Newborns discriminate the rhythm of multisyllabic stressed words

Developmental Psychology1997, MM. 33, No. 1,3-11 Copyright 1997 by the American Psychological Association, Inc.

0012-l649/97/$3.00

Newborns Discriminate the Rhythm of Multisyllabic Stressed Words

Alessandra SansaviniUniversity of Bologna

Josiane BertonciniLaboratoire de Sciences Cognitives et Psycholinguistique,

Centre National de la Recherche Scientifique

Giuliana GiovanelliUniversity of Bologna

Three experiments were run to test whether newborns were able to discriminate different stresspatterns in multisyllabic stressed Italian words that varied both in consonants and in number ofsyllables. A high-amplitude sucking procedure was adopted in which the experimental group heard2 sets of stimuli alternating minute by minute, whereas the control group heard only a single set ofstimuli. The results showed that stress patterns were discriminated in 2 disyllabic phoneticallyunvaried words (Experiment 1), in 2 trisyllabic consonant-varied words (Experiment 2), and in 2sets of disyllabic words varied in consonants within and between words (Experiment 3). Thealternation procedure proved to be suitable for examining newboms' abilities of discrimination andcategorization. It also lowered the participants* rejection rate compared with the classic habituationprocedure. The present results suggest that newborns are sensitive to words' rhythm, as carried bystress patterns, and that this prosodic information is salient even in the presence of substantialconsonant variation.

Studies on newborns and very young infants have shownthat they are sensitive to the global prosody of natural speech.Newborns discriminate and prefer the mother's voice to otherfemale voices (DeCasper & Fifer, 1980). The same preferencehas been found in 6-week-old infants as long as mothers' utter-ances are correctly intonated (Mehler, Bertoncini, Barriere, &Jassik-Gerschenfeld, 1978). Newborns as well as 2-month-oldsalso discriminate and prefer the maternal language comparedwith a foreign language (Mehler et al., 1988; Moon, Cooper, &Fifer, 1993), and they prefer a female voice to a male voice ora musical phrase (Giovanelli 1991; Giovanelli et al., 1990).Moreover, newboms (Cooper & Aslin, 1990) as well as 2-, 4-,

Alessandra Sansavini and Giuliana Giovanelli, Department of Psy-chology, University of Bologna, Bologna, Italy; Josiane Bertoncini, La-boratoire de Sciences Cognitives et Psycholinguistique, CNRS, Paris,France.

This research was supported by Italian CNR (National Council ofResearch) Grant 924561.08 and University of Bologna Grant 9200291.

We thank Professors Marino Bosinelli, George Butterworth, and Na-than Fox and the three anonymous reviewers for their thoughtful andhelpful comments; Professors GianPaolo Salvioli and Paola DallaCasaand the obstetricians of Bologna S. Orsola Hospital for the possibilityof testing newborns at the hospital and the parents who, along withtheir infants, took part in the experiments; Daniele Maurizzi, Jean-LucManguin, and Michel Dutat for their technical assistance; Caroline Floc-cia for informatic suggestions; and Britte von Ooijen for reading thepaper.

Correspondence concerning this article should be addressed to Ales-sandra Sansavini, Department of Psychology, University of Bologna,Viale Berti Pichat 5,40127, Bologna, Italy. Electronic mail may be sentvia Internet to [email protected].

5-, 1-, and 9-month-old infants (Pegg, Werker, & McLeod, 1992;Werker & McLeod, 1989) presented with infant-directed (ID)and adult-directed speech prefer to listen to ID speech, whichemphasizes such prosodic characteristics as intonation, pauses,stress, tone, and duration of vowels (Bsrnald, 1985).

The preference for the prosody of the maternal language be-comes more specific at about 6 months. At this age, infantsdiscriminate and prefer to listen to sequences of words of thematernal language rather than to those of a foreign languagewhen the two languages actually differ in their global prosody,whereas this preference appears only at 9 months when thetwo languages mostly differ in their phonetic and phonotacticproperties (Jusczyk, Friederici, Wessels, Svenkerud, & Jusczyk,1993). Moreover, 7- to 10-month-old infants prefer to listen toclauses of the maternal language in which artificial pauses areinserted between rather than within clauses (Hirsh-Pasek et al.,1987), and 9-month-old infants prefer to listen to phrases ofthe maternal language in which artificial pauses are insertedbetween rather than within phrases (Jusczyk, Hirsch-Pasek^ etal., 1992). It thus appears that, in their first months of life,infants pay attention to and discriminate the global prosody ofspeech and become attuned to the prosodic patterns of the mater-nal language. This tuning also appears in their first linguisticproductions; 13-month-old French infants produce more length-ened final syllables and more ascending intonations than age-matched American infants (Levitt & Wang, 1991). These samecharacteristics have been found in 18-month-old French infants'productions, whereas age-matched Japanese infants producedmore descending intonations and did not lengthen final syllables(Halle\ Boysson-Bardies, & Vihman, 1991).

