cost 150 ms inresponse time for articulation if the vowels arenecessary for correct pronunciation. Note that, incontrast to the unambiguous words of Experiment3, the difference in RTs between dominant andsubordinate al~ernatives of ambiguous wordscannot reveal the extent of lexical involvement.First, this difference cannot be accurately labeledas a frequency effect. The two phonologicalalternatives of each homograph differed only intheir "dominance, and thus could have been bothfrequent or both nonfrequent. Moreover, there is aqualitative difference between frequency effects
that arises from lexical search and the dominanceeffect that results from the conflict between twophonological alternatives of heterophonichomographs. Thus, the slower RTs of subordinatealternatives, at least in the naming task, wereprobably due not to a longer lexical search for lowfrequency words, but to a change in thearticulation program. If subjects first consideredpronouncing the dominant alternative after theambiguous consonants were presented, the laterappearance of the vowel marks that specified thesubordinate alternative would have caused achange in their articulation plan, resulting inslower RTs.
General Discussion
The present study investigated the relative useof assembled and addressed phonology in namingunpointed and pointed printed Hebrew words.Experiment 1 demonstrated that the lexical statusof the stimulus had greater effect in unpointedthan in pointed print. Experiment 2 confirmedthat semantic priming facilitated naming to alesser extent in the shallow pointed orthographythan in the deeper unpointed orthography, eventhough nonwords were not included in the stimulus list. Experiment 3 and 4 examined the effect ofdelaying the vowel mark presentation on lexicaldecision and naming, in order to assess their contribution and importance in the two tasks.Because the vowel marks allow fast conversion ofthe graphemic structure into a phonologic representation using prelexical conversion rules, theirdelay constitutes an experimental manipulationthat reflects the cost of assembling phonologyfrom print. The two experiments showed that, although both naming and lexical decision could be
Prelexical and Postlexical Strategies in Reading: Evidence from a Deep and a Shallow Orthographl/ 167
performed without considering the vowel marks,subjects were inclined to use them to a greaterextent in the naming task when the cost inresponse delay was low.
These results provide strong support for theweaker version of the ODH. The disagreement between the alternative view and the ODH revolvesaround the extent of using assembled phonologyin shallow orthographies. It is more or less unanimously accepted that in deep orthographies readers prefer to use the addressed routine in naming.This is because the opacity of the relationship between orthographic structure and phonologicform, which is characteristic of deep orthographies, prevents readers from assembling a prelexical phonological representation through the useof simple GPC rules. Thus, the major debate concerning the validity of the ODH often takes placeon the territory of shallow orthographies, in orderto show their extensive use of addressed phonology in naming. What lies behind the alternativeview, therefore, is the assumption (even axiom)that the use of the addressed routine for namingconstitutes the least cognitive effort for any readerin any alphabetic orthography.
The advantage of examining the validity of the. ODH by investigating reading in pointed and unpointed Hebrew is therefore multiple. First, it iswell established that Hebrew readers are used toaccessing the lexicon by recognizing the word's orthographic structure. The phonologic informationneeded for pronunciation is then addressed fromthe lexicon (Frost & Bentin, 1992b). The questionof interest, then, is: what reading strategies areadopted by Hebrew readers when they are exposed to a shallower orthography? Do they adopt aprelexical strategy even though it is not-the natural strategy for processing most Hebrew readingmaterial? A demonstration of the extensive use ofthe assembled routine in pointed Hebrew therefore provides strong support for the ODH. If readers of Hebrew prefer the use of the assembledroutine, surely habitual readers of shallower orthographies would have a similar preference.
The second advantage in using the twoorthographies of Hebrew is that it allows themanipulation of a within-language betweenorthography design. This design has amethodological advantage over studies thatcompare different languages. The interpretation ofdifferences in reading performance between twolanguages as reflecting subjects' use of pre- vs.post-lexical phonology can be criticized onmethodological grounds. The correspondencebetween orthography and phonology is only one
dimension on which two languages differ. Englishand Serbo-Croatian, for example, differ ingrammatical structure and in the size andorganization of the lexicon. These confoundingfactors, it can be argued, may also affect subjects'performance. The comparison of pointed andunpointed orthography in one language, Hebrew,allows these pitfalls to be circumvented.
