A WORD SUPERIORITY EFFECT IN A PHONETICALLY PRECISE ORTHOGRAPHY

/

G. Lukatela,+ B. Lorenc,+ P. Ognjenovic,+ and M. T. Turvey++

Abstract. Other things being equal, a letter is identified moreaccurately and rapidly in the context of a word than in the contextof a nonword. This word-superiori ty effect has been demonstratedmany times wi th materials conforming to English orthography. Thepresent experiment, using the probe letter-recogni tion procedure,demonstrates the same effect for the Serbo-Croatian orthography. Inthat the English and Serbo-Croatian orthographies distinguishmarkedly in the level at which they systematically reference thespoken language, it appears that the word-superiority effect is notowing to orthographic idiosyncracies. Analysis of the effect inSerbo-Croatian suggests that it is not completely accountable for interms of inter letter probability structure and that word-specificfactors may be involved.

Under the same conditions, a letter is identified more rapidly and moreaccurately in the context of a word than in the context of a nonword. Thisletter-in-context or word-superiori ty effect is now a well-established factfor fluent readers of the English orthography (Baron, 1978). Arguably, fluentreaders of English relate more efficiently to English words than to letterstrings wi th which they have had no experience because they have learnedsomething about the structure of written English in general and/or theproperties of English words in particular. What has been learned to enhanceword perception cannot be precisely pinpointed. Nevertheless, several kindsof knowledge can be proposed as potential candidates, for example, meaning,whole-word familiarity, word-specific associations with sounds, spelling rulesand familiarity with spelling patterns (Baron, 1978). Questions as to theaspect or aspects of word processing that these kinds of knowledge influenceare largely unresolved, although most recent evidence appears to rule out thefeature analysis of component letters (Krueger & Shapiro, 1979; Massaro, 1979;Staller & Lappin, 1979).

The major focus of the present paper is a simple question: Does the wordsuperiority effect hold for an orthography that differs nontrivially from theorthography of English? Orthographies work as transcriptions of languagebecause the patterning of symbols in written text bears a systematic relation­ship to some corresponding patterning in the spoken language. The orthographyof English is principally (but not exclusively) systematic with reference to

+University of Belgrade++Also University of Connecticut.

This research was supported in part by NICHD Grant HD-08495to the University of Belgrade and in part by NICHD Grant HD-01994 to HaskinsLaboratories

LABORATORIES: Status on Research (1980) ]263

the morphophonemics of the spoken language, while the orthography of Serbo­Croatian is principally (but not exclusively) systematic with reference to the(classically defined) phonemics of the spoken language (see Lukatela & Turvey,1980; Lukatela, Popadic, Ognjenovic, & Turvey, 1980)0 We might expect tofind, therefore, differences between the reading-related processes exhibitedby fluent readers of English and those exhibited by fluent readers of Serbo­Croatian 0 For fluent readers of Serbo-Croatian, lexical decision is mediatedby phonetic recoding (Lukatela et alo, 1980); in contrast, fluent readers ofEnglish tend to access the lexicon in nonphonological terms (Coltheart,Besner, Jonasson, & Davelaar, 1979)0 With respect to a distinction drawn byBaron and Strawson (1976), fluent readers of Serbo-Croatian may be dispropor­tionatel y "Phoenician" (that is, treat the written word as an alphabetictranscription), while fl uent of ish may be yn Chinese" (that is, treat the wri tten word as a logographic transcription) e

Though the latter contrast is exaggerated, it makes the point that thephonemically oriented Serbo-Croatian orthography and the morphophonemicallyoriented Engl ish orthography may give emphasis to different aspects of thewritten form of the word and thus motivate the acquisition of, and adependency on, different kinds of knowledge for word perception 0 theletter-in-context or word-superiority effect is indigenous to the ishorthography (and to orthographies of like kind) and is due to the fact thatthe processing of written English often demands the use of recoding unitslarger than the single letter. We doubt that there is such a restriction onthe word-superiori ty effect, but the question of the effect's dependency onthe orthography must be asked neverthelesso

The question was addressed through the firstintroduced Reicher (1 ). A horizontally of letters isbriefly exposed and followed immediately a mask ( the region of theletter ) with two letters located above and below theof a letter in the The ect t s task is y to choosewhich of the two letters occ the posi tion 0 Of interest is howletter recognition varies with the nature of the letter string0