One can question whether the sensitivity to the prosodic struc-ture of speech can help infants to segment speech into words,

SANSAVINI, BEKT0NCIN1, AND GIOVANELLI

which is a crucial requirement for lexical acquisition. Retrievingwords in speech is not a trivial task because speech is a continu-ous acoustic signal, highly variable in speaker's voice and into-nation, consisting of chains of coarticulated phonemes, withoutclear physical boundaries between words. Because, at least inthe first year of life, infants cannot rely on syntactic or onsemantic cues to identify words in speech, a hypothesis is thatthey use prosodic cues to solve this problem (Jusczyk, 1995;Morgan, 1986; Peters, 1983).

Some prosodic characteristics of speech, such as intonationand pauses, mark sentences, clauses, and phrases. Others, suchas pitch and stress, mark words. Languages differ in the way inwhich their prosodic characteristics are organized. With regardto stress, which increases vowels' physical parameters (i.e.,duration, fundamental frequency, and intensity), in some lan-guages it is fixed (e.g., in French, on the last syllable of words),whereas in others it can have different positions, among whichone may be dominant (e.g., in Italian, the more frequent positionof stress is on the penultimate syllable of words, i.e., stressedsyllable plus unstressed syllable in disyllabic words and un-stressed syllable plus stressed syllable plus unstressed syllablein trisyllabic words; Bertinetto, 1980). Moreover, in Italian,stress mostly increases vowel duration (Bertinetto, 1981),whereas in English it increases vowel fundamental frequencyabove all (Lehiste, 1970).

Adults can rely on the redundant prosodic cues of phrases,such as pitch discontinuities and lengthening of segment dura-tion, to learn an artificial complex grammar (Morgan, Meier, &Newport, 1987). Native English-speaking adults actually useprosodic cues, such as strong (stressed) and weak (unstressed)syllables, to segment continuous spoken English into words(Cutler & Butterfield, 1992). Native Italian-speaking adults gen-erally represent the structure of words of their maternal languagein terms of its more frequent stress pattern (Colombo, 1992).

With regard to infants' segmentation and representation ofwords, some authors (Gleitman & Wanner, 1982) proposed thatstressed syllables attract infants' attention and could help themto identify word boundaries and to analyze the phonetic compo-nents of the syllables. Other studies have shown that 9-month-old infants prefer to listen to words of the maternal languagecarrying the more frequent stress pattern rather than to wordscarrying a less frequent stress pattern (Jusczyk, Cutler, & Re-danz, 1993). Moreover, infants who are just starting to speakprimarily utter the stressed and final syllables of words (i.e.,those syllables having lengthened vowels); only at a later agedo they also pronounce the unstressed nonfinal syllables (Blas-dell & Jensen, 1970; Echols & Newport, 1992; Gerken, Lan-dau, & Remez, 1990). Stressed syllables are also uttered morecorrectly than unstressed syllables (Echols, 1993), and stressedwords are produced earlier than unstressed words (Landau &Gleitman, 1984).

These studies indicate that after a few months of linguisticexperience infants have figured out the prosodic structure of thematernal language that helps them to retrieve meaningful unitsin speech. It may be asked whether newborns are already sensi-tive to those prosodic cues that mark such units and can usethem for speech segmentation.

A recent study on French newborns showed that they discrim-

inate between tokens of /mati/ taken from the word boundaryof two-word sequences like "panorama typique" and word-internal /mati/ tokens taken from single words like "mathema-n'quement" (Christophe, Dupoux, Bertoncini, & Mehler, 1994).The authors concluded that newborns are capable of detectingword boundaries using acoustic cues, such as lengthening ofphonemes at the end and at the beginning of words. Anotherstudy on French newborns (Bijeljac-Babic, Bertoncini, &Mehler, 1993) showed that they discriminate disyllabic (e.g., /ri 'fo/, /zu'ti/) from trisyllabic (e.g., /mazo'pu/, /keso'pa/)1 ,whereas they do not discriminate disyllabic items having fourphonemes (e.g., / i 'blo/, /a'zur/) from disyllabic items havingsix phonemes (e.g., /al'prim/, /sul 'dri/). From these results,Bijeljac-Babic et al. (1993) hypothesized that newborns canperceive the syllabic units of spoken words. Provided that Frenchis a syllable-timed language and has a fixed stress, it might alsobe supposed that newborns discriminate the constant rhyth-micity of disyllabic items from that of trisyllabic items.

Thus, a complementary hypothesis to those mentioned earliermay be that newborns are predisposed to attend to the rhythmicalstructure of speech imposed by the variation of prosodic ele-ments. In particular, newborns would be sensitive to the thealternation of stressed and unstressed syllables (stress patterns)and, more specifically, of stressed and unstressed vowels. Stresspatterns might thus guide both the segmentation of speech andthe representation of its lexical structure (Sansavini, 1994).

