Journal of Memory and Language
Journal of Memory and Language 54 (2006) 283–299
www.elsevier.com/locate/jml
Early segmentation of fluent speech by infants acquiring French: Emerging evidence for crosslinguistic differences Thierry Nazzi *, Galina Iakimova, Josiane Bertoncini, Se´verine Fre´donie, Carmela Alcantara Laboratoire Cognition et De´veloppement, CNRS—Universite´ Paris 5, Paris, France Received 25 March 2005; revision received 12 October 2005 Available online 18 January 2006
Abstract Four experiments explored French-learning infants’ ability to segment words from fluent speech. The focus was on bisyllabic words to investigate whether infants segment them as whole words or segment each syllable individually. No segmentation effects were found in 8-month-olds. Twelve-month-olds segmented individually both the final syllables and, under appropriate test conditions, the initial syllables of these bisyllabic words, but failed to segment bisyllabic words as whole units. The opposite pattern was observed at 16 months: final syllables were not segmented, while there was evidence that the words were segmented as whole units. The present findings are consistent with the proposal that the syllable is a unit of prosodic segmentation in French, therefore introducing evidence from a syllable-based language in support of the more general hypothesis that the emergence of segmentation abilities differs crosslinguistically as a function of the rhythmic class of the language in acquisition. 2005 Elsevier Inc. All rights reserved. Keywords: Speech segmentation; Lexical acquisition; Crosslinguistic comparisons; French; Rhythm; Syllable
The segmentation of fluent speech into words constitutes a critical aspect of speech processing. It allows adults and infants to determine the sequence of lexical units that make up the utterances they hear. Moreover, because most speech to infants is fluent speech rather than isolated words, speech segmentation has to play a central role in the early acquisition of words by infants. The ability to segment fluent speech into word forms has been shown to emerge during the first year of life, around 8 months of age (Jusczyk & Aslin, 1995). However, the bulk of the research on this issue has been con-
*
Corresponding author. Fax: +33 1 55 20 59 85. E-mail address:
[email protected] (T. Nazzi).
ducted in English. Therefore, determining the pattern of emergence of segmentation abilities in different languages has become a necessity. Because prosodic information, and more specifically rhythmic properties, appears to play a central role in early segmentation in English, the present paper focuses on the early segmentation abilities of infants growing up hearing a language with radically different rhythmic properties: French. Learning the sound patterns of individual words is a requirement for the acquisition of both the lexicon (these sound patterns being associated to their meanings) and syntax (as theories of syntax acquisition assume that infants process sentences as sequences of individuated words). Therefore, it is crucial for infants to have the capacity to access the isolated forms of the words they
0749-596X/$ - see front matter 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.jml.2005.10.004
284
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
hear. This issue would be trivial if the boundaries between the words spoken to infants were clearly marked at the acoustic level, or if words were (often) presented in isolation. However, there are no obvious pauses between words in connected speech, and no clear and systematic marking of word boundaries at the acoustic level (Cole & Jakimik, 1978, 1980; Klatt, 1979, 1989). Moreover, a few studies have evaluated the presence of isolated word forms in the input received by infants acquiring English (Aslin, 1993; Brent & Siskind, 2001) or in the input received by a Dutch/German bilingual infant (van de Weijer, 1998). The results of these studies suggest that most speech to infants consists of multiword utterances, words spoken in isolation accounting for no more than 10% of all the words present in the analyzed samples. These isolated forms might help infants’ acquisition of these words, as supported by the finding that the frequency with which a word is presented in isolation (rather than the total frequency of that word) partly predicts whether it will be produced several months later (Brent & Siskind, 2001). However, the need for procedures to segment continuous speech remains given that not every type of word appears in isolation (e.g., grammatical words), and that many of the words that appear in isolation correspond to fillers (‘‘yes,’’ ‘‘hmm,’’ . . .), vocatives (‘‘baby’s name’’) and social expressions (‘‘hi!,’’. . .), as found by van de Weijer (1998). Although there are no pauses between consecutive words in the signal, infants could rely on several subtle linguistic cues that signal boundaries between words and cohesion of consecutive sounds within words: prosodic cues (how stress and pitch information are affected by position within words), allophonic cues (the realization of some phonemes according to their position within words), phonotactic cues (constraints on phoneme order within words), and statistical/distributional information regarding consecutive syllables (higher transitional probabilities within words compared to between words). None of these cues systematically marks word boundaries, but it is possible that using them in conjunction may provide sufficient information for successful segmentation (Christiansen, Allen, & Seidenberg, 1998). Several studies have established that young Englishlearning infants are sensitive to prosodic (Jusczyk, Cutler, & Redanz, 1993; Turk, Jusczyk, & Gerken, 1995), allophonic (Hohne & Jusczyk, 1994), and phonotactic (Jusczyk, Friederici, Wessels, Svenkerud, & Jusczyk, 1993; Jusczyk, Luce, & Charles-Luce, 1994; Mattys, Jusczyk, Luce, & Morgan, 1999; see also Friederici & Wessels, 1993, for Dutch; Sebastia´n-Galle´s & Bosch, 2002, for Catalan) markers of word boundaries in their native language. Other studies have investigated infants’ use of these various linguistic cues for segmentation by exploring whether and how they segment multisyllabic words from fluent speech. The methodology used for the
majority of these studies is the same as what was used in the seminal study by Jusczyk and Aslin (1995), which established that English-learning infants start segmenting monosyllabic words between 6 and 7.5 months of age. The present study also uses the same methodology. In this paradigm, infants are first familiarized with whole words or syllables from these words, and then they are presented with passages containing the familiarized words and with passages that do not. Evidence for segmentation is demonstrated by longer orientation times to the passages containing the familiarized words. From the studies on English-learning infants emerges the following developmental pattern. At about 7–8 months of age, infants use prosodic information to segment fluent speech into sequences of syllables that begin with a strong syllable, i.e., trochaic units (Jusczyk, Houston, & Newsome, 1999b; for further evidence, see also Curtin, Mintz, & Christiansen, 2005; Echols, Crowhurst, & Childers, 1997; Houston, Santelmann, & Jusczyk, 2004; Johnson & Jusczyk, 2001; Morgan & Saffran, 1995; Nazzi, Dilley, Jusczyk, Shattuck-Hufnagel, & Jusczyk, 2005). Given that most English bisyllablic words have a strong–weak stress pattern (Cassidy & Kelly, 1991; Cutler & Carter, 1987; Kelly & Bock, 1988), this prosodic segmentation procedure (similar to the metrical segmentation strategy used by adults, c.f. Cutler, Mehler, Norris, & Segui, 1986; Cutler & Norris, 1988; McQueen, Norris, & Cutler, 1994) would allow English-learning infants, from a very young age, to appropriately segment most bisyllabic words. Indeed, Jusczyk et al. (1999b) found that English-learning infants segment trochaic (strong–weak) nouns (e.g., candle) by 7.5 months of age, but missegment iambic (weak–strong) nouns (e.g., guitar) at that age, a boundary being placed between the initial/weak syllable and the final/strong syllable (e.g., gui/tar, c.f. Jusczyk et al., 1999b). Finally, note that the proposal of this prosodic segmentation procedure is compatible with the data on monosyllabic word segmentation (Jusczyk & Aslin, 1995) given that these words were strong syllables. Distributional regularities of the order of syllables (henceforward, syllabic order information) in the speech signal were also found to be a crucial cue for early segmentation. For example, 7.5-month-olds tested on passages containing strong–weak words such as doctor were found to show a segmentation effect if familiarized with the whole words, but not if familiarized with their initial syllables, e.g., doc (Jusczyk et al., 1999b). Moreover, using an artificial language paradigm in which infants are presented with a continuous sequence made-up of randomly ordered repetitions of 4 trisyllabic pseudo-words, 8-month-olds were found to group syllables into cohesive word-like units on the basis of the transitional probabilities between consecutive syllables (Saffran, Aslin, & Newport, 1996; though see Perruchet & Vinter, 1998, for an alternative interpretation of these
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
results, and Brent & Cartwright, 1996; Dahan & Brent, 1999, for an alternative model). Importantly, the data obtained using the Jusczyk and Aslin (1995) paradigm suggest that English-learning infants first segment speech according to prosodic information, syllabic order information being then used within the prosodically defined units (Jusczyk et al., 1999b). Indeed, the prosodic boundary placed between the two syllables of a weak–strong word (e.g., guitar) appears to block 7.5-month-olds’ use of syllabic order information (the fact that gui and tar always appeared consecutively), resulting in the segmentation of the sole strong syllable. Similarly, if a weak–strong word is always followed by the same weak syllable (e.g., guitar_is), 7.5month-olds place a word boundary between the first two syllables (via prosodic information) and group together the last two syllables (via syllabic order information), resulting in an incorrect segmentation (e.g., gui/taris). These findings suggest that English-learning 7.5-month-olds use prosody to perform a first-pass parsing of continuous speech into smaller units that constitute the basis of further analyses of the signal (however, see the debate between Johnson & Jusczyk, 2001; Thiessen & Saffran, 2003, regarding data on this issue obtained with the artificial language paradigm).1 Later in development, the role of prosody and/or the relation between the use of prosodic and other segmentation cues appears to change. Evidence for this comes from findings that 10.5-month-olds can segment weak– strong words correctly and do not incorrectly segment strong–weak units across word boundaries when presented with a weak–strong word that is consistently followed by a weak syllable (Jusczyk et al., 1999b). This could be a reflection of these infants’ use of syllabic order information to detect the cohesiveness of two consecutive syllables even when they cross a prosodically placed boundary, which would suggest that by 10.5 months of age, infants weight syllabic order information more heavily than prosodic cues. It could also reflect the
285
fact that 10.5-month-olds have started to use additional word boundaries such as allophonic (Jusczyk, Hohne, & Bauman, 1999a), phonotactic (Mattys & Jusczyk, 2001a), and phonological phrase boundary (Gout, Christophe, & Morgan, 2004; Nazzi et al., 2005) cues.2 In summary, the studies described above have begun to trace the picture of the development of a combination of segmentation abilities during the first year of life. Prosody and syllabic order regularities are used as early as 8 months, with evidence that infants use a prosodybased segmentation procedure in their initial parsing of the signal and then use syllabic order information within the prosodically defined units. By 10.5 months, the relative weight of these two cues has changed, and infants have started using additional cues (e.g., phonotactic, allophonic). However, this pattern was established for English-learning infants. Given that these infants segment speech using language-specific cues (prosodic, phonotactic, allophonic cues) that signal word boundaries differently across different languages, it is likely that the pattern of emergence of segmentation abilities will differ across languages. Moreover, the question of the developmental origin of the prosodic segmentation procedure needs to be addressed to explain how it could be used at the onset of speech segmentation. One way to address this developmental issue and make specific crosslinguistic predictions regarding early prosodic segmentation is to consider a body of research, conducted in different domains, that converge towards the notion that segmentation differs according to the global rhythmic properties of one’s native language. The idea that there are different rhythmic classes of languages, in particular the stress-, syllabic-, and mora3-timed language classes, goes back many decades (Abercrombie, 1967; Pike, 1945), and has recently started receiving new linguistic evidence in support of it (Arvaniti, 1994; Den Os, 1988; Fant, Kruckenberg, & Nord, 1991; Nazzi, 1997; Ramus, Nespor, & Mehler, 1999; Shafer, Shucard, & Jaeger, 1999).