Taken together, the four experiments suggestthat the presentation of vowel marks in Hebrewencourages the reader to generate a prelexicalphonologic representation for naming. The use ofthe assembled routine was detected by examiningboth frequency and semantic priming effects innaming. Experiment 3 provides an importantinsight concerning the preference for prelexicalphonology. When the vowels appearedsimultaneously with the consonants, thefrequency effect in naming was small andnonsignificant, again suggesting minimal lexicalinvolvement. Thus the results of this conditionreplicate the findings of Experiment l.
Two methodological factors should be consideredin interpreting the results of Experiments 3 and 4.The first relates to the experimental demandcharacteristics of the lagged pointed presentation.It could be argued that this unnatural presentation of printed information encouraged subjects to wait for the vowels to appear and consequently to use them. In fact, why not adopt astrategy of waiting for all possible information tobe provided? Although this could be a reasonablesolution for efficient performance in these experiments, the results of the lexical decision taskmake it very unlikely that this simple strategywas adopted by our subjects. In this task subjectswere not inclined to wait for the vowel marks.This result was most conspicuous in Experiment4, in which the lag effect was minimal. This outcome suggests that the lagged presentation didnot induce a uniform strategy of vowel mark processing, but that subjects adopted a flexible strategy characterized by a gradual pattern of relyingmore and more on the vowel marks (prelexicalphonology) as a function of the task and of theambiguity of the stimuli.
The differential effect of lag on ambiguous andunambiguous words in the two tasks suggests thatboth the assembled and the addressed routineswere used for generating phonology from print. Onone hand subjects used explicit vowel informationemploying prelexical transformation rules. This isreflected in the greater effect of lag on namingrelative to lexical decision latencies, and in thesmaller frequency effect of experiment 3 at the
168 Frost
shorter SOAs. On the other hand, it is clear thatthe phonologic structure of unambiguous wordswas generated using the addressed routine aswell. This is reflected in the differential effect oflagging the vowel information on namingambiguous and unambiguous words: There weresmaller effects of lag on naming unambiguousthan ambiguous words. This confirms thatsubjects did not wait for the vowel marks in orderto pronounce the unambiguous words, as long asthey waited in order to pronounce the ambiguouswords. The flexibility in using the two routines fornaming unambiguous words was affected,however, by the "cost" of the vowel information.When no cost was involved in obtaining the vowelinformation (simultaneous presentation), theprelexical routine was preferred, as reflected bythe small frequency effect. In contrast, when theuse of vowel information involved a higher cost,given the delayed presentation of the vowelmarks, a gradually greater reliance on theaddressed routine was observed. This is wellreflected by the larger frequency effect at thelonger lags. This pattern stands in sharp contrastto the lexical decision task, in which the frequencyeffect was stable and very similar across lags.
The second methodological issue to be considered in interpreting the results of Experiments 3and 4 is the presence of nonwords in the experiments. One factor that has been recently proposedto account for the conflicting results concerningthe effect of orthographic depth is the inclusion ofnonwords in the stimulus list (Baluch & Besner,1991). In their study, Baluch and Besner presented native speakers of Persian with opaqueand transparent Persian words and showed thatdifferences in semantic facilitation in the namingtask appeared only when nonwords were includedin the stimulus list. When the nonwords wereomitted, no differences in semantic facilitationwere found. Baluch and Besner concluded that theinclusion of nonwords in the list encourages subject to adopt a prelexical strategy of generatingphonology from print, because the phonologicstructure of nonwords cannot be lexically addressed but only prelexically assembled.