To verify this hypothesis, it is first necessary to show thatnewborns perceive and discriminate stress patterns. A few stud-ies have shown that 2-month-old infants discriminate the stresspattern of two disyllabic phonetically unvaried items (/'baba/vs. /ba'ba/; Spring & Dale, 1977) and consonant-varied items(/'bada/ vs. /ba'da/; Jusczyk & Thompson, 1978), but no stud-ies on neonatal discrimination of stress patterns exist. The aimof our study is, therefore, to verify whether 2-day-old newbornsdiscriminate stress patterns of multisyllabic words and how pho-netic and structural variations of words may affect this capacity.

To examine newborns' discrimination of stress patterns, amodified version of the classic habituation procedure was cho-sen. In this procedure, already used by Cowan, Suomi, andMorse (1982) with 2-month-old infants, two stimuli are alter-nated for the experimental group, whereas a single stimulus isrepeatedly presented to the control group. If the high-amplitudesucking (HAS) rate of the experimental group remains highduring the test and gradually diverges from that of the controlgroup, it is inferred that infants detect the difference between thealternating stimuli. More specifically, the control group shouldhabituate to the unchanged stimulus and lower its HAS rate overthe course of the experiment. Instead, the experimental groupshould notice the difference between the alternating stimuli.Thus, if the experimental group delays habituation and main-tains a high HAS rate throughout the task, this would indicatethat newborns react to the change of stimuli. By alternating thestimuli minute by minute in the experimental group and not

1 In order to indicate the Stress pattern, items are written accordingto International Phonetic Association rules: Stressed syllables are pre-ceded by the accent sign ( ' ) (e.g., 'iblo is stressed on the first syllable,whereas i'blo is stressed on the final syllable).

NEONATAL DISCRIMINATION OF STRESSED WORDS

exceeding a total duration of 9 min, this procedure should de-crease the high rejection rate typically obtained in other studieson newborns that have used the classic habituation procedure.Our study, therefore, also aims to verify whether this procedureis suitable to test discrimination in newborns.

Experiment 1

Assuming that stress is one of the prosodic characteristics ofspeech to which newborns pay attention (Sansavini, Berton-cini, & Giovanelli, 1994), this first experiment tests the discrim-ination of stress patterns of two disyllabic words simple instructure (CVCV) and identical in their phonetic segments, asSpring and Dale (1977) did with 2-month-old infants.

Method

Participants. Sixty-six healthy and neurologically normal Italiannewborns were recruited at Bologna University Hospital. They were 48to 72 hours old and had a birth weight of 2,900 to 4,500 g, a gestationalage of 39 to 42 weeks, a minimum score of 8 on the 1-min Apgar scoreand 10 on the 5-min Apgar. Newboms were excluded from the sampleif they did not maintain a state of quiet alertness (e.g., sleeping, crying,agitated awakening, continuous fluctuations between sleep and wake)or stopped nonnutritive sucking (e.g., refusing the pacifier, not sucking)for 2 consecutive min or for 3 nonconsecutive min. Forty participants(20 in the experimental group and 20 in the control group) completedthe test, remaining awake and sucking during baseline and the 9 min oftest.

Apparatus. A multifunction electronic custom-designed deviceserved as interface between an IBM-PC compatible computer, a GouldP23 pressure transducer (connected to a standard pacifier), and twoTASCAM Porta05 tape recorders (connected to an amplified loud-speaker). The electronic device detected the sucking responses, trans-formed them into digital output, and, when sucking amplitude wasgreater than a predefined threshold, opened the connection between therecorders and the loudspeaker, allowing the stimuli to be delivered. Theelectronic device changed the delivery of the stimuli from one recorderto the other every minute; the two recorders played two different tapesfor the experimental group but two copies of a single tape for the controlgroup.

Stimuli. The stimuli were two disyllabic stressed words followingthe prosodic, phonetic, and phonotactic rules of the Italian language:/'mama/, stressed on the penultimate syllable, and /ma'ma/, stressed onthe last syllable (stimuli are written according to International PhoneticAssociation rules; stressed syllables are preceded by the accent sign).Because we wanted to use natural, and not synthesized, stimuli, wordswere naturally produced by a native Italian-speaking female.

Each stimulus was recorded within a carrier phrase that ensured aconstant flat intonation of the stimuli by putting the target item in themiddle of phrase without emphatic stress (e.g., /Una 'mama bella/).This was done because intonation tends to vary at the beginning andend of phrases, but not in the middle, unless an emphatic focus isvoluntarily put on single words (Nespor & Vogel, 1986). Phrases weredigitally recorded on a computer equipped with an Orose board andanalyzed to their acoustical parameters using a spectrographic analysisprogram. To control for duration, fundamental frequency, and intensity,several tokens were recorded, and those two that were most alike in allthose parameters were recorded on the tapes and used in the experiment(for the physical details of the stimuli see the Appendix). The onset-to-onset stimulus interval on the tape was 1.100 ms, with about 450 ms ofsilence from the offset of one stimulus to the onset of the next.