1
Thiessen and Saffran (2003), following up on Johnson and Jusczyk’s (2001) finding that 9-month-olds rely more on stress than syllabic order information to segment words when tested in the artificial language paradigm, replicated these results at 9 months, but found the opposite pattern at 7 months. They concluded that syllabic order information is used earlier than prosodic information. Although this may well be true in this experimental context, it is important to keep in mind that the artificial languages to be learned in those studies are made up of only 4 words that all have the same number of syllables (3 syllables). Although Thiessen and Saffran’s (2003) results show that infants can track syllabic order information and use it after a few minutes of exposure to this very simple language, it is unclear whether infants would benefit from syllabic order information at such a young age in the context of a natural language made up of thousands of words of varied syllabic length.
2
Other cues have been found to influence segmentation: coarticulation at 8 months (Johnson & Jusczyk, 2001) and, between 8 and 13 months, the nature (consonant versus vowel) of the initial phoneme of the word (Mattys & Jusczyk, 2001b; Nazzi et al., 2005) and pitch accent information (Nazzi et al., 2005). 3 The mora is a rhythmic unit that can either be syllabic or subsyllabic. In English, a mora roughly corresponds to a CV syllable with a short vowel (e.g., ‘‘the’’ as opposed to ‘‘thee’’, which has a long vowel). In Japanese, CV syllables with long vowels and syllables with final nasals (like the first syllable in ‘‘Honda’’) or final geminate consonants (like the first syllable in ‘‘Nissan’’) have two morae. Note that the relation between these three units is hierarchical: the stress unit is made up of syllables that are themselves made up of morae.
286
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
These rhythmic classes have proved useful for explaining crosslinguistic differences in the way adults segment speech and access the lexicon by suggesting a link between the adults’ behavior and the global rhythmic properties of their native language. In particular, the syllable appeared as the segmentation unit for adults speaking French (Mehler, Dommergues, Frauenfelder, & Segui, 1981; Peretz, Lussier, & Be´land, 1998), Spanish and Catalan (Sebastia´n-Galle´s, Dupoux, Segui, & Mehler, 1992) while the trochaic stress unit appeared to be used in English (Cutler et al., 1986; Cutler & Norris, 1988; McQueen et al., 1994) and Dutch (Vroomen, van Zon, & de Gelder, 1996). Furthermore, young infants have been found to discriminate languages according to how they fall within rhythmic classes (Nazzi, Bertoncini, & Mehler, 1998; Nazzi, Jusczyk, & Johnson, 2000; Nazzi & Ramus, 2003; Ramus, Hauser, Miller, Morris, & Mehler, 2000). Given the adequacy of the rhythmic and segmentation units in the languages previously investigated, it has been proposed that this bias to attend to global rhythmic patterns allows the emergence of the prosodic/rhythmic segmentation procedure appropriate to the (rhythmic class of the) native language (Nazzi et al., 1998, 2000; Mehler, Dupoux, Nazzi, & Dehaene-Lambertz, 1996). Although still unknown, the mechanisms that could allow infants to exploit rhythmic differences to specify the segmentation unit appropriate to their native language are currently under investigation (see current research based on an adaptive dynamical model, McLennan, 2005). In this rhythmic-based framework, the trochaic stress unit should be the unit of early prosodic segmentation in stress-based languages, and the syllable should be the unit of early prosodic segmentation in syllable-based languages, thus in French. Note that our claim relates to the size of the segmentation unit rather than its structure. Thus, for a French bisyllabic word (e.g., toucan), we predict that it will be segmented as two syllabic units (as opposed to smaller moraic units or larger stress units). However, we do not make any predictions regarding the structure of these syllabic units (e.g., whether toucan is segmented as tou/can or as touc/an), although the present study will provide some data pertaining to this issue (see General discussion). Let us now consider the evidence in favor of the rhythmic segmentation proposal. Although the early trochaic segmentation bias found in English-learning infants (Jusczyk et al., 1999b) fits into this framework, none of the few studies that have investigated early word segmentation in languages other than English allows us to evaluate our predictions. These studies mainly focused on establishing the ages at which infants are able to segment different types of words. With respect to stress-based languages, Dutch-learning infants were found to start segmenting strong–weak words between 7.5 and 9 months of age (Houston, Jusczyk, Kuijpers,
Coolen, & Cutler, 2000; Kooijman, Hagoort, & Cutler, 2005; Kuijpers, Coolen, Houston, & Cutler, 1998), while German-learning infants appeared to start segmenting monosyllabic words between 6 and 8 months of age (Ho¨hle & Weissenborn, 2003).4 French is the only syllable-based language that has been studied, and there is still very little known about early word segmentation in this language. Moreover, the studies available so far offer contrasting results. On one hand, data on European-French learning infants showed segmentation of monosyllabic words by 7.5 months of age but no evidence for segmentation of bisyllabic words between 7.5 and 11 months of age (Gout, 2001). On the other hand, data on Canadian-French learning infants showed segmentation of bisyllabic words as early as 8 months of age (Polka & Sundara, 2003). This discrepancy in results (which will be discussed in more detail in the General discussion) calls for further studies to be conducted on French, especially given that neither of these studies evaluated whether the two syllables of the bisyllabic words are segmented individually as predicted by the syllable-based segmentation hypothesis. The aim of the present study was therefore to test one prediction of the rhythmic hypothesis, namely that the syllable is the unit of early prosodic segmentation in syllable-based French. To test this prediction, it is crucial to evaluate how infants segment words made up of (at least) two syllables to be able to distinguish segmentation at the syllabic and lexical levels. At the onset of speech segmentation, we predict that infants’ prosodic segmentation procedure will place boundaries between every two consecutive syllables, and that no other information (e.g., syllabic order information) will be used to group together these two syllables (as consistent with Jusczyk et al., 1999b, results on English). Thus, we predict that French-learning infants will not initially segment whole bisyllabic words from fluent speech; rather, they will segment both syllables of bisyllabic words as individual, independent units. A few months later in development, the importance of the use of prosodic information to segment will diminish in comparison
4
The words used in that study were function words. It might seem contrary to our hypothesis that German-learning infants would segment such words so early, given that functional words tend not to be accented in the speech signal. Note however that, as pointed out by Ho¨hle and Weissenborn (2003), functional words in German are not as unstressed as their English equivalent, and might not count as weak syllables in larger strong-initial stress units. This possibility thus raises several issues for future research, for example, whether Englishlearning 8-month-olds would segment functional words (we predict they would not), and more importantly whether German-learning infants would also segment the weak syllables of multisyllabic words.
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
to other cues, and we will observe the segmentation of whole bisyllabic words as a result of infants’ use of other cues (e.g., syllabic order or phonotactic cues). This developmental scenario was evaluated in two steps. In the first step, we explored the development of French-learning infants’ ability to segment bisyllabic French words, as attested by their ability to recognize them after being familiarized either by the whole word (Experiment 1), or by their more salient, final syllable (Experiment 2). This research protocol, which replicates that used by Jusczyk et al. (1999b) to study the emergence of the segmentation of iambic words (Experiments 7 and 12 for whole words, Experiments 8 and 13 for final syllables), is motivated by the fact that although it is technically incorrect to say that French words are iambic, French words are nevertheless acoustically more similar to iambic words than to trochaic words as a consequence of the fact that they tend to have lengthened final syllables. The results from both experiments should specify the age at which segmentation effects can be found in French using our stimuli and this procedure, and the relation between whole word versus syllabic segmentation at different points in development. Based on our proposal, we predicted that whole word segmentation should occur later in development than final syllable segmentation. The prediction was more open regarding the question of whether, once infants start segmenting bisyllabic words as whole words, they stop segmenting the two syllables individually, or whether syllable-based prosodic segmentation remains important enough to allow for both kinds of segmentation. The second step of the research was to determine whether at the age at which infants segment the final syllables of the target words, they also segment their initial syllables (Experiments 3 and 4).