The results of Experiment 3 and 4 do notsupport the view that the nonwords induced apure prelexical strategy. If this were so, the effectsof lag would have been similar for words and fornonwords. Subjects would have waited for thevowels of every stimulus to appear and the lageffect for unambiguous words, ambiguous words,and nonwords would have been similar, about 150ms. This clearly was not the outcome we obtained.
Subjects waited the full 150 ms to pronounce thenonwords and the ambiguous words but waitedonly half as much time to pronounce theunambiguous words. This confirms a mixedstrategy in reading.
However, the argument proposed by Baluch andBesner deserves serious consideration. Not only isits logic compelling, but also the effect ofnonwords on reading strategies is welldocumented in several studies. The crux of thedebate is, however, whether this argument canaccount for the observed effects of orthographicdepth found in the various studies described above(e.g., Frost et al., 1987; Katz & Feldman, 1983).The evidence for that claim seems to be much lesscompelling. There is no argument that nonwordsinduce a prelexical strategy of reading in allorthographies (e.g., Hawkins, Reicher, & Rogers,1976; and see McCusker, Hillinger, & Bias, 1981,for a review). However, note that the weak ODHproposes that whatever the effect of nonwordinclusion is, it would be different in deep and inshallow orthographies. Thus, evidence for oragainst the weak ODH could only be supported bystudies directly examining the differential effect ofnonwords in various orthographies. Such a designwas indeed employed by Baluch and Besner(1991), but their conclusions hinge on the nonrejection of the null hypothesis. In contrast totheir results, several studies have provided clearevidence supporting the weak ODH. For example,Frost et al. (1987) showed that the ratio ofnonwords had a dramatically different effect onnaming in Hebrew, in English, and in SerboCroatian. Different effects of nonwords inclusionon semantic facilitation in naming in the deepEnglish and the shallow Italian, were alsoreported by Tabossi and Laghi (1992).
The results of Experiment 2 also stand in sharpcontrast to Baluch and Besner's conclusions. InExperiment 2 different semantic priming effectswere found between pointed and unpointedHebrew, even though nonwords were not includedin the stimulus list. How can this difference be accounted for? A possible explanation for this discrepancy could be related to the strength of theexperimental manipulation employed in the two.studies. Baluch and Besner used opaque andtransparent words within the unpointed-deepPersian writing system. The phonological transparency of their words was due to the inclusions ofletters that convey vowels. However, becausethese letters can also convey consonants in a different context, they may have introduced someambiguity in the print that is characteristic of
PrelexicaI and PostIexicaI Strategies in Reading: Evidence from a Deep and a Shallow Orthography 169
deep orthographies, hereby encouraging the use ofthe addressed routine. In contrast, the experimental manipulation of using pointed and unpointedHebrew print is much stronger. Hence, differential effects of semantic facilitation in namingemerged in Hebrew, providing strong support forthe weak ODH.
The debate concerning the ODH, however, cannot simply revolve around the interpretation ofvarious findings without adopting a theoreticalframework for reading in general. Interestingly,while contrasting the ODH with the alternativeview, one might find very similar definitions thatcapture these conflicting approaches. Proponentsof the alternative view would argue that, regardless of the characteristics of their orthographyreaders in different languages seem to adopt remarkably similar mechanisms in reading (e.g.,Besner & Smith, 1992; Seidenberg, 1992). Thus,the alternative view attempts to offer a universalmechanism that portrays a vast communality inthe reading process in different languages.Surprisingly, proponents of the ODH take a verysimilar approach. They suggest a basic mechanism for processing print in all languages, whichis finely tuned to the particular structure of everylanguage (Carello et al., 1992). The major discussion is, therefore, what exactly this mechanism is.