Method

from thein the exects received their

theeno-

The subj ectsPsychology at the

of a courseelementary education in

ic al prior to the, & Turv ) •

264

positioned at the center of the display. The mask pattern subtended 21'vertical by 2°17' horizontal to coincide perfectly with the region occupied bythe letter string. The response alternatives subtended 1· 34' vertically fromthe top part of the upper letter to the bottom part of the lower letter. Thelight background regions of the target and mask fields were equated at 10cd/m2 •

There were four kinds of target stimuli: single letters, five-letterwords 9 five-letter nonwords wi th vowels (" pseudowords"), and five-letternonwords wi thout vowels ("nonwords").. Thirty-two instances of each kind wereconstructed. Six instances of each kind were used in the preliminaries to theexperiment and twenty instances of each kind were used in the experimentproper

In the fashion of Reicher (1969) and Wheeler (1970) the words and theirresponse alternatives were selected so that the wrong alternative, if substi­tuted for the probed letter, also made a word with a frequency of occurrenceroughly equivalent to that of the target word. Frequency equivalence wasdetermined according to the frequency count of Dj. Kostic (Note 1). Thus, ifthe target word were TACKA (point), and the alternatives for the first letteras the probed letter were T and M, then the substitution of T by M would giveMACKA (cat).

The words were of five different consonant(c)-vowel(v) structures, CVCVC,CCVCV, VCCVC, VCVCV, CVCCV, which were represented in the set of twenty words,respectively, seven times, seven times, twice, twice, and twice. The differ­ent consonant-vowel structures were necessitated by the requirements that (1)only consonants were probed in the four kinds of stimuli (the nonwords werecomposed only of consonants) and (2) each letter position was probed equallyoften. Table 1 gives the words and pseudowords together wi th the responseal ternatives. Each of the twenty pseudowords was constructed from its wordmate by changing two letters without altering the consonant-vowel structure.Which two letters were changed depended on the particular consonant-vowelstructure of the word as is evident from inspection of Table 1. Moreover, thepartiCUlar letter substitutes chosen were selected to keep the pronounceabili­ty of a word and its pseudoword partner approximately equivalent e This"pronounceability" stricture also determined the selection of the incorrectresponse alternative. The response alternatives for an individual pseudowordwere the same as for its word mate

The nonwords were constructed by a random drawing of consonants under theconstraint that no letter could be wi thin a letter string The

stimuli were all consonants and occurred in themiddle of the slide

A sequences of slideschannel (Scientific , Model GB)cri tical member one ofnearer of the two indexed "lower" and the farther of

sequence of slides consisted of the,a ation field of msec exposure was

three­to the

Theindexed

to afollowed by

265

Table 1

Words, pseudowords and response alternatives withtarget letters specified

266

WORDS

BRANA

LIlAR- I

SRECA

VRA.IA

IlEAl

NAPAD

U~ICA

TRAyA

-SAVEl

-METAL

OBRAl

GLAVA

BOMBA

KANAL

- /

pONoC

°EERA

BRADA

PSEUDOWORDS

!:JREKA

LEIoR

SRISA

VLI!A

IGREl

NALID

UbElA

-TLEyA

SAGI~

MEBOL.

GLOTA

BU!1KA

KASOk

- '"PANue

OfINA

RESPONSE ALTERNATIVES

H,G

T,M

R,V

T,N

R,L

N,l

L,D

V,K

l,T

L,K

B,D

G,S

M,R

L,P

N,M

P V

L,T

D,V

T

a slide containing one or five letters. The duration of this letter-string ortarget slide was tailored to the individual subject and therefore variableacross subjects but constant for a given subject wi thin the sequences ofslides. Immediately following the termination of the target slide, that is,at an inter-stimulus interval of 0 msec, a slide containing a randompatterning of lines (that overlapped the letters of the sl and twoletters was for a duration of 1.5 sec. One of the two letters wasabove the t while the other was below it. These two letterswere aligned vertically and ,located so as to correspond to the position of oneof the letters in the target slide. The subject's task was to press one ofthe two keys to identify which of the two letters, the upper or the lower, wasthe letter occurring in that position of the target slide. One of the letteralternatives was always correct.

