JOURNAL OF MEMORY AND LANGUAGE 35, 666–688 (1996) ARTICLE NO. 0035 A Rhythmic Bias in Preverbal Speech Segmentation JAMES L. MORGAN Brown University Four studies using a variant of the conditioned head turning procedure, in which response latencies to extraneous noises occurring at different junctures within synthetic syllable strings served as the dependent variable, investigated 6- and 9-month-old infants’ representations of familiar and novel syllable pairs manifesting diverse rhythmic patterns. Familiar bisyllables tended to be perceived similarly by both age groups: These bisyllables were perceived as being cohesive, without regard to whether they manifested trochaic (longer – shorter) or iambic (shorter – longer) rhythm. Novel bisyllables were perceived differently by the two age groups. Six-month- olds appeared to perceive segmentally novel, but rhythmically familiar, bisyllables as cohesive, whereas they failed to perceive rhythmically novel, but segmentally familiar bisyllables in the same fashion. Similar patterns of results were obtained with trochaic and iambic bisyllables. Nine-month-olds, however, were differentially sensitive to different rhythmic patterns. Older infants appeared to perceive novel bisyllables as cohesive only when they manifested trochaic rhythm, whether the bisyllables were segmentally or rhythmically novel. Nine-month-olds’ behavior is consistent with the possibility that they have adopted a metrical strategy for segmenta- tion, as A. Cutler (1990 and elsewhere) has argued is the case for novice and expert English speakers alike. The plausibility of such a strategy with respect to the nature of child-directed speech is discussed, as are possible consequences for acquisition of such a strategy. q 1996 Academic Press, Inc. To infants embarking on learning language, fluent speech are often unintelligible (Bard & Anderson, 1983; Pollack & Pickett, 1964). input speech must initially be perceived as a wordless babble. In speech, words are not Nevertheless, location and recognition of words in fluent speech is critical to language physically separated, as they are on the printed page. Nor does any universal recipe for ‘word’ learning, for it is only as a result of these processes that componential meanings of ut- exist: Words may comprise one syllable or many, one sequence of sounds or another, one terances can be understood or patterns of syn- tax can be discovered. Moreover, from the stress pattern or an alternate. To make matters yet more complex, words are mutable: When advent of productive language and the bur- geoning of language comprehension, it may be words come into contact with one another in fluent speech, the sounds at their edges may surmised that infants have substantially solved the segmentation problem and begun to build change (following language-specific patterns of sandhi ), and their rhythmic patterns may a lexicon before the end of the first year. The processes by which these occur are only be- be altered (Chomsky & Halle, 1968; Selkirk, 1984), to the degree that words excised from ginning to be understood. This paper presents a series of studies exploring whether prever- bal, English-learning infants exhibit a bias to- This work was supported by NIMH Grant MH 48892. I ward language-predominant rhythmic patterns thank John Mertus for providing technical and programming in segmenting short sequences of syllables — support and Nancy Allard, Shinina Butler, Yael Harlap, Ei- leen Hoff, Rachel Kessinger, Allison Klein, and Michelle comparable to phonological words—from Rogers for assistance in testing subjects and coding video larger strings of syllables. tapes. Address correspondence and reprint requests to James The hypothesis that particular rhythmic pat- L. Morgan, Department of Cognitive and Linguistic Sci- terns are exploited in speech segmentation by ences, Brown University, Box 1978, Providence RI 02912. E-mail: James_[email protected]. novice and expert language users alike has 666 0749-596X/96 $18.00 Copyright q 1996 by Academic Press, Inc. All rights of reproduction in any form reserved.
Transcript
JOURNAL OF MEMORY AND LANGUAGE 35, 666–688 (1996)ARTICLE NO. 0035
A Rhythmic Bias in Preverbal Speech Segmentation
JAMES L. MORGAN
Brown University
Four studies using a variant of the conditioned head turning procedure, in which responselatencies to extraneous noises occurring at different junctures within synthetic syllable stringsserved as the dependent variable, investigated 6- and 9-month-old infants’ representations offamiliar and novel syllable pairs manifesting diverse rhythmic patterns. Familiar bisyllablestended to be perceived similarly by both age groups: These bisyllables were perceived as beingcohesive, without regard to whether they manifested trochaic (longer–shorter) or iambic (shorter–longer) rhythm. Novel bisyllables were perceived differently by the two age groups. Six-month-olds appeared to perceive segmentally novel, but rhythmically familiar, bisyllables as cohesive,whereas they failed to perceive rhythmically novel, but segmentally familiar bisyllables in thesame fashion. Similar patterns of results were obtained with trochaic and iambic bisyllables.Nine-month-olds, however, were differentially sensitive to different rhythmic patterns. Olderinfants appeared to perceive novel bisyllables as cohesive only when they manifested trochaicrhythm, whether the bisyllables were segmentally or rhythmically novel. Nine-month-olds’behavior is consistent with the possibility that they have adopted a metrical strategy for segmenta-tion, as A. Cutler (1990 and elsewhere) has argued is the case for novice and expert Englishspeakers alike. The plausibility of such a strategy with respect to the nature of child-directedspeech is discussed, as are possible consequences for acquisition of such a strategy. q 1996
Academic Press, Inc.
To infants embarking on learning language, fluent speech are often unintelligible (Bard &Anderson, 1983; Pollack & Pickett, 1964).input speech must initially be perceived as
a wordless babble. In speech, words are not Nevertheless, location and recognition ofwords in fluent speech is critical to languagephysically separated, as they are on the printed
page. Nor does any universal recipe for ‘word’ learning, for it is only as a result of theseprocesses that componential meanings of ut-exist: Words may comprise one syllable or
many, one sequence of sounds or another, one terances can be understood or patterns of syn-tax can be discovered. Moreover, from thestress pattern or an alternate. To make matters
yet more complex, words are mutable: When advent of productive language and the bur-geoning of language comprehension, it may bewords come into contact with one another in
fluent speech, the sounds at their edges may surmised that infants have substantially solvedthe segmentation problem and begun to buildchange (following language-specific patterns
of sandhi), and their rhythmic patterns may a lexicon before the end of the first year. Theprocesses by which these occur are only be-be altered (Chomsky & Halle, 1968; Selkirk,
1984), to the degree that words excised from ginning to be understood. This paper presentsa series of studies exploring whether prever-bal, English-learning infants exhibit a bias to-
This work was supported by NIMH Grant MH 48892. Iward language-predominant rhythmic patternsthank John Mertus for providing technical and programmingin segmenting short sequences of syllables—support and Nancy Allard, Shinina Butler, Yael Harlap, Ei-
leen Hoff, Rachel Kessinger, Allison Klein, and Michelle comparable to phonological words—fromRogers for assistance in testing subjects and coding video larger strings of syllables.tapes. Address correspondence and reprint requests to James
The hypothesis that particular rhythmic pat-L. Morgan, Department of Cognitive and Linguistic Sci-terns are exploited in speech segmentation byences, Brown University, Box 1978, Providence RI 02912.
E-mail: [email protected]. novice and expert language users alike has
6660749-596X/96 $18.00Copyright q 1996 by Academic Press, Inc.All rights of reproduction in any form reserved.
AID JML 2447 / a004$$$$$1 09-25-96 11:50:42 jmla AP: JML
667PREVERBAL SPEECH SEGMENTATION
been championed in a series of papers by Cut- that infants possess capacities prerequisite forapplication of a segmentation strategy basedler and her colleagues. Cutler (1990; Cutler &
Norris, 1988) has proposed that English on speech rhythm. Demany, MacKenzie, andVurpillot (1977) have shown that, by 2speakers may employ a Metrical Segmenta-
tion Strategy, according to which word bound- months, infants can discriminate auditory pat-terns differing in rhythm, and Trehub andaries may be posited preceding strong sylla-
bles (stressed syllables, with relatively long Thorpe (1988) have shown that, by 6 months,infants can categorize sequences on the basisdurations and full vowels). Cutler and Carter
(1987) have argued that this strategy provides of rhythmic attributes. Spring and Dale (1977)and Jusczyk and Thompson (1978) havea plausible first stage in segmentation, inas-
much as an overwhelming preponderance of shown that infants can discriminate simplesyllabic sequences differing in stress patterns.English content words in fluent speech—in
excess of 85%—begin with strong syllables. Jusczyk, Cutler, and Redanz (1993) have dem-onstrated that, by 9 months, English-exposedSpeakers of languages with rhythmic proper-
ties different than those of English may pursue infants prefer to listen to lists of words withstrong–weak patterns rather than to wordsalternative segmentation strategies (Cutler,
Mehler, Norris, & Segui, 1986; Otake, Ha- with weak–strong patterns. This last resultshows that infants are not only attending totano, Cutler, & Mehler, 1993).
Cutler has adduced a variety of psycholin- rhythmic patterns themselves, but also to theirrelative frequencies of occurrence in input. Inguistic evidence to show that adult English
speakers may employ such a metrical segmen- addition, evidence from older children’s pro-ductions suggests that they are susceptible totation strategy. Cutler and Clifton (1984), for
example, found that mis-stressed words with the same sorts of mis-segmentations as thosedocumented by Cutler and Butterfield (1992),unreduced vowels were more easily recog-
nized if they were mis-stressed to comport as the following exchange between the authorand his daughter (3;5) illustrates:with the English-predominant strong–weak
pattern (e.g., raccoon) than if they were mis-(in the supermarket)
stressed to fit the opposite pattern (e.g.,Child: What is that flower?
rabbıt). Cutler and Norris (1988) demon-Father: That’s called an azalea.
strated that monosyllabic words embedded inChild: Let’s get Mommy some zaleas.
nonsense bisyllables are more readily detectedwhen the bisyllables manifest strong–weak Although this set of findings is compatible
with the possibility that English-exposed in-patterns than when the bisyllables manifeststrong–strong patterns. Analyses of ‘slips of fants may employ a metrical strategy for seg-
mentation, none of these results show that in-the ear’ by Cutler and Butterfield (1992)showed that a large majority of errors can be fants actually do so. Consider how this hy-
pothesis might be tested in preverbal infants.analyzed as resulting from postulation ofboundaries before strong syllables (for exam- A corollary of the metrical segmentation strat-
egy hypothesis is that weak syllables (un-ple, ‘‘it was illegal’’ may be misperceived as‘‘it was an eagle’’). stressed syllables with relatively short dura-
tions and, often, reduced vowels) will tend toCutler (1996) has argued that an importantadvantage of a metrical segmentation strategy, be perceived as continuing, rather than begin-
ning, words. From this, the prediction followsin contrast to segmentation models that relyexclusively on matching of input to existing that, other things being equal, strong–weak
(trochaic, or longer–shorter) sequences of syl-lexical representations (e.g., Cole & Jakimik,1978), is that it provides a means for novice lables should be perceived as being more co-
hesive than weak–strong (iambic, or shorter–language learners to begin to locate words influent speech. Several studies of infant audi- longer) sequences of syllables.