The basic assumption of the ODH concerns therole of phonology in reading. It postulates as abasic tenet that all writing systems are phonological in nature and their primary aim is to conveyphonologic structures, i.e. words, regardless of thegraphemic structure adopted by each system (seeDe Francis, 1989; Mattingly, 1992, for a discussion). Thus, the extraction of phonologic information from print is the primary goal of the reader,whether skilled or beginner. It is the emphasisupon the role of phonology in the reading processthat bears upon the importance of prelexicalphonology in print processing. The results of thepresent study furnish additional support for anincreasingly wide corpus of research that providesevidence confirming the role of prelexical phonology in reading. This evidence comes not only fromshallow orthographies like Serbo-Croatian (e.g.,Feldman & Turvey, 1983; Lukatela & Turvey,1990), but also from deeper orthographies likeEnglish, using a backward masking paradigm(e.g., Perfetti et al., 1988) or a semanticcategorization task with pseudohomophonic foils(Van Orden, 1987; Van Orden, Johnston, & Halle,1988). Recently, Frost and Bentin (l992a) showedthat phonologic analysis of printed heterophonichomographs in the even deeper unpointed Hebrew
orthography precedes semantic disambiguation.In another study, Frost and Kampf (1993) showedthat the two phonologic alternatives of Hebrewheterophonic homographs are automatically activated following the presentation of the ambiguousletter string.
The results of the present study converge withthese findings. They provide an opportunity toexamine the ODH and the alternative view not inthe linguistic environment of shalloworthographies, but of deep orthographies. Ifprelexical phonology plays a significant role in thereading of pointed Hebrew by readers who aretrained to use mainly the addressed routine forphonological analysis, then the plausibleconclusion is that, in any orthography, assembledphonology plays a much greater role in readingthan the alternative view would assume.
REFERENCESBalota, D. A., & Chumbley, J. 1. (1984). Are lexical decisions a
good measure of lexical access? The role of word frequency inthe neglected decision stage. Journal of Experimental Psychology:Human Perception and Performance, 10, 340-357.
Baluch, B., & Besner, D. (1991). Strategic use of lexical andnonlexical routines in visual word recognition: Evidence fromoral reading in Persian. Journal of Experimental Psychology:Learning, Memory, and Cognition, 17, 644-652.
Bentin, S., Bargai. N., & Katz, 1. (1984). Orthographic andphonemic coding for lexical access: Evidence from Hebrew.Journal of Experimental Psychology: Learning Memory & Cognition,10,353"368.
Bentin, S., & Frost, R. (1987). Processing lexical ambiguity andvisual word recognition in a deep orthography. Memory &Cognition, 15, 13-23.
Besner, D., & Hildebrant (1987). Orthographic and phonologicalcodes in the oral reading of Japanese Kana. Journal ofExperimental Psychology: Learning Memory 6' Cognition, 13,335-343.
Besner, D., & Smith, M. C (1992). Basic processes in reading: Isthe orthographic depth hypothesis sinking? In R. Frost & 1.Katz (Eds.), Orthography, phonology, morphology, and meaning(pp. 45-66). Amsterdam: Advances in Psychology, ElsevierScience Publishers.
Besner, D., Patterson, K., Lee, 1., & Hildebrant N. (1992). Twoforms of Japanese Kana: Phonologically but NOTorthographically interchangeable. Journal of ExperimentalPsychology: Learning Memory & Cognition.
Carello, C, Lukatela, G., & Turvey, M. T. (1988). Rapid naming isaffected by association but not syntax. Memory & Cognition, 16,187-195.
Carello, C, Turvey, M. T., & Lukatela, G. (1992). Can theories ofword recognition remain stubbornly nonphonological? In R.Frost & 1. Katz (Eds.) Orthography, phonology, morphology, andmeaning (pp. 211-226). Amsterdam: Advances in Psychology,Elsevier Science Publishers.
Colombo, 1., & Tabossi, P. (1992). Strategies and stressassignment: Evidence from a shallow orthography. In R. Frost& 1. Katz (Eds.), Orthography, phonology, morphology, andmeaning (pp. 319-340). Amsterdam: Advances in Psychology,Elsevier Science Publishers.
