content 2001 content2001tn

8/3/2019 Content 2001 Content2001tn

1/23

Journal of Memory and Language 45, 177199 (2001)doi:10.1006/jmla.2000.2775, available online at http://www.academicpress.com on

0749-596X/01 $35.00Copyright 2001 by Academic Press

All rights of reproduction in any form reserved.

177

Boundaries versus Onsets in Syllabic Segmentation

Alain Content

Laboratory of Experimental Psychology, Universit Libre de Bruxelles, Brussels, Belgium

Ruth K. Kearns

MRC Applied Psychology Unit, Cambridge, United Kingdom

and

Uli H. Frauenfelder

Laboratory of Experimental Psycholinguistics, Universit de Genve, Geneva, Switzerland

This study investigated the explicit syllabification of CVCV words in French. In a first syllablereversal exper-iment, most responses corresponded to the expected canonical CV.CV segmentation, but a small proportion in-cluded the intervocalic consonant in both the first and second syllables, a result previously interpreted for Englishas indicating ambisyllabicity. Two further partial-repetition experiments showed that listeners systematically in-clude the consonant in the onset of the second syllable, but also often include it in the offset of the first syllable.In addition, the assignment of the intervocalic consonant to the first and second syllables was differentially sensi-tive to the sonority of the consonant and to its spelling. We argue that the findings are inconsistent with the tradi-tionally held boundary conception and instead support the view that distinct processes are involved in locating the

onsets and the offsets of syllables. Onset determination is both more reliable and more dominant. Finally, we pro-pose that syllable onsets serve as alignment points for the lexical search process in continuous spoken wordrecognition. 2001 Academic Press

Key Words: syllable; syllabification; continuous speech recognition; spoken word recognition; speech percep-tion; lexical segmentation.

Researchers have long considered the sylla-ble to be an important structural unit in lan-guage perception and production. The syllablehas been invoked as a key to the invariance

problem in phonetic perception because it is

considered the natural domain of coarticulationphenomena (Cole & Scott, 1974; Liberman &Studdert-Kennedy, 1978; Massaro, 1974), itplays a role in some leading models of speech

production (Levelt, Roelofs, & Meyer, 1998;Levelt & Wheeldon, 1994), it has been arguedto constitute a primitive in the evolution of lan-guage (MacNeilage, 1998), it has been shown tobe more easily available to conscious metalin-guistic manipulation than other linguistic unitssuch as phonemes (e.g., Alegria, Pignot, &Morais, 1982; Liberman, Shankweiler, Fisher,& Carter, 1974; Morais, Bertelson, Cary, & Ale-

gria, 1986; Morais, Cary, Alegria, & Bertelson,1979), and it has been proposed as a potentialprocessing unit in visual word recognition (Car-reiras, Alvarez, & De Vega, 1993; Prinzmetal,Treiman, & Rho, 1986; Spoehr & Smith, 1973;Taft, 1979).

Somewhat paradoxically, despite the largeamount of research that the syllable has gener-

We thank Sverine Perrenoud for assistance with Experi-ment 1; Nick Clements, Nicolas Dumay, Ronald Peereman,and Niels Schiller for useful discussions and comments onprevious drafts; and Ccile Fougeron for help with the stim-ulus analyses. Data from Experiments 1 and 2 were pre-sented at the VIIIth Conference of the European Society forCognitive Psychology in Rome, 1995. This research wassupported by grants from the FNRS, Switzerland (1114-039553.93 and 1113-049698.96); the Direction gnrale de

la Recherche scientifiqueCommunaut franaise de Bel-gique (A.R.C. 96/01-203); and Academic Research Collabo-ration Programmmes between the British Council, theC.G.R.I. (Belgium), and the Swiss F.N.R.S.

Address correspondence and reprint requests to AlainContent, Laboratoire de Psychologie Exprimentale, ULB-LAPSE CP191, Avenue F.D. Roosevelt, 50, B-1050 Brux-elles, Belgique. Fax: 32-2-650-2209. E-mail: [email protected].


2/23

178 CONTENT, KEARNS, AND FRAUENFELDER

ated and its ubiquitous role in psycholinguistictheorizing,thereisnoconsensusabouttheprinci-plesorrulesthatdescribehowaphoneticstringissegmented intosyllables.Although listenersgen-erally agree on the number of syllables in an ut-

terance, their syllabification choices do not al-ways concord. The research reported here startsfrom this observation. Our aim is to investigateFrench syllabification by examining speakersexplicit segmentation preferences in metalin-guistic manipulation games. Typically in suchexperiments,participants hearan input sequence,which they transform according to specific in-structions. For instance, in one common experi-

mental technique (Treiman & Danis, 1988a),participants are required to repeat the syllablesmaking up bisyllabic words in reverse order.

The paradox outlined above is also found inthe linguistic literature. The syllable has becomeincreasingly important in phonology since theearly 1970s (e.g., Hooper, 1972), and mostphonological theories now recognize the sylla-ble as a fundamental unit. Several arguments

justify the use of the syllable as a primitivein phonological descriptions (Hooper, 1972;Selkirk, 1982). Syllables constitute the naturaldomain of many phonological processes, so thatphonological regularities or alternations receivesimpler and more economical descriptions whenthe syllable environment is specified. The pro-posal that syllables have internal structure, char-acterized in terms of hierarchical constituents,

provides an elegant and parsimonious accountof a number of language-universal facts aboutphonotactics. Finally, many scholars have ar-gued that syllables are indispensable for the de-scription of suprasegmental phenomena such asstress and tone (see Blevins, 1995).

Despite the central status of the syllable incontemporary phonology, few, if any, languageshave received an accepted and complete account

of syllabification. In a review of the syllabifica-tion of French, Laeufer (1985) contrasted twogeneral approaches to syllabification, based ei-ther on language-specific distributional con-straints or on universal and abstract phoneticproperties of segments. She illustrated these ap-proaches by comparing five existing proposalsand showed that they disagree on the syllabifica-

tion of a large proportion of biconsonantal clus-ters. Similarly, in a discussion of French syllabi-fication, Laks (1995, p. 54) reiterated the com-mon claim that only syllable quantity isunambiguous: In French, as in most languages,

the location and the perception of the syllabicnuclei is unproblematic, while precise locationof the syllable boundaries is not. Analogous is-sues have been raised for English. For instance,Treiman and Danis (1988a) introduced their re-search as follows: with a word like demon,for instance, is the /m/ part of the first syllable,part of the second syllable, or part of both? Dif-ferent theories of syllabification offer different

answers (p. 87).One complicating factor is that some phono-logical descriptions assume that an intervocalicconsonant may belong to the preceding syllableas well as to the following syllable, as illus-trated in the latter citation. This double affilia-tion is referred to as ambisyllabicity. Kahn(1980) popularized this notion to account for di-verse phonological phenomena in the pronunci-

ation of English and specified the conditionsunder which an intervocalic consonant shouldbe considered ambisyllabic. Thus, in theseanalyses, some segments are not uniquely as-signed to one syllable.

The linguistic notion of ambisyllabicity raisessome important issues about the psycholinguis-tic process of syllabic segmentation. Psycholin-guists have generally taken ambisyllabicity in

English to imply that syllable boundaries areambiguous and hence hard to locate. However,this interpretation presupposes that listeners syl-labify speech by locating the boundary betweentwo syllables. Since ambisyllabicity implies thatthe offset of the first syllable does not coincidewith the onset of the second syllable, as indi-cated by common notations such as CV[C]V, itappears logically incompatible with this con-

ception. This incompatibility was pointed out bySelkirk (1982, p. 355), who noted: Clearly, am-bisyllabicity, if it exists, would provide a furtherargument against the boundary approach to thesyllable: a syllable boundary cannot be simulta-neously before and after some segment of thestring (see de Cornulier, 1978, for similar re-marks).


3/23

SYLLABLE SEGMENTATION 179

In this article, we propose an alternative to theboundary approach in which the requirementthat the offset of one unit corresponds to theonset of the next is abandoned. Specifically, wepresent a view of syllable segmentation in

which locating syllable offsets and onsets con-stitute separate operations. We first argue that, ingeneral, determining beginnings and endings ofconstituents in any input that unfolds in timemay rely on different sensory cues and mentalprocesses. Then, we suggest that the existingsyllabification studies provide preliminary evi-dence in support of such a view.

With speech, as with other types of sequential

sensory input, one function of segmentation isto assign successive packets of proximal sen-sory information to the distal events to whichthey belong. This general function can be de-composed into two operationsdetermining theend of one event and determining the beginningof the following. In the boundary view, thesetwo operations are conflated. This may well beappropriate for many perceptual inputs, when

the distal events are strictly sequential. Indeed,if the two events are in strict succession and thecorresponding information does not overlap intime, any cue indicating the ending of the firstcan also be taken as informative about the be-ginning of the second.

However, the coupling of onset and offset de-cisions is not a logical necessity. In fact, thereare cases in which a decision can be made about

the beginning of a new event without necessar-ily making a commitment about the ending ofthe previous event. In the case of parallel oroverlapping transmission of information, thecues that indicate the beginning of the new eventdo not correspond to the cues indicating the endof the previous one. An example can be found insyntactic parsing, in which the creation of a newsubconstituent (e.g., a prepositional phrase or a

center-embedded relative clause) does not re-quire closure of the dominating constituent.Regarding syllabification, we thus propose

that locating onsets and locating offsets consti-tute distinct operations, and we expect that theresults of the two operations do not always co-incide. We further assume that when there is aconflict between onset and offset preferences,

the onset decision dominates because it consti-tutes a primary and more fundamental process.We henceforth call this view the onset hypoth-esis.