287
the familiarized words, and two containing repetitions of two other words. Given previous results obtained in French and in other languages suggesting that the age of emergence of segmentation abilities varies from around 7.5 months to later than 11 months, three age groups were tested: 8-, 12- and 16-month-olds. Method Participants Forty-eight infants from French-speaking families were tested and their data included in the analyses: sixteen 8-month-olds (mean age = 8.10 months; range: 7.25–9.06; 6 girls, 10 boys), sixteen 12-month-olds (mean age = 12.10 months; range: 11.29–12.27; 11 girls, 5 boys), sixteen 16-month-olds (mean age = 16.17 months; range: 16.01–16.29; 5 girls, 11 boys). The data of 10 additional 8-month-olds were not included in the analyses: 4 infants for fussiness or crying, and 6 infants for having at least 1 orientation time in the test phase shorter than 1.5 s (this criterion was used to ensure that infants heard at least one instance of the target words for each passage). The data of 6 additional 12-montholds were not included in the analyses: 3 infants for fussiness or crying; 1 infant for not turning to lights; 1 infant for having at least 1 orientation time in the test phase shorter than 1.5 s; 1 infant for having a segmentation index (defined as the difference between the mean orientation times to passages containing the familiarized words and those containing the new words) more than 2 SDs above or below the group mean. The data of 10 additional 16-month-olds were not included in the analyses: 5 infants for fussiness or crying; 4 infants for having at least 1 orientation time in the test phase shorter than 1.5 s; 1 infant for having a segmentation index more than 2 SDs above or below the group mean.
Experiment 1 This experiment evaluated when French-learning infants start showing evidence of segmenting bisyllabic words from fluent speech after being familiarized with the whole words. The procedure used is the version of the head-turn preference procedure (HPP) implemented by Jusczyk and Aslin (1995) to test early word segmentation. In this procedure, infants can either be familiarized with isolated words and then tested with passages containing or not the familiarized words, or the other way round. We decided to use the words-first paradigm, as it has been used more often (a quick survey of the literature shows that the words-first order has been used in 43 published experiments, compared to 11 experiments for the passages-first experiment), and studies having directly compared both presentation orders always found converging results. Therefore, infants were familiarized with repetitions of two bisyllabic words, and then tested on four passages, two containing repetitions of
Stimuli Recordings were made in a sound-attenuated booth. A female talker, who was a native speaker of French, recorded four 6-sentence passages (see Appendix A), one passage for each of the four target bisyllabic nouns. The selected nouns were: [pytwa] (French: putois; English: skunk), [tuka˜] (French and English: toucan), [ba˜do] (French: bandeau; English: headband), and [gido˜] (French: guidon; English: handlebar). Each noun appeared in every sentence of its appropriate passage. The talker was asked to pronounce the stimuli as if speaking to an infant. All the passages were 12.50 s long. The target bisyllabic words in the sentences had an average duration of 397 ms ([pytwa]: 400 ms; [tuka˜]: 389 ms; [ba˜do]: 410 ms; [gido˜]: 391 ms). Acoustic analyses on individual syllables were conducted on the initial and final syllables to obtain relative measures for duration (including prevoicing and between-syllable silences), mean intensity (RMS) and mean F0 (both calculated on
288
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
the vocalic portion of the syllable). The results are summarized in Table 1. They show that the second syllables of these bisyllabic words were significantly longer, had greater intensity and higher F0 than their first syllables. For each noun, the same talker also produced a list of 15 isolated occurrences for use in the familiarization phase. The speaker produced the tokens with some variation. The duration of each list was 12.50 s. The target bisyllabic words spoken in isolation had an average duration of 479 ms ([pytwa]: 485 ms; [tuka˜]: 467 ms; [ba˜do]: 496 ms; [gido˜]: 467 ms). Again, acoustic analyses on individual syllables were conducted on the initial and final syllables to obtain relative measures for duration (including prevoicing and between-syllable silences), mean intensity (RMS) and mean F0 (calculated on the vocalic portion of the syllables). The results are summarized in Table 2. They show that the second syllables of these bisyllabic words were significantly longer, though there were no differences between the two syllables on mean intensity and mean F0. Procedure, and apparatus The experiment was conducted in a three-sided test booth made of pegboard panels. Except for a small section of pre-existing holes in the front panel used for monitoring the infant’s headturns, the panels were
Table 1 Acoustic measurements for the two syllables of the bisyllabic words from the test passages (6 occurrences per word) Duration (ms)
Intensity (dB)
F0 (Hz)
[pytwa] [tuka˜] [ba˜do] [gido˜]
144 144 204 166
256 246 206 225
67.0 62.7 74.7 69.8
73.1 74.4 70.9 70.8
190 189 181 204
239 231 209 232
Mean F test
164
233
68.5
72.3
191
228
***
***
Results of F(1, 20) tests: *p < .05;
***p
*
< .001.
Table 2 Acoustic measurements for the two syllables of the bisyllabic words spoken in isolation (15 occurrences per word) Duration (ms)
Intensity (dB)
F0 (Hz)
[pytwa] [tuka˜] [ba˜do] [gido˜]
125 125 190 158
360 342 306 308
69.8 67.6 74.6 70.7
71.2 71.0 68.4 70.2
183 194 165 175
176 195 177 176
Mean F test
170
329
70.7 n.s.
70.2
179 n.s.
181
***
Results of F(1, 56) tests:
***p
< .001, n.s. p > .05.
backed with white cardboard to prevent the infant from seeing behind the panels. The test booth had a red light and a loudspeaker (SONY xs-F1722) mounted at eye level on each of the side panels and a green light mounted on the center panel. Directly below the center light a 5-cm hole accommodated the lens of a video camera used to record each test session. A white curtain suspended around the top of the booth shielded the infant’s view of the rest of the room. A PC computer terminal (COD) and response box were located behind the center panel, out of view of the infant. The response box, which was connected to the computer, was equipped with a series of buttons. The box was controlled by an observer hidden behind the center panel, who looked through a peephole and pressed the buttons of the response box according to the direction of the infant’s headturns, thus starting and stopping the flashing of the lights and the presentation of the sounds. Information about the direction and duration of the headturns and the total trial duration were stored in a data file on the computer. The observer was not informed as to the condition to which the infant was assigned. Moreover, both the observer and the infant’s caregiver wore earplugs and listened to masking music over tight-fitting closed headphones, which prevented them from hearing the stimuli presented. The version of the HPP set-up by Jusczyk and Aslin (1995) was used in the present study. Each infant was held on a caregiver’s lap. The caregiver was seated in a chair in the center of the test booth. Each trial began with the green light on the center panel blinking until the infant had oriented in that direction. Then, the center light was extinguished and the red light above the loudspeaker on one of the side panels began to flash. When the infant made a turn of at least 30 in the direction of the loudspeaker, the stimulus for that trial began to play. Each stimulus was played to completion (i.e., when the fifteen isolated occurrences of a given word, or the six sentences in a given passage had been presented) or stopped immediately after the infant failed to maintain the 30 headturn for 2 consecutive seconds (200 ms fade-out). The stimuli were stored in digitized form on the computer, and were delivered by the loudspeakers via an audio amplifier (Marantz PM4000). If the infant turned away from the target by 30 in any direction for less than 2 s and then turned back again, the trial continued but the time spent looking away was not included in the orientation time. Thus, the maximum orientation time for a given trial was the duration of the entire speech sample. The flashing red light remained on for the entire duration of the trial. Each experimental session began with a familiarization phase in which infants heard repetitions of two of the targets on alternating trials until they accumulated 20 s of orientation times to each. If the infants achieved the familiarization criterion for one item, but not for the
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
Design In each age group, half of the infants were familiarized with the nouns [gido˜] and [tuka˜], and the other half with the nouns [ba˜do] and [pytwa].
9,00
8,00
7,00
6,00 orientation times (s)
other, the trials continued to alternate until the criterion was achieved for both. The side of the loudspeaker from which the stimuli were presented was randomly varied from trial to trial. The test phase began immediately after the familiarization criterion was attained. It consisted of four test blocks (in each of which the four 6-sentence passages were presented) for the 8-month-olds, and two test blocks for the 12- and 16-month-olds. The order of the different passages within each block was randomized.
289
5,00 familiar new 4,00
3,00
Results and discussion Familiarization phase For all three age groups, mean orientation times during familiarization were calculated for the infants in both familiarization conditions. A 2-way ANOVA with the main between-subject factors of age (8, 12, and 16 months) and familiarization condition (familiarization with [gido˜]—[tuka˜] versus [ba˜do]—[pytwa]) was conducted. All effects and interactions failed to reach significance, suggesting that the duration of familiarization was equivalent at all ages and in all familiarization conditions received (mean orientation time was 26.96 s for 8-month-olds, 28.34 s for 12-month-olds, and 27.01 s for 16-month-olds). Test phase Mean orientation times to the passages containing the familiarized nouns and to the passages containing the new bisyllabic nouns were calculated for each infant. The data for the three age groups are presented in Fig. 1. A 3-way ANOVA with the main between-subject factors of age (8, 12, and 16 months) and condition (familiarization with [gido˜]—[tuka˜] versus [ba˜do]—[pytwa]) and the main within-subject factor of familiarity (familiar versus new) was conducted. The effect of familiarity approached significance, F(1, 42) = 4.03, p = .051, indicating that the infants tended to have longer orientation times to the passages with the familiarized nouns (M = 7.02 s, SD = 1.84) than to those with the new nouns (M = 6.62 ms, SD = 1.70). However, there was a significant familiarity · age interaction, F(2, 42) = 4.53, p = .017, indicating that the effect of familiarity changed with age. No other effect or interaction reached significance. To specify the familiarity · age interaction, planned comparisons were conducted. The effect of familiarity failed to reach significance at 8 months, F(1, 14) < 1, indicating that these infants had similar orientation times to the passages with the familiarized nouns
2,00
1,00
0,00 8 months
12 months
16 months
Fig. 1. Mean orientation times (s) to the test passages containing the familiarized bisyllabic words (familiar) or the new words (new). The error bars indicate the standard error of the mean. Left panel: 8-month-old infants; central panel: 12month-old infants; right panel: 16-month-old infants (Experiment 1).