Recent work by Morgan (1994; Morgan &tory and speech perception provide evidence
AID JML 2447 / a004$$$$$2 09-25-96 11:50:42 jmla AP: JML
668 JAMES L. MORGAN
Saffran, 1995) investigating infants’ represen-tations of synthetic syllable strings providesevidence that speech rhythm is an importantfactor influencing the perceived cohesivenessof short sequences of syllables. This researchuses a NOISE DETECTION method whereinhead turning latencies to extraneous noises(short buzzes) occurring in different juncturesin syllable strings are measured. The expecta-tion underlying this method is that responsesto buzzes occurring between a pair of syllablesperceived as constituting a single unit will beslower than will be responses to buzzes oc-curring between two syllables not perceivedas constituting a unit. This reflects the fact that
FIG. 1. Nine-month-olds. Head turning latencies fromperceptual/representational units are cohesive,Morgan and Saffran, Experiment 2.the property that gives rise to the tendency of
such units to resist interruption.Experiments by both Morgan (1994) and
Morgan and Saffran (1995) show that infants and Saffran found that familiarized trochaicand iambic sequences were both perceived bybetween 6 and 9 months use rhythmic proper-
ties of stimulus strings as an initial means of 9 month old infants as being cohesive; as Fig.1 shows, both conditions displayed significantgrouping pairs of syllables. With increased fa-
miliarization, 9-month-olds balance sequential nonzero differences in latencies to buzzes oc-curring inside (‘‘within unit’’) and outsideand rhythmic properties of the stimulus strings
in forming an integrated basis for grouping (‘‘between units’’) these syllable pairs.These results might provide some evidencesyllables. In contrast, 6-month-olds continue
to rely on rhythmic characteristics of stimulus that 9-month-olds are not yet using a metricalstrategy for segmentation; 9-month-olds’ pref-strings in grouping particular syllables, even
when the sequential properties of the stimulus erences for trochaic bisyllables (Jusczyk, Cut-ler, & Redane 1993) may not reflect particularstrings are incompatible with grouping those
syllables. processing biases. However, these results aresubject to several alternative explanations.However, one study in Morgan and Saffran
(1995) provides some evidence that fails to First, Morgan and Saffran’s failure to observedifferences between trochaic and iambic se-support the metrical segmentation strategy hy-
pothesis, in particular the prediction that tro- quences might be due to a failure to manipu-late relevant stimulus properties. Rhythmicchaic sequences should be more cohesive than
iambic sequences. Two between-subjects patterns in their stimuli were produced byvarying the durations of individual syllablesstimulus conditions from Morgan and Saffran,
Experiment 2, are relevant to present con- only. In contrast, in spontaneous speech,strong and weak syllables typically differ notcerns. In both of these, rhythmic and sequen-
tial regularities across the set of stimulus only in duration, but also in vowel quality,pitch contour, and amplitude. These propertiesstrings supported grouping of a particular pair
of adjacent syllables into a single unit. In one may not all be required for distinguishing be-tween strong and weak syllables. For example,condition, the rhythmic pattern across the pair
was trochaic (a longer syllable followed by a Ratner (1985) reports analyses of maternalchild-directed speech in which content wordshorter syllable); in the second condition, the
rhythmic pattern was iambic (a shorter sylla- (strong syllable) vowels and function word(weak syllable) vowels were spectrally simi-ble followed by a longer syllable). Morgan
AID JML 2447 / a004$$$$$2 09-25-96 11:50:42 jmla AP: JML
669PREVERBAL SPEECH SEGMENTATION
lar; in these data, vowel duration was a power- both the trained and the opposing, untrainedrhythmic patterns.ful predictor of syllable type. Moreover, in
To the extent that the metrical segmentationstudies of adult segmentation, duration hasstrategy hypothesis correctly predicts that tro-been shown to be a more powerful influencechaic rhythm tends to enhance the perceivedthan vowel quality, pitch contour, or ampli-cohesiveness of syllable sequences, asymmet-tude (Nakatani & Schaffer, 1978). Neverthe-ries across stimulus conditions should result.less, differences in infants’ processing of tro-In particular, novel bisyllables should be per-chaic and iambic stimuli may only appearceived as being more cohesive when theywhen they are presented with stimuli in whichmanifest trochaic rhythm, and familiar bisyll-these natural correlations are preserved.ables manifesting a novel rhythmic patternSecond, the infants in Morgan and Saffran’sshould be perceived as being more cohesivestudy had received considerable exposure towhen the novel pattern is trochaic. Examina-the stimulus strings before testing. It is possi-tion of infants at two ages may permit theble that any initial tendency toward perceivingdetermination of whether enhanced cohesive-a trochaic sequence as more cohesive may beness of trochaic syllable sequences emergesneutralized by sufficient familiarization withbetween 6 and 9 months, in tandem with thean iambic sequence. After all, words manifest-emergence of preference for trochaic se-ing iambic patterns are possible in English;quences documented by Jusczyk, Cutler, andfor mature speakers, a metrical segmentationRedanz (1993).strategy is a bias rather than an absolute con-
straint on processing. It would be counterpro- EXPERIMENT 1ductive for language learners to adopt an abso-
The noise detection method is a variant oflute version of a metrical strategy for segmen-the conditioned head turning technique (Kuhl,tation. Thus, a more relevant test of whether1985) that draws its inspiration from click de-
infants employ a metrical strategy for segmen-tection methods formerly employed in investi-
tation might involve examination of the per-gating adult psycholinguistic representations
ceived cohesiveness of novel syllabic se-(see Fodor, Bever, & Garrett, 1974, for an
quences. Note that the stimuli used by Jusc- extensive historical review). Unlike the typi-zyk, Cutler, and Redanz (1993) included only cal procedure of that time, which used sub-low frequency words, to which infants were jects’ recollections of click locations as theunlikely to have been previously exposed. primary dependent measure, the noise detec-Thus, the preference they demonstrated was a tion method is based on measurements of re-preference for novel trochaic sequences over sponse latencies to extraneous noises (shortnovel iambic sequences. The goal of the stud- buzzes) occurring in different junctures in syl-ies presented below is to further investigate lable strings. (A handful of studies investigat-infants’ representations of novel syllabic se- ing adult sentence processing, including Ab-quences. rams & Bever, 1969; Bond, 1972; Flores
In the first two studies, 6- and 9-month- d’Arcais, 1978; and Holmes & Forster, 1970,old infants were familiarized to a key pair of used on-line click detection methods compara-syllables manifesting one particular rhythmic ble to the method used here. Those studiespattern or another (either trochaic or iambic) found differences in response latency as aand were later tested on both the familiar pair function of objective click location: Clicksof syllables and a novel pair of syllables mani- within syntactic units were responded to morefesting the trained rhythmic pattern. In the sec- slowly than were clicks between units. Theseond two studies, 6- and 9-month-old infants results are consistent with the predictionsare trained in the same fashion as in the first made here.) Under this method, infants aretwo experiments, but were later tested on the conditioned to turn their heads in response to
short buzzes which are initially presented insame pair of syllables when they appeared in
AID JML 2447 / a004$$$$$2 09-25-96 11:50:42 jmla AP: JML
670 JAMES L. MORGAN
silent intervals between multisyllabic strings. monitor loudspeaker located in the testingroom with the infant. Two smoked PlexiglasAfter training, buzzes are presented between
different pairs of syllables within the strings, boxes containing motorized animal figuresthat provided reinforcement were locatedand differences in response latencies ac-
cording to noise location are used to index above the loudspeaker.Test sessions were recorded (video only)infants’ perceptual organization of the stimu-
lus strings. Across stimulus conditions, pro- with a Panasonic AG6040 time-lapse VCR,which was interfaced to the computer. At thesodic and sequential properties of the multi-
syllabic strings are to be manipulated to beginning of each trial, the computer wroteidentifying information to the video tapepromote or discourage perceptions of cohe-
siveness between particular pairs of syllables. which was later used in coding the video tapesfor head turn latencies.
Method Stimuli. A set of stimulus syllables wererecorded and digitally edited for use in thisSubjects. Infants approximately 9 months
of age were recruited from the Providence experiment. Original versions of the syllableswere spoken in isolation with a flat intonationmetropolitan area. Thirty-two infants, approx-
imately half male and half female, served as contour by a voice-trained adult female. Ed-ited stimuli included two tokens each of [ko],subjects; 16 infants were assigned at random
to each of the two stimulus conditions. Sub- [ga], [pu], and [b`c], and one token each of[de] and [ti]. Desired durations were producedjects had a mean age of 0;9.05 at the first
session (range 0;8.20 to 0;9.22). The two ses- by excising entire glottal periods, from zero-crossing to zero-crossing, so that no transientssions required for each infant were scheduled
so that the time between the two sessions was were introduced. Shorter versions of [ko],[ga], [pu], and [b`c] (henceforth denoted byno more than 10 days.
Fifty-nine infants were tested in order to [ko], [ga], [pu], and [b`c]) were 350 ms,whereas longer versions of [ko], [ga], [pu],attain the final sample; infants were excluded
from the study for the following reasons: Fail- and [b´c] (henceforth denoted by [ko], [ga],[pu], and [b´c]) were 500 ms in duration.ure to complete initial shaping within 40 trials
or to meet the predetermined training criterion Shorter and longer versions of these syllablesdid not differ in either amplitude or fundamen-within 30 trials in the initial session (14), in-
terference from uncooperative siblings (1), tal frequency contour. The tokens of [de] and[ti] were edited to be 425 ms in duration. Thedifficulties scheduling subsequent testing ses-
sions within a 10 day period (4), and failure extraneous buzz was created by digitally con-catenating clicks from a child’s toy cricket.to contribute at least two latencies to each cell
in the analysis (8). Subjects heard the syllables in trisyllabicstrings. Each string included one filler sylla-Apparatus. Infants were tested in a sound-
proofed laboratory room. An experimenter in ble, either [de] or [ti], which appeared in eitherinitial or final position. Each string also in-an adjoining control room monitored the in-
fant’s behavior via a Panasonic closed-circuit cluded one key pair of syllables which eitherconsisted of one token of [ko] or [ko] and onetelevision system. Trial duration, stimulus pre-
sentation, and delivery of reinforcement were token of [ga] or [ga], or else one token of [pu]or [pu] and one token of [b`c] or [b´c]. Thecontrolled by custom designed software run-
ning on an IBM PS/2 equipped with a Data syllables constituting the key pair were alwaysadjacent to one another and always occurredTranslations DT2901 D–A board. Stimuli
were presented through a 5000 Hz TTE in the same relative order. Within the strings,individual syllables were separated by 50 ms411AFS low pass filter, a custom programma-
ble attenuator, an Onkyo P-304 preamplifier, silences; strings themselves were separated by1000 ms silences.and an Onkyo M-504 power amplifier that was
connected to an Electrovoice Sentry 100A Two between-subject stimulus conditions
AID JML 2447 / a004$$$$$2 09-25-96 11:50:42 jmla AP: JML
671PREVERBAL SPEECH SEGMENTATION
TABLE 1
TRISYLLABIC STRINGS INCLUDED IN STIMULUS CONDITIONS, EXPERIMENTS 1 AND 2a
Condition Phase Stimulus stringsb
Trochaic meterSubcondition Ac Training gakode gakoti degako tigako
Testing gakode gakoti degako tigakob´cpuD de deb´cpuD
Subcondition B Training b´cpuD de b´cpuD ti deb´cpuD tib´cpuDTesting b´cpuD de b´cpuD ti deb´cpuD tib´cpuD
gakode degakoIambic meter
Subcondition A Training gakode gakoti degako tigakoTesting gakode gakoti degako tigako
b`cpude deb`cpuSubcondition B Training b`cpude b`cputi deb`cpu tib`cpu
a Accute accents (´) indicate longer syllables (500 ms); grave accents (`) indicate shorter syllables (350 ms);syllables without accents are medium length (425 ms).