The primacy of onsets is motivated by general

functional considerations related to the sequen-tial nature of the speech signal. By definition,onsets signal the beginning of a new event,which may require the immediate allocation ofattentional or memory resources (Mattys, 1997).In contrast, decisions about the offset of anevent do not have the same urgency and mayeven benefit from being delayed to resolve localambiguity in the input. Our argument about

onset primacy is in line with influential propos-als in the spoken word recognition literature.Most current models attribute a special status toword onsets (Frauenfelder & Peeters, 1990;Marslen-Wilson, 1987; McClelland & Elman,1986). Moreover, the Metrical SegmentationStrategy (Cutler & Norris, 1988) assumes thatthe listener uses strong syllable onsets to initiatelexical searches. In both instances, the need to

process sensory information as it arrives givesonsets a privileged status.The theoretical assumptions outlined above

have direct implications for the interpretation ofmetalinguistic syllabification tasks, which weclassify into three types. For some tasks, onlyonset or offset determination is relevant. One in-stance is the part-reduplication task used by Fal-lows (1981), in which participants are instructed

to reduplicate either the first or the second partof the words. Presumably, only offset determi-nation is relevant to deciding where the first partends, and only onset determination is relevant todecide where the second part of the word starts.Thus in the present framework, part-reduplica-tion appears to provide a relatively direct indexof the result of each operation, and it shouldmaximize the putative differences between

onset and offset decisions.Tasks of the second type require participantsto combine the results of the two operations.Perhaps the clearest instance is the pause-inser-tion task (Gillis & De Schutter, 1996), in whichparticipants have to repeat the spoken input andinsert a pause at the syllable boundary. The lo-cation of the pause depends on both the offset


4/23


of the first part and of the second part.The third type of task, illustrated by the sylla-

ble-reversal paradigm, requires participants toproduce both parts of a word in temporal prox-imity. The task does not strictly require that par-

ticipants determine a unique boundary, sincethe offset of the first response part can in princi-ple differ from the onset of the second part.However, we assume that participants tend toobey an implicit response-dependency con-straintconsisting in avoiding the repetition of aphonemic segment. This constraint may maskpotential dissociations between onset and offsetdetermination. Given our assumption that onset

decisions dominate, the response-dependencyconstraint should lead offset decisions to con-form more often to onset decisions.

Some evidence from metalinguistic segmen-tation experiments may be interpreted as sup-porting the uncoupling of onset and offset deci-sions. In the experiments most directly relevantto our research, Treiman and Danis (1988a)used the syllable inversion task with bisyllabic

words varying along several dimensions: stresspattern (either initial or final stress), intervocalicconsonant class (liquids, nasals, or obstruents,i.e., plosives and fricatives), and orthography(single or double consonants). They observedthat American listeners did not syllabify thesewords consistently. The intervocalic consonantwas sometimes placed in the first syllable(MELON, /mln//ln.m/) and sometimes in

the second (/mln//n.ml/). In addition,they observed responses which they called 1-2, where the consonant was placed in both syl-lables (/mln//ln.ml/). These ambisyl-labic responses1 were more frequent withliquid and nasal consonants than obstruents, fol-lowing the sonority hierarchy. An additionalfactor affecting participants responses was thespelling of the intervocalic consonant. Partici-

pants were much more likely to give ambisyl-labic responses when the consonant was spelledwith two letters than when it was spelled with asingle letter (e.g., COMMA vs LEMON). In line

with our view, ambisyllabic responses also oc-curred, to some extent, for words that from alinguistic point of view are not considered to in-clude a phonologically ambisyllabic consonant(i.e., words with stress on the second syllable).

Similar observations have been described forEnglish with other metalinguistic tasks (Der-wing, 1992; Fallows, 1981).

The observation of ambisyllabic responses isnot restricted to English. Schiller, Meyer, andLevelt (1997) tested Dutch participants with thesyllable reversal task. For items with a shortvowel in the first syllable (e.g., LETTER, /ltr/and LICHAAM, /lIxa:m/), which are considered

ambisyllabic in phonological analyses of Dutch,participants generally assigned the intervocalicconsonant to both syllables (/tr.lt/ and/xa:m.lIx/). Gillis and De Schutter (1996), usinga variant of the pause-insertion technique, testedpreliterate and literate Flemish children and alsoobserved about 30% ambisyllabic syllabifica-tion for words in which the first syllable wasstressed and contained a short vowel. A small

percentage of ambisyllabic responses was alsoobtained for words with long vowels in the firstsyllable and second-syllable stress, for which nophonological ambisyllabicity is posited.

Unlike for Germanic languages, ambisyllab-icity does not play a major role in phonologicaldescriptions of French. It is only invoked in veryspecific contexts and it is not clear whether itmeans double affiliation or ambiguous affilia-

tion. Thus, according to Laks (1995, p. 54),Ambisyllabicity, mainly in heavy internal con-sonantal clusters such as fricative-stop-liquid/str/ but also in fricative-nasal /sm/, comes intoplay. However, some evidence of ambisyllabicresponses was recently obtained with singletonintervocalic consonants. Beaudoin (1998) useda task in which participants matched FrenchCVCV words to three alternative orthographi-

cally represented segmentations. The main aimof the study was to compare Canadian Frenchand English native speakers with second lan-guage learners. About 20% ambisyllabic re-sponses were reported for monolingual Frenchspeakers tested on French stimuli.

In summary, there is already some evidencefor ambisyllabic responses in several languages.

1In line with common usage, we henceforth use the termambisyllabic responses as a descriptive label for Treimanand Danis (1988a) 1-2 responses in which participants as-sign a given consonant to both the first and second syllable.


5/23


Overall, the proportion of ambisyllabic re-sponses is relatively low (except when doubleletters are found in the orthography, e.g.,Treiman & Danis, 1988a). Such responses areobserved both for stimuli that phonological

analyses assume to be ambisyllabic, but also, ina smaller proportion, for stimuli which do notadmit an ambisyllabic analysis on linguisticgrounds.

The occurrence of ambisyllabic responsesprovides some tentative support for our view ofsyllabification. Indeed, taken at face value, am-bisyllabic responses reflect the fact that theonset of the second syllable does not correspond

to the offset of the first. However, the observa-tion of ambisyllabic responses is not in itself asufficient reason to abandon the simple bound-ary hypothesis. For instance, one could arguethat ambisyllabic responses are due to boundaryambiguity. Thus, when participants are unsurewhere to place the syllable boundary, they mightalternate more or less randomly between twopotential solutions and produce ambisyllabic re-

sponses.Several tests may help differentiate betweenthe two conceptions of syllabic segmentationoutlined above. In the boundary view, any factorinfluencing offset decisions should similarly af-fect onset decisions, since onset and offsetpoints represent two faces of the same coin. Incontrast, the onset hypothesis allows for dissoci-ations in the effect of different factors on onset

and offset decisions. One factor that has beenshown to influence syllabification is the sonorityof the intervocalic consonant. The boundaryview predicts that sonority should influenceonset and offset decisions equally. If sonority af-fects the placement of a unique boundary, an in-creased number of first-syllable responses of theCVC type for more sonorous consonants shouldbe mirrored by a corresponding decreasing ten-

dency to include the consonant at the onset ofthe second part.A second test builds on the analysis of task

demands presented above and involves compar-ing performance in a syllable reversal task toperformance in partial repetition tasks. Unlikethe latter, which we argued provides a directmeasure of either onset or offset preferences,

the reversal task requires participants to produceboth parts of the stimulus in temporal proximityand is thus subject to the response-dependencyconstraint. Hence, in such a task, participantsshould syllabify more often according to the

dominant onset location. Thus, the two syllabifi-cation accounts make different predictionsabout the effect of task requirements on syllabi-fication. The onset hypothesis predicts little orno influence of task on the distribution of onsetdecisions, whereas offset decisions should varyaccording to the response-dependency con-straint. According to the boundary hypothesis,the task manipulation should not differentially

influence the pattern of syllabication responses.In the present experiments, we examined thesyllabification of French words and manipu-lated sonority, spelling, and task requirementsin order to test these two hypotheses. To test theonset hypothesis under the most stringent con-ditions, we used a type of experimental stimu-lussingleton intervocalic consonants inFrenchfor which all linguistic analyses un-

equivocally assign the consonant to the secondsyllable. In our view, ambisyllabic responsesreveal general, language-independent propertiesof syllabic segmentation processes and are notnecessarily caused by specific phonological andrhythmic characteristics of languages. If onsetand offset decisions indeed correspond to dis-tinct operations, we expect to observe ambisyl-labic responses even when they do not corre-

spond to phonological ambisyllabicity. Wecompared performance in the syllable-reversaltask, in Experiment 1, with a partial repetitiontask in Experiment 2. Experiment 3 used aslightly modified methodology to assess in amore direct manner the influence of responsedependency.

EXPERIMENT 1

Experiment 1 tested the influence of two fac-tors, consonant category and spelling, on onsetand offset determination. To that end we exam-ined the performance of French listeners in asyllable-reversal task using words with a singleintervocalic consonant.

The experiment was designed to provide a di-rect comparison with Experiment 2 of Treiman


6/23


and Danis (1988a), which investigated the syl-labification of analogous words in English. Asmentioned above, Treiman and Danis found thatboth consonant sonority and spelling affectedsyllabification. A replication of this reversal ex-

periment in French provides a test of the gener-ality of the role of sonority and spelling in syl-labification. However, since we wanted to assesswhether consonant category and spelling differ-entially affected onset and offset decisions, first-syllable and second-syllable responses werescored separately for the inclusion or exclusionof the intervocalic consonant.