(M = 7.52 s, SD = 1.88) and to those with the new nouns (M = 7.40 ms, SD = 1.34). Only seven of the sixteen 8-month-olds oriented longer to the passages with the familiarized nouns. The effect of familiarity also failed to reach significance at 12 months, F(1, 14) < 1, indicating that these infants had similar orientation times to the passages with the familiarized nouns (M = 6.28 s, SD = 1.77) and to those with the new nouns (M = 6.43 ms, SD = 1.58). Only eight of the sixteen 12-month-olds oriented longer to the passages with the familiarized nouns. However, the effect of familiarity was significant at 16 months, F(1, 14) = 6.45, p = .024, indicating that 16-month-old infants had longer orientation times to the passages with the familiarized nouns (M = 7.25 s, SD = 1.73) than to those with the new nouns (M = 6.02 ms, SD = 1.93). This pattern of preference was found for 11 of the 16 infants. In the present experiment, French-learning infants were familiarized with two bisyllabic words and then tested with passages either containing these bisyllabic words or containing new bisyllabic words. No segmentation evidence was found at either 8 or 12 months of age, while evidence for word segmentation was found at 16 months of age. These results therefore establish that
290
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
the ability to segment our bisyllabic words emerges between 12 and 16 months of age, suggesting an important lag in bisyllabic word segmentation compared to the previous findings obtained for English- and Dutchlearning infants (Houston et al., 2000; Jusczyk et al., 1999b). The present experiment however does not allow us to specify the reasons for this lag. Moreover, the interpretation of this pattern of failure to segment followed by success remains unclear. Our proposal that Frenchlearning infants initially use a syllable-based prosodic segmentation procedure, paired with previous findings on English-learning infants showing that segmentation is observed only when there is a match between the size of the familiarization and segmented units (Jusczyk et al., 1999b), predicted that younger infants (segmenting individual syllables) would fail to show segmentation effects in this experiment, while older infants (segmenting whole words) would show such effects. Therefore, to clarify the pattern of results obtained in the present experiment, we needed to test the following two predictions of the syllable-based segmentation hypothesis. The first one is that infants younger than 16 months familiarized with either the final or the initial syllables of the target words should show segmentation due to a match between the familiarization and segmented units. The second is that 16-month-olds should fail to show segmentation under the same familiarization conditions, unless syllable-based segmentation remains important enough at that age. The following experiment tested these predictions by focusing on the final syllables of the bisyllabic words, syllables which tend to be more salient in French because they are lengthened compared to non-final syllables.
Experiment 2 The present experiment explored whether Frenchlearning 8-, 12-, and 16-month-olds (ages tested in Experiment 1), familiarized with the final syllables of two bisyllabic words, show segmentation effects when hearing the four passages used in Experiment 1: two passages containing the bisyllabic words corresponding to the familiarization syllables, and two passages containing new bisyllabic words. Given our proposal of early syllable-based segmentation in French and the segmentation changes between 12 and 16 months found in Experiment 1, we predicted that we would find evidence of final syllable segmentation at 12 months. The results for the 16-month-olds will depend on the relative strength of the prosodic cue relative to the other segmentation cues. Finally, the results for the 8-month-olds will depend on whether segmentation emerges before/ around 8 months as found for English (Jusczyk & Aslin, 1995; Jusczyk et al., 1999b) and German (Ho¨hle &
Weissenborn, 2003) or whether it is slightly delayed as in Dutch (Houston et al., 2000; Kooijman et al., 2005; Kuijpers et al., 1998). Method Participants Forty-eight infants from French-speaking families were tested and their data included in the analyses: sixteen 8-month-olds (mean age = 8.17 months; range: 7.21–8.28; 9 girls, 7 boys), sixteen 12-month-olds (mean age = 12.17 months; range: 12.04–13.00; 5 girls, 11 boys), sixteen 16-month-olds (mean age = 16.14 months; range: 16.00–16.26; 6 girls, 10 boys). The data of 11 additional 8-month-olds were not included in the analyses (fussiness or crying: 7; at least 1 looking time in the test phase shorter than 1.5 s: 3; segmentation index more than 2 SDs above or below the group mean: 1). The data of 7 additional 12-month-olds were not included in the analyses (fussiness or crying: 5; at least 1 looking time in the test phase shorter than 1.5 s: 1; segmentation index more than 2 SDs above or below the group mean: 1). The data of 10 additional 16-montholds were not included in the analyses (fussiness or crying: 7; at least 1 looking time in the test phase shorter than 1.5 s: 3). Stimuli The four passages used in the test phase of Experiment 1 were used again in the test phase of this experiment. However, the familiarization stimuli were different. Infants were familiarized with the final syllables of two of the target bisyllabic words, either [twa] and [ka˜], or [do] and [do˜]. All of these syllables were real French words (respectively, French: toit, English: roof; French and English: camp; French: dos; English: back; French: don, English: gift). All of these items had been recorded as isolated words by the same French talker in the same recording session as the passages, and using the same procedure for producing a list of 15 isolated occurrences. The duration of each list was 12.50 s. The final syllables spoken in isolation had an average duration of 294 ms ([twa]: 308 ms; [ka˜]: 263 ms; [do]:
Table 3 Acoustic measurements for the final syllables spoken in isolation (15 occurrences per syllable) Duration (ms)
Intensity (dB)
F0 (Hz)
[twa] [ka˜] [do] [do˜]
308 263 304 299
70.3 71.0 71.2 71.1
195 193 198 184
Mean
294
70.7
192
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
304 ms; [do˜]: 299 ms). Again, acoustic analyses were conducted to obtain measures for duration (including prevoicing), mean intensity (RMS) and mean F0 (calculated on the vocalic portion of the syllables). The results are summarized in Table 3.
Results and discussion Familiarization phase For all three age groups, mean orientation times during familiarization were calculated for the infants in both familiarization conditions. A 2-way ANOVA with the main between-subject factors of age (8, 12, and 16 months) and familiarization condition (familiarization with [twa]—[ka˜] versus [do]—[do˜]) was conducted. All effects and interactions failed to reach significance, suggesting that the duration of familiarization was equivalent at all ages and in all familiarization conditions received (mean orientation time was 27.03 s for 8month-olds, 28.47 s for 12-month-olds, and 28.54 s for 16-month-olds). Test phase Mean orientation times to the passages containing the familiarized and new bisyllabic nouns were calculated for each infant. The data for the three age groups are presented in Fig. 2. We conducted the same analyses as for Experiment 1, that is: (a) a 3-way ANOVA with the main between-subject factors of age (8, 12, and 16 months) and condition (familiarization with [twa]— [ka˜] versus [do]—[do˜]) and the main within-subject factor of familiarity (familiar versus new), and (b) planned comparisons at the three different ages. First, the 3-way ANOVA revealed a significant effect of age, F(2, 42) = 7.87, p = .001, due to the fact that orientation times tended to decrease with age (for the 8-montholds: M = 7.54 s, SD = 1.26; for the 12-month-olds: M = 6.54 s, SD = 1.66; for the 16-month-olds: M = 5.60 s, SD = 1.20). No other effect or interaction reached significance. Second, planned comparisons at 8 months failed to show a significant effect of familiarity, F(1, 14) < 1, indicating that these infants had similar orientation times to the passages containing the nouns corresponding to the familiarized final syllables (M = 7.44 s, SD = 1.29) and to those with the new nouns (M = 7.63 ms, SD = 1.39). Only seven of the sixteen 8-month-olds ori-
9,00
8,00
7,00
6,00 orientation times (s)
Procedure, apparatus, and design The procedure and apparatus were identical to that in Experiment 1. Again, 8-month-olds received 16 test passages while 12- and 16-month-olds received 8 test passages. Half of the infants were familiarized with [twa] and [ka˜] (final syllables of [pytwa] and [tuka˜]), the other half with [do] and [do˜] (final syllables of [ba˜do] and [gido˜]).