As can be seen in the Appendix, /'mama/ is slightly longer than/ma'ma/. This happens because in the Italian language words stressedon the penultimate syllable consist of two long syllables (the stressednonfinal syllable and the unstressed but final syllable), whereas wordsstressed on the last syllable consist of a short (the unstressed nonfinal)syllable and a long (the stressed final) syllable. However, if words hadbeen compressed or lengthened, they would have lost their naturalrhythm. In favor of the use of natural stimuli, it can also be pointed outthat, in the study by Bijeljac-Babic et al. (1993), newborns discrimi-nated disyllabic (CVCV) from trisyllabic items (CVCVCV) not onlywhen the stimuli were naturally produced (the disyllabic being slightlyshorter than the trisyllabic ones) but also when the stimuli were com-pressed or extended in such a way that they could not be distinguishedon the basis of their duration.

Procedure. The test was conducted in a quiet room. Newborns weretested individually during a state of quiet alertness, about 2.5 hours afterfeeding. They were placed in a reclining position in a special infantbath chair that reduces head movements. A loudspeaker was placed infront of the infant at about 12 in. (30 cm). A pacifier, connected to thepressure transducer, was given to the infant. To make the situation asnatural as possible, the pacifier was not held by an assistant or by amechanical arm. As soon as the infant was adapted to the situation (i.e.,accepted the pacifier and sucked in a state of quiet alertness), the testbegan. If the infant lost the pacifier once or twice during the test, anassistant put it in the infant's mouth. Infants who repeatedly refused thepacifier were excluded from the study.

The dependent variable was the nonnutritive sucking response. Thecomputer recorded both the frequency and amplitude of sucking, distin-guishing sucking lower and higher than a predetermined amplitudethreshold. The threshold was fixed before the experiment for all new-borns, as Siqueland and De Lucia (1969) did, in such a way that about80% of the sucking exceeded the threshold (HAS) and triggered stimulusdelivery. This threshold guaranteed that only effective sucking triggeredthe stimuli, but at the same time it did not demand too great an effortto hear the stimuli.

During a 1-min baseline, sucking was recorded and no stimulus wasdelivered. Then the test began. As done by Cowan et al. (1982), twoalternating stimuli were presented to the experimental group, whereas asingle stimulus was presented to the control group. The two stimuli werealternated minute by minute, and not every 30 s as Cowan et al. (1982)did, in the belief that newborns have longer latencies than 2-month-oldsin processing stimulation. Using this procedure of alternation, newbornsdid not have to satisfy any habituation criterion to hear the stimuli. Thestimuli were delivered contingent on HAS because, as shown by Wil-liams and Golenski (1978), Trehub and Chang (1977), and Floccia,Christophe, and Bertoncini (in press), a connection between HAS anddelivery of stimulation is necessary to observe linguistic discriminationin very young infants. The test lasted 9 min, and the order of presentationof the stimuli was counterbalanced. Half of the experimental groupheard /'mama/ followed by /ma'ma/; for the other half this was thereverse. Similarly, half of the control group heard only /'mama/, andthe other half heard only /ma'ma/.

It was expected that at baseline and on the first minute of test theHAS rate of the experimental group did not differ from that of thecontrol group. Then, the HAS rate of the experimental group shouldremain high and gradually diverge from that of the control group infants,who were expected to show habituation.

Results

The statistical analyses chosen for evaluating the effects ofthe stimuli on newborns' HAS rates were linear regression andanalysis of variance (ANOVA). Tb determine whether there was

SANSAVINI, BEKT0NCIN1, AND GIOVANELLI

Expar.Contr.

38.8541.3

40.144.5

31.83S.6

34.2633.06

35.536.65

34.436.2

34.938.9

39.9529.9

37.730.85

38.9529.65

. + Contr.

Figure 1. Stress patterns discrimination in two disyllabic phoneticallyunvaried words. Values represent mean high-amplitude (ampl.) suckingrates of experimental (exper.) and control (contr.) newborn groups dur-ing baseline (BL) and 9 min of test.

any significant increase or decrease in HAS rates throughoutthe test, a linear regression analysis was performed for eachgroup on the data of the 9 min of the test. To ensure that thetwo groups did not differ at the beginning, an ANOVA wasperformed on the baseline. To find out whether and at whichpoint of the test there were significant differences between thetwo groups, an ANOVA was performed on the first 3 min (first3-min period), on the second 3 min (second 3-min period),and on the third 3 min (third 3-min period) of test. The choiceto divide the period of test into three blocks and to performstatistical analyses on them has been adopted also by otherauthors who used nonnutritive sucking as a response (e.g., De-Casper & Fifer, 1980; Moon, Bever, & Fifer, 1992; Moon et al.,1993). Tb determine whether the HAS rates of the two groupsdiverged during the course of the test, the interaction betweengroup and time was examined. Mean HAS rates for baselineand each minute of test are reported in Figure 1 for both groups.

Linear regression. The linear regression analysis was sig-nificant in the control group in the 9-min test, F(\, 178) =10.42,/? = .001, R2 = .05, $ = - . 23 ; the control group's HASrate significantly decreased during the test, whereas that of theexperimental group did not vary significantly from the beginningto the end of the test (F < 1).