Method

Stimuli. The experimental items were 80monomorphemic bisyllabic words with a singleintervocalic consonant. They were divided intofour groups that varied in the nature of the inter-vocalic consonant. The consonant could be aliquid (/l/ or /R /), a fricative (/s/ or /f/), a nasal(/n/ or /m/), or a plosive (/p/, /t/, or /k/). In halfof the words within each group the intervocalic

consonant was spelled either by a single letter orby two nonidentical letters forming a singlegrapheme (e.g., BARON and SAPHIR), and in theother half it was represented by two identicalletters (MARRON and SUFFIRE). The stimuli arelisted in the Appendix. In addition to the testitems there were 80 fillers with two intervocalicconsonants (e.g., BALCON and JOURNAL). Be-cause the syllable boundary is usually placed

between the two consonants in such words, theirinclusion would prevent participants from al-ways segmenting after the first vowel.

The stimuli were randomized, divided intofour blocks, and recorded by a native speaker ofFrench onto digital audiotape, along with fivemonomorphemic warmup items (JARDIN, SAPIN,GALERIE, BOUTIQUE, and BOLET). Stimuli on thetapes were separated by approximately 3 s of

silence.In order to verify that the spelling manipula-tion was not confounded with phonetic varia-tions, an expert phonetician measured theduration of the two vowels and intervocalicconsonant in the 80 experimental stimuli.Analyses of variance were run with spellingand consonant category as factors. While there

were significant differences across consonantcategories on first vowel duration and conso-nant duration, no hint of any difference relatedto spelling was observed.

Participants. Twenty-four students at the Uni-

versity of Geneva participated. In this and thefollowing experiments, all participants were na-tive speakers of French without hearing deficits,and they received course credits for their partic-ipation.

Procedure. The procedure and instructionswere taken from Treiman and Danis (1988a,Experiment 2). The task was presented as a lan-guage game in which participants were required

to reverse the two parts of the word they hadjust heard. The experimenter gave some exam-ples where the syllable boundary coincidedwith a morpheme boundary (if I say CREVE-COEUR, you should say COEUR-CREVE; if I sayTRICYCLE, you should say CYCLE-TRI; if I sayMALSAIN, you should say SAIN-MAL). Partici-pants then received the five warm-up items.They were not told to respond quickly, but they

were constrained to respond before the presen-tation of the next stimulus on the tape. At theend of the warm-up, participants were givenfeedback concerning their responses. For exam-ple, they were reminded that all sounds in theword had to be repeated when the word was re-versed. To ensure that participants did not relyon formally learned rules of syllabification, noexplicit reference to syllables was made in the

instructions. Responses were recorded for laterverification and also transcribed online by theexperimenter. The order of presentation of thefour blocks was randomized for each partici-pant.

After the reversal task was completed, theparticipants were asked to write down each ofthe 80 test words as the experimenter read them.This was done to check that participants knew

the standard spelling of the intervocalic conso-nant.

Results

The responses were analyzed in two differentways. First, we examined the responses for thefirst and the second syllable of the stimuli sep-arately, each being coded in terms of whether it


7/23


included the intervocalic consonant or not.Thus, responses corresponding to the first syl-lable of the stimulus (henceforth called first-syllable responses) were coded as CV or CVCindependently of what was produced for the

other syllable (e.g., for the word PALAIS, /pal/,/pa/ vs /pal/, respectively). Similarly, second-syllable responses were coded as CV or V (/l/vs //). Responses in which one or morephonemes were incorrectly reproduced werecoded as errors.

Second, to compare our results with previousstudies, we also coded each combined responsein terms of the assignment of the intervocalic

consonant. Thus, as in the analyses conductedby Treiman and Danis, responses were scored as1 if the consonant was placed in the first sylla-ble (PALAIS, /pal//.pal/), 2 if the conso-nant was placed in the second syllable(/(pal//l.pa/), and 1-2 if it was placed inboth syllables (/pal//l.pal/).

In all experiments, items that yielded errorson more than 20% of the trials were removed

from the analyses, as were participants forwhom 10% or more of responses were errors.The word BALLET was removed as it has a hom-ophone (BALAI) with a different spelling of theliquid. In addition, if a participant spelled theintervocalic consonant of a word incorrectly,his/her response for that item was removedfrom the analyses. Four items (MANEGE,GAMMA, COMMERE, and RACCORD) and two par-

ticipants were discarded based on the error cri-teria. Table 1 shows the mean proportions of re-sponses (by subjects) according to the twocoding schemes.

In the following analyses we used the propor-tion of CV responses for the first syllable andfor the second syllable as the dependent vari-able. As can be seen in Table 1, CV and CVC re-sponses for the first part and CV and V re-

sponses for the second part account for virtuallyall of the data, so that the complementary analy-ses of the proportion of CVC first-syllable re-sponses and V second-syllable responses wouldbe redundant. We used the CV responses be-cause they correspond to the canonical syllabifi-cation (CV.CV). Two analyses of variance wereconducted, one with subjects as the random fac-

tor (F1) and one with items as the random factor(F2). The means were calculated by subject andby item as described above. An arcsine transfor-mation was then performed in order to deal withthe problem of unequal variance across experi-

mental conditions. The independent variableswere the response part (first syllable or secondsyllable), the type of intervocalic consonant(liquid, nasal, fricative, or plosive), and thespelling of the word (one letter or two).

Overall, the responses reflect the syllabifica-tion agreed upon by phonologists. Thus, for thesecond syllable, 97.6% of the responses beganwith the intervocalic consonant. For the first

syllable, 83.8% of responses were the expectedCV response. The difference between first- andsecond-syllable responses was significant[F1(1, 21) 70.84, p .0001; F2(1, 67) 130.69, p .0001]. The effect of consonantcategory was significant [F1(3, 63) 10.81,p .0001; F2(3, 67) 4.48,p .01], as wasthe effect of spelling [F1(1, 21) 9.94, p .005; F2(1, 67) 6.83,p .05], but both were

qualified by interactions with response part[F1(3, 63) 25.33, p .0001; F2(3, 67) 8.60, p .0001 for consonant category re-sponse part, F1(1, 21) 9.01, p .01; F2(1,67) 2.61, p .11 for spelling by responsepart]. No interaction was observed betweenconsonant category and spelling (both Fs 1),and the three-way interaction was significantby subjects only [F1(3, 63) 3.42, p .05;

F2(3, 67) 1.34, ns]. Separate analyses onfirst- and second-syllable responses were run toanalyze the effect of consonant category andspelling in more detail.

First-syllable responses. As is shown in Table1, fewer CV responses were observed for themore sonorous consonants [F1(3, 63) 21.79,p .0001; F2(3, 67) 7.88,p .0001]. Posthoc Tukey/Kramer HSD tests across item sets

with a .05 significance level confirmed the dif-ference between liquids and obstruents (frica-tives as well as plosives). Slightly fewer CV re-sponses were obtained for items in which theconsonant was represented by two letters thanfor the items in which the consonant was repre-sented by one letter [F1(1, 21) 11.86, p .005; F2(1, 67) 6.47,p .02]. No interaction


8/23


of consonant type and spelling was observed.Second-syllable responses. There were no

significant effects in this analysis. Subjects con-sistently assigned the intervocalic consonant tothe second syllable, independently of the conso-nant type and of the orthographic representationof that consonant.

Additional analyses were based on the com-bined responses, using the proportion of ambi-syllabic 1-2 responses as the dependentvariable. Overall, the mean proportion of ambi-syllabic responses was 13.8%. Predictably, theANOVAs confirmed the effect of consonanttype [F1(3, 63) 25.47, p .0001; F2(3,67) 9.91,p .0001] and spelling [F1(1, 21) 10.83,p .005; F2(1, 67) 5.09,p .05].

Post hoc Tukey tests indicated a significantlyhigher proportion of ambisyllabic responses forliquids than for fricatives or plosives, withnasals intermediate. The interaction betweenspelling and consonant type was significant inthe subject analysis only [F1(3, 63) 2.86,p .05; F2(3, 67) 1.02, ns], apparently re-flecting the fact that the influence of spellingwas particularly marked for nasals.

Discussion

As expected, the dominant response patterncorresponded to the canonical CV.CV syllabifi-cation. However, the data were not totally ho-mogeneous. About 14% of responses for thefirst syllable were of the CVC type, and theseresponses were always accompanied by CV re-

sponses for the second syllable. These syllabifi-cation choices substantiate the claim made inthe introduction that ambisyllabic responsesoccur even for stimuli that do not admit an am-bisyllabic analysis on phonological grounds. Inline with the onset hypothesis, separate analysesof first- and second-syllable responses showed

that the influence of consonant category andspelling was limited to first-syllable responses.An additional aim of the present experiment

was to compare the syllabification of singletonintervocalic consonants by French speakers withthat by American English speakers. Table 2presents the equivalent data from Treiman andDaniss Experiment 2. In order to compare theirresults to ours, only the data for words with

stress on the second syllable are shown, as all bi-syllabic words in French have stress on the sec-ond syllable. The proportions of CV and CVCfirst-syllable responses and CV and V second-syllable responses were calculated fromTreiman and Daniss Table 4. For both Frenchand English, only first-syllable responses varysubstantially with sonority, and the effects ofsonority are qualitatively similar. One difference

between the two languages concerns the effectof spelling, which is much larger in English.An-other minor difference is that the predominanceof CV second-syllable responses appearsslightly less marked in English than in French.Nevertheless, the overall pattern of results isstrikingly similar for French and American Eng-lish words with stress on the second syllable.