b All four Training strings were used as background during Testing, but, for the conditions illustrated, only familiarstrings with [de] were used in test trials.
c Subconditions C–D were similar to A–B, respectively, except that [ti], rather than [de] appeared in the novel testitems; subconditions E–H were similar to A–D, respectively, except that the order of syllables within key pairs wasreversed, e.g., [koga] and puB b`c] rather than [gako] and b´cpu].
were employed. Each condition included a novel key pairs of syllables. Eight subcondi-tions were included, to each of which 2 infantstraining set of four trisyllabic strings, across
which the key syllable pair consistently were assigned at random, and across whichkey training syllables, key syllable order, andmanifested one of the two rhythmic patterns
used. This set of strings was used in initial use of [de] or [ti] in test items were counter-balanced. A representative subset of these sub-training, as background stimuli between test
trials, and as test stimuli in half of the test conditions are shown in Table 1. This condi-tion tested the cohesiveness of a novel trocha-trials (see Procedure for additional details).
Strings with key pairs comprising syllables ically accented syllable sequence relative to afamiliar trochaically accented syllable se-not included in the training strings provided
test stimuli for the remaining test trials. quence.In the Iambic Meter condition, the rhythmicStimulus conditions are described below and
illustrated in Table 1. pattern across the key pair of syllables wasalways iambic: a shorter syllable followed byIn the Trochaic Meter condition, the rhyth-
mic pattern over the key pair of syllables was a longer syllable. Again, both rhythmic anddistributional information provided evidencealways trochaic: a longer syllable followed by
a shorter syllable. Across the training set of that the pair of syllables constituted a unit,and distributional evidence indicated that [de]strings, both this consistent rhythmic pattern
and distributional (or sequential) information and [ti] each constituted independent units.Following training, infants were tested oncontributed evidence that the pair of syllables
constituted a unit. The distributional evidence strings with either familiar or novel key pairsof syllables. Eight subconditions were in-also signaled that [de] and [ti] each formed
independent units. Following training, infants cluded, some of which are shown in Table 1.This condition tested the cohesiveness of awere tested on strings with either familiar or
AID JML 2447 / a004$$$$$3 09-25-96 11:50:42 jmla AP: JML
672 JAMES L. MORGAN
novel iambically accented syllable sequence four training strings to present in each inputblock; three repetitions of the string normallyrelative to a familiar iambically accented syl-
lable sequence. occurred in each block. Upon judging the in-fant’s attention to be focused at midline, theAs in Morgan and Saffran, trochaic and
iambic rhythmic patterns were produced here experimenter could request a trial; the trialbegan at the earliest possible point followingby varying only the durations of individual
syllables. This provides a stringent test of the an inter-string interval and included one blockof three repetitions of a string. In some in-metrical segmentation hypothesis. If differ-
ences in infants’ processing of trochaic and stances, the trial began in the middle of a back-ground stimulus block, in which case theiambic rhythms appear when they are pre-
sented with stimuli in which only syllable du- string in that block was repeated across thetrial. In other instances, the trial began at theration distinguishes strong and weak syllables,
such differences should appear a fortiori when end of a background stimulus block, in whichcase the software simply advanced to the nextstimuli incorporate additional properties nor-
mally correlated with meter. stimulus block.During the shaping phase, in each trial aProcedure. Infants participated in one train-
ing session and one testing session. During buzz was inserted in the middle of the first1000 ms inter-string interval. Infants were al-both sessions infants were seated on their par-
ents’ laps at a small table. An assistant seated lowed a window of 2.5 s from the onset ofthe buzz in which to turn toward the loud-directly across from the infant maintained the
infant’s attention at midline by displaying and speaker in order to receive reinforcement;head turns occurring after this period were notmanipulating an assortment of toys. The ex-
perimenter in the control room observed the reinforced. Initially, the buzz was 100 ms induration and 12 dB louder than the speechinfant on a video monitor. This experimenter
initiated trials upon observing the infant’s at- stimuli; across shaping trials, after every twoconsecutive correct trials, the buzz was short-tention to be focused at the midline. Rein-
forcement was delivered contingent on the in- ened in 20 ms steps to 40 ms, and its intensitywas lowered in 4 dB steps until it was equalfant responding to a buzz with a headturn to-
ward the loudspeaker, which was located 90 to that of the speech. The shaping phase wascontinued for 40 trials or until the infantdegrees from midline on the infant’s left. The
experimenter served as sole judge of whether turned his or her head on two consecutive tri-als with the buzz at its minimum duration.the infant turned its head during trials, de-
pressing a button to signal to the computer Following successful shaping, the criterionphase began. In this phase of the training ses-that a headturn had occurred. (The video cam-
era was located directly above the loud- sion, the computer randomly selected trials tobe either target (buzz) trials or control (no-speaker, so the experimenter judged whether
the infant looked into the camera.) Reinforce- buzz) trials. Infants were tested until they hadeither attained the predetermined criterion byment began when the experimenter signaled a
headturn, and lasted for 3 s. To preclude bias, responding correctly on seven of eight consec-utive trials (by turning on target trials and notthroughout all sessions, parent, assistant, and
experimenter all listened to music over turning on control trials) or completed 30 tri-als. Infants not attaining the criterion withinaround-the-ear headphones (Sony MDR-CD
550). the specified number of trials were excludedfrom further participation; infants who at-During the Training Session, infants heard
the trisyllabic strings presented at a comfort- tained the criterion returned for a testing ses-sion.able listening level (60–62 dB (B) SPL). Ex-
ample stimulus sequences from each phase of The Testing Session began with a short re-view (requiring four head turns) recapitulatingthe training session are illustrated in Table 2.
The software randomly chose which of the the shaping phase of the Training Session—
AID JML 2447 / a004$$$$$3 09-25-96 11:50:42 jmla AP: JML
a Possible sequences are given here only for one subcondition in the Trochaic Meter (Trochaic Training) condition,in which [gako] served as the key training syllable pair.
b Superscripted numbers indicate repetitions. f indicates the location of the buzz within a trial.
the buzz was presented in the first inter-string hence, responses to extraneous noises occurringbetween these syllables should be relativelyinterval, and the duration and amplitude of the
buzz were reduced following each head turn. slow. In contrast, across the set of stimulusstrings, the familiar syllables flanking between-Following this, the testing phase began. In this
phase, between trials, infants heard the same units buzzes vary in segmental content, relativeorder, and syllable duration. These propertiesstrings in the same fashion as before. Test
trials differed from training trials in that the should discourage perception of these pairs ofsyllables as single units; hence responses to ex-40 ms buzz occurred in one of the 50 ms inter-
syllable intervals within the first string, eitherbetween the syllables constituting the key pair(within unit) or between one of these syllables TABLE 3and a filler syllable (between units). As in the
EXAMPLES OF TESTING TRIALS IN EXPERIMENTS 1 & 2a
training trials, infants were allowed a periodof 2.5 s beginning with the onset of the buzz Trial type Trial Stimulib
in which to turn in order to receive reinforce-Familiarment. The string presented in each of the trials
Within-unitwas predetermined by the software; hence, un-first-interval gafkode gakode gakode
like the Training Session, interrupted stimulus Within-unitblocks were not extended through the trials. second-interval degafko degako degako
Between-unitsThe testing phase included 24 randomly or-first-interval defgako degako degakodered trials in total. Crossing of three binary
Between-unitsfactors—Familiarity (familiar vs novel stimu-second-interval gakofde gakode gakode
lus), Position (first inter-syllable interval vs Novelsecond inter-syllable interval buzz location), Within-unit
first-interval b´cfpude gakode gakodeand Unit Status (within-unit vs between-unitsWithin-unitbuzz location)—yields eight trial types, each
second-interval deb´cfpu degako degakoof which occurred three times. Example stim-Between-units
ulus sequences illustrating each trial type are first-interval defb´cpu degako degakoprovided in Table 3. Between-units
second-interval b´cpufde gakode gakodeNote that across the set of stimulus strings, thefamiliar syllables flanking within-unit buzzes—
a Possible sequences are given here only for Subcondi-the key pair of syllables—are invariant in seg-tion A in the Trochaic Meter condition, in which [gako]
mental content, relative order, and rhythmic pat- served as the key syllable pair in training, and all stringstern. These properties should promote percep- in test trials included [de] as the filler syllable.
b f Indicates the location of the buzz within a trial.tion of this pair of syllables as a single unit;
AID JML 2447 / a004$$$$$3 09-25-96 11:50:42 jmla AP: JML
674 JAMES L. MORGAN
traneous noises occurring between these sylla-bles should be relatively fast. Differences in la-tencies to within-unit versus between-unitsbuzzes thus provide a measure of the degree towhich the key syllables are perceived as coher-ing with one another to form a unit.