The dependent measure was the accuracy of the subject's choice betweenthe two response alternatives. A level of performance was sought, therefore,at which a subject recognized the probed-for letters above chance but notperfectly. To this purpose, the collection of data for was precededby a session during which the subject was familiarized with the taskand during which the determined the duration of the slideexposure at which the subject vs performance was seventy-five

accurate

session was divided into two the first phasetime of the stimuli was held constant at 100 msec and the

feedback on the accuracy of his or her choice. In thestimulus duration was reduced until a duration

was reached. Further sequencesof the criterial duration with

necessarYG Across ects thefrom to msec

to the ect wi th theduration and with the different

theaccuracy of

to assess theor decreases

crisession sequences were

target exposure at the individually determinedof stimuli distributed

second

The number ofwas enteredwhich showed.001. Thewere64052.difference.02) ,G01 ) •

for each subject for each stimulus typeof variance ( x Stimulus ),

, F 3, ) 12.,.E <four stimul us

nonwords,ficant

<

Let usfor

accountedparse into

Vene

267

for English. There are t of course t consistent mappings but they are oftenabstract and they generally relate SYmbols to the morphophonemic andnot to the phonetic level of the language.. Moreover Ii their applicationgenerally involves lexical reference.. in is not a singlephoneme as it is in or must recognizethat in mishap the two letters are morpheme boundary.Knowledge of parts of in addition to is necessary forthe pronunciation of at the end of words (compare the verbs deflate,integrate with the nouns A more straightforward rule isthat which ascribes the e, i or y plus a consonant orjuncture. Because of the it is often necessaryfor a speaker of ish to of a word that anotherfinds perplex by ind and order of thealphabetic constituents.. In contrast Serbo-Croatian can commun-icate the spelling in almost all speaking the word moreslowly.. The point is that fund rules required forspelling English has no in Serbo-Croatian and thus if suchknowledge were a critical in the effect, then nosuch effect should be Serbo-Croatian

Consider a further but related reason that derives from doubts as to thevalue of reforming the ish in the direction of greaterphonetic specificity (cf @ Gibson & Levin, ) .. Ii the efficientrecognition of ( words is based in the intra-word redun-dancies generated increase the phonetic precisionof a writing away these clues to a word's nature.The orthography of allows skilled readers to obtain grammatical andsemantic information about from forms ( Chomsky,1970). This is because ish preserves the similarity ofwords (for anxious 9 , whereas an oriented tophonetics would Ii this commitment to and etymology.Thus in Serbo-Croatian of the same word undergo ortho-graphic modification in transcriptionfrom the to the nominativeand dative forms, word Given theseconsiderations one could entertain an of the kind: Meaningis a type of that determines the word effect. Butmeaning is less d internal structure of Serbo-Croatian words than of ish words" Atthe time of , a reader ofSerbo-Croatian is 'Consequently. underdistinction is less

Of course,reason for belfactortothe

268

if anyto s

sufficientserve

whatkinds

Serbo­However t

given that fluent readers of Serbo-Croatian did letters in wordsbetter than letters in nonwords and pseudowords, let us to considerthe reasons why did so. Wi th to the differencebetween the words and the single letters, it suffices to note that when singleletter is the poorer of the two (eog@, Carr, Lehmkuhle, Kottas,Astor-Stetson, & Arnold t 1976), it i due to(Estes, 1 ) In our the letters in thesame of the

words were better than the nonwords may not require anfactors in that the pseudowords were similarly

However, that the words were t in turn, better than themean an to factors may be required

for a full account $ The in of words and pseudowordsover nonwords can considered from two perspectives: One emphasizes generalorthographic distinctions and the other emphasizes general (non-orthographic)

and conceptual distinctions between the two kinds of letter patterns.Thus the of written Serbo-Croatian (for example the tendency toalternate and vowels, the limited number of consonant runs of twoand three in the words and and not present in therandom consonant that were the nonwords may be the source of the

d Yet recourse to the regulari ties of the wri ttenlanguage may be there are nonlinguistic factors that woulddistinguish the words and pseudowords from the nonwords in ways that arepotentially the

letters--vowels the words andof the nonwords and

one letters--consonants--was There is much evidenceto show that information facilitates the detection of invisual search 1 1; 1 Jonides & Glei tman, 19721 Lukatela ) Sometimes referred to a tv tv

effect evidence that this be an ill-chosenlabel Denotable relations may well the reliable discrimina-tion of vowels from consonants ( ler & , 1979· White, 1977). At allevents, enhanced of letters words and in wi th