TABLE 1

Mean Percentages of Responses of Various Types for Experiment 1

First syllable Second syllable Combined responses

Stimulus type Example CV CVC CV V 1 2 1-2 Other

Liquid 1 letter COLERE 73.3 26.2 98.9 0.7 0.7 73.3 25.6 0.5Liquid 2 letters COLLINE 71.7 27.2 98.3 0.5 0.5 71.7 26.6 1.2Nasal 1 letter PONEY 86.8 9.0 95.2 0.6 0.6 86.8 8.4 4.2Nasal 2 letters COMMANDE 75.8 21.0 96.8 0 0 75.8 21.0 3.2Fricative 1 letter FACILE 94.8 3.7 98.1 0.5 0.5 94.8 3.3 1.5Fricative 2 letters CHIFFON 88.6 8.3 96.9 0 0 88.6 8.3 3.1Plosive 1 letter BOCAL 92.2 7.4 99.5 0 0 92.2 7.4 0.5Plosive 2 letters SUPPORT 87.5 10.2 97.1 0.6 0.6 87.5 9.6 2.3Means 83.8 14.1 97.6 0.3 0.3 83.8 13.8 2.0


9/23


As stated in the introduction, the proportionof ambisyllabic responses may be limited intasks such as pause insertion and reversal be-cause of the response-dependency constraint.According to this view, the proportion of ambi-syllabic responses should increase for syllabifi-cation tasks in which this constraint is weak-ened. An alternative hypothesis is that the smallproportion of ambisyllabic responses does not

reflect listeners syllabification preferences, butrather that it is caused by the complexity of thereversal task. Intuitively, syllable reversal seemsto require two operations: the segmentation ofthe word into two components and the manipu-lation of the order of the components. This intu-ition is supported by the occurrence of a few er-rors which seem to be due to the manipulationcomponent rather than the segmentation compo-

nent (e.g., BOLIDE, /blid//libd/). Develop-mental evidence indicates that reversal tasksmay indeed be harder than other types of met-alinguistic tasks (Yopp, 1988). According to thisinterpretation, ambisyllabic responses shoulddisappear if the task is made less demanding.Thus, in Experiment 2, we used a variant of thepartial repetition task which weakened responsedependency but could still be compared directly

to Experiment 1.EXPERIMENT 2

Experiment 2 used the same stimuli as in Ex-periment 1, but participants were tested on twoseparate occasions 1 week apart. They were re-quired to produce either the first or the secondpart of each stimulus word. This task thus pro-

vided an opportunity to replicate the main find-ings of Experiment 1, using a different syllabifi-cation paradigm.

In addition, Experiment 2 allowed us to as-sess the influence of response dependency onsyllabification responses. If the occurrence ofCVC first-syllable responses in Experiment 1 isdue to the complexity of the reversal task, thensuch responses should disappear in this experi-

ment. Conversely, if these responses are limitedby response dependency, an increase in CVCfirst-syllable responses should be observed.

Method

Stimuli. The stimuli and tapes were the sameas those used in Experiment 1.

Participants. Twenty-eight students at theUniversity of Geneva participated.

Procedure. Participants took part in two ses-sions separated by approximately 1 week. Theywere required to repeat one part of the word in afirst session and the other part of the word in theother session. As in Experiment 1, the task waspresented as a language game and explainedthrough the use of examples and practice items,without referring explicitly to the notion of syl-lable. All participants received both conditions

for all items. Half the participants performed thefirst-part repetition condition in the first sessionand the others performed the second-part repeti-tion condition in the first session. The order ofpresentation of the four experimental blockswas varied across participants.

Following the second session, the experi-menter dictated the words in order to record,

TABLE 2

Mean Percentages of Responses of Various Types in Treiman and Danis (1988a) Experiment 2(Second-Syllable-Stress Words)

First syllable Second syllable

Stimulus type Example CV CVC CV V 1-2

Liquid 1 letter RELATE 78 22 88 12 10Liquid 2 letters BALLOON 35 65 94 6 59Nasal 1 letter CANOE 84 16 92 8 8Nasal 2 letters CONNECT 29 71 93 7 64Obstruent 1 letter GUITAR 93 7 94 6 1Obstruent 2 letters BUFFET 47 54 95 6 48


10/23


as for Experiment 1, whether the participantsknew the standard spelling of the intervocalicconsonant.

Results

The same coding system was used as for thefirst experiment. Two participants for whom 10%of responses or more were errors were removedfrom the analysis. There were no items for whichmore than 20% of responses were errors. As inExperiment 1, analyses of variance were con-ducted on the arcsine-transformed proportions ofCV responses. The independent variables werethe response part (first or second syllable), the

type of intervocalic consonant (liquid, nasal,fricative, or plosive), and the spelling of the word(single or double letter). Mean proportions ofeach type of response are shown in Table 3.

Overall, the statistical analyses closelymatched those of Experiment 1. There weremore CV responses for the second syllable(93.3%) than for the first [54.0%; F1(1, 25) 72.85, p .0001; F2(1, 71) 440.49, p

.0001]. The effect of consonant category wassignificant [F1(3, 75) 15.92, p .0001;F2(3, 71) 12.15, p .0001], as was the ef-fect of spelling [F1(1, 25) 10.06, p .005;F2(1, 71) 3.91, p .052], but both werequalified by interactions with response part[F1(3, 75) 9.39, p .0001; F2(3, 71) 4.61, p .005 for consonant category re-sponse part, F1(1, 25) 19.77, p .0005;

F2(1, 71) 12.16, p .001 for spelling by re-

sponse part]. The interaction between conso-nant category and spelling was significant bysubjects only [F1(3, 75) 5.04, p .005;F2(3, 71) 1.56, ns], and no three-way inter-action was observed (both Fs 1). Separate

analyses on first- and second-syllable responseswere run to analyze the effect of consonant cat-egory and spelling in more detail.

First-syllable responses. Fewer CV responseswere observed for the more sonorous conso-nants [F1(3, 75) 17.06, p .0001; F2(3,71) 13.46,p .0001]. Post hoc comparisonsindicated that the rate of CV responses waslower for liquids than for nasals and lower for

nasals than for fricatives, with plosives interme-diate between nasals and fricatives.Fewer CV responses were produced for two-

letter than for one-letter items [F1(1, 25) 19.36,p .001; F2(1, 71) 13.73,p .001].The interaction of consonant type and spellingwas only significant in the subject analysis[F1(3, 75) 3.27,p .05; F2(3, 71) 0.8, ns],again reflecting the trend for the spelling effect

to be more marked for nasals than for any othertype of consonant.Second-syllable responses. As in Experiment

1, there were no significant effects in this analy-sis. Participants consistently assigned the inter-vocalic consonant to the second syllable, inde-pendently of the consonant type and of theorthographic representation of that consonant.

The influence of the task: Reversal vs repeti-

tion. One objective of this experiment was to

TABLE 3

Mean Percentages of Responses of Various Types in Experiment 2

First syllable Second Syllable

CV CVC CV V 1-2

Liquid 1 letter 45.3 54.7 90.5 9.5 48.1Liquid 2 letters 35.6 64.4 95.3 4.7 61.7Nasal 1 letter 61.8 37.8 92.2 7.4 35.7Nasal 2 letters 38.3 61.7 91.8 8.2 57.8Fricative 1 letter 71.9 28.1 93.6 6.4 26.7Fricative 2 letters 60.2 39.8 93.7 6.3 37.7Plosive 1 letter 63.2 36.8 93.8 6.2 34.8Plosive 2 letters 55.9 44.1 95.2 4.8 41.6Means 54.0 45.9 93.3 6.7 43.0


11/23


compare the syllabification patterns obtained inthe repetition and reversal tasks. We conducted2 2 analyses of variance taking the task (re-versal vs repetition) as one factor and responsepart (first vs second syllable) as the other. The

dependent variable was the arcsine-transformedproportion of CV responses. The reversal datawere taken from Experiment 1 (see Fig. 1).

There was a main effect of response part[F1(1, 46) 130.69, p .0001; F2(1, 74) 272.23,p .0001]. In both tasks, the proportionof second-syllable CV responses was higherthan the proportion of first-syllable CV re-sponses. Hence in both experiments, second-

syllable responses appeared more consistentwith the canonical syllabic segmentation thanfirst-syllable responses. The main effect of task[F1(1, 46) 19.74, p .0001; F2(1, 74) 228.07, p .0001] occurred because, overall,the proportion of canonical responses washigher in the reversal task of Experiment 1 thanin the repetition task of Experiment 2. More im-portantly with regards to our predictions, the in-

teraction was highly significant [F1(1, 46) 11.72,p .005; F2(1, 74) 67.67,p .0001].This reflects the fact that the effect of task was

largely restricted to first-syllable responses.

Discussion

The main finding in this experiment was thedramatic increase in the proportion of CVC

first-syllable responses with respect to Experi-ment 1 (14.1% vs 45.9%). These responses con-stitute a robust phenomenon and cannot be con-sidered an artifact of the reversal task. Theincrease in CVC first-syllable responses, with-out a corresponding change in second-syllableresponses, is in line with the onset hypothesisbut not with the syllable-boundary hypothesis.According to the latter view there should be per-

fect consistency between first-part and second-part responses, since both are determined by thesame criterionthe location of the boundarybetween the two parts. Conversely, if onset andoffset decisions are distinct, the responses forone part might be affected by different variablesthan the responses for the other part. Indeed, inboth Experiment 1 and Experiment 2, the firstsyllable offset choices are variable, whereas the

second-syllable onset decisions remain stable.In both experiments, only offset decisions aresensitive to sonority and spelling.

FIG. 1. Mean proportion of CV first-syllable responses and CV second-syllable responses in Experiment 1(reversal task) and Experiment 2 (repetition task).


12/23


EXPERIMENT 3

Experiment 3 aimed to verify in a more directmanner that the effect of response dependencyis limited to first-syllable responses. This wouldsupport the onset hypothesis by demonstrating

that onset decisions are relatively unaffected bytask variables and that onset decisions dominateoffset decisions. We compared the independentrepetition of either the first or the second part, sothat no response dependency was induced, withthe repetition of the two parts either in the origi-nal (forward repetition) or reversed order (back-ward repetition). We used a slightly differentversion of the reversal and repetition tasks by in-

troducing a sentence frame [Je dis (part 1) etpuis (part 2), I say (part 1) and then (part 2)]since otherwise the forward repetition conditionwould have amounted to plain repetition of theinput.