All test trials were coded off-line from thevideo tapes. Off-line coders were blind to theon-line scoring of the experimenter, the natureof each trial, and the conditions to which sub-jects were assigned. During coding, the VCRwas interfaced to the computer and controlledby custom software. Using the identifying in-formation written on the tape during the exper-iment, the computer advanced the videotape
FIG. 2. Nine-month-olds. Means and standard errors ofto the beginning of the buzz in each trial. Thedifferences in head turning latencies in Experiment 1 tocoder then played the tape forwards and back-buzzes occurring inside versus outside key syllable pairswards in slow motion (at speeds ranging downfor familiar and novel test stimuli. Large positive differ-
to 1 field—1/60 s—per s) and identified the ences are consistent with perception of the syllable pairspoint at which the infant began to rotate its as cohesive units.head in a smooth movement that culminatedin a complete head turn toward the camera.The computer automatically calculated the yielded no significant main effects or inter-amount of time elapsed between this point and action terms involving the Position factorthe beginning of the buzz. Twenty-five percent (first inter-syllable interval vs second inter-of the head turn latencies were coded a second syllable interval buzz location). Therefore,time to assess coder reliability. treatment means were collapsed across lev-
Agreement between on-line experimenters els of Position, and this factor was not usedand off-line coders on whether head turns oc- in subsequent analyses. Means and SEs ofcurred exceeded 95%. Intra-coder reliability difference scores for familiar and novelon coding latencies was quite high: More than stimuli for each of the stimulus conditions90% of these latencies differed by no more are shown in Fig. 2.than two frames of the video tape (67 ms). Familiar stimuli. Thirteen of 16 subjects in
Root mean square averages were computed the Trochaic Meter condition responded morefor head turning latencies within each of the slowly on average to ‘within unit’ buzzes thaneight Familiarity by Position by Unit Status trial to ‘between units’ buzzes; 12 of 16 subjectstypes. RTs of less than 100 ms were excluded. in the Iambic Meter condition did likewise.Subjects were excluded if they did not have at Overall, subjects required 569 ms (SD Å 108least two usable RTs in each cell (maximum ms) to initiate head turns on familiar within-three). Difference scores were then computed unit trials, and 516 ms (SD Å 138 ms) tofor each Familiarity by Position cell by sub- initiate head turns on familiar between-unitstracting the between-units mean from the trials. Both groups of subjects displayed sig-within-units mean. These difference scores pro- nificant positive difference scores: for Tro-vide measures of the relative cohesiveness of chaic Meter, t(15) Å 2.75, p õ .05, and forthe key syllable pair for the different Familiarity Iambic Meter, t(15) Å 2.15, p õ .05. Differ-by Position trial types. ence scores did not vary significantly acrossResults and Discussion the two stimulus conditions, t(30) Å 0.33, NS.
This pattern of results is similar to that pre-A preliminary set of Stimulus Conditionby Familiarity by Position omnibus analyses viously obtained in Morgan and Saffran, Ex-
AID JML 2447 / a004$$$$$3 09-25-96 11:50:42 jmla AP: JML
675PREVERBAL SPEECH SEGMENTATION
periment 2 and suggests that no processing trials by infants in the Trochaic Meter condi-tion cannot be attributed to their having ig-advantage exists for familiar trochaic over fa-
miliar iambic sequences. nored the segmental changes introduced inthese trials. Rather, the pattern of results inNovel stimuli. Once again, 13 of 16 subjects
in the Trochaic Meter condition responded this experiment is compatible with the possi-bility that 9-month-old, English-exposed in-more slowly on average to within-unit buzzes
than to between-units buzzes. However, only fants are biased toward perceiving trochaic bi-syllables as constituting cohesive portions of8 of 16 subjects in the Iambic Meter condition
did likewise. Subjects in the Trochaic Meter the speech stream—a bias tantamount to useof a metrical strategy for segmentation.condition required 665 ms (SD Å 96 ms) to
initiate head turns on novel within-unit trialsEXPERIMENT 2and 612 ms (SD Å 102 ms) to initiate head
turns on novel between-units trials. The mean Several recent studies have demonstratedchanges in speech perception occurring overdifference score for the Trochaic Meter condi-
tion was significantly greater than zero, t(15) the second half of the first year, many of whichappear to reflect attunement to properties ofÅ 2.82, p õ .05. Subjects in the Iambic Meter
condition required 623 ms (SD Å 88 ms) to the native language. For example, Werker andTees (1984) showed that 6-month-olds caninitiate head turns on novel within-unit trials
and 635 ms (SD Å 101 ms) to initiate head discriminate non-native speech contrasts that10-month-olds cannot, and Jusczyk, Frieder-turns on novel between-units trials; but the
mean difference score for the Iambic Meter ici, Wessels, Svenkerud, and Jusczyk (1993)showed that 9-month-olds prefer to listen tocondition was not significantly different from
zero, t(15) Å 00.56, NS. Difference scores words exemplifying native, rather than non-native, sound sequences, whereas 6-month-varied significantly across the two conditions,
t(30) Å 2.58, põ .05. The different patterning olds show no preference. Experiment 2 askswhether comparable changes occur with re-of results across familiar and novel stimuli
was also evidenced by a significant Stimulus gard to rhythmic biases in speech segmenta-tion, testing 6-month-olds under the proce-Condition by Familiarity interaction in the
omnibus analysis of difference scores, F(1,30) dures of Experiment 1.Two recent findings are particularly ger-Å 4.85, p õ .05.
The results of this experiment replicate in mane to making predictions in the present ex-periment. First, Morgan and Saffran (1995)part those of Morgan and Saffran, Experiment
2: Infants appeared to perceive the familiar- found that rhythmic regularities holding overa set of stimulus strings were sufficient for 6-ized key syllable pairs as equally cohesive re-
gardless of whether they manifested trochaic month-olds to group pairs of syllables to-gether. Thus, for familiar stimuli, 6-month-or iambic rhythm. However, infants’ re-
sponses to novel syllable pairs did vary de- olds may be expected to behave in a fashionsimilar to that of the 9-month-olds in Experi-pending on rhythmic pattern: Whereas novel
trochaic bisyllables appeared to be as cohesive ment 1, providing evidence that familiar tro-chaic and iambic bisyllables are both per-as familiar trochaic bisyllables, novel iambic
bisyllables were significantly less cohesive ceived as cohesive. Second, Jusczyk, Cutler,and Redanz (1993) found that 6-month-old,than were familiar iambic bisyllables. The
comparable overall increases in head turning English-exposed infants show no preferencefor either trochaic or iambic novel bisyllables,latency on trials with novel stimuli (99 ms for
Trochaic Meter, t(15) Å 6.16, p õ .01; 83 ms whereas 9-month-olds prefer trochaic novelbisyllables. Thus, for novel stimuli, 6-month-for Iambic Meter, t(15) Å 5.97, põ .01) show
that infants in both conditions noticed the olds may be expected to behave differentlythan did the 9-month-olds in Experiment 1,change in stimulus. Therefore, the continued
relatively slow responses on novel within-unit failing to provide evidence that novel trochaic
AID JML 2447 / a004$$$$$3 09-25-96 11:50:42 jmla AP: JML
676 JAMES L. MORGAN
and iambic bisyllables are perceived as differ-entially cohesive.
Method
Subjects. Infants approximately 6 monthsof age were recruited from the Providencemetropolitan area. Thirty-two infants, approx-imately half male and half female, served assubjects; 16 infants were assigned at randomto each of the two stimulus conditions. Sub-jects had a mean age of 0;6.13 at the firstsession (range 0;6.1 to 0;7.0). The two ses-sions required for each infant were scheduledso that the time between the two sessions wasno more than 10 days.
FIG. 3. Six-month-olds. Means and standard errors ofSixty-six infants were tested in order to at- differences in head turning latencies in Experiment 2 to
tain the final sample; infants were excluded buzzes occurring inside versus outside key syllable pairsfor familiar and novel test stimuli. Large positive differ-from the study for the following reasons: fail-ences are consistent with perception of the syllable pairsure to complete initial shaping within 40 trialsas cohesive units.or to meet the predetermined training criterion
within 30 trials in the initial session (19), dif-ficulties scheduling subsequent testing ses-sions within a 10 day period (5), and failure for the Iambic Meter condition, t(15) Å 2.44,
põ .05, but not for the Trochaic Meter condi-to contribute at least two latencies to each cellin the analysis (10). tion, t(15) Å 1.75, NS. Difference scores did
not vary significantly across the two stimulusApparatus, stimuli, and procedure. The ap-paratus, stimuli, and procedure used in this conditions, t(30) Å 00.52, NS.
Novel stimuli. Fourteen of 16 subjects instudy were identical to those used in Experi-ment 1. the Trochaic Meter condition responded more
slowly on average to within-unit buzzes thanResults and Discussion to between-units buzzes. Twelve of 16 sub-
jects in the Iambic Meter condition did like-As in Experiment 1, preliminary analysesshowed no significant effects involving the wise. Overall, subjects required 623 ms (SD
Å 113 ms) to initiate head turns on novelPosition factor, so treatment means were onceagain collapsed across levels of Position, and within-unit trials and 566 ms (SD Å 100 ms)
to initiate head turns on novel between-unitsthis factor was not used in subsequent analy-ses. Means and SEs of difference scores for trials. Although only the mean difference
score for the Trochaic Meter condition wasfamiliar and novel stimuli for each of the stim-ulus conditions are shown in Fig. 3. significantly greater than zero, t(15) Å 2.56,
p õ .05, difference scores did not vary sig-Familiar stimuli. Ten of 16 subjects in theTrochaic Meter condition, and 13 of 16 sub- nificantly across the two stimulus conditions,
t(30) Å 0.91, NS. The omnibus analysis ofjects in the Iambic Meter condition, respondedmore slowly on average to within-unit buzzes difference scores did not yield a significant
Stimulus Condition by Familiarity interaction,than to between-units buzzes. Overall, sub-jects required 599 ms (SD Å 113 ms) to initi- F(1,30) õ 1, NS (neither of the main effects
were significant in the omnibus analysis,ate head turns on familiar within-unit trials,and 547 ms (SD Å 105 ms) to initiate head either).