Analysis of variance. The HAS rate of the two groups didnot differ significantly at baseline or in the first 3-min periodor in the second 3-min period ( F < 1), but it differed signifi-cantly in the third 3-min period, F ( l , 38) = 4.14, p = .048.The interaction between group and time was significant for acomparison of the first and the third 3-min periods, F ( l , 38)= 14.1, p = .0005, and for a comparison of the second and thethird 3-min periods, F ( l , 38) = 12.47, p = .001. An ANOVAwas also performed within each group to check whether theorder of presentation in the experimental group and the type ofstimulus in the control group affected newborns' responses. Nosignificant effects were found (F < 1).

Discussion

The results showed that the HAS rate of the experimentalgroup remained fairly constant over the course of the test,whereas that of the control group significantly decreased. TheHAS rates of the two groups diverged significantly in the third3-min period of the test, from these results, it can be inferredthat the experimental group maintained a high interest for thestimulation that changes rhythm every minute, whereas the con-trol group habituated to the rhythmically invariant stimulation.Newborns, therefore, like 2-month-olds (Spring & Dale, 1977),discriminate two disyllabic phonetically unvaried words thatdiffer in their stress pattern.

These results also showed that this procedure, characterizedby the connection between HAS and delivery of stimulation andby the alternation between the two stimuli for the experimentalgroup, was suitable for testing discrimination in newborns. Dis-crimination was inferred by the progressive and significant di-vergence between the HAS rates of the experimental and controlgroups and by the maintenance of a higher HAS rate in theexperimental group. The rate of infants who completed the test(60.7%) was higher than that generally obtained with the classichabituation procedure (see, e.g., Bertoncini, Bijeljac-Babic, Jus-czyk, Kennedy, & Mehler, 1988). This result may be explainedby two factors. First, a fixed duration of 9 min seems to be areasonable time for newborns' limited ability to maintain atten-tion (Weiss & Zelazo, 1991). Second, the minute-by-minutealternations seem to give the newborns more opportunity todetect the characteristic on which the two stimuli differ.

The HAS rate of the experimental group, although remaininghigh, did not increase significantly during the course of thepresent experiment, unlike that of 2-month-old infants observedby Cowan et al. (1982). Hoccia et al. (in press), testing new-borns with the same procedure, also obtained results similar toours. A possible explanation is that newborns are not able tomodulate their HAS rate as easily as 2-month-old infants.

Experiment 2

In the first experiment, infants were presented with two disyl-labic words in which the same syllable /ma/ was duplicated. Inthis second experiment, our question is whether a more complexcontext affects newborns' discrimination of stress patterns.Complexity, meant as the variation of words' phonetic segmentsand number of syllables, seems to affect phonetic discrimina-tion. It has been shown that 1- to 4-month-old infants fail todiscriminate a phonetic contrast embedded in two trisyllabicconsonant-varied words (/atapa/ vs. /ataba/), whereas they suc-ceed with two disyllabic words (/apa/ vs. /aba/; Trehub, 1976);2-month-old infants discriminate a phonetic contrast embeddedin trisyllabic words only if it is carried by a syllable having anemphatic stress (/ma'rana/ vs. /ma'lana/; Karzon, 1985); 6-month-old infants succeed better in phonetic discriminationwhen presented with redundant contexts (/tibati/ vs. /tiduti/)than when presented with more complex contexts (/tibako/ vs. /tiduko/; Goodsitt, Morse, Ver Hoeve, & Cowan, 1984).

Does complexity (i.e., variation of phonetic segments andnumber of syllables in words), affect stress pattern discrimina-

NEONATAL DISCRIMINATION OF STRESSED WORDS

tion as well as phonetic discrimination? It has been shown that2-month-old infants discriminate the stress patterns of two con-sonant-varied disyllabic words (/'bada/ vs. /ba'da/; Jusczyk &Thompson, 1978), and 5- to 11-month-old infants discriminateduration and intensity of final vowels in consonant-varied mono-syllabic (/mad/), disyllabic (/samad/), and trisyllabic (/masa-mad/) items (Bull, Eilers, & Oiler, 1984; Eilers, Bull, Oiler, &Lewis, 1984). However, those authors did not verify whetherthe discrimination of vowel duration and intensity continued toappear when the target syllable was not the final one. We maysuppose that final syllables are easier to discriminate than non-final syllables because the former are marked by vowel lengthen-ing and are followed by a pause.

Therefore, the aim of this second experiment was to examinewhether newboms discriminate stress patterns in phoneticallyvaried trisyllabic words having nonfinal stress patterns (i.e.,carrying stress on the antipenultimate syllable (stressed syllableplus unstressed syllable plus unstressed syllable) or on the pen-ultimate syllable (unstressed syllable plus stressed syllable plusunstressed syllable). These are both common stress patterns inItalian trisyllabic words. To avoid too much phonetic variation,the phonetic context was varied in only one type of phoneticsegment. As it was done in the studies described previously on2-month-old infants, consonants were varied and vowels werekept constant.