The experiment had four conditionsrepeti-tion of the two parts in either forward or back-ward order, repetition of the first part only, andrepetition of the second part only. Hence, the de-

sign of the experiment enabled us to compareresponses for each syllable when producedalone or combined with the other. We predictedthat first-syllable CVC responses should vary asa function of response dependency, whereassecond-syllable responses should stay invariant.Thus, the proportion of first-syllable CVC re-sponses should be higher for the repetition ofthe first part only [je dis (part 1)] than for the

repetition of both parts, whatever the order. Anadditional aim was to test participants whospoke a different dialect of French to check thatthe previous findings were not specific to the di-alect spoken in Geneva.

Method

Stimuli. The stimuli and tapes were the same

as those used for the previous experiments.Participants. Forty students at the Universityof Brussels participated in the experiment.

Procedure. Participants had to produce eitherone part or both parts of the stimulus words in asentence frame. Four conditions were used: rep-etition of both parts [Je dis (syllable 1) et puis(syllable 2), I say (syllable 1) and then (sylla-

ble 2)], repetition of both parts in reverse order[Je dis (syllable 2) et puis (syllable 1)], repeti-tion of the first part only [Je dis (syllable 1)],and repetition of the second part only [Je dis(syllable 2)]. Participants were randomly as-

signed to one condition. Within each conditionthe order of presentation of the four experimen-tal blocks was randomized.

Participants were given written instructionspresenting the task as a language game and theyreceived the same examples and practice itemsas in previous experiments. Following the seg-mentation task, the experimenter dictated theexperimental words in order to check partici-

pants knowledge of spelling.Results

As in previous experiments, first-syllable andsecond-syllable responses were coded sepa-rately according to whether they included the in-tervocalic consonant. All other responses werecoded as errors. There were no participants forwhom more than 10% of responses were errors

and no items for which more than 20% of re-sponses were errors. As the error rate was verylow (maximum of 2% in any one condition),these responses are not discussed further. As inprevious experiments, syllabification responsesfor words whose intervocalic consonant hadbeen incorrectly spelled were removed for eachparticipant. In addition, three items (FISSILE,RAFFUT, and BUCCAL) were discarded because

they generated a large number of spelling errors,leaving too few valid observations per group tobe meaningful.

The response dependency effect. The firstanalysis aimed at testing the influence of re-sponse dependency. We predicted that responsedependency should specifically affect first-syl-lable response. Hence, fewer CV first-syllableresponses should be obtained in the first-sylla-

ble-only condition than in the forward-repeti-tion condition, whereas for the second syllablethe proportion of CV responses should remainhigh when comparing the second-syllable-onlycondition with the backward-repetition condi-tion. We conducted 2 2 between-subjectsanalyses of variance with dependency as onefactor (single- vs double-response conditions)


13/23


and response part (first vs second syllable) asthe other. For the double-response conditions,we used the data from the forward conditionfor the first-syllable response and from thebackward condition for the second-syllable re-

sponse. Hence we compared responses for Jedis (syllable 1) and Je dis (syllable 2) eitherfollowed by the other part or not.

As shown in Fig. 2, the results conform to ourprediction. There were fewer CV first-syllableresponses than CV second-syllable responses[F1(1, 36) 14.59, p .0005; F2(1, 75) 32.73, p .0001], and single-response condi-tions tended to produce fewer CV responses

than double-response conditions [F1(1, 36) 2.20,p .15; F2(1, 75) 39.65, p .0001].However, both main effects are essentially dueto the lower rate of CV responses for the firstsyllable in the single-response condition. Theinteraction was significant in the item analysis[F(1, 75) 158.15,p .0001] and just missedsignificance in the subject analysis [F(1, 36) 3.93,p .055]. Pairwise comparisons between

conditions for each response part confirmed thatthere were significantly fewer CV first-syllableresponses in the single than in the double-

response conditions [F1(1, 36) 6.00,p .02;F2(1, 75) 21.01,p .0001]. The slight differ-ence in the opposite direction for second-sylla-ble responses was significant in the item analy-sis only [F1 1; F2(1, 75) 17.15,p .0001].

Effects of consonant type and spelling. Sepa-rate analyses were run on each of the four con-ditions to examine the influence of consonanttype and spelling. As in previous experiments,post hoc Tukey/Kramer tests were used to as-sess the effect of consonant type in more detail(see Tables 46).

In the forward repetition condition, therewere main effects of consonant type [F1(3,

27) 30.98,p .0001; F2(3, 68) 10.64,p.0001] and spelling [F1(1, 9) 9.38,p .05;F2(1, 68) 18.00,p .0001] for first-syllableresponses. The interaction reached significancein the subject analysis only [F1(3, 27) 10.47,p .0001; F2(3, 68) 2.46,p .10]. Post hoccomparisons indicated a lower rate of CV re-sponses for liquids than for any of the otherthree categories. For the second syllable, there

was a main effect of consonant type [F1(3, 27) 4.56,p .01; F2(3, 68) 4.99,p .005]. Post hoc tests indicated that the rate of

FIG. 2. Mean proportion of CV first-syllable responses and CV second-syllable responses in Experiment 3 for the sin-gle-response and double-response conditions.


14/23


CV responses was significantly lower for liquidsthan for nasals and fricatives.

In the backward repetition condition, forfirst-syllable responses, there was an effect ofconsonant type [F1(3, 27) 4.51, p .05;F2(3, 68) 5.75, p .005] and a nonsignifi-cant trend for spelling. Post hoc comparisonsindicated a lower rate of CV responses for liq-

uids and nasals relative to fricatives. No signifi-cant effect was obtained for the second-syllableresponses.

In the single-response conditions, for the firstsyllable, the effect of consonant type [F1(3,27) 5.01, p .01; F2(3, 68) 4.89, p .005] was significant. The effect of spelling ap-proached significance by subjects [F1(1, 9) 3.41, p .10] and was significant by items

[F2(1, 68) 5.62, p .05]. Double-letter

words led to fewer CV responses than single-letter words. According to post hoc tests, fewerCV responses were obtained for liquids thanfor obstruents, with nasals intermediate. Fi-nally, for the second syllable, there was a maineffect of consonant type, significant by itemsonly [F1(3, 27) 2.49, p .10; F2(3, 68) 3.50,p .05] and a main effect of orthography[F1(1, 9) 6.08, p .05; F2(1, 68) 3.88,p .05]. Double-letter words led to more CVresponses than single-letter words. However,none of the post hoc tests reached significance.

In summary, Experiment 3 confirmed that re-sponse dependency is a major determinant offirst-syllable responses. Indeed, the proportionof CVC first-syllable responses was much higher

in the first-syllable-only condition, in which re-

TABLE 4

Mean Percentage of Responses of Various Types in Experiment 3 (Forward Repetition Condition)


CV CVC CV V

Liquid 1 letter 63.5 35.6 86.4 10.6Liquid 2 letters 36.9 63.3 89.7 8.0Nasal 1 letter 90.8 9.2 97.8 2.2Nasal 2 letters 59.6 38.1 94.1 3.6Fricative 1 letter 94.8 3.2 96.9 2.1Fricative 2 letters 78.8 20.1 97.5 0Plosive 1 letter 74.8 24.2 90.7 8.4Plosive 2 letters 73.9 25.1 96.4 2.5Means 71.6 27.4 93.7 4.7

TABLE 5

Mean Percentages of Responses of Various Types in Experiment 3 (Backward Repetition Condition)


CV CVC CV V

Liquid 1 letter 57.6 40.3 79.0 20.1Liquid 2 letters 48.8 50.3 82.4 15.2Nasal 1 letter 66.4 31.8 84.9 14.1Nasal 2 letters 47.9 47.9 79.1 16.6Fricative 1 letter 86.2 12.5 88.5 10.1Fricative 2 letters 69.1 31.3 83.5 16.6Plosive 1 letter 62.7 35.3 74.3 23.7Plosive 2 letters 68.8 28.9 84.3 14.5Means 63.4 34.8 82.0 16.4


15/23


sponse dependency is eliminated, than in theforward-repetition condition, and no such differ-ence was obtained forsecond-syllable responses.

In contrast with previous experiments, smalleffects of consonant category and spelling wereobtained for second-syllable responses, in thesingle-response condition as well as in the for-

ward repetition condition. However these effectswere much smaller than those observed for thefirst syllable and thus do not contradict ourmajor claims.

GENERAL DISCUSSION

In the introduction we presented two oppos-ing views of syllabic segmentation. The bound-ary view assumes that syllabification involves

determining the unique juncture point betweentwo constituents. In contrast, according to theonset hypothesis, determining the initial andfinal edges of syllables constitutes two distinctprocesses, and onset determination prevails overoffset determination.

We proposed two tests of the onset hypothesisby examining the syllabification of singleton in-tervocalic consonants by French native listeners.

The first examined whether onset and offset de-cisions were equally affected by stimulus char-acteristics known to influence syllabification.The boundary hypothesis predicts that any fac-tor influencing offset determination should in-fluence onset determination similarly, whereasthe onset hypothesis allows for differential ef-fects. The second test assessed the influence of

task demands on onset and offset decisions. Wepredicted that second-syllable responses, whichrequire locating syllable onsets, should not beaffected by task variations. In contrast, first-syllable responses, which reflect offset deci-sions, may well be biased by task demands.