291
5,00 familiar new 4,00
3,00
2,00
1,00
0,00 8 months
12 months
16 months
Fig. 2. Mean orientation times (s) to the test passages containing the words corresponding to the familiarized final syllables (familiar) or the new words (new). The error bars indicate the standard error of the mean. Left panel: 8-monthold infants; central panel: 12-month-old infants; right panel: 16month-old infants (Experiment 2).
ented longer to the passages corresponding to the familiarized final syllables. On the contrary, the effect of familiarity was significant at 12 months, F(1, 14) = 8.52, p = .011, indicating that 12-month-old infants had longer orientation times to the passages corresponding to the familiarized final syllables (M = 6.96 s, SD = 1.93) than to those with the new nouns (M = 6.13 ms, SD = 1.55). This pattern of preference was found for 11 of the 16 infants. Finally, the effect of familiarity failed to reach significance at 16 months, F(1, 14) < 1, indicating that these infants had similar orientation times to the passages corresponding to the familiarized final syllables (M = 5.63 s, SD = 1.79) and to those with the new nouns (M = 5.57 ms, SD = 1.63). Only nine of the sixteen 16-month-olds oriented longer to the passages corresponding to the familiarized final syllables. Finally, we conducted a 4-way ANOVA on the combined data of Experiments 1 and 2. This analysis confirmed the overall decrease in orientation times as age increases (age effect: F(2, 84) = 7.67, p < .001). It also revealed a significant effect of familiarity, F(1, 84) = 4.25, p = .04, and importantly, a three-way interaction between age, experiment and familiarity, F(2, 84) = 4.19, p = .02, confirming that the ages at which a significant effect of familiarity is obtained differs according to
292
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
whether the infants were familiarized with the whole bisyllabic words (significant effect at 16 months) or only their final syllables (significant effect at 12 months). In the present experiment, following Jusczyk et al.’s (1999b) protocol, infants were familiarized with the final syllables of two bisyllabic words, and then heard passages that either contained the corresponding words, or contained new bisyllabic words. As previously demonstrated by these authors, a segmentation effect in such an experimental condition should be observed if the final syllables of the bisyllabic words in the passages are segmented individually. A significant segmentation effect was found at 12 months, but was present neither at 8 months nor at 16 months. Based on our developmental proposal regarding the early use of syllables in speech segmentation in French and the pattern of results found in Experiment 1, we had predicted that French-learning 12-month-olds would segment the two syllables individually. This prediction is supported by the present data. We had also raised the issue of whether the 16-month-olds were really segmenting the bisyllabic words as whole units and/or whether they would still show a syllabic segmentation effect, a result which would have suggested the importance of the use of prosodic segmentation cues at that age. Given the results of Experiment 1, the absence of final syllable segmentation indicates that 16-month-olds segment bisyllabic words as whole units, and suggests that these infants rely more heavily on other segmentation cues (e.g., syllable order or phonotactic information) than on the prosodic cue. Segmentation was not found at 8 months of age which, combined with Experiment 1, suggests that 8month-olds neither segmented our bisyllabic words as whole words nor did they segment the final syllables. This suggests a delay compared to results found for English (Jusczyk & Aslin, 1995; Jusczyk et al., 1999b) and German (Ho¨hle & Weissenborn, 2003), although they might be similar to Dutch-learning infants who failed to segment words before 9–10 months of age (Houston et al., 2000; Kooijman et al., 2005; Kuijpers et al., 1998). We will return to the discussion of these findings in relation to other data previously obtained for French in the General discussion. Returning to the effect of linguistic rhythm on early word segmentation, the proposal that French-learning infants initially use a syllable-based segmentation procedure predicted not only that infants should initially segment final (lengthened) syllables individually, but also that they should segment initial (shorter) syllables individually. Given that our results established final syllable segmentation at 12 months of age but not at either 8 or 16 months of age, the next two experiments evaluated the segmentation of initial syllables in 12-month-olds only.
Experiment 3 Following up on the previous experiment, which established that French-learning 12-month-olds segment the final syllable of bisyllabic words, this experiment tested whether French-learning 12-month-olds also segment the initial syllable of bisyllabic words, as predicted if segmentation at this age is syllable-based. The procedure is the same as in the previous experiment, except that the infants are now familiarized with the initial syllables of two bisyllabic words, and then tested with the four passages used in Experiments 1 and 2: two passages containing the bisyllabic words corresponding to the familiarization syllables, and two passages containing new bisyllabic words. Method Participants Sixteen 12-month-old infants from French-speaking families were tested and their data included in the analyses (mean age = 12.18 months; range: 12.00–13.07; 10 girls, 6 boys). The data of 10 additional infants were not included in the analyses (fussiness or crying: 7; not turning to lights: 1; at least 1 looking time in the test phase shorter than 1.5 s: 1; segmentation index more than 2 SDs above or below the group mean: 1). Stimuli The four passages used in the test phase of Experiment 1 were again used in the test phase of this experiment. However, the familiarization stimuli were different. Infants were presented with the initial syllables of two of the target bisyllabic words, either [py] and [tu], or [ba˜] and [gi]. All of these syllables were real French words (respectively, French and English: pus; French: tout, English: whole; French: banc, English: bench; French: gui, English: mistletoe). All of these items had been recorded as isolated words by the same French talker in the same recording session as the passages, and using the same procedure for producing a list of 15 isolated occurrences. The duration of each list was 12.50 s. The initial syllables spoken in isolation had an average duration of 308 ms ([py]: 301 ms; [tu]: 255 ms; [ba˜]: 322 ms; [gi]: 355 ms). Again, acoustic analyses were conducted to obtain measures for duration (including prevoicing), mean intensity (RMS) and mean F0 (calculated on the vocalic portion of the syllables). The results are summarized in Table 4. Procedure, apparatus, and design The procedure and apparatus were identical to that in Experiment 1 and 2, infants receiving 8 test passages. Half of the infants were familiarized with [py] and [tu]
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299 Table 4 Acoustic measurements for the initial syllables spoken in isolation (15 occurrences per syllable) Duration (ms)
Intensity (dB)
9,00
F0 (Hz)
8,00
301 255 322 355
71.1 71.6 69.9 70.2
266 269 217 252
7,00
Mean
308
70.7
251
6,00
Results and discussion Familiarization phase Mean orientation times during familiarization were calculated for the infants in both familiarization conditions. Mean familiarization time in this experiment was 28.69 s, and there was no difference between the two familiarization conditions, t(14) < 1.
o r i en t a t i o n ti m e s ( s )
[py] [twa] [ba˜] [gi]
(initial syllables of [pytwa] and [tuka˜]), the other half with [ba˜] and [gi] (initial syllables of [ba˜do] and [gido˜]).
293
5,00 familiar new 4,00
3,00
2,00
1,00
Test phase Mean orientation times to the passages containing the bisyllabic nouns corresponding to the familiarized initial syllables and to the passages containing the new bisyllabic nouns were calculated for each infant. The data are presented in Fig. 3, left panel. A 2-way ANOVA with the main between-subject factor of condition (familiarization with [py]—[tu] versus [ba˜]—[gi]) and the main within-subject factor of familiarity (familiar versus new) was conducted. There was no effect of familiarity, F(1, 14) < 1, indicating that the infants had similar orientation times to the passages containing the nouns corresponding to the familiarized initial syllables (M = 8.24 s, SD = 2.36) and to those with the new nouns (M = 7.95 ms, SD = 1.99). Nine infants out of 16 showed longer orientation times to the passages corresponding to the familiarized initial syllables. The absence of a familiarity effect in the present experiment does not support our hypothesis that French-learning 12-month-olds segment both initial and final syllables of French bisyllabic words individually. However, several factors suggest that such a failure may not be a result of not segmenting the initial syllables. Indeed, in these experiments, evidence of segmentation requires that infants both segment the continuous signal according to a unit the size of the unit presented during familiarization, and also match this segmented representation to the representation of the familiarized unit. Therefore, the absence of a segmentation effect for the initial syllables could either signal that they were not reliably segmented, or that they were segmented but unreliably matched to the familiarization stimulus. This latter possibility is supported by the
0,00 initial syllables
spliced-out initial syllables
Fig. 3. Mean orientation times (s) to the test passages containing the words corresponding to the familiarized syllables (familiar) or the new words (new) for 12-month-olds. The error bars indicate the standard error of the mean. Left panel: initial syllables (Experiment 3); right panel: spliced-out initial syllables (Experiment 4).
fact that the matching process of the initial syllables could have been made more difficult for the two following reasons. First, the initial syllables embedded in the passages were less salient than their final counterparts (see Table 1). Second, the acoustic distance between the realization of the syllables in isolation and in the passages was greater for the initial syllables (mean duration difference: 144 ms, 87.8% increase, see Tables 1 and 4) than for the final syllables (mean duration difference: 61 ms, 26.2% increase, see Tables 1 and 3). The following experiment was designed to test this hypothesis.