170 Frost
DeFrancis, J. (1989). Visible speech: The diverse oneness of writingsystems. Honolulu: University of Hawaii Press.
Feldman, L. B., & Turvey, M. T. (1983). Word recognition inSerbo-Croatian is phonologically analytic. Journal ofExperimental Psychology: Human Perception and Performance, 9,228-298.
Frost, R (1992). Orthography and phonology: The psychologicalreality of orthographic depth. In M. Noonan, P. Downing, & S.Lima (Eds.), The linguistics of literacy (pp. 255-274).Amsterdam/Philadelphia: John Benjamins Publishing CO.
Frost, R, Katz, L., & Bentin, S. (1987). Strategies for visual wordrecognition and orthographical depth: A multilingualcomparison. Journal of Experimental Psychology: HumanPerception and Performance, 13, 104-115.
Frost, R, & Katz, L. (1989). Orthographic depth and theinteraction of visual and auditory processing in wordrecognition. Memory & Cognition, 17, 302-311.
Frost, R, & Bentin, S. (1992a). Processing phonological andsemantic ambiguity: Evidence from semantic priming atdifferent SOAs. Journal of Experimental Psychology: Learning,Memory, and Cognition. 18,58-68.
Frost, R, & Bentin, S. (1992b). Reading consonants and guessingvowels: Visual word recognition in Hebrew orthography. In RFrost & L. Katz (Eds.), Orthography, phonology, morphology, andmeaning (pp. 27-44). Amsterdam: Advances in Psychology,Elsevier Science Publishers.
Frost, R, & Kampf, M. (1993). Phonetic recoding ofphonologically ambiguous printed words. Journal ofExperimental Psychology: Learning, Memory, and Cognition, 1-11.
Hawkins, H. L., Reicher, G. M., & Rogers, M. (1976). Flexiblecoding in word recognition. Journal of Experimental Psychology:Human Perception and Performance, 2, 235-242.
Katz, L. & Feldman L. B. (1981). Linguistic coding in wordrecognition. In: A.M. Lesgold & CA. Perfetti (Eds.), Interactiveprocesses in reading. Hillsdale, NJ: Erlbaum.
Katz, L., & Feldman L. B. (1983). Relation between pronunciationand recognition of printed words in deep and shalloworthographies. Journal of Experimental Psychology: Learning,Memory, and Cognition., 9, 157-166.
Katz, L., & Frost, R. (1992). Reading in different orthographies: theorthographic depth hypothesis. In: R. Frost & L. Katz (Eds.),Orthography, phonology, morphology, and meaning (pp. 67-84).Advances in Psychology, Elsevier, North-Holland.
Klima, E. S. (1972). How alphabets might reflect language. In F.Kavanagh & I. G. Mattingly (Eds.). Language by ear and by eye.The MIT Press, Cambridge, Massachusetts, and London,England.
Laudanna, A., & Caramazza, A. (1992). Morpho-lexicalrepresentations and reading. Paper presented at the fifthconference of the European Society for Cognitive Psychology,Paris, France.
Liberman, I. Y., Liberman, A. M., Mattingly, I. G., & Shankweiler,D. (1980). Orthography and the beginning reader. In J. F.Kavanagh & R. L. Venezky (Eds.), Orthography, reading, anddyslexia. Austin, TX: Pro-Ed.
Lukatela, G., Popadic, D., Ognjenovic, P., & Turvey, M. 1. (1980).Lexical decision in a phonologically shallow orthography.Memory & Cognition, 8, 415-423.
Lukatela, G., Feldman, L. 8., Turvey, M. 1., Carello, C, & Katz, L.(1989). Context effects in bi-alphabetical word perception.Journal of Memory and Language, 28, 214-236.
Lukatela, G., & Turvey, M. 1. (1990). Automatic and prelexicalcomputation of phonology in visual word identification.European Journal ofCognitive Psychology, 2, 325-343.