In the first test, we manipulated two factors,

the type of intervocalic consonant and itsspelling. In most experimental conditions, clearand significant effects of both factors were ob-served on first-syllable responses (syllable off-set) only. We recently replicated this responsepattern with French children, using a differentstimulus set (Content, Frauenfelder, & Dumay,1999). We adapted the part repetition task usedin the present experiments to test 5- and 8-year-

old children. In both age groups, only first-syllable responses were affected by consonantsonority. In addition, in the literate group, first-syllable responses varied as a function of spelling. Neither factor influenced second-sylla-ble responses.

The second test also provided support for theonset hypothesis. We compared the responsepatterns obtained in the syllable reversal task

(Experiment 1) with those in the repetition task(Experiment 2). A significant interaction be-tween task and response part was observed, dueto fewer first-syllable CV responses in Experi-ment 2. Experiment 3 provided a more straight-forward test of the response dependency effectby comparing single-response and double-response conditions directly. Again, participants

TABLE 6

Mean Percentage of Responses of Various Types in Experiment 3 (Single-Response Conditions)


CV CVC CV V

Liquid 1 letter 36.6 62.4 83.3 14.7Liquid 2 letters 28.0 70.9 85.3 11.4Nasal 1 letter 52.4 46.5 82.5 14.4Nasal 2 letters 35.8 64.2 86.8 13.2Fricative 1 letter 52.8 46.3 92.0 8.0Fricative 2 letters 40.9 59.2 93.0 7.1Plosive 1 letter 47.8 48.1 85.7 14.3Plosive 2 letters 48.7 49.1 93.8 6.3Means 42.9 55.8 87.8 11.2


16/23


were more likely to give CVC first-syllable re-sponses when instructed to produce only thefirst part of the stimuli than when they had toproduce both the first and the second syllables.No such variation was observed for second-

syllable responses.In sum, the results of both tests are consistentwith the view that locating syllable onsets andoffsets obey different constraints. The findingsprovide empirical support for the onset hypothe-sis. In what follows, we discuss the observedpattern of onset and offset segmentation choicesin light of linguistic and psycholinguistic evi-dence. We then consider the implications of the

onset hypothesis for theories of lexical segmen-tation and word recognition in continuousspeech and for the issue of differences and simi-larities in lexical processing across languages.

Regarding syllable onset preferences, the sys-tematic assignment of the intervocalic conso-nant to the onset of the second syllable observedin the present experiments is consistent withgeneral principles of syllabification (e.g.,

Selkirk, 1982; Venneman, 1988; but seeBlevins, 1995; Breen & Pensalfini, 1999, forsome possible counterexamples). According tothe Obligatory Onset Principle (OOP; Hooper,1972) an intervocalic consonant is affiliatedwith the following vowel as long as there is noviolation of phonotactic constraints. This gen-eral preference conforms to cross-linguistic reg-ularities in the distribution of syllable types (see

Blevins, 1995, for a review).According to linguistic descriptions, the OOPmay conflict with other constraints. One suchconstraint, known as the Weight Law (Venne-man, 1988) or Stress Principle, concerns the in-fluence of lexical stress. Syllables bearing pri-mary stress tend to be heavy and include eitherlong vowels or a filled coda position. Anotherphonotactic constraint, sometimes termed the

Branching Rhyme Constraint (Schiller et al.,1997), states that open syllables with short/laxvowels are not allowed. Neither of these gener-alizations applies in French. Open syllable typesclearly predominate (Goldman, Content, &Frauenfelder, 1996), and stress is not lexicallycontrastive and falls on the final syllable in listpronunciation. Given that any singleton conso-

nant used here is a possible word onset, theOOP predicts the systematic affiliation of theconsonant to the onset position, as we observed.

The frequent assignment of the intervocalicconsonant to the onset of the second syllable

has also been observed in most previous studiesof singleton consonant syllabification, althoughthe pattern varies somewhat across languages.In Dutch, both Gillis and De Schutter (1996),with 5- and 8-year-olds, and Schiller et al.(1997), with adults, examined the syllabifica-tion of singleton consonants in bisyllabic wordsand manipulated stress and vowel length. Inboth studies, nearly all responses conformed to

the OOP in all conditions. Thus, Schiller andcoworkers (1997, p. 124, authors italics) con-clude that although stressed syllables attractcoda consonants ... We cannot argue that astressed second syllable takes away the inter-vocalic consonant from the first syllable be-cause all second syllables, stressed or un-stressed, were provided with an onset.Derwing (1992) described a similar tendency

for Swiss German.In contrast, the available evidence in Englishsuggests that stress and vowel quality affect bothfirst-syllable and second-syllable responses. InTreiman and Daniss (1988a) syllable reversalexperiments, the consonant was less often as-signed to the second syllable in first- than in sec-ond-syllable stress words (Experiment 2, 56%vs 91%) and within first-syllable stress words

for short/lax than for long/tense vowels (Experi-ment 3, 61% vs 76%). Similarly, in Derwings(1992) study, in which participants were re-quired to choose between three different syllabi-fications (coda of first syllable, onset of secondsyllable, or ambisyllabic), coda assignment wasthe predominant choice for words with first-syllable stress which contained a lax vowel and aliquid or nasal intervocalic consonant.

Thus it appears that in various languages,conflicts between the OOP and other syllabifica-tion constraints are generally resolved by honor-ing the OOP and assigning the intervocalic con-sonant to the second syllable. The English dataseem to depart from this generalization to somedegree, suggesting that other factors besidesstress and vowel length might play a role. One


17/23


potential factor worth examining in further re-search is vowel reduction, the tendency to neu-tralize vowels in unstressed syllables, which is afrequent phonological process in English but notin Dutch.

Regarding syllable offset preferences, the in-fluence of consonant sonority observed in thisstudy is consistent both with descriptive regular-ities and psycholinguistic evidence. Clements(1992) argued that there is a universal prefer-ence across languages for syllables in which thesonority decreases minimally from the vowelpeak to the right edge. Accordingly, sonorousconsonants, such as liquids and nasals, consti-

tute more preferred codas than less sonorousconsonants such as fricatives or plosives. De-spite the predominant open-syllable structure ofFrench words, the distribution of onset and codaconsonants in the lexicon seems to support thesonority sequencing generalization. We esti-mated the frequency of single onset and singlecoda consonants from the LEXOP database(Peereman & Content, 1999). As shown in Table

7, obstruents are more likely to appear at onsetthan at coda position, and conversely, liquids aremore likely to appear at coda than at onset posi-tion. While more subtle regularities in CV andVC co-occurrence patterns do exist (see, forEnglish, Kessler & Treiman, 1997), it appearsthat the sonority sequencing generalization cap-tures one important phonotactic fact of French.

The preference for more sonorous conso-

nants in coda position is reflected in experi-mental data as well as in some aspects of lan-guage use. Such a trend appeared in Treimanand Daniss (1988a) data. As shown in our re-analysis, for second-syllable-stress words, cleareffects of spelling and sonority were presentfor first-syllable responses. Gillis (1996) also

observed large effects of sonority in both 5-and in 8-year-old children. Other psycholin-guistic evidence, based on speech errors inchildren and aphasic patients (Christman &Depaolis, 1996; Romani & Calabrese, 1998),

word games (Treiman, 1983, 1984; Treiman,1986), and phoneme recombination errors inshort-term memory tasks (Treiman & Danis,1988b; Treiman, Straub, & Lavery, 1994), alsoindicates that more sonorous consonants aremore cohesive with the preceding vowel.

The observation that first-syllable responsesare sensitive to spelling is again consistent withprevious findings (Gillis & De Schutter, 1996;

Schiller et al., 1997; Treiman & Danis, 1988a).However, the effect of spelling was stronger inTreimans experiments than in the present study.One may wonder whether these effects are dueto the contamination of syllabification processesby spelling or hyphenation rules or rather tosubtle differences in pronunciation reflected inthe orthography. We believe that the latter possi-bility is unlikely. Although there is a recent

trend in some dialects of French to pronouncedouble letters as geminates, this type of pronun-ciation is considered very formal, it is generallylimited to some prefixed forms, and is not con-sidered contrastive. Furthermore, in phoneticanalyses of the stimuli used in the present study,we found no trace of geminate realizations oforthographically doubled consonants.

In conclusion, the main results of the present

experiments are in line with previous findings insimilar syllabification experiments as well aswith linguistic descriptions of syllabification.Neither the predominant affiliation of the inter-vocalic consonant with the second syllable northe trend to produce more closed-syllable re-sponses for more sonorous consonants seems

TABLE 7

Relative Frequency of Major Consonantal Classes in Onset or Coda Position

Type count Token count

Consonant category Onset Coda Onset Coda

Obstruents .74 .63 .73 .27Nasals .13 .11 .13 .12Liquids .13 .26 .14 .62


18/23


surprising as such. Perhaps the most interestingfinding is that the latter trend is observed evenwhen everything concurs in assigning the inter-vocalic consonant unequivocally to the secondsyllable. Our results, together with the Dutch

data (Gillis & De Schutter, 1996; Schiller et al.,1997), suggest that the constraints that deter-mine syllable offsets do not systematically influ-ence syllable onset decisions.

The Role of the Syllable in Spoken Word

Recognition

The early discussions about the function ofthe syllable in speech processing focused on the

question of its psychological reality as a per-ceptual unit (Savin & Bever, 1970). However,it has become clear that this question is toovague and needs to be refined. In this regard,Cutler and Norris (1988, p. 114) introduced animportant distinction between segmentation andclassification. They stated: The process of clas-sification is logically distinct from the processof segmentation. Segmentation means making a

division at some point in the signal. Classifica-tion means identifying units occurring in thesignal. In order to classify speech into any se-quence of units (phonemes, syllables, or feet)the recognizer must indeed segment the speechsignal at the boundaries of these units. But thereverse is not true: It is possible to segmentspeech without classifying it.