in nonwords have been due to abil distinguish(consonants) from the (vowel

number letters to be Stallerone instance that this

269

present data. First, current knowledge does not permit a systematic equatingof words and pseudowords on the many non-semantic, non-lexical dimensions ofpotential relevance to perceiving letter strings (for example, the frequenciesof letter groups, the frequencies with which letter groupings occur in certainpositions within the letter string). Second, methods vary in their sensitivi­ty to the word-superiority effect and where the difference between words andnonwords is relatively small, that between words and pseudowords is usuallynonexistent. Type of mask (Johnston & McClelland, 1973), visual angle of thedisplay (Purcell, Stanovich, & Spector, 1978) and the onset asynchrony betweenletter string presentation and mask presentation (Michaels & Turvey, 1979)contribute significantly to the magnitude of the word-superiority effect.

The difference between words and pseudowords was significant in thepresent experiment. Is it a genuine word-specific effect? The answer is noteasily given, largely because of the first reason noted above--ignorance ofwhether all the non word-specific dimensions were equated between the two setsof stimul i. Nevertheless, when general factors are considered, such asfrequency of letter patterns and geometric properties of the letter strings,there remains some reason for believing that specific factors such as meaning,lexical membership or whole-word familiarity (Baron, 1978) may have contribut­ed to the word/ pseudoword difference Wi th respect to geometric properties,Staller and Lappin (1979) have shown that the symmetry and directional ofletters are significant to the perceptibility of letters in letter contexts.In the present experiment, where a symmetrical letter (e.g", M, T) in a wordwas changed in the construction of its pseudoword pair, the letter was changedhal f of the time into another symmetrical letter and half of the time into aright-facing letter (e.g, G,L) Likewise, right- letters were convert-ed into another right- letter hal of the time and intoletter the other hal f of the time. So at least in terms of these twodimensions, symmetry and directional of individual letters, the words andpseudowords were numerically

A potentially more and likely source of difference is theconditional probabilities among the letter Changing two letters of aword to produce a pseudoword may have the to which letterpairings conformed to the Consulting Tomic's (1978) digramcy analysis of 1,250,000 tokens, the conditional frequencies of letterin the forward direction (that is, the frequency the letter E occurs givenletter i before it) were determined for each letter string Since the stringswere five letters in length there were four conditional for eachletter string; these four were summed for each indiv id ual str of letters"For the words of the experiment the overall mean of the individualsums was 26, 1 an overall mean of 17, for the soMoreover, of the 0 words and s, the word member was ofhigher summed conditional in seventeen of the It wouldtherefore, that the wordl difference in theaccountable for in terms in the interletterstructure e A further anal however, thatstructure may not be the com

A correlation compseudowords and the number

0513, J? < .05), mean

270

the errors. In contrast, a similar correlation computed for the word stimuliproved insignificant (r : -.005). The possibility that interletter probabili­ty characteristics may have contributed more significantly to letter recogni-tion in pseudowords than in words consistent with other observations in theliterature. Thus. Engel (1974) that the relationship between inter-letter probabilities and the accuracy of letter was mostfor low words, and Rice and Robinson (1975) showed that theinfluence of mean digram on lexical decision latencies was restrict-ed to rare words. An analysis Whaley (1978) concurs with these observa-tions: Whereas general factors such as interletter probabil i ty structurecontribute to the of letter stri that are nonwordsor pseudowords and perhaps to the of relatively new or unfamiliarwords, they contribute to the perception ofwords. In word perception the are overridden by the specificaspects such as richness of ng and famil In the absence offurther analysis on general we may, therefore, draw the qualifiedconclusions that the effect of the present experiment is aword-specific effect.

It remains for us to make one final remark way of reinforcing a pointabove wi th to the data. The Serbo-Croatian language isbiased heavily toward open A perusal of the Tomie (1978) normsreveals that consonant-vowel and vowel-consonant pairs are by far the mostfrequent, with consonant-consonant rare& A crude compari-son suggests that the relative of consonant pairs and consonanttriples in English is ( , Conrad, &: Thompson t 1960, comparedwith Tomie, 1978)& This difference between the interletter structure of thetwo languages may account for the word/nonword difference inexperiment was in tUde than that compar-able experiments with ish materials. In the wi thSerbo-Croatian the difference was 11 compared to the differ-ence commonly for is on the order of 10 orless. Nonword letter of randomly selected consonantsare considerably more structure of English words than theyare like the internal words. Structurallying the difference between words and (all-consonant) nonwords isSerbo-Croatian than it is in ish@

To summarize, evidence hasthe Serbo-Croatianthe English orthographySerbo-Croatianwhile the ish

271

REFERENCE NOTE

Kostic, Dj. Frequency of of words in Serbo-Croatian.Unpublished paper, Institute of Experimental Phonetics and Speech Pathol­ogy, Belgrade, Yugoslavia, 1965.