Method

Participants. Of 39 healthy newboms who were tested, 24 (12 inthe experimental and 12 in the control group) completed the test. (Thecriteria of selection and validity of subjects are the same as in Experi-ment 1.)

Stimuli. The stimuli were two trisyllabic words, differently stressedand varied in consonants, naturally produced by a native Italian-speakingfemale in the same way as in Experiment 1. Stimuli were stressed onthe antipenultimate syllable (Z'tacala/) and on the penultimate syllable/ta'cala/, respectively. The onset-to-onset stimulus interval in the tapewas 1.250 ms, with about 450 ms of silence from the offset of onestimulus to the onset of the next. For more details about the physicalcharacteristics of the stimuli, see the Appendix.

Apparatus and procedure. The apparatus and procedure were thesame as in Experiment 1.

Results

The same statistical analyses as in Experiment 1 were per-formed. Mean HAS rates for baseline and each minute of testfor both groups are reported in Figure 2.

Linear regression. The linear regression analysis was sig-nificant in the experimental group in the 9-min test, F( 1, 106)= 5.62, p = .019, R2 = .05, /? = .22; the experimental group'sHAS rate significantly increased during the test, whereas thatof the control group did not vary significantly from the begin-ning to the end of the test (F < 1).

Analysis of variance. The HAS rate of the two groups didnot differ significantly at baseline or in the first 3-min period(F < 1), but it differed significantly in the second, F ( l , 22)= 6.8, p - .016, and third 3-min periods, F ( l , 22) = 8.49,p = .008. The interaction between group and time for a compari-

55

?50

oM

•£45

1X w

35

Exper.Contr.

*

BL

43.1835.41

1

40.6638.41

/

/ \

2

46.3343.33

\

V

3

48.3337.41

* /\ /V

y

4

44.4140.83

A//

*\5

51.9138.26

t\ /

V

8

47.8335.26

r

s

7

62.2536.33

8

51.539.16

9

53.9136.08

-"-Exper. 4- Contr.

Figure 2. Stress patterns discrimination in two trisyllabic consonant-varied words. Wues represent mean high-amplitude (ampl.) suckingrates of experimental (exper.) and control (contr.) groups during baseline(BL) and 9 min of test.

son of the first and the third 3-min periods was marginallysignificant, F( 1,22) = 3.53, p = .07, whereas a similar compar-ison of the second and the third 3-min periods yielded no sig-nificant differences ( F < 1). An ANONA. was also performedwithin each group to check whether the order of presentationin the experimental group and the type of stimulus in the controlgroup affected newboms' responses. No significant effects werefound ( F < 1).

Discussion

The results showed that the HAS rate of the experimentalgroup increased significantly over the course of the test anddiverged significantly from the HAS rate of the control groupin the second and third 3-min periods of the test. As in Experi-ment 1, more than 60% of the examined participants completedthe test.

Newboms are thus able to discriminate two trisyllabic wordsthat are differently stressed and varied in consonants. This sug-gests that they are sensitive to the rhythm not only of disyllabicbut also of trisyllabic consonant-varied words having nonfinalstress patterns.

However, whereas in Experiment 1 the significant divergencebetween the HAS rates of the two groups was caused by adecrease in the control group's HAS rate, in this case it resultedfrom an increase in the experimental group's HAS rate. Thislatter trend is similar to that found by Cowan et al. (1982)with 2-month-old infants. It might be that trisyllabic consonant-varied words, such as those used in this second experiment,activate neonatal sucking more than disyllabic phonetically un-varied words, such as those used in Experiment 1 because theformer are rhythmically and phonetically more complex thanthe latter.

Experiment 3

Experiments 1 and 2 showed that newboms discriminate thestress patterns of two disyllabic and trisyllabic items regardless

8 SANSAVINI, BEKFONCINI, AND GIOVANELLI

of consonant variation. However, fluent speech is constituted bysequences of words in which the phonetic variation is not onlyinside words but also between words. If stress is a prosodic cueuseful for segmenting speech into words, stress patterns shouldbe detected and discriminated by newborns also when they arepresented with sets of words differing in stress position andvaried in phonetic segments (i.e., stimuli closer to a naturallinguistic context). In this latter case, newborns should perceivewords depending on their stress pattern and respond to prosodicrather than to segmental variation, even though it has been shownthat they are able to discriminate consonant contrasts at leastwhen presented with two monosyllabic items (Bertoncini, Bijel-jac-Babic, Blumstein, & Mehler, 1987). Categorization of pro-sodic characteristics may guide language acquisition byallowing infants to figure out the prosodic structures of thematernal language (e.g., intonation patterns, stress patterns) use-ful for segmenting speech into units (Jusczyk, Cutler, & Redanz,1993).

In this third experiment, therefore, newborns had to recognizethat several words, although phonetically different, had the samestress pattern (e.g., belong to the same prosodic category) anddiffered from others, equally varied, that had a different stresspattern. Thus, this third experiment examined whether newbornsdiscriminate stress patterns of two sets of disyllabic words var-ied in all of the consonants. As in Experiments 1 and 2, vowelswere kept unvaried.