The distinction has a direct bearing on the

role of the syllable and may help clarify thepresent discussion. Indeed, authors such asMehler and colleagues (e.g., Mehler, Dupoux, &Segui, 1990) have argued that the syllable con-stitutes a primary classification unit for Frenchspeakers. Others, such as Cutler and Norris(1988), have insisted on the role of the syllablein segmentation. The present data are compati-ble with the latter view that syllable onsets help

listeners segment the speech flow. They areharder to reconcile with theories assumingprelexical syllabic categorization.

Mehler, Dupoux, and Segui (1990) assumethat the signal is recoded and categorizedprelexically in terms of syllable-sized units. Intheir view, speech is segmented into elemen-tary units that roughly correspond to the sylla-

ble. [...] Syllabic frames are recognized by abank of syllabic analyzers. [...] Phonemes donot play a direct role in speech perception butare derived from a prelexical code, namely fromthe syllabic frames (p. 255). Perhaps the most

influential argument in favor of prelexical syl-labic classification units comes from the findingof an effect of syllable structure in the fragmentdetection task. In a seminal study, Mehler, Dom-mergues, Frauenfelder, and Segui (1981) re-ported that French listeners were faster to detecttargets which correspond exactly to the first syl-lable of a stimulus word (e.g., ba in balance orbal in balcon) than to detect targets which cor-

respond either to more or less than the first syl-lable (ba in balcon or bal in balance).Follow-up experiments (Cutler, Mehler, Norris,& Segui, 1986) comparing the performance ofEnglish and French speakers tested on bothFrench and English materials revealed a syllableeffect for French speakers on English materials.In contrast, English participants showed no sucheffect for either the French or the English mate-

rials. This finding demonstrates that the effectcan be attributed not just to differences in the re-spective acoustic/phonetic realizations of thewords, but also to differences in the perceptualprocess for the two languages.

The syllable detection paradigm has sincebeen used for various cross-linguistic compar-isons (see, e.g., Kolinsky, 1998, for a review;Tabossi et al., 2000). These data have con-

tributed to the wide acceptance of the notionthat the syllable plays a central role as a classifi-cation unit in the processing of French.Nonetheless, we believe that they do not provideconclusive evidence to support such a claim.Several authors have argued that the syllable ef-fect in fragment detection experiments may notreflect early perceptual processing (Frauen-felder & Kearns, 1996; Kolinsky, 1998) and

might involve postlexical matching/decisionprocesses (McQueen, 1998). In fact, in a re-analysis of the original data, Dupoux (1993)showed that the target/carrier interaction wasonly observed for slower participants. Kolinsky(1998) further noted that the detailed pattern ofreaction times observed in the fragment detec-tion experiment for French does not exactly fit


19/23


with the predictions of a syllabic classificationaccount. In particular, she argued that a largerRT difference than that actually observed wouldbe expected between CV and CVC targets inCV carriers (e.g., ba and bal in balance) be-

cause in the latter condition, responses shouldbe delayed until the second syllable has beenrecognized.

In a set of syllable detection experimentsusing pseudoword target-bearing sequenceswith French-speaking adults, we failed to ob-serve the syllable effect in most of the condi-tions tested (Content, Meunier, Kearns, &Frauenfelder, in press). Moreover, the amount

of signal that participants required in a gatingtask to identify the first vowel and the pivotalconsonant correlated extremely strongly withthe detection latencies to CV and CVC tar-gets, respectively. Such correlations indicatethat the incoming sensory information is usedimmediately as it arrives and suggest that asubsyllabic classification unit mediates thedetection process.

Other techniques have provided further evi-dence of listeners sensitivity to syllabic struc-ture. One technique is based on the manipula-tion of attentional expectations. Pallier (1993)showed that phoneme detection was facilitatedwhen participants were led to expect the targetphonemes to occur in a fixed syllabic position.No such effect was obtained in control condi-tions in which the expectation was based on the

absolute position of the target phonemes. Simi-lar effects of attentional allocation have alsobeen observed in English (Finney, Protopapas,& Eimas, 1996; Pitt, Smith, & Klein, 1998). An-other technique relies on illusory combinationsof fragments of dichotic stimuli. Using CVCVstimuli, Kolinsky, Morais, and Cluytens (1995)examined the probability of illusory percepts re-sulting from the recombination of various parts

of the two inputs. They observed that syllableblends were more frequent than blends involv-ing two adjacent but not tautosyllabic segments.While the findings based on both techniquesclearly demonstrate that listeners are sensitive tosyllabic structure in processing incomingspeech, they do not decide between a syllabicsegmentation and a syllabic classification hy-

pothesis. In sum, we would argue that there islittle unequivocal evidence to support the hy-pothesis that syllables serve as prelexical codingunits in any language.

According to our view, rather than serving as

classification units, syllables are involved insegmentation. More precisely, syllable onsetshelp the listener segment the speech flow. Thisonset segmentation process might serve variousfunctions. As suggested in the introduction,onset segmentation might index the beginningof new events in the continuous speech streamand determine the allocation of resources.Given that syllable onsets and word onsets fre-

quently coincide, a syllable onset segmentationstrategy may also provide a useful heuristic fordetermining potential word beginnings. In thisview, the detection of syllable onsets wouldhelp lexical segmentation by providing the lexi-cal search mechanism with privileged align-ment points in the signal (or its prelexical repre-sentation). According to this heuristic, anysyllable onset is taken to be a possible word

onset.The findings reported here support this viewby showing the robustness of onset decisions inspeakers explicit segmentation choices. How-ever, because they are based on offline metalin-guistic tasks, it is unlikely that they tap directlyinto early perceptual processing. Various au-thors (e.g., Kolinsky, 1998) have warned againstdirectly inferring perceptual processing from

deliberate and conscious decision processes.While a detailed description of the mechanismsdetermining participants behavior in syllabifi-cation experiments is currently not available, weassume, in line with other researchers usingmetalinguistic tasks, that perceptual processesand representations may partially determinethese explicit choices.

Other data from online word recognition

tasks support the hypothesis that syllable onsetsare detected during perceptual processing andhelp determine the initial cohort of lexical can-didates. In a cross-modal semantic priming ex-periment, Vroomen and de Gelder (1997, Ex-periment 3) examined the influence ofalignment between embedded words and theirnonword carriers. They found significant facili-


20/23


tation when the prime word was embedded atthe onset of the carrier (VEL, animal skin, inVELK, as a prime for HUID, human skin) but nofacilitation when it was embedded at the offset(WIJN, wine, in TWIJN, as a prime for ROOD,

red).In a recent word spotting study, Dumay,Frauenfelder, and Content (submitted) assessedthe cost of a misalignment between the onset orthe offset of the target words and the syllableboundary of the target-bearing carriers. Frenchparticipants had to detect words embedded inbisyllabic nonsense carriers such that the onsetof the target could be aligned or not with the

onset of the second syllable of the target-bear-ing sequence (e.g., LAC in /zyn.lak/ and/zy.lak/, respectively). In another condition,the target words were embedded in initial posi-tion to assess the effect of offset misalignment(e.g., LAC in /lak.tyf/ vs /la.klyf/). Based on thesyllable onset segmentation hypothesis, we ex-pected that onset misalignment would createmore interference. In line with our prediction,

the negative effect of a misalignment wasgreater for the onset condition than for the off-set condition, the latter effect failing to reachstatistical significance.

Similar findings have been reported by Mc-Queen (1998) with Dutch listeners. In thatstudy, participants had more trouble detectingwords embedded at the end of nonsense carriersequences when the onset of the words was not

aligned with the syllable boundary (e.g., ROK in/fi.drok/ vs /fim.rok/). The alignment effect wasalso found for words embedded at the beginningof carriers (PIL, in /pilm.rm/ vs /pil.vrm/).However, for both miss rate and response laten-cies, the final misalignment effect was muchsmaller, especially when the influence ofacoustic differences was taken into account. Theclose match between Dumay et al.s (1998) re-

sults in French and McQueens (1998) results inDutch suggests that similar processes are atwork in both languages.

Similarities and Differences in Processing

across Languages

As Cutler and Otake (1996) have argued, apsycholinguistic model of speech processing

must attempt to account for universal character-istics of the human language production andlanguage comprehension system, but at thesame time to explain how language specificstructural features affect the processing of indi-

vidual languages (p. 2). Indeed, the differencesbetween the performance of English and Frenchlisteners in the fragment detection task led Cut-ler et al. (1986) to propose that listeners employlanguage-specific segmentation proceduresbased on the rhythmic properties of their nativelanguage and that a syllable-based access pro-cedure is more appropriate for French than forEnglish.

In contrast, the present study reveals strikingsimilarities between the syllabification of sin-gleton intervocalic consonants by French andEnglish speakers, at least when words with sim-ilar stress patterns (second-syllable stress) arecompared. In Treiman and Daniss (1988a) ex-periments as well as in the present study, themajority of second-syllable responses includethe consonant, and only first-syllable responses

vary substantially with sonority and spelling.The similarity of the results across languagessuggests that both languages share an analogousonset detection process.

One finding in English that appears to contra-dict this claim is the distribution of responsesfor words with first- rather than second-syllablestress. This large variability in onset decisionappears at first to be problematic for our account

since we claim that onset detection is primaryand highly reliable. The contradiction is moreapparent than real, however, if one additionallyassumes that only strong-syllable onsets aresalient and reliably detected. It is precisely thisclaim that Cutler and Norris put forward as theMetrical Segmentation Strategy (Cutler & Nor-ris, 1988). According to this view, lexicallookup is initiated only at the onset of strong

syllables, which contain full vowels.Weak sylla-bles, which are unstressed and contain reducedvowels, do not trigger lexical searches. Most ifnot all of Treiman and Daniss (1988a) first-syllable-stress words had a strong-weak metri-cal pattern, as typical for bisyllabic words inEnglish. Therefore, no salient onset point wouldbe marked for the second syllable. In French, by


21/23


contrast, there is no analogous process of vowelreduction, so even unstressed vowels are fullyrealized and one may assume that all syllablesare strong. Hence, the restriction of the onsethypothesis to strong syllables appears to provide

a parsimonious account of both the English andthe French data and underscores the similaritiesin processing between French and English morethan the differences.