Unlike 9-month-olds, 6-month-old English-turns on familiar between-units trials. A sig-nificant positive difference score was found exposed infants do not appear to be biased
AID JML 2447 / a004$$$$$4 09-25-96 11:50:42 jmla AP: JML
677PREVERBAL SPEECH SEGMENTATION
toward perceiving trochaic bisyllables, whether tween expected and observed rhythmic pat-terns, but also lexical access may be mistak-familiar or novel, as being more cohesive por-
tions of the speech stream than iambic bisyl- enly initiated at the second syllable. On thesecond explanation, listeners may be biasedlables. This comports with Jusczyk, Cutler,
and Redanz’s (1993) finding that 6-month- toward imposing trochaic segmentation (oragainst imposing iambic segmentations) onolds showed no preference for listening to
trochaic, rather than iambic, bisyllables. In strings that they hear. Thus, iambs mis-stressed as trochees might be segmented ascontrast to Experiment 1, it is not clear that
introducing novel stimulus syllables was an single strong–weak units that correspond seg-mentally, if not rhythmically, to known unitseffective manipulation. First, in this experi-
ment, infants’ patterns of responding were (or words), whereas trochees mis-stressed asiambs might be segmented as sequences ofvery similar across familiar and novel sylla-
bles. Second, whereas 9-month-olds’ re- two units (the first weak, the second strong),neither corresponding segmentally to knownsponses to novel syllables were on average
about 90 ms slower than their responses to units. The pattern of reaction times observedby Cutler and Clifton was more compatiblefamiliar syllables, 6-month-olds displayed a
much smaller difference. This raises the possi- with the first explanation than the second.The present study is a miniaturized versionbility that in the noise detection task, younger
infants may become ‘‘locked in’’ to certain of Cutler and Clifton (1984), using 9-month-old infants as subjects. Infants were first famil-rhythmic patterns, to the point of failing to
notice or attend to segmental changes in the iarized with a particular bisyllable embodyingeither trochaic or iambic rhythm. They werestimuli. This comports with Morgan and Saf-
fran’s previous finding that 6-month-olds tend then tested to determine whether the familiarversion of the bisyllable and the mis-stressedto focus on rhythmic stimulus properties in
grouping syllables, providing evidence that bi- version of the bisyllable both are perceived asrelatively cohesive. In light of the results ofsyllables with consistent rhythmic patterns are
perceived as cohesive, even when the segmen- Experiment 1, it was expected that infantswould perceive familiarized bisyllables as be-tal content of those bisyllables varies.ing cohesive, regardless of the rhythmic pat-
EXPERIMENT 3 tern they manifested. However, because Ex-periment 1 suggests that 9-month-old infantsCutler and Clifton (1984) investigated En-
glish-speaking adults’ recognition of mis- exposed to English are biased to perceive tro-chaic bisyllables as being particularly cohe-stressed familiar words, finding that words
with unreduced vowels that were misstressed sive, it was also expected that infants wouldperceive novel versions of the bisyllables asto be trochaic were recognized faster than
were words mis-stressed to be iambic. In prin- being more cohesive when they were mis-stressed trochaically than when they are mis-ciple, this could come about for two reasons,
one having to do with lexical access, the other stressed iambically.The results of Experiment 1 suggest thathaving to do with segmentation per se.1 On
the first explanation, iambs mis-stressed as the most likely interpretation of the expectedpattern of results would appeal to infants’ bi-trochees might be recognized more quickly
than trochees mis-stressed as iambs because ases for imposing trochaic segmentations oninput strings. Recall that in Experiment 1, seg-strong syllables are more likely to initiate lexi-
cal access than are weak syllables (Cutler & mentally novel trochaic bisyllables were per-ceived by 9-month-olds as being cohesive. Be-Norris, 1988). Thus for trochees mis-stressed
as iambs, not only is there a mismatch be- cause these bisyllables were segmentallynovel, they could not have corresponded toany existing ‘‘lexical’’ representation; hence1 I thank one of the reviewers for pointing out the dis-
tinction between these possibilities. their cohesiveness was not related to processes
AID JML 2447 / a004$$$$$4 09-25-96 11:50:42 jmla AP: JML
678 JAMES L. MORGAN
of lexical access. Thus, the explanation pre- token of either [de] or [ti]. The tokens of [ko]and [ga], which constituted the key pair, wereferred by Cutler and Clifton for their adult
findings is not applicable to infants’ pro- always adjacent to one another and alwaysoccurred in the same relative order. The to-cessing of syllable strings.kens of [de] and [ti] served as filler syllables
Method and always appeared in either initial or finalposition. Within the strings, individual sylla-Subjects. Infants approximately 9 months
of age were recruited from the Providence bles were separated by 50 ms silences; stringsthemselves were separated by 1000 ms si-metropolitan area. Thirty-two infants, approx-
imately half male and half female, served as lences.Two between-subject stimulus conditionssubjects; 16 infants were assigned at random
to each of the two stimulus conditions. Sub- were employed. Each condition included atraining set of four trisyllabic strings, acrossjects had a mean age of 0;9.17 at the first
session (range 0;9.9 to 0;10.4). The two ses- which the key syllable pair consistently mani-fested one of the two rhythmic patterns used.sions required for each infant were scheduled
so that the time between the two sessions was This set of strings was used in initial training,as background stimuli between test trials, andno more than 10 days.
Forty-nine infants were tested in order to as test stimuli in half of the test trials (see‘‘Procedure’’ for additional details). Theattain the final sample; infants were excluded
from the study for the following reasons: fail- training set for the opposite condition pro-vided test stimuli for the remaining test trials.ure to complete initial shaping within 40 trials
or to meet the predetermined training criterion Stimulus conditions are described below andillustrated in Table 4.within 30 trials in the initial session (11), dif-
ficulties scheduling subsequent testing ses- In the Iambic Training condition, the rhyth-mic pattern across the key pair of syllables insions within a 10 day period (2), and failure
to contribute at least two latencies to each cell the training set of trisyllabic strings was iam-bic: a shorter syllable followed by a longerin the analysis (4).
Apparatus. The apparatus used in this study syllable. Across this set of strings, both thisconsistent rhythmic pattern and distributionalwas identical to that used in Experiments 1
and 2. (or sequential) information contributed evi-dence that the pair of syllables constituted aStimuli. The stimuli used in this experi-
ment included two digitally edited tokens unit. The distributional evidence also signaledthat [de] and [ti] each formed independenteach of [ko] and [ga], and one token each of
[de] and [ti], originally spoken in isolation units. Following training, infants were testedon strings with key pairs manifesting eitherwith a flat intonation contour by a voice-
trained adult female. Shorter versions of [ko] iambic or trochaic rhythms. Four subcondi-tions were included, to each of which fourand [ga] (henceforth denoted by [ko] and
[ga]) were 350 ms, whereas longer versions infants were assigned at random, and acrosswhich key syllable order and use of [de] orof [ko] and [ga] (henceforth denoted by [ko]
and [ga]) were 500 ms in duration. Shorter [ti] in test items were counterbalanced. Theseare illustrated in Table 4. This condition testedand longer versions of these syllables did
not differ in either amplitude or fundamental the cohesiveness of a trochaically accentedsyllable sequence relative to a familiar, iam-frequency contour. The tokens of [de] and
[ti] were edited to be 425 ms in duration. bically accented syllable sequence.In the Trochaic Training condition, theSee Morgan (1994) for additional details
concerning stimulus properties. rhythmic pattern over the key pair of syllablesin the training set of strings was trochaic: aSubjects heard the syllables in trisyllabic
strings. Each string included one token of [ko] longer syllable followed by a shorter syllable.Again, both rhythmic and distributional infor-or [ko], one token of [ga] or [ga], and one
AID JML 2447 / a004$$$$$4 09-25-96 11:50:42 jmla AP: JML
679PREVERBAL SPEECH SEGMENTATION
TABLE 4
Trisyllabic Strings Included in Stimulus Conditions, Experiments 3 and 4a
Condition Phase Stimulus strings
Iambic trainingSubcondition Ab Training gakode gakoti degako tigako
Testing gakode gakoti degako tigakogakode degako
Subcondition B Training kogade kogati dekoga tikogaTesting kogade kogati dekoga tikoga
kogade dekogaTrochaic training
Subcondition A Training gakode gakoti degako tigakoTesting gakode gakoti degako tigako
gakode degakoSubcondition B Training kogade kogati dekoga tikoga
Testing kogade kogati dekoga tikogakogade dekoga
a Acute accents (´) indicate longer syllables (500 ms); grave accents (`) indicate shorter syllables (350 ms); syllableswithout accents are medium length (425 ms).
b Subconditions C–D were similar to A–B, respectively, except that [ti], rather than [de], appeared in the noveltest items.
mation provided evidence that the pair of syl- tations for the rhythmic pattern manifested bythe second test string. Example stimulus se-lables constituted a unit, and distributional ev-
idence indicated that [de] and [ti] each consti- quences illustrating each trial type in the test-ing phase are provided in Table 5.tuted independent units. Following training,
infants were tested on strings with key pairsResults and Discussionmanifesting either trochaic or iambic rhythms.
Four subconditions were included, as shown As in Experiments 1 and 2, a preliminaryset of Stimulus Condition by Familiarity byin Table 4. This condition tested the cohesive-
ness of an iambically accented syllable se- Position omnibus analyses were conducted onthe difference scores. In this experiment, how-quence relative to a familiar, trochaically ac-
cented syllable sequence. ever, Position did interact with other factors;specifically, the Stimulus Condition by Posi-Procedure. The procedure in the Training
Session was identical to that followed in Ex- tion interaction was significant for novel stim-uli, F(1,30) Å 6.74, p õ .05. In light of this,periments 1 and 2. See Table 2 for illustrations
of example stimulus sequences from each subsequent analyses of responses to novelstimuli were carried out separately for firstphase of this session. The procedure followed
in the Testing Session was similar to that de- and second inter-syllable interval trials.Means and SEs of difference scores for famil-scribed earlier, except that the buzz appeared
within the second string in each trial. This iar stimuli collapsed across levels of Position,and for novel stimuli within each level of Posi-modification was adopted because infants
would have no means of discerning the rhyth- tion are shown in Fig. 4.Familiar stimuli. Eleven of 16 subjects inmic pattern across the key pair of syllables at
the point the buzz occurred if the buzz were the Iambic Training condition responded moreslowly on average to within-unit buzzes thanpresented within the first string. In this experi-
ment, the rhythmic pattern manifested by the to between-units buzzes; 14 of 16 subjects inthe Trochaic Training condition did likewise.first test string served to prime infants’ expec-
AID JML 2447 / a004$$$$$4 09-25-96 11:50:42 jmla AP: JML
680 JAMES L. MORGAN
TABLE 5 likewise. Subjects in the Iambic Training con-dition required 604 ms (SD Å 112 ms) to initi-EAMPLES OF TESTING TRIALS IN EXPERIMENTS 3 AND 4a
ate head turns on within-unit trials, and 545Trial type Trial stimulib ms (SD Å 89 ms) to initiate head turns on
between-units trials. Subjects in the TrochaicFamiliar
Training condition required 574 ms (SD ÅWithin-unit102 ms) to initiate head turns on within-unitfirst-interval gakode gafkode gakode
Within-unit trials, and 598 ms (SD Å 78 ms) to initiatesecond-interval degako degafko degako head turns on between-units trials. Neither
Between-units mean difference score was significantlyfirst-interval degako defgako degako
greater than zero: for Iambic Training, t(15)Between-unitsÅ 1.85, p Å .08, and for Trochaic Training,second-interval gakode gakofde gakodet(15) Å 01.11, NS. However, difference
Novelscores did vary significantly across the twoWithin-unitconditions, t(30) Å 2.16, põ .05. Further evi-first-interval gakode gafkode gakode
Within-unit dence for different patterning of results acrosssecond-interval degako degafko degako conditions for familiar stimuli versus first-po-
Between-units sition novel stimuli was provided by a sig-first-interval degako defgako degako
nificant Stimulus Condition by Familiarity in-Between-unitsteraction in the omnibus analysis of differencesecond-interval gakode gakofde gakodescores, F(1,30) Å 4.76, p õ .05. Responses
a Possible sequences are given here only for Subcondi- to first-position novel stimuli fit the predictiontion A of the Trochaic Training condition, in which [gako] that infants would perceive mis-stressed ver-served as the key syllable pair in training, and all strings
sions of the bisyllables to be more cohesivein test trials included [de] as the filler syllable.when they were mis-stressed trochaically thanb f Indicates the location of the buzz within a trial.when they were mis-stressed iambically.