Lupker, J. S. (1984). Semantic priming without association: Asecond look. Journal of Verbal Learning and Verbal Behavior, 23,709-733.
McCusker, L. X., Hillinger, M. L., & Bias, R G. (1981). Phonologicrecoding and reading, Psychological Bulletin, 89, 217-245.
Mattingly, I. G. (1992). Linguistic awareness and orthographicform. In R Frost & L. Katz (Eds.), Orthography, phonology,morphology, and meaning (pp. 11-26). Amsterdam: Advances inPsychology, Elsevier Science Publishers.
Neely, J. H. (1991). Semantic priming effects in visual wordrecognition. In D. Besner & G. W. Humphreys (Eds.), Basic processes in reading: Visual word recognition. Hillsdale, NJ: Erlbaum.
Perfetti, C A., Bell, L. C, & Delaney, S. M. (1988). Automatic(prelexical) phonetic activation in silent word reading:Evidence from backward masking. Journal of Memory andLanguage, 27, 59-70.
Perfetti, CA., Zhang,S., & Berent, I. (1992). Reading in Englishand Chinese: Evidence for a universal phonological principle.In R. Frost & L. Katz (Eds.), Orthography, phonology, morphology,and meaning (pp. 227-248). Amsterdam: Advances inPsychology, Elsevier Science Publishers.
Scheerer, E. (1986). Orthography and lexical access. In G. Augst(Ed.), New trend in graphemics and orthography. (pp. 262-286).Berlin: De Groyter.
Sebastian-Galles, N. (1991). Reading by analogy in a shalloworthography. Journal of Experimental Psychology: HumanPerception and Performance, 17, 471-477.
Seidenberg, M. S. (1985). The time course of phonological codeactivation in two writing systems. Cognition, 19, 1-30.
Seidenberg, M. S. (1992). Beyond orthographic depth in reading:Equitable division of labor. In R Frost & L. Katz (Eds.),Orthography, phonology, morphology, and meaning (pp. 85-118).Amsterdam: Advances in Psychology, Elsevier SciencePublishers.
Seidenberg, M. S. & Vidanovic, S. (1985). Word recognition inSerbo-Croatian and English: Do they differ? Paper presented atthe Twenty-fifth Annual Meeting of the Psychonomic Society,Boston.
Tabossi, P., & Laghi, L. (1992). Semantic priming in thepronunciation of words in two writing systems: Italian andEnglish. Memory & Cognition, 20, 303-313.
Turvey, M. 1., Feldman. L. B., & Lukatela, G. (1984). The SerboCroatian orthography constrains the reader to a phonologicallyanalytic strategy. In L. Henderson (Ed.), Orthographies andreading: Perspectives from cognitive psychology, neuropsychology,and linguistics (pp. 81-89). Hillsdale, NJ: Erlbaum.
Van Orden, G. C (1987). A ROWS is a ROSE: Spelling, sound andreading. Memory & Cognition, 15,181-198.
Van Orden, G. C, Johnston, J. C, & Halle, B. L. (1988). Wordidentification in reading proceeds from spelling to sound tomeaning. lournal of Experimental Psychology: Learning, Memory,and Cognition, 14,371-386.
FOOTNOTES'Journal of Experimental Psychology: Learning, Memory, andCognition, in press.
I The error rates presented in this study should be evaluated withcare because the criteria for assessing an error in pointed andunpointed print are different. While in unpointed print anypronunciation consistent with the consonants only is consideredcorrect, in pointed print any pronunciation that is inconsistentwith the explicit vowel information is considered an error. Thisis of special significance while evaluating the nonwords data. Inunpointed print subjects have much greater flexibility in namingnonwords than in pointed print, hence the larger error rates inthe pointed condition.
2But note that the effect of delaying vowels on naming latenciesdoes not linearly decrease with lag, as should be predicted bythe increase of frequency effect. In order to obtain a moreordered relationship more lag conditions should have been used.