Conclusions

The present study provides evidence againstthe commonly accepted boundary view of syl-labic segmentation. By demonstrating clear dis-

sociations between syllable onset and offset de-cisions, it supports the alternative hypothesisaccording to which listeners rely on syllable on-sets for continuous speech segmentation. We be-lieve that these findings are important becausethey lead to a view of speech perception inwhich syllable onsets provide privileged entryor reference points for segmentation and lexicalaccess. The findings also suggest that there may

be more similarities than previously thought be-tween syllabic segmentation processes in Ro-mance and Germanic languages. Indeed, our re-sults point to the existence of general, perhapseven universal, speech segmentation routines.

APPENDIX

List of Stimuli Used in the Experiments

BARON MARRONCORAN TORRENTFORAIN LORRAINSOURIS NOURRICAROTTE CARROSSEPALAIS BALLETBOLIDE COLLINELILAS SYLLABEDLUGE CELLULE

COLRE POLLENLAMA GAMMACHAMADE MAMMOUTHDOMAINE SOMMAIRECOMTE COMMRERAMEAU COMMANDEMANGE CANNELLECANOT PANNEAUBNIT TENNIS

PHONO TONNEAUPONEY MONNAIE

RCENT DESSEINFACILE BASSINELICITE FISSILESOUCI POUSSIN

BACILLE PASSIFCAF RAFFUTLOUFOQUE BOUFFONTYPHON DIFFUSSIPHON CHIFFONSAPHIR SUFFIRE

POTIN BOTTINVETO GHETTOCHAPELLE RAPPELCOUPOLE SUPPORT

JUPON NIPPONCAPOTE RAPPORTBIQUET TICKETVICAIRE NICKELTACOT RACCORDBOCAL BUCCAL

REFERENCES

Alegria, J., Pignot, E., & Morais, J. (1982). Phonetic analy-sis of speech and memory codes in beginning readers.

Memory & Cognition, 10, 451456.Beaudoin, M. (1998). Dcoupage syllabique en franaiscomme langue seconde. Canadian Modern Language

Review, 54, 354375.Blevins, J. (1995). The syllable in phonological theory. In

J. A. Goldsmith (Ed.), The handbook of phonologicaltheory (pp. 206244). Cambridge, MA: Blackwell.

Breen, G., & Pensalfini, R. (1999). Arrernte: A languagewith no syllable onsets.Linguistic Inquiry, 30, 125.

Carreiras, M., Alvarez, C. J., & De Vega, M. (1993). Sylla-ble frequency and visual word recognition in Spanish.

Journal of Memory and Language, 32, 766780.Christman, S. S., & Depaolis, R. A. (1996). Sonority as a

constraint on word identification processes. AppliedPsycholinguistics, 17, 273291.

Clements, G. N. (1992). The sonority cycle and syllableorganization. In W. U. Dressler, H. C. Luschutzky,O. E. Pfeiffer, & J. R. Rennison (Eds.), Phonologica:Proceedings of the 6th International Phonology Meet-

ing (pp. 6376). Cambridge, UK: Cambridge Univ.Press.

Cole, R. A., & Scott, B. (1974). Toward a theory of speech

perception. Psychological Review, 81, 348374.Content, A., Frauenfelder, U., & Dumay, N. (1999). Seg-

mentation syllabique chez lenfant. In Actes des Journes dEtudes LinguistiqueLa syllabe (pp.6974). Nantes: Universit de Nantes.

Content, A., Meunier, C., Kearns, R. K., & Frauenfelder, U.H. (2001). Sequence detection in pseudowords inFrench: Where is the syllable effect? Language andCognitive Processes, in press.


22/23


Cutler, A., Mehler, J., Norris, D., & Segui, J. (1986). Thesyllables differing role in the segmentation of Frenchand English. Journal of Memory and Language, 25,385400.

Cutler, A., & Norris, D. (1988). The role of strong syllablesin segmentation for lexical access. Journal of Experi-

mental Psychology: Human Perception and Perfor-mance, 14, 113121.

Cutler, A., & Otake, T. (1996). Phonological structure andits role in language processing. In T. Otake & A. Cutler(Eds.), Phonological structure and language process-ing: Cross-linguistic studies (pp. 112). New York:Mouton de Gruyter.

de Cornulier, B. (1978). Syllabes et suites de phonmes enphonologie du franais. In B. de Cornulier & F. Dell(Eds.), Etudes de phonologie (pp. 3170). Paris:C.N.R.S.

Derwing, B. L. (1992). A pause-break task for elicitingsyllable boundary judgments from literate and illiteratespeakers: Preliminary results for five diverse lan-guages.Language and Speech, 35, 219235.

Dumay, N., Frauenfelder, U. H., & Content, A. (submitted).The role of the syllable in lexical segmentation in

French: Word-spotting data.Dupoux, E. (1993). The time course of prelexical process-

ing: The syllabic hypothesis revisited. In G. Altmann &R. Shillcock (Eds.), Cognitive models of speech proc-essing (pp. 81111). Hillsdale, NJ: Erlbaum.

Fallows, D. (1981). Experimental evidence for English syl-labification and syllable structure. Journal of Linguis-tics, 17, 309317.

Finney, S. A., Protopapas, A., & Eimas, P. D. (1996). Atten-tional allocation to syllables in American English.

Journal of Memory and Language, 35, 893909.Frauenfelder, U. H., & Kearns, R. (1996). Sequence moni-

toring. Language and Cognitive Processes, 11,665674.

Gillis, S., & De Schutter, G. (1996). Intuitive syllabification:Universals and language specific constraints. Journalof Child Language, 23, 487514.

Goldman, J. P., Content, A., & Frauenfelder, U. H. (1996).Comparaison des structures syllabiques en franais et

en anglais. Paper presented at the 21mes JournesdEtude sur la Parole, Avignon.

Hooper, J. B. (1972). The syllable in phonological theory.Language, 48, 525540.

Kahn, D. (1980). Syllable-based generalizations in Englishphonology. New York: Garland.

Kessler, B., & Treiman, R. (1997). Syllable structure and thedistribution of phonemes in English syllables. Journalof Memory and Language, 37, 295311.

Kolinsky, R. (1998). Spoken word recognition: A stage-pro-cessing approach of language differences. European

Journal of Cognitive Psychology, 10, 140.Kolinsky, R., Morais, J., & Cluytens, M. (1995). Intermedi-

ate representations in spoken word recognitionEvi-dence from word illusions. Journal of Memory and

Language, 34, 1940.

Laeufer, C. (1985). Some language-specific and universalaspects of syllable structure and syllabification: Evi-

dence from French and German. Unpublished Ph.D.dissertation, Cornell University.

Laks, B. (1995). A connectionist account of French syllabi-fication.Lingua, 95, 5176.

Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1998). A the-ory of lexical access in speech production.Behavioraland Brain Sciences, 22, 138.

Levelt, W. J. M., & Wheeldon, L. (1994). Do speakers haveaccess to a mental syllabary? Cognition, 50, 239269.

Liberman, A. M., & Studdert-Kennedy, M. (1978). Phoneticperception. In R. Held, H. Leibowitz, & H. L. Teuber(Eds.),Handbook of sensory physiology (Vol. VIII, pp.143178). Heidelberg: Springer-Verlag.

Liberman, I. Y., Shankweiler, D., Fisher, W. F., & Carter, B.(1974). Explicit syllable and phoneme segmentation in

the young child. Journal of Experimental Child Psy-chology, 18, 201212.

MacNeilage, P. F. (1998). The frame/content theory of evo-lution of speech production.Behavioral and Brain Sci-ences, 21, 499546.

Marslen-Wilson, W. D. (1987). Functional parallelism inspoken word recognition. Cognition, 25, 71102.

Massaro, D. W. (1974). Perceptual units in speech recogni-tion. Journal of Experimental Psychology, 102,199208.

Mattys, S. (1997). The use of time during lexical processing

and segmentation: A review. Psychonomic Bulletin &Review, 4, 310329.

McClelland, J. L., & Elman, J. L. (1986). The TRACEmodel of speech perception. Cognitive Psychology, 18,186.

McQueen, J. M. (1998). Segmentation of continuous speechusing phonotactics.Journal of Memory and Language,39, 2146.

Mehler, J., Dommergues, J.-Y., Frauenfelder, U., & Segui, J.(1981). The syllables role in speech segmentation.

Journal of Verbal Learning and Verbal Behavior, 20,

298305.Mehler, J., Dupoux, E., & Segui, J. (1990). Constraining

models of lexical access: The onset of word recogni-tion. In G. Altmann (Ed.), Cognitive models of speech processing: Psycholinguistic and computa-

tional perspectives (pp. 236261). Cambridge, MA:MIT Press.

Morais, J., Bertelson, P., Cary, L., & Alegria, J. (1986). Lit-eracy training and speech segmentation. Cognition, 24,4564.

Morais, J., Cary, L.,Alegria, J., & Bertelson, P. (1979). Doesawareness of speech as a sequence of phones arisespontaneously? Cognition, 7, 323331.

Pallier, C., Sebastian-Galls, N., Felguera, T., Christophe,A., & Mehler, J. (1993). Attentional allocation withinthe syllabic structure of spoken words. Journal of

Memory and

content 2001 content2001tn

Documents