Overall, subjects required 588 ms (SD Å 105ms) to initiate head turns on familiar within-unit trials, and 526 ms (SDÅ 93 ms) to initiatehead turns on familiar between-units trials.Both groups of subjects displayed significantpositive difference scores: for Iambic Train-ing, t(15) Å 2.39, p õ .05, and for TrochaicTraining, t(15) Å 2.74, p õ .05. Differencescores did not vary significantly across the twostimulus conditions, t(30) Å 0.43, NS. Thispattern of results is similar to those of Experi-ments 1 and 2, and again suggests that noprocessing advantage exists for familiar tro-chaic over familiar iambic sequences.
First position novel stimuli. Twelve of 16subjects in the Iambic Training condition (forwhom novel stimuli were trochaic) responded
FIG. 4. Nine-month-olds. Means and standard errors ofmore slowly on average to first intervaldifferences in head turning latencies in Experiment 3 towithin-unit buzzes than to first interval be-buzzes occurring inside versus outside key syllable pairs
tween-units buzzes. However, only 6 of 16 for familiar and novel test stimuli. Large positive differ-subjects in the Trochaic Training condition ences are consistent with perception of the syllable pairs
as cohesive units.(for whom novel stimuli were iambic) did
AID JML 2447 / a004$$$$$4 09-25-96 11:50:42 jmla AP: JML
681PREVERBAL SPEECH SEGMENTATION
Second position novel stimuli. Only 7 of virtue of a rhythmic change or a segmentalchange) are perceived as being more cohesive16 subjects in each condition responded more
slowly on average to second interval within- than are comparable novel iambic sequences.unit buzzes than to second interval between-units buzzes. Overall, subjects required 521 EXPERIMENT 4ms (SD Å 99 ms) to initiate head turns on
The results of Experiment 2 suggested that,within-unit trials, and 526 ms (SD Å 86 ms)
in tasks comparable to that used here, 6-to initiate head turns on between-units trials.
month-old infants may get locked on to rhyth-Neither mean difference score was signifi-
mic properties of speech stimuli to the extentcantly greater than zero: for both conditions,
that they fail to attend to segmental changest(15) õ 0, NS, nor did difference scores vary
in those stimuli. On this view, 6-month-oldssignificantly across the two conditions, t(30)
in Experiment 2 perceived segmentally novelÅ 0.56, NS.syllable pairs (manifesting a familiar rhythm)
Responses on second-position novel stimulito be as cohesive as segmentally familiar pairs
clearly do not fit the prediction that mis-because they overlooked their novelty. When
stressed trochaic stimuli would be perceivedrhythmic changes are introduced, however, a
as being more cohesive that mis-stressed iam-different pattern of results should emerge: 6-
bic stimuli. A likely explanation for thismonth-olds may be likely to perceive segmen-
hinges on the design variation that was intro-tally familiar syllable pairs as lacking cohe-
duced in this experiment, namely, that buzzessiveness when they appear in a novel rhythm.
occurred within the second string in each trial.If 6-month-olds process trochaic and iambic
On novel trials, the key syllable pair in therhythms in similar fashion, as previous results
first string manifested the unfamiliar rhythmicfrom Jusczyk, Cutler, and Redanz (1993a), as
pattern. This was done to prime infants to ex-well as results from Experiment 2, suggest,
pect the unfamiliar rhythmic pattern in the sec-then changes from trochaic to iambic, or iam-
ond string, but it may well have also cuedbic to trochaic, should yield similar patterns
infants that a trial was about to occur (theof responses.
change in rhythm was a perfect cue in thisregard). In corroboration of this, note that in-
Methodfants’ latencies were very short in novel trialsin this experiment, particularly in second-in- Subjects. Infants approximately 6 months
of age were recruited from the Providenceterval trials. Possibly, anticipatory respondingmay have washed out effects of stimulus metropolitan area. Thirty-two infants, approx-
imately half male and half female, served asrhythm in these trials. Including a number ofno-buzz catch trials might have reduced such subjects; 16 infants were assigned at random
to each of the two stimulus conditions. Sub-anticipatory responding; unfortunately, suchtrials would also have reduced the novelty of jects had a mean age of 0;6.22 at the first
session (range 0;6.10 to 0;7.6). The two ses-the unfamiliar rhythmic pattern.Results from both familiar trials and first- sions required for each infant were scheduled
so that the time between the two sessions wasinterval novel trials, however, do fit the pre-dicted patterns. Data from the familiar trials no more than 10 days.
Seventy-one infants were tested in order toprovide additional evidence that familiar tro-chaic and iambic sequences are processed attain the final sample; infants were excluded
from the study for the following reasons: fail-similarly by 9-month-olds—neither is per-ceived as being more or less cohesive than the ure to complete initial shaping within 40 trials
or to meet the predetermined training criterionother. Data from the first-interval novel trialscomport with the findings of Experiment 1: within 30 trials in the initial session (18), dif-
ficulties scheduling subsequent testing ses-For 9-month-old English-exposed infants,novel trochaic sequences (whether novel by sions within a 10 day period (8), and failure
AID JML 2447 / a004$$$$$4 09-25-96 11:50:42 jmla AP: JML
682 JAMES L. MORGAN
2.33, p õ .05, but not for the Iambic Trainingcondition, t(15) Å 1.87, p Å .08. Differencescores did not vary significantly across the twostimulus conditions, t(30) Å 0.42, NS.
Novel stimuli. Only 6 of 16 subjects in theIambic Training condition and 7 of 16 subjectsin the Trochaic Training condition respondedmore slowly on average to within-unit buzzesthan to between-units buzzes. Overall, sub-jects required 695 ms (SD Å 94 ms) to initiatehead turns on novel within-unit trials, and 682ms (SD Å 77 ms) to initiate head turns onnovel between-units trials. Mean differencescores were not significantly different thanzero for either condition; in both cases, t(15)
FIG. 5. Six-month-olds. Means and standard errors of õ 1, NS.differences in head turning latencies in Experiment 4 to
Six-month-old English-exposed infants ap-buzzes occurring inside versus outside key syllable pairspear to be equally affected by any change infor familiar and novel test stimuli. Large positive differ-
ences are consistent with perception of the syllable pairs rhythm, whether it is from trochaic to iambic,as cohesive units. or from iambic to trochaic. Regardless of the
direction of change, 6-month-olds apparentlyfail to perceive novelly accented bisyllablesas being cohesive, even though the segmentalto contribute at least two latencies to each cell
in the analysis (13). content of the bisyllable is familiar. In contrastto Experiment 2, it is clear that introducingApparatus, stimuli, and procedure. The ap-
paratus, stimuli, and procedure used in this novel stimulus rhythms was an effective ma-nipulation, as evidenced both by infants’ dif-study were identical to those used in Experi-
ment 3. ferent patterns of responding to familiar andnovel rhythms and by infants’ relatively long
Results and Discussion response latencies to stimuli with novelrhythms. The results of this study are consis-As in Experiments 1 and 2, preliminary
analyses showed no significant effects involv- tent with both the results of Experiment 2 andthose of Jusczyk, Cutler, and Redanz (1993a),ing the Position factor. Thus, treatment means
were collapsed across levels of Position, and showing that at 6 months trochaic rhythm isnot privileged in speech segmentation.this factor was not used in subsequent analy-
ses. Means and SEs of difference scores forGENERAL DISCUSSIONfamiliar and novel stimuli for each of the stim-
ulus conditions are shown in Fig. 5. The results of the present set of studies sup-port three conclusions. First, any initial biasesFamiliar stimuli. Ten of 16 subjects in the
Iambic Training condition, and 11 of 16 sub- 6- and 9-month-old infants may have for pref-erentially extracting or processing trochaicjects in the Trochaic Training condition, re-
sponded more slowly on average to within- versus iambic syllable sequences are neutral-ized by sufficient exposure to those sequences.unit buzzes than to between-units buzzes.
Overall, subjects required 627 ms (SD Å 90 In all four experiments, infants providedequivalent evidence of having represented fa-ms) to initiate head turns on familiar within-
unit trials, and 560 ms (SDÅ 97 ms) to initiate miliar trochaic or iambic bisyllables as cohe-sive units. Repeated exposure to the bisyll-head turns on familiar between-units trials. A
significant positive difference score was found ables used in training may have induced in-fants to form a rudimentary sort of ‘‘lexicalfor the Trochaic Training condition, t(15) Å
AID JML 2447 / a004$$$$$4 09-25-96 11:50:42 jmla AP: JML
683PREVERBAL SPEECH SEGMENTATION
entry’’ for the familiar bisyllable; on any ac- conclusively rule out the possibility of anearly, universal perceptual bias in processingcount, segmentation must rely at least in part
on matching of input to existing lexical repre- of trochaic sequences. Such evidence mightbe obtained by testing infants exposed to lan-sentations. On this view, it is not surprising
to find that such matching may override initial guages in which initial stress in multisyllabicwords is exceptional (such as French or He-processing biases.
Second, at 6 months, infants do not appear brew) under the procedures used here. If thepresent results are indeed due to 9-month-to process novel trochaic bisyllables differ-
ently than novel iambic bisyllables. This result olds’ exposure to English, then 9-month-old,Hebrew-exposed infants ought to yield com-is consistent with Jusczyk, Cutler, and Re-
danz’s (1993) failure to find any rhythmic plementary patterns of responses, perceivingnovel iambic sequences as constituting morepreference in 6-month-olds. At 6 months (at
least in the sort of task used here), infants cohesive units than novel trochaic sequences.In the absence of such evidence, it stilldo appear to be biased to attend to rhythmic
properties, rather than segmental properties, of seems most plausible that the bias observedhas its roots in infants’ linguistic experience.speech stimuli. Thus, 6-month-olds provided
evidence of having represented both novel tro- Certainly, it would be a most curious coinci-dence if an innate bias for processing metricalchaic and novel iambic bisyllables as cohesive
units, when the bisyllables were novel by vir- sequences would manifest itself over just thesame period that changes in perception of seg-tue of segmental changes. However, when the
bisyllables were novel by virtue of rhythmic mental properties of speech begin to reflectlinguistic experience. Evidence for suchchanges, 6-month-olds provided evidence that
they represented neither novel trochaic nor changes has been provided by Best (1993);Best, McRoberts, and Sithole (1988); Jusczyk,novel iambic bisyllables as cohesive units.