Experiment 4 Like Experiment 3, the present experiment evaluated 12-month-olds’ ability to segment the initial syllables of bisyllabic words. The procedure used was exactly the same with one crucial difference: infants were familiarized with initial syllables spliced-out from the test passages rather than recorded in isolation. This was done
294
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
to test whether acoustic distance between the initial syllables used during familiarization and test in Experiment 3 could explain why segmentation was not demonstrated by infants in that experiment. Contrary to the explanation in terms of a non-segmentation of initial syllables, the acoustic-based matching explanation predicts reliable segmentation effects for spliced-out syllables, due to facilitated matching of initial syllables across familiarization and test. Method Participants Sixteen 12-month-old infants from French-speaking families were tested and their data included in the analyses (mean age: 12.17 months; range: 11.03–13.10; 12 girls, 4 boys). The data of 6 additional infants were not included in the analyses (fussiness or crying: 1; at least 1 looking time in the test phase shorter than 1.5 s: 5). Stimuli The four passages used in the test phase of Experiments 1, 2 and 3 were used again in the test phase of this experiment. However, the familiarization stimuli were different. For each passage, the initial syllables of the six occurrences of the target words were spliced-out. For each initial syllable, a list of 15 repetitions was constructed by using each token two or three times. The duration of each list was 12.50 s. Procedure, apparatus, and design The procedure and apparatus were identical to that in Experiments 1, 2 and 3 infants receiving 8 test passages. Half of the infants were familiarized with spliced-out [py] and [tu] (initial syllables of [pytwa] and [tuka˜]), the other half with spliced-out [ba˜] and [gi] (initial syllables of [ba˜do] and [gido˜]). Results and discussion Familiarization phase Mean orientation times during familiarization were calculated for the infants in both familiarization condition. Mean familiarization time in this experiment was 27.95 s. There was no significant difference between the two familiarization conditions, t(14) < 1. Test phase Mean orientation times to the passages containing the bisyllabic nouns corresponding to the familiarized spliced-out initial syllables and to the passages containing the new bisyllabic nouns were calculated for each infant. The data are presented in Fig. 3, right panel. A 2-way ANOVA with the main between-subject factor of condition (familiarization with spliced-out [py]—[tu] versus spliced-out [ba˜]—[gi]) and the main within-sub-
ject factor of familiarity (familiar versus new) was conducted. There was a significant effect of familiarity, F(1, 14) = 7.63, p = .015, indicating that the infants had longer orientation times to the passages corresponding to the familiarized spliced-out initial syllables (M = 7.03 s, SD = 1.72) than to those with the new nouns (M = 6.05 s, SD = 1.48). This pattern of preference was found for 12 of the 16 infants. There was no effect of condition, F(1, 14) < 2.56, p = .13, and no interaction between familiarity and condition, F(1, 14) < 1. The present experiment shows that French-learning 12-month-olds are able to segment the initial syllable of bisyllabic words and recognize them after familiarization with syllables that are not too acoustically different from those embedded in the test passages. This result suggests that the unreliable segmentation effect found in Experiment 3 when using familiarization initial syllables produced in isolation was likely due to difficulty matching familiarization and test stimuli when the acoustic differences between these stimuli are too pronounced, rather than to not being able to segment the initial syllables of bisyllabic French words.5 Thus, taken together, the results of Experiments 3 and 4 support the hypothesis that French-learning 12-month-olds segment individually both syllables of a bisyllabic word, as expected if they are using a syllable-based segmentation procedure.
General discussion The present study explored the emergence of segmentation abilities in French-learning infants. It was guided by the general hypothesis that there would be differences in how segmentation abilities emerge between languages of different rhythmic classes. More specifically, it tested the hypothesis that the syllable would be the unit of prosodic segmentation in syllable-based French just like the trochaic unit appeared to be the unit of prosodic segmentation in stress-based English (Curtin et al., 2005; Echols et al., 1997; Houston et al., 2004; Johnson & Jusczyk, 2001; Jusczyk et al., 1999b; Morgan & Saffran, 1995; Nazzi et al., 2005). Our strategy to obtain evidence for our proposal was twofold. First, relying on the research protocol used by Jusczyk et al. (1999b) to study the segmentation of iambic words, we explored whether 8-, 12-, and 16-month-old infants segment bisyllabic words as whole units or whether they segment the final syllables individually 5
The reduced saliency of initial syllables in the passages could also have contributed to the failure in Experiment 3. However, this factor does not seem to be crucial when acoustic distance is neutralized, as done here using syllables spliced-out from the test stimuli; its contribution could be increased in conditions in which the acoustic distance between familiarization and test would fall in between those used in Experiments 3 and 4.
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
(Experiments 1 and 2). No segmentation effects of any kind were found at 8 months. At 12 months, the evidence suggests that infants segment the final syllables of the target bisyllabic words rather than segmenting the target words as whole units. At 16 months, on the other hand, the findings suggest that bisyllabic words are segmented as whole units, with no further evidence of final syllable segmentation. The second step of our research was to extend the finding of syllabic segmentation at 12 months to the initial syllables. Experiment 3 failed to provide such evidence, although a trend in the predicted direction was obtained. However, a significant effect of initial syllable segmentation was found in Experiment 4, in which we removed the large acoustic differences between the familiarization and test stimuli, suggesting that the initial failure was a familiarization/test matching problem rather than a segmentation one. The developmental data obtained in the present study support the proposal that the syllable is the unit of early prosodic segmentation in French, indicating that French-learning infants initially segment speech into syllable-sized units. The determination of the exact syllabic structure of the segmented units was beyond the main scope of the present study, but the fact that segmentation effects could be found for both the initial (e.g., tou) and final syllables (e.g., can) suggests that 12-month-olds segmented the bisyllabic word toucan as tou/can. One possible way to strengthen this suggestion, which is consistent with evidence that English-learning 8- to 13-month-olds are sensitive to the syllabic structure of the words to be segmented (Mattys & Jusczyk, 2001b), would be to evaluate the related prediction that French-learning 12-month-olds would not show a segmentation effect for the target word toucan if familiarized with the syllable touc. More generally, the present results bring support from a syllable-based language to the more general proposal that the emergence of prosodic segmentation is dependent on the rhythmic type of the language in acquisition. Given that the results for English-learning infants (Jusczyk et al., 1999b) are compatible with the notion that these infants are using the rhythmic unit of this language to segment speech, we now have evidence from two languages of different rhythmic types for this proposal. Future research will have to extend the present line of research to more languages to further confirm the congruence of rhythmic and segmentation units across languages predicted by our rhythmic-based proposal. Preliminary data on German has shown evidence of an early strong–weak segmentation bias similar to that found for English in this other stress-based language (Ho¨hle & Weissenborn, 2005). Additionally, if the above interpretation is correct, the present results are compatible with previous results for English (in particular, Jusczyk et al., 1999b), obtained
295
using the same natural speech procedure, regarding the evolution of the weight of prosodic information for segmentation. Evidence from both languages suggest that prosodic information plays a predominant role at the onset of segmentation and that its predominance later decreases. But while studies conducted in English have pointed out (some of) the cues that come to be used in older infants (e.g., syllabic order, phonotactic and allophonic information), no such studies have yet been conducted for French, an important task left for future research. Moreover, as discussed earlier, studies on the acquisition and segmentation of simple artificial languages showed that young infants do use syllabic order distribution. Some findings even support the claim that under certain circumstances infants use syllabic order information before they use prosodic information (Thiessen & Saffran, 2003; see also footnote 1). Given the apparent differences between the results from the natural versus artificial language studies, it will be interesting to test in the future that effects of syllable-based segmentation can also be found using the artificial language paradigm. Furthermore, the pattern of results of Experiments 2– 4 suggests that French-learning 12-month-olds are sensitive, to a certain degree, to correlates of syllabic accentuation in French. This finding is congruent with studies showing that lexical accentuation affects the early representations of initial and final syllables in French, in particular their phonetic specificity. Such evidence comes from production studies (see De Boysson-Bardies, 1996, for data on infants in the second year of life), but also from early perception studies (see Halle´ & de Boysson-Bardies, 1996, Vihman, Nakai, DePaolis, & Halle´, 2004, for data on 11-month-olds). This sensitivity to accentuation could be a trace of newborns’ sensitivity to the acoustic correlates of accentuation (Nazzi, Floccia, & Bertoncini, 1998; Sansavini, Bertoncini, & Giovanelli, 1997). However, this sensitivity may decline with age in French learners as a result of the fact that this acoustic dimension is not used contrastively at the lexical level in French. This idea is consistent with results which failed to show evidence of lexical position/accentuation on the degree of phonetic specificity in 20-month-olds’ newly learnt lexical representations (Nazzi, 2005), and with studies showing that French-speaking adults are less sensitive to lexical accentuation than Spanish-speaking adults (Dupoux, Pallier, Sebastia´n, & Mehler, 1997; Dupoux, Peperkamp, & Sebastia´n-Galle´s, 2001). Finally, we need to evaluate two issues: the timing of emergence of segmentation abilities revealed by the present study relative to the timing found in other languages, and the pattern of segmentation effects obtained here relative to that obtained in previous studies on early segmentation by French-learning infants. We will discuss these issues in turn. Regarding the age at which segmentation abilities emerge in French-learning infants, our data point
296
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