Several studies testing infants' categorization used the classichabituation procedure with two sets of stimuli instead of twosingle stimuli (Bertoncini et al., 1988; Jusczyk, Bertoncini, Bi-jeljac-Babic, Kennedy, & Mehler, 1990); a set of stimuli, sharinga common property, was presented for habituation and anotherset was presented for reaction to novelty. In the same way,Experiment 3, using the procedure of alternation, used two setsof stimuli. This experiment also aimed to verify whether thechosen procedure is suitable for testing newborns' discrimina-tion and categorization abilities when two sets of stimuli insteadof two single stimuli are used.

Method

Participants. Of the 66 healthy newborns tested, 46 ( 23 in the exper-imental group and 23 in the control group) completed the test. (Thecriteria of selection and validity of subjects are the same as in Experi-ments 1 and 2.)

Stimuli. The stimuli were two sets of eight disyllabic words naturallyproduced by a native Italian-speaking female in the same way as inExperiments 1 and 2. In each group the stimuli had the same stresspattern but were varied in all of the consonants, so that each word wasnot constituted by a duplication of the same syllable. The first set ofstimuli was stressed on the penultimate syllable (/'daga/, /'nata/,/'mara/, /"baga/, /'lama/, /'taca/, /'rama/, /'gaba/), whereas the secondset was stressed on the last syllable (/da'ga/, /na'ta/, /ma'ra/, /ba'ga/,/la'ma/, /ta'ca/, /ra 'ma/, /ga 'ba/) . The onset-to-onset stimulus inter-val in the tape was 1.150 ms, with about 450 ms of silence from theoffset of one stimulus to the onset of the next. The order of the stimuliwas randomized throughout the tape. For more details about the physicalcharacteristics of the stimuli see the Appendix.

Apparatus and procedure. The apparatus and procedure were thesame as in Experiments 1 and 2.

50 -

3 46h

Ib

35

s

,.-*. • • •>Sfct

\

- . ^ \

^ ^ \

t ' \

\

A-

\V\\

BL

Exper.Contr.

41.3943.21

44.1746

44.7344.17

42.3945.47

46.2639.96

44.2141.21

45.S243.39

43.7836.13

47.636.52

45.7836.17

•*" Exper. + Contr.

Figure 3. Stress pattern discrimination in two sets of disyllabic conso-nant-varied words. Values represent mean high-amplitude (ampl.) suck-ing rates of experimental (exper.) and control (contr.) groups duringbaseline (BL) and 9 min of test.

Results

The same statistical analyses as in Experiments 1 and 2 wereperformed. Mean HAS rates for baseline and each minute oftest are reported in Figure 3 for both groups.

Linear regression. The linear regression analysis was sig-nificant in the control group in the 9-min test, F ( l , 205) =9.78, p = .002, R2 = .045, /? = - . 2 1 ; the HAS rate of thecontrol group significantly decreased during the test, whereasthat of the experimental group did not vary significantly fromthe beginning to the end of the test ( F < 1).

Analysis of variance. The HAS rate of the two groups didnot differ significantly at baseline or in the first or second 3-min periods (F < 1), but it differed significantly in the third3-min period, F ( l , 44) = 8.45, p = .005. The interaction be-tween group and time was significant for a comparison of thefirst and the third 3-min periods, F ( l , 44) = 8.89, p = .004,and approached a significant p value for a comparison of thesecond and third 3-min periods, F( 1,44) = 3.32, p = .074. AnANOVA was also performed within each group to check whetherthe order of presentation in the experimental group and the typeof stimulus in the control group affected newborns' responses.No significant effects were found (F < 1).

Discussion

The results showed that the HAS rate of the experimentalgroup remained fairly constant over the course of the test,whereas that of the control group significantly decreased. TheHAS rates of the two groups diverged significantly in the third3-min period of the test. Newborns in the experimental group,therefore, discriminated stress patterns of two sets of disyllabicwords varied in all of the consonants, whereas newborns in thecontrol group habituated to a set of disyllabic words varying inconsonants but sharing a common stress pattern.

This third experiment suggests that newborns can extract and

NEONATAL DISCRIMINATION OF STRESSED WORDS

discriminate a common rhythmical property (i.e., the stress pat-tern), regardless of consonant variations and perceive words incategories on the basis of their stress pattern.

As regards the methodology, the alternation procedure workedas well with two sets of stimuli as with two single stimuli. As inExperiments 1 and 2, more than 60% of the examined newbornscompleted the test, and the difference between the groups issignificant in the third 3-min period of the test.

The HAS rate of the experimental group remained high butdid not increase significantly during the test, as observed inExperiment 1. This result seems to confirm that newborns donot increase their HAS rate easily while they are listening todisyllabic words. Thus, it is likely that the increasing of theHAS rate observed in the experimental group of Experiment 2was mostly due to the rhythmical complexity of trisyllabicwords and not to their consonant variation because no similarincrease was observed in this third experiment in which wordswere consonant-varied. As shown by Cairns and Butterfield(1975), newborns seem to discriminate syllables as well assinging voices or musical phrases. However, they increase non-nutritive sucking only with the two latter stimuli, which arerhythmically more complex than the former (Cairns & But-terfield, 1975).