Third, at 9 months, infants are biased to- Friederici, Wessels, Svenkerud, and Jusczyk(1993); Jusczyk, Luce, and Charles-Luceward perceiving novel trochaic bisyllables as
forming more cohesive units than novel iam- (1994); Kuhl, Williams, Lacerda, Stevens, andLindblom (1992); and Werker and Teesbic bisyllables. This bias is evident whether
the bisyllables are novel by virtue of segmen- (1984). The assumption that the trochaic biasis not innate, however, requires an account oftal changes (as in Experiment 1) or by virtue
of rhythmic changes (as in Experiment 3). how experience (e.g., with English) mightgive rise to this bias.These results are consistent with Jusczyk, Cut-
ler, and Redanz’s (1993) finding of a prefer- Perhaps the simplest account is that infantsmight generalize the properties of primitiveence for trochaic bisyllables at 9 months and
suggest that English-learning infants may perceptual units to word-like units. Utter-ances—stretches of speech bounded by si-adopt a metrical strategy for segmentation be-
fore the end of the first year. lences—are suitable candidates for primitiveunits; note that boundaries of utterances nec-How might infants come to adopt a trochaic
bias? One possibility is that this bias reflects essarily coincide with word boundaries (ex-cept in instances of dysfluencies, which area universal stage of language acquisition. Such
a hypothesis has been advanced with regard quite rare in child-directed speech). On thisexplanation, if utterances typically begin withto children’s early speech production by Allen
and Hawkins (1980) and recently echoed, in strong syllables and end with weak syllables,it would be reasonable for learners to hypothe-part, by Wijnen, Krikhaar, and den Os (1994).
However, analyses of early production across size that smaller units within utterances beginand end with comparable types of syllables.a broad range of languages provide evidence
that such productions are conditioned by met- Strategies that involve scanning from oneedge of a prosodic unit to the other, such asrical properties of specific input languages
(Demuth, 1996). No evidence yet exists to that proposed in Dresher and Kaye (1990) for
AID JML 2447 / a004$$$$$5 09-25-96 11:50:42 jmla AP: JML
684 JAMES L. MORGAN
TABLE 6
METRICAL PROPERTIES OF PERIPHERAL SYLLABLES IN MATERNAL UTTERANCES
All utterances Two-syllable utterances
Child N Strong–weak Weak–strong N Strong–Weak Weak–strong
setting of the Foot Rhythm parameter, will terances to properties of words, they might inyield comparable results. If such parameters some cases be inclined to adopt an iambicare to be set before an extensive lexicon has bias, which apparently does not occur.been established, the unit that is scanned will An alternative possibility is that the trochaicgenerally not be equivalent to a word, but will bias in infants learning English emerges frommore likely be an utterance. the initial stages of word-learning. On this
However, the syntactic structure of English view, infants begin with a general expectationmilitates against input utterances manifesting that rhythm will play a role in defining linguis-the metrical pattern desired under this class of tic units (and perhaps some notion that theexplanation: Phrases (and hence utterances) binary stress foot may be a significant rhyth-often begin with function-word specifiers mic unit). Evidence of early attention to rhyth-(such as determiners, which are typically weak mic properties of speech (e.g., Mehler, Jusc-syllables) and often end with content-word zyk, Lambertz, Halsted, Bertoncini, & Amiel-heads. A computational analysis of maternal Tison, 1988) supports this contention. Aninput to Adam and Eve in the Brown corpora appropriate rhythmic bias (for trochaic rather(Brown, 1973; MacWhinney & Snow, 1985) than iambic stress feet) then arises as a resultshows that 64% of maternal utterances began of early, limited distributional analyses of in-with a function word and 66% of maternal put. Infants may occasionally extract bisyll-utterances ended with a content word, of ables, some iambic, some trochaic, from inputwhich the overwhelming preponderance were speech to entertain as possible words. If themonosyllabic. As shown in Table 6, further two syllables in question occur together withanalyses of accent patterns reveal that input some frequency in subsequent input, and ifutterances do not generally begin and end with neither of the two syllables occurs in combina-strong and weak syllables, respectively, tion with large numbers of different syllables,whether all utterances, or only bisyllabic utter- then the putative unit will be retained; other-ances, are considered.2 Thus, if children at- wise, it will be dropped. Trochaic bisyllablestempted to extrapolate from properties of ut- that an infant acquiring English might extract
are likely to occur again in the input, inasmuchas English bisyllabic words (or word-stems)2 These analyses were based on orthographic transcrip-
tions and therefore provide only approximate character- tend to be trochaic. Some iambic bisyllablesizations of speech children hear. For purposes of these that an infant might extract, such as thedog,analyses, all monosyllabic function words were consid-
will also occur repeatedly in the input. How-ered to be weak syllables, all other monosyllabic wordsever, the first syllables of these will often bewere considered to be strong syllables, and all multisyl-
labic words were considered to be stressed in accordance function words and will thus typically occur inwith their citation forms. Function words in utterance- combination with very many different secondfinal position are often realized as strong syllables: What’s syllables, a pattern supporting their analysisthat or Put it down. Therefore, the ratio of SW to WS
as independent linguistic elements.utterances is probably somewhat smaller than indicatedby the data in Table 6. An analysis of bisyllabic utterances that re-
AID JML 2447 / a004$$$$$5 09-25-96 11:50:42 jmla AP: JML
685PREVERBAL SPEECH SEGMENTATION
curred in maternal speech to Adam and Eve the appearance of these changes in speech per-ception at 9 or 10 months may ultimately rest(either in isolation or within larger utterances)
supports the plausibility of a distributional on broader changes in infants’ cognition oc-curring at this epoch.strategy.3 Virtually all recurring two-word bi-
syllabic utterances included at least one func- Regardless of how it may originate, infants’use of a trochaic strategy for segmentationtion word; sufficient evidence existed for these
to be analyzed as two-element sequences. In ought to have implications for how acquisitionproceeds. An obvious prediction, one that isrecurring one-word bisyllabic utterances, tro-
chaic types outnumbered iambic types by ra- strongly confirmed for English (Gerken, 1994)and Dutch (Wijnen, Krikhaar, & den Os,tios of 5 to 1 for Adam and 9 to 1 for Eve.
In essence, this strategy harnesses the infor- 1994), is that infants’ early words should pre-dominately conform to the strong–(weak)mation available in the input lexicon. For ex-
ample, in maternal utterances to both Adam rhythmic pattern. These data are ambiguous,however; this pattern is equally well explainedand Eve, 85% of bisyllabic word tokens, and
92% of bisyllabic word types, manifested by appeals to the nature of the lexicons of thelanguages in question or to biases in speechstrong–weak rhythms. Note that use of baby-
talk forms in English, such as doggy or kitty, production. Somewhat more subtly, use of ametrical strategy for segmentation may influ-in which monosyllabic nouns are converted to
canonical strong–weak bisyllables, causes the ence children’s acquisition of grammaticalmorphemes: Frequent appearance in metricalstrong–weak bias found in adult-directed
speech to be exaggerated in child-directed environments from which they are easily seg-mented should facilitate acquisition of particu-speech. Ferguson (1964) observed that com-
parable affixes are found in baby talk registers lar morphemes.Consider, for example, the relationship be-of a variety of languages. In other languages,
such as French, monosyllables are redupli- tween parental yes–no questions and chil-dren’s acquisition of auxiliary verbs. Severalcated to produce baby-talk forms exhibiting
canonical metrical patterns. In short, a distri- studies have found a significant positive rela-tionship for children learning English (Fur-butional analysis of English input proceeding
from initial bisyllabic segmentations will tend row, Nelson, & Benedict, 1979; Gleitman,Newport, & Gleitman, 1984; Newport, Gleit-to reinforce a trochaic bias and discourage an
iambic bias, whereas a comparable distribu- man, & Gleitman, 1977; Scarborough & Wy-coff, 1986); indeed, this is the single mosttional analysis of, say, Hebrew, would have
the opposite result. robust finding in the literature on input andacquisition. Newport, Gleitman, and GleitmanChanges in perception of segmental proper-
ties of speech reported by Werker and Tees advanced one explanation for this relation-ship, suggesting that normally unstressed aux-(1984), Jusczyk, Friederici, Wessels, Svenk-
erud, and Jusczyk (1993), and others are also iliaries become salient in initial position inyes–no questions because they are stressed.traceable to infants’ distributional analyses of
input. On this view, the appearance of a tro- However, no acoustic measurements of utter-ance-initial auxiliaries in child-directedchaic bias appears contemporaneously with
these other changes in speech perception be- speech have supported this claim.The metrical segmentation strategy hypoth-cause they all result from the same underlying
process. Morgan and Saffran (1995) argue that esis provides an alternative explanation for therelation between parental yes–no questionsdistributional analyses rest on the ability to
integrate disparate forms of information; thus, and children’s acquisition of auxiliaries: Ut-terance-initial auxiliaries contribute to acqui-sition because they are easily segmentable3 This analysis included bisyllabic utterances that re-(more generally, functors consisting of singlecurred within the same recording session and covered
transcripts 1–10 for each child. weak syllables will be most easily segmented
AID JML 2447 / a004$$$$$5 09-25-96 11:50:42 jmla AP: JML
686 JAMES L. MORGAN
when they appear in utterance-initial posi- whether the isolated sequences of sounds arewords.tion). The leading edge of an utterance-initial
Recent work in our laboratory suggests thatauxiliary will be readily identifiable becauseadditional biases, relating to phonotactic prop-it will typically follow a pause (Broen, 1972).erties of the native language, may appear be-The second word of the sentence may beginfore the end of the first year: At 10 months,with a strong syllable or a weak syllable. If itinfants perceive bisyllables with common syl-begins with a strong syllable, under the metri-lable-to-syllable transitions (e.g., ‘‘monkey’’)cal segmentation strategy, this syllable will beas more cohesive than bisyllables with infre-taken as signaling a word boundary, and thequent syllable-to-syllable transitions (e.g.,initial auxiliary will fall out as residue. If the‘‘reptile’’). Whether and how a trochaic seg-second word begins with a weak syllable, thementation strategy might interact with othermetrical segmentation strategy will provide nosegmentation biases is unknown at present.impetus to group it with the utterance-initialUnderstanding how early speech segmenta-(weak) auxiliary, and the initial auxiliary maytion proceeds remains an important problem,again fall out as residue. Contrast this withbecause it is the representations resulting fromthe segmentability of utterance-medial auxil-segmentation that provide the data for chil-iaries: Under the metrical segmentation strat-dren’s discovery of the structure and meaningegy, such auxiliaries, as weak syllables, willof their native language.tend to be aggregated with the most closely
preceding strong syllable and will not be REFERENCES
readily identifiable as individual words. Note ABRAMS, K., & BEVER, T. (1969). Syntactic structuremodifies attention during speech perception and rec-that this explanation predicts that frequent ap-ognition. Quarterly Journal of Experimental Psy-pearance in utterance-initial position may fa-chology, 21, 280–290.cilitate acquisition of other normally un-
ALLEN, G., & HAWKINS, S. (1980). Phonological rhythm:stressed grammatical morphemes. Definition and development. In G. Yeni-Komshian,
J. Kavanagh, & C. Ferguson (Eds.), Child Phonol-A trochaic strategy, of course, cannot pro-ogy, Vol. 1 (pp. 227–256). New York: Academicvide a complete account of early speech seg-Press.mentation. Not all trochees are words, nor are
BARD, E., & ANDERSON, A. H. (1983). The unintelligibil-all words trochaic. As noted above, matching ity of speech to children. Journal of Child Language,of input sequences to existing (lexical) repre- 10, 265–292.