towards an age of emergence between 8 and 12 months (segmentation of individual syllables, Experiments 2–4). This is slightly older than what has been found for English (between 6 and 7.5 months, Jusczyk & Aslin, 1995) and German (between 6 and 8 months, Ho¨hle & Weissenborn, 2003), but not necessarily different from what has been found for Dutch (between 7.5 and 9 months, Houston et al., 2000; Kuijpers et al., 1998). Given the relatively small size of the differences observed, and the fact that we do not yet know all the factors that influence segmentation (see for example, the lag of 6 months found between the segmentation of English strong–weak nouns versus verbs, Jusczyk et al., 1999b; Nazzi et al., 2005), all of these results can be taken as evidence that segmentation abilities emerge crosslinguistically some time around 8 months (see also discussion next paragraph). However, the emergence lag is larger for the comparison of French bisyllabic words (between 12 and 16 months, Experiment 1) and English weak– strong words (between 7.5 and 10.5 months), both of which require the use of cues other than the prosodic cues to be segmented. These results thus call for future studies evaluating the contribution to segmentation in French of cues previously found to impact on segmentation in English. Let us now discuss our data in relation to other data on French-learning infants. The data on EuropeanFrench learning infants by Gout (2001) are consistent with our data regarding the segmentation of whole bisyllabic words, in that both studies fail to show such segmentation before 12 months of age. On the other hand, our failure to find evidence of segmentation for the final syllables of bisyllabic words at 8 months seems to be in contradiction with Gout’s (2001) finding of a segmentation effect for monosyllabic words at 7.5 months. However, the difference could be explained by coarticulation differences between Gout’s (2001) monosyllabic targets and our bisyllabic targets, in line with results found for English-learning 8-month-olds (Johnson & Jusczyk, 2001), which again would suggest that segmentation onset in French is not particularly delayed compared to the other languages previously tested. On the other hand, the data on Canadian-French learning infants showing segmentation of bisyllabic words as early as 8 months of age (Polka & Sundara, 2003) appear, at first sight, in contradiction with our results on European-French infants and with our proposal for early syllable-based segmentation in French. However, such a conclusion seems premature given the limited number of experiments conducted on early segmentation in French, and given data showing syllabic segmentation in both European- (Mehler et al., 1981) and Canadian-French (Peretz et al., 1998) adults. This conclusion is even more premature as the role of the syllable has in fact not yet been investigated for CanadianFrench infants. If segmentation of individual syllables
were to be found for learners of Canadian-French before 8 months, then the presently observed differences would be easily explainable in terms of differences in the timing of emergence of segmentation abilities between the two French dialects. Crucial data on Canadian-French learning infants are therefore missing. In further support of the timing difference explanation suggested above, there is evidence that there are prosodic differences between the two French dialects, and in particular, that there is more intonational variability in Canadian- than in European-French (Me´nard, Ouellon, & Dolbec, 1999). Given recent findings with English-learning infants that pitch information might influence segmentation (Nazzi et al., 2005), these dialectal differences could have had an impact on early segmentation.6 One possible impact is that the presence of more intonational variability in Canadian-French would have drawn the attention of Canadian-French infants to this dimension and allow its use for segmentation at a younger age. By 8 months, these infants would then be segmenting on the basis of both syllabic and intonational information even when hearing a dialect (EuropeanFrench) that makes a lesser use of this intonational dimension, contrary to European-French infants who would only use syllabic information. A second possible implication of these dialectal differences is that Canadian-French (like English) allows the production of more exaggerated infant-directed speech than European-French (to which cultural differences in the way to address infants could also contribute). This could partly explain why the infant-directed speech we were able to elicit from our speaker had a faster speech rate and less intonational variations than the infant-directed speech used in previous studies of word segmentation in English, and in Polka and Sundara’s (2003) study (L. Polka, personal communication).7 Given evidence 6 Note that in Nazzi et al. (2005), there was an important lag between the onset of verb segmentation (between 10.5 and 13.5 months) and that of noun segmentation (between 6 and 10.5 months). However, the pattern of results was similar for both lexical categories. In particular, at the prosodic level, verbs with a weak–strong stress pattern were at a disadvantage compared to verbs with a strong–weak pattern as previously found for nouns. 7 Speech rate (computed by dividing the duration of the stimuli by their number of syllables) was of about 4.4 syllables per second for our stimuli, compared to about 2.5 for the weak– strong stimuli used by Jusczyk et al. (1999b) and the infantdirected stimuli used by Thiessen, Hill, and Saffran (2005). The mean F0 of our speaker was 201 Hz (range = 68–371 Hz), closer to the 230 Hz value (range = 140–260 Hz) of the adultdirected speech used by Thiessen et al. (2005) than to the 292 Hz value (range = 140–480 Hz) of their infant-directed stimuli. Finally, the average pitch peaks of our stimuli was 308 Hz, a value in between those of the adult- (252 Hz) and infant-directed (406 Hz) stimuli of Thiessen et al. (2005).
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
from Thiessen et al. (2005) that infant-directed speech facilitates 7-month-olds’ ability to segment the words of an artificial language, it is likely that the EuropeanFrench infants in our study were at a segmentation disadvantage compared to infants in other studies, including that of Polka and Sundara (2003), which could have resulted in a segmentation lag compared to what was observed for Canadian-French infants. In order to more clearly evaluate the possible contribution of intonational factors to early segmentation in French, it will be important to conduct comparative studies using the same stimuli with both European- and Canadian-French infants. To conclude, the main finding of the present study is the demonstration of syllabic segmentation in (European-) French-learning infants. These results point towards crosslinguistic differences in the pattern of emergence of segmentation abilities, and support the hypothesis that the emergence of segmentation abilities differs according to the rhythmic type of infants’ native language, the syllable being the unit of early prosodic segmentation in French. From a developmental point of view, as mentioned earlier, the emergence of these different procedures could be rooted in the sensitivity to rhythmic classes already present in newborns (Nazzi et al., 1998, 2000; Nazzi & Ramus, 2003; Ramus et al., 2000), a prosodic bootstrapping proposal initially suggested by Nazzi et al. (1998) and receiving here further support. Acknowledgments This research was supported by a grant from the Fyssen Foundation to T.N., and a grant from the European Science Foundation EUROCORES program ‘‘The Origin of Man, Language and Languages’’ to T.N. We thank Derek Houston for comments on a previous version of this paper. Appendix A. Passages used in the test phase of Experiments 1 to 4 Un putois a encore essaye´ de fuir. Il voulait voir mon putois du zoo. Mais il n’a trouve´ que le vieux putois. Ce putois n’e´tait pas tre`s content. Elle pensait au joli putois. Alors le grand putois s’est e´nerve´. Mon guidon est original. Il est plus re´sistant qu’un vieux guidon. La selle et ce guidon me plaisent beaucoup. Ce joli guidon doit eˆtre repeint. Il faut qu’un guidon soit bien fixe´. Elle aurait voulu un grand guidon. Un toucan mangeait des grains de ble´. Elle aimait beaucoup le vieux toucan. Il trouvait mon toucan des plus beaux. Le tre`s joli toucan e´tait ce´le`bre. Ce toucan savait bien chanter. Il est devenu un si grand toucan. Mon bandeau se plie tre`s facilement. Elle veut un joli bandeau de cette sorte. Il faudrait jeter le vieux bandeau. Tu sais que ce bandeau me manque beaucoup. Il ne trouve pas de grand bandeau. Un bandeau est toujours a` la mode.
297
References Abercrombie, D. (1967). Elements of general phonetics. Edinburgh: University of Edinburgh Press. Arvaniti, A. (1994). Acoustic features of Greek rhythmic structure. Journal of Phonetics, 22, 239–268. Aslin, R. N. (1993). Segmentation of fluent speech into words: learning models and the role of maternal input. In B. de Boysson-Bardies et al. (Eds.), Developmental neurocognition: Speech and face processing in the first year of life (pp. 305–315). Kluwer. Brent, M. R., & Cartwright, T. A. (1996). Distributional regularity and phonotactic constraints are useful for segmentation. Cognition, 61, 93–125. Brent, M. R., & Siskind, J. M. (2001). The role of exposure to isolated words in early vocabulary development. Cognition, 81, B33–B44. Cassidy, K. W., & Kelly, M. H. (1991). Phonological information for grammatical category assignments. Journal of Memory and Language, 30, 348–369. Christiansen, M. H., Allen, J., & Seidenberg, M. S. (1998). Learning to segment speech using multiple cues: A connectionist model. Language and Cognitive Processes, 13, 221–268. Cole, R. A., & Jakimik, J. (1978). Understanding speech: How words are heard. In G. Underwood (Ed.), Strategies of information processing (pp. 67–116). New York, NY: Academic Press. Cole, R. A., & Jakimik, J. (1980). How are syllables used to recognize words? Journal of the Acoustical Society of America, 67, 965–970. Curtin, S., Mintz, T. H., & Christiansen, M. H. (2005). Stress changes the representational landscape: evidence from word segmentation. Cognition, 96, 233–262. Cutler, A., & Carter, D. M. (1987). The predominance of strong initial syllables in the English vocabulary. Computer Speech and Language, 2, 133–142. Cutler, A., Mehler, J., Norris, D. G., & Segui, J. (1986). The syllable’s differing role in the segmentation of French and English. Journal of Memory and Language, 25, 385–400. Cutler, A., & Norris, D. (1988). The role of strong syllables in segmentation for lexical access. Journal of Experimental Psychology: Human Perception and Performance, 14, 113–121. Dahan, D., & Brent, M. R. (1999). On the discovery of novel wordlike units from utterances: An artificial-language study with implications for native-language acquisition. Journal of Experimental Psychology: General, 128, 165–185. De Boysson-Bardies, B. (1996). Comment la parole vient aux enfants. Paris, France: Editions Odile Jacob. Den Os, E. (1988). Rhythm and tempo of Dutch and Italian: a contrastive study. Utrecht, Holland: Drukkerij Elinkwijk BV. Dupoux, E., Pallier, C., Sebastia´n, N., & Mehler, J. (1997). A distressing ‘‘deafness’’ in French. Journal of Memory and Language, 36, 406–421. Dupoux, E., Peperkamp, S., & Sebastia´n-Galle´s, N. (2001). A robust method to study stress ‘‘deafness’’. Journal of the Acoustical Society of America, 110, 1606–1618. Echols, C. H., Crowhurst, M. J., & Childers, J. B. (1997). The perception of rhythmic units in speech by infants and adults. Journal of Memory and Language, 36, 202–225.