REFERENCES

Baddeley, A., Conrad, R., & Thompson, W. Latter structure of the Englishlanguage.. Nature, 1960, 186, 414-416.

Baron, J. The word superiori ty effect. In W" K. Estes (Ed.), Handbook oflearning and cognitive processes VI. Hillsdale, NJ: Erlbaum, 1978.

Baron, J. t & Strawson, C. Use of orthographic and word-specific knowledge inreading words aloud. Journal of Experimental Psychology: HumanPerception and Performance, 1976, ~, 386-393.

Brand, J. Classification without identification in visual search. QuarterlyJournal of Experimental Psychology, 1971, 23, 178-186.

Carr, T. H., Lehmkuhle, S. W., Kottas, B., Astor-Stetson, E. C., & Arnold, D.Target position and practice in the identification of letters in varyingcontexts: A word superiority effect. ! Psychophysics, 1976,.l2., 412-416.

Chomsky, N. Phonology and reading. In H. Levin & J. P. Williams (Eds.),Basic studies on reading. New York: Basic Books, 1970.

Col theart , M., Besner, D., Jonasson, J. T., & Davelaar, E. Phonologicalencoding in the lexical decision task. Quarterly Journal of Experimental

1979, 11, 489-508.letter ---frequency and

ished doctoral

Engel, Go R. On the functional relationship between word identification andletter probability. Canadian of Psychology, 1974, 28, 300-309.

Estes, W. K. The locus of inferential and perceptual processes in letteridentification o of 1 1122-145.

Gibson, E. J., & Levin, H. Cambridge, Mass.: TheMIT Press, 1975.

Ingling, No Categorization: A rapid information processing.

o

decision

serial visualfeature extraction.

in visual1972, 1

in&

Let ter detectionword

as a letter or a dJonides, Jo, & Gleitman, H.

location without289-298.

&

272

&

structure and its relation to75-105.

The Hague: Mouton, 1970 ..of Verbal

__~ Psychology, 1970, 1,

Identification and categorization in visual search ..~~ , 1971, 648-657 ..

Lukatela, G., Savie, M., ognjenovic, P.. , & Turvey, M.. T. On the relationbetween processing the Roman and the Cyrillic alphabets: A preliminaryanalysis with bi-alphabetical readers.. Language and Speech, 1978, ~,

113-141 ..Lukatela, G., & Turvey, M. T.. Some experiments on the Roman and Cyrillic

alphabets of Serbo-Croatian.. In J .. Kavanaugh & R.. Venezky (Eds .. ),Orthography, reading and Bal timore, Md .. : University ParkPress, 1980.

Massaro, D. W. Letter information and orthographic context in word percep-tion.. Journal of Experimental Psychology: Human PerceptionPerformance, 1979, 5, 595-609.

Michaels, C.. F., & Turvey, M. T.. Central sources of visual masking: Indexingstructures supporting seeing at a single, brief glance.. PsychologicalResearch, 1979, ~,1 1..

Purcell, D.. G.. , Stanovich, K.. E., & Spector, A.. Visual angle and the wordsuperiority effect. Memory! Cognition 1978, 6, 3-8.

Reicher, G. M. Perceptual recognition as a function of meaningfulness ofstimulus material.. of 1969, .§l, 275-280.

Staller, J. D., & Lappin, J .. S.. Word and nonword superiority effects in aletter detection task. Perception! Psychophysics, 1979, ,47-54~

Tomie, T. StatistiCSka analiza srpskohrvatskog teksta pomocu racunara[statistical analysis of Serbo-Croatian text by computer]. Institut zajezike i knji~evnost u Sarajevu, Odelj e za jezike p Posebna izdanja,1978, 4, 221-235, Sarajevo ..

Venezky, R.- English orthography: Itssound. Reading Research ""'-:'-.....,.".__,J!..

Venezky, R. The structure ofWhaley, C. P..--Word-nonword- _...::::::..._-

and Verbal BehaVior, 1978, 17, 143-154 ..Wheeler, D.. D. Processes in wor~recognition@

59-85 ..Whi te, M. J.

273