BEST, C. T. (1993). Emergence of language-specific con-sentations must also play a role; clearly, rec-straints in perception of non-native speech: A win-ognition of sequences of speech sounds as ex-dow on early phonological development. In B. deemplars of lexical entries (which after all is the Boysson-Bardies, S. de Schoen, P. Jusczyk, P.
goal which segmentation serves) must involve McNeilage & J. Morton (Eds.), Developmental neu-rocognition: Speech and face processing during thesuch matching. In establishing a lexicon, orfirst year of life (pp. 289–304). Dordrecht: Kluwer.learning new words, infants might be able to
BEST, C. T., MCROBERTS, G. W., & SITHOLE, N. M.apply a trochaic strategy as a means of initially(1988). Examination of the perceptual reorganization
isolating sequences of speech sounds that have for speech contrasts: Zulu click discrimination bya high probability of constituting English English-speaking adults and infants. Journal of Ex-
perimental Psychology: Human Perception and Per-words. But whether a sequence of sounds isformance, 14, 345–360.actually a word also depends on its distribu-
BOND, Z. S. (1972). Phonological units in sentence per-tional regularities: The sequence can consti- ception. Phonetica, 25, 129–139.tute a word only if it occurs in a predictable BROEN, P. (1972). The verbal environment of the lan-
guage-learning child. Monograph of the Americanform in a variety of contexts. On this view,Speech and Hearing Association (No. 17).application of a trochaic strategy can help to
BROWN, R. (1973). A first language. Cambridge, MA:maximize the efficiency (by limiting theHarvard Univ. Press.
scope) of the required distributional/sequen- CHOMSKY, N., & HALLE, M. (1968). The sound patternof English. New York: Harper & Row.tial analysis ultimately required to determine
AID JML 2447 / a004$$$$$5 09-25-96 11:50:42 jmla AP: JML
687PREVERBAL SPEECH SEGMENTATION
COLE, R. A., & JAKIMIK, J. (1978). Understanding speech: (1984). The current status of the motherese hypothe-sis. Journal of Child Language, 11, 43–79.How words are heard. In G. Underwood (Ed.), Strat-
egies of information processing (pp. 67–116). New HOLMES, V. M., & FORSTER, K. I. (1970). Detection ofYork: Academic Press. extraneous signals during sentence recognition. Per-
ception and Psychophysics, 7, 297–301.CUTLER, A. (1990). Exploiting prosodic probabilities inspeech segmentation. In G. T. M. Altman (Ed.), Cog- JUSCZYK, P. W., CUTLER, A., & REDANZ, L. (1993). In-
fants’ sensitivity to predominant stress patterns innitive models of speech processing (pp. 105–121).English. Child Development, 64, 675–687.Cambridge, MA: MIT Press.
JUSCZYK, P. W., FRIEDERICI, A. D., WESSELS, J. M.,CUTLER, A. (1996). Prosody and the word boundary prob-SVENKERUD, V. Y., & JUSCZYK, A. M. (1993). In-lem. In J. L. Morgan & K. Demuth (Eds.), Signal tofants’ sensitivity to the sound patterns of native lan-syntax (pp. 87–99). Mahwah, NJ: Erlbaum.guage words. Journal of Memory and Language, 32,CUTLER, A., & BUTTERFIELD, S. (1992). Rhythmic cues402–420.to speech segmentation: Evidence from juncture mis-
JUSCZYK, P. W., LUCE, P., & CHARLES-LUCE, J. (1994).perception. Journal of Memory and Language, 31,Infants’ sensitivity to phonotactic patterns in the na-218–236.tive language. Journal of Memory and Language, 33,CUTLER, A., & CARTER, D. M. (1987). The predominance630–645.of strong initial syllables in the English vocabulary.
JUSCZYK, P. W., & THOMPSON, E. (1978). Perception ofComputer Speech and Language, 2, 133–142.a phonetic contrast in multisyllabic utterances by 2-CUTLER, A., & CLIFTON, C. (1984). The use of prosodicmonth-old infants. Perception and Psychophysics,information in word recognition. In H. Bouma & D.23, 105–109.G. Bouwhuis (Eds.), Attention and performance X
KUHL, P. K. (1985). Methods in the study of infant speech(pp. 183–196). Hillsdale, NJ: Erlbaum.perception. In G. Gottleib & N. Krasnegor (Eds.),CUTLER, A., MEHLER, J., NORRIS, D., & SEGUI, J. (1986).Measurement of audition and vision in the first yearThe syllable’s differing role in the segmentation ofof postnatal life: A methodological overview. Nor-French and English. Journal of Memory and Lan-wood, NJ: Ablex.guage, 25, 385–400.
KUHL, P. K., WILLIAMS, K. A., LACERDA, F., STEVENS,CUTLER, A., & NORRIS, D. G. (1988). The role of strongK. N., & LINDBLOM, B. (1992). Linguistic experiencesyllables in segmentation for lexical access. Journalalters phonetic perception in infants by 6 months ofof Experimental Psychology: Human Perception andage. Science, 255, 606–608.
Performance, 14, 113–121.MACWHINNEY, B., & SNOW, C. E. (1985). The child lan-
DEMANY, L., MACKENZIE, B., & VURPILLOT, E. (1977).guage data exchange system. Journal of Child Lan-
Rhythmic perception in early infancy. Nature, 266,guage, 12, 271–296.
DEMUTH, K. (1996). The prosodic structure of earlyBERTONCINI, J., & AMIEL-TISON, C. (1988). A pre-
words. In J. L. Morgan & K. Demuth (Eds.), Signal cursor of language acquisition in young infants. Cog-to syntax (pp. 171–184). Mahwah, NJ: Erlbaum. nition, 29, 143–178.
DRESHER, B. E., & KAYE, J. D. (1990). A computational MORGAN, J. L. (1994). Converging measures of speechlearning model for metrical phonology. Cognition, segmentation in preverbal infants. Infant Behavior34, 137–195. and Development, 17, 389–403.
FERGUSON, C. (1964). Baby talk in six languages. Ameri- MORGAN, J. L., & SAFFRAN, J. R. (1995). Emerging inte-can Anthropologist, 66, 103–114. gration of sequential and suprasegmental information
FLORES D’ARCAIS, G. B. (1978). The perception of com- in preverbal speech segmentation. Child Develop-plex sentences. In W. J. M. Levelt & G. B. Flores ment, 66, 911–936.d’Arcais (Eds.), Studies in the perception of lan- NAKATANI, L. H., & SCHAFFER, J. A. (1978). Hearingguage. New York: Wiley. ‘‘words’’ without words: Prosodic cues for word per-
FODOR, J. A., BEVER, T. G., & GARRETT, M. F. (1974). ception. Journal of the Acoustical Society ofThe psychology of language. New York: McGraw– America, 63, 234–245.Hill. NEWPORT, E. L., GLEITMAN, H., & GLEITMAN, L. R.
FURROW, D., NELSON, K., & BENEDICT, H. (1979). Moth- (1977). Mother, I’d rather do it myself: Some effectsers’ speech to children and syntactic development: and non effects of maternal speech style. In C. E.Some simple relationships. Journal of Child Lan- Snow & C. A. Ferguson (Eds.), Talking to children:guage, 6, 423–442. Language input and acquisition. Cambridge: Cam-
GERKEN, L. A. (1994). Young children’s representation bridge Univ. Press.of prosodic phonology: evidence from English- OTAKE, T., HATANO, G., CUTLER, A., & MEHLER, J.speakers’ weak syllable omissions. Journal of Mem- (1993). Mora or syllable? Speech segmentation inory and Language, 33, 19–38. Japanese. Journal of Memory & Language, 32, 358–
378.GLEITMAN, L. R., NEWPORT, E. L., & GLEITMAN, H.
AID JML 2447 / a004$$$$$6 09-25-96 11:50:42 jmla AP: JML
688 JAMES L. MORGAN
POLLACK, I., & PICKETT, J. M. (1964). The unintelligibil- TREHUB, S. E., & THORPE, L. A. (1989). Infants’ percep-tion of rhythm: Categorization of auditory sequencesity of excerpts from conversations. Language &
Speech, 6, 165–171. by temporal structure. Canadian Journal of Psychol-ogy, 43, 217–229.RATNER, N. B. (1985). Dissociations between vowel dura-
tions and formant frequency characteristics. Journal WERKER, J., & TEES, R. C. (1984). Cross-language speechperception: Evidence for perceptual reorganizationof Speech and Hearing Research, 28, 255–264.
SCARBOROUGH, H., & WYCOFF, J. (1986). Mother, I’d during the first year of life. Infant Behavior and De-velopment, 7, 49–63.still rather do it myself: some further noneffects of
‘motherese.’ Journal of Child Language, 13, 431– WIJNEN, F., KRIKHAAR, E., & DEN OS, E. (1994). The(non)realization of unstressed elements in children’s438.
SELKIRK, E. O. (1984). Phonology and syntax. Cam- utterances: evidence for a rhythmic constraint. Jour-nal of Child Language, 21, 59–83.bridge, MA: MIT Press.
SPRING, D. R., & DALE, P. S. (1977). Discrimination of(Received July 15, 1994)linguistic stress in early infancy. Journal of Speech &
Hearing Research, 20, 224–232. (Revision received August 11, 1995)
AID JML 2447 / a004$$$$$6 09-25-96 11:50:42 jmla AP: JML