298
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299
Fant, G., Kruckenberg, A., & Nord, L. (1991). Durational correlates of stress in Swedish, French and English. Journal of Phonetics, 19, 351–365. Friederici, A. D., & Wessels, J. M. I. (1993). Phonotactic knowledge and its use in infant speech perception. Perception Psychophysics, 54, 287–295. Gout, A. (2001). Etapes pre´coces de l’acquisition du lexique. [Early steps in lexical acquisition]. Unpublished Dissertation. Ecole des Hautes Etudes en Sciences Sociales, Paris, France. Gout, A., Christophe, A., & Morgan, J. (2004). Phonological phrase boundaries constrain lexical access II. Infant data. Journal of Memory and Language, 51, 548–567. Halle´, P. A., & de Boysson-Bardies, B. (1996). The format of representation of recognized words in infants’ early receptive vocabulary. Infant Behavior and Development, 19, 463–481. Ho¨hle, B., & Weissenborn, J. (2003). German-learning infants’ ability to detect unstressed closed-class elements in continuous speech. Developmental Science, 6, 122–127. Ho¨hle, B., & Weissenborn, J. (2005). Word segmentation in German-learning infants. In Paper presented at the international workshop—Early word segmentation: a crosslinguistic approach taking advantage of Europe’s linguistic diversity. February 25–26, Paris, France. Hohne, E. A., & Jusczyk, P. W. (1994). Two-month-old infants’ sensitivity to allophonic differences. Perception Psychophysics, 56, 613–623. Houston, D. M., Jusczyk, P. W., Kuijpers, C., Coolen, R., & Cutler, A. (2000). Cross-language word segmentation by 9month-olds. Psychonomics Bulletin Review, 7, 504–509. Houston, D. M., Santelmann, L. M., & Jusczyk, P. W. (2004). English-learning infants’ segmentation of trisyllabic words from fluent speech. Language and Cognitive Processes, 19, 97–136. Johnson, E. K., & Jusczyk, P. W. (2001). Word segmentation by 8-month-olds: When speech cues count more than statistics. Journal of Memory and Language, 44, 1–20. Jusczyk, P. W., & Aslin, R. N. (1995). Infants’ detection of the sound patterns of words in fluent speech. Cognitive Psychology, 29, 1–23. Jusczyk, P. W., Cutler, A., & Redanz, N. (1993). Preference for the predominant stress patterns of English words. Child Development, 64, 675–687. Jusczyk, P. W., Friederici, A. D., Wessels, J., Svenkerud, V. Y., & Jusczyk, A. M. (1993). Infants’ sensitivity to the sound patterns of native language words. Journal of Memory and Language, 32, 402–420. Jusczyk, P. W., Hohne, E. A., & Bauman, A. (1999a). Infants’ sensitivity to allophonic cues for word segmentation. Perception & Psychophysics, 62, 1465–1476. Jusczyk, P. W., Houston, D. M., & Newsome, M. (1999b). The beginning of word segmentation in English-learning infants. Cognitive Psychology, 39, 159–207. Jusczyk, P. W., Luce, P. A., & Charles-Luce, J. (1994). Infants’ sensitivity to phonotactic patterns in the native language. Journal of Memory and Language, 33, 630–645. Kelly, M. H., & Bock, J. K. (1988). Stress in time. Journal of Experimental Psychology: Human Perception and Performance, 14, 389–403. Klatt, D. H. (1979). Speech perception: A model of acousticphonetic analysis and lexical access. Journal of Phonetics, 7, 279–312.
Klatt, D. H. (1989). Review of selected models of speech perception. In W. Marslen-Wilson (Ed.), Lexical representation and process (pp. 169–226). Cambridge, MA: MIT Press. Kooijman, V., Hagoort, P., & Cutler, A. (2005). Electrophysiological evidence for prelinguistic infants’ word recognition in continuous speech. Cognitive Brain Research, 24, 109–116. Kuijpers, C., Coolen, R., Houston, D., & Cutler, A. (1998). Using the head-turning technique to explore cross-linguistic performance differences. Advances in Infancy Research (Vol. 12, pp. 205–220). Stamford, CT: Ablex. Mattys, S., & Jusczyk, P. W. (2001a). Phonotactic cues for segmentation of fluent speech by infants. Cognition, 78, 91–121. Mattys, S., & Jusczyk, P. W. (2001b). Do infants segment words or recurring contiguous patterns? Journal of Experimental Psychology: Human Perception and Performance, 27, 644–655. Mattys, S., Jusczyk, P. W., Luce, P. A., & Morgan, J. L. (1999). Phonotactic and prosodic effects on word segmentation in infants. Cognitive Psychology, 38, 465–494. McLennan, S. (2005). A dynamic adaptive model of linguistic rhythm. In Proceedings of the I.S.C.A. Workshop on Plasticity in Speech Perception (p. 130), London, UK, June 15–17, 2005. McQueen, J. M., Norris, D., & Cutler, A. (1994). Competition in spoken word recognition: Spotting words in other words. Journal of Experimental Psychology: Learning, Memory and Cognition, 20, 621–638. Mehler, J., Dommergues, J. Y., Frauenfelder, U., & Segui, J. (1981). The syllable’s role in speech segmentation. Journal of Verbal Learning and Verbal Behavior, 20, 298–305. Mehler, J., Dupoux, E., Nazzi, T., & Dehaene-Lambertz, G. (1996). Coping with linguistic diversity: The infant’s viewpoint. In J. L. Morgan & K. Demuth (Eds.), Signal to syntax (pp. 101–116). Mahwah, NJ: Erlbaum. Me´nard, L., Ouellon, C., & Dolbec, J. (1999). Prosodic markers of regional group membership: The case of the French of Quebec versus France. In Proceedings of the XIVth International Congress of Phonetic Sciences (ICPhS 99), San Francisco, pp. 1601–1604. Morgan, J. L., & Saffran, J. R. (1995). Emerging integration of sequential and suprasegmental information in preverbal speech segmentation. Child Development, 66, 911–936. Nazzi, T. (1997). Du rythme dans l’acquisition et le traitement de la parole. [Rhythm in the acquisition and processing of speech]. Unpublished Ph.D. Ecole des Hautes Etudes et Sciences Sociales, Paris. Nazzi, T. (2005). Use of phonetic specificity during the acquisition of new words: differences between consonants and vowels. Cognition, 98, 13–30. Nazzi, T., Bertoncini, J., & Mehler, J. (1998). Language discrimination by newborns: Toward an understanding of the role of rhythm. Journal of Experimental Psychology: Human Perception and Performance, 24, 756–766. Nazzi, T., Dilley, L. C., Jusczyk, A. M., Shattuck-Hufnagel, S., & Jusczyk, P. W. (2005). English-learning infants’ segmentation of verbs from fluent speech. Language and Speech, 48, 279–298. Nazzi, T., Floccia, C., & Bertoncini, J. (1998). Discrimination of pitch contours by neonates. Infant Behavior and Development, 21, 779–784.
T. Nazzi et al. / Journal of Memory and Language 54 (2006) 283–299 Nazzi, T., Jusczyk, P. W., & Johnson, E. K. (2000). Language discrimination by English learning 5-month-olds: Effects of rhythm and familiarity. Journal of Memory and Language, 43, 1–19. Nazzi, T., & Ramus, F. (2003). Perception and acquisition of linguistic rhythm by infants. Speech Communication, 41, 233–243. Peretz, I., Lussier, I., & Be´land, R. (1998). The differential role of syllabic structure in stem completion for French and English. European Journal of Cognitive Psychology, 10, 75–112. Perruchet, P., & Vinter, A. (1998). PARSER: a model for word segmentation. Journal of Memory and Language, 39, 246–263. Pike, K. (1945). The intonation of American English. Ann Arbor, MI: University of Michigan Press. Polka, L., & Sundara, M. (2003). Word segmentation in monolingual and bilingual infant learners of English and French. In Proceedings of the 15th international congress of phonetic sciences (pp. 1021–1024). Barcelona, Spain. Ramus, F., Hauser, M. D., Miller, C., Morris, D., & Mehler, J. (2000). Language discrimination by human newborns and by cotton-top tamarin monkeys. Science, 288, 349–351. Ramus, F., Nespor, M., & Mehler, J. (1999). Correlates of linguistic rhythm in the speech signal. Cognition, 73, 265–292. Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274, 1926–1928. Sansavini, A., Bertoncini, J., & Giovanelli, G. (1997). Newborns discriminate the rhythm of multisyllabic stressed words. Developmental Psychology, 33, 3–11.
299
Sebastia´n-Galle´s, N., & Bosch, L. (2002). Building phonotactic knowledge in bilinguals : role of early exposure. Journal of Experimental Psychology: Human Perception and Performance, 28, 974–989. Sebastia´n-Galle´s, N., Dupoux, E., Segui, J., & Mehler, J. (1992). Contrasting syllabic effects in Catalan and Spanish. Journal of Memory and Language, 31, 18–32. Shafer, V. L., Shucard, D. W., & Jaeger, J. J. (1999). Electrophysiological indices of cerebral specialization and the role of prosody in language acquisition in three-monthold infants. Developmental Neuropsychology, 15, 73–109. Thiessen, E. D., Hill, E. A., & Saffran, J. R. (2005). Infantdirected speech facilitates word segmentation. Infancy, 7, 53–71. Thiessen, E. D., & Saffran, J. R. (2003). When cues collide: Use of stress and statistical cues to word boundaries by 7- to 9month-old infants. Developmental Psychology, 39, 706–716. Turk, A. E., Jusczyk, P. W., & Gerken, L. (1995). Do Englishlearning infants use syllable weight to determine stress? Language and Speech, 38, 143–158. van de Weijer, J. (1998). Language input for word discovery. Doctoral dissertation. University of Nijmegen (MPI Series in Psycholinguistics, 9). Vihman, M. M., Nakai, S., DePaolis, R. A., & Halle´, P. (2004). The role of accentual pattern in early lexical representation. Journal of Memory and Language, 50, 336–353. Vroomen, J., van Zon, M., & de Gelder, B. (1996). Metrical segmentation and inhibition in spoken word recognition. Journal of Experimental Psychology: Human Perception and Performance, 21, 98–108.