General Discussion

Stress is a prosodic characteristic that seems to attract new-borns' attention. Newborns discriminate the stress patterns oftwo phonetically unvaried disyllabic words, of two consonant-varied trisyllabic words, and of two sets of consonant-varieddisyllabic words. These results suggest that newborns are able toperceive the rhythm of disyllabic and trisyllabic stressed words,regardless of consonant variations, and to categorize words onthe basis of their stress pattern.

The chosen procedure—with its connection between HASand delivery of stimulation, minute-by-minute alternation of twostimuli or two sets of stimuli, fixed duration of 9 min, relativelyundemanding amplitude threshold, and absence of any arbitrarycriterion of habituation—is suitable for examining discrimina-tion and categorization in newborns. Besides, compared withthe classic habituation procedure, it allows a higher percentageof the examined participants (at least 60%) to reach the end ofthe test and to hear both stimuli in successive minutes. Discrimi-nation is inferred by the difference between the HAS rates ofthe experimental and control groups. In fact, the experimentalgroup's HAS rate, which does not differ from that of the controlgroup at the beginning, gradually diverges and becomes signifi-cantly higher than the control group's HAS rate. In particular,the difference between the two groups becomes significant inthe third 3-min period of the test.

It may be observed that newborns, unlike 2-month-old infants(Cowan et al., 1982), do not easily increase their HAS rates.However, in Experiment 2, an increase of the experimentalgroup's HAS rate was observed. This increase might be due tothe rhythmical complexity of trisyllabic words, greater than thatof disyllabic words, rather than to consonant variations because,in Experiment 3, although consonants were varied, no significantincrease in the experimental group's HAS rate was observed.

Further experiments might explore how nonnutritive sucking,being a rhythmical behavior, can be modified by listening torhythmically complex stimuli.

In conclusion, stress seems to be a salient prosodic cue inspeech perception, suggesting that newborns might use it tosegment speech in rhythmical units, such as stressed words,whose rhythm depends on the position of stress and number ofsyllables. Discrimination and categorization of stress patternsand, later on, recognition of the more frequent stress pattern ofthe maternal language, might constitute a basis for representingthe lexical structure of speech.

Newborns' processing of stress patterns does not seem to beaffected by consonant variations. However, this does not implythat segmental information is not perceived or represented bynewborns. In particular, the perception of stress patterns mightbe linked to that of vowels. Vowels carry prosodic information,such as stress, and have clearer acoustical cues, such as longerduration and higher fundamental frequency and intensity, thanconsonants. Moreover, vowels constitute the necessary and suf-ficient nucleus of the syllable. Some experiments indicate thatnewborns, when presented with a list of monosyllables, detecta change of vowel but not of consonant alone (Bertoncini etal., 1988). Thus, vowel varations seem to be perceptually moresalient than consonant variations for newborns. Because stressis carried by vowels, we suggest that vowel variations mightaffect the processing of stress patterns (Sansavini, 1994). Therelationship between the processing of stress patterns and vow-els is thus to be investigated in further research.

Moreover, it might be worthwhile to examine how the initialpredisposition to discriminate stress patterns is reorganized afterthe first months of life, as a function of the role played by stresspatterns in the maternal language (e.g., fixed stress in somelanguages, as in French, which has just one stress pattern; vary-ing stress in other languages, as in Italian and English, whichhave different stress patterns). More specifically, because 2-month-old infants show finer abilities than newborns in syllabicrepresentations (Bertoncini et al., 1988) and can memorize theserepresentations (Jusczyk, Pisoni, & Mullenix, 1992), it wouldbe worth verifying whether they also start to be biased by theway in which stress is used in the maternal language and whetherthis eventual bias affects the capacity of discriminating stresspatterns, of segmenting speech into words, and of discriminatingphonetic contrasts.

References

Bertinetto, P. M. (1980). The perception of stress by Italian speakers.Journal of Phonetics, 8, 385-395.

Bertinetto, P.M. (1981). Strutture prosodiche dell'Italiano. Firenze,Italy: Accademia deJla Crusca.

Bertoncini, J., Bijeljac-Babic, R., Blumstein, S., & Mehler, J. (1987).Discrimination in newborns of very short CV's. Journal of the Acous-tical Society of America, 82, 31-37.

Bertoncini, J., Bijeljac-Babic, R., Jusczyk, P. W., Kennedy, L., & Mehler,J. (1988). An investigation of young infants' perceptual representa-tions of speech sounds. Journal of Experimental Psychology: General,117, 21-33.

Bijeljac-Babic, R., Bertoncini, J., & Mehler, J. (1993). How do four-day-old infants categorize multisyllabic utterances? DevelopmentalPsychology, 29, 711-721.

10 SANSAVINI, BEKTONCINI, AND GIOVANELLI

BlasdeLl, R., & Jensen, P. (1970). Stress and word position as determi-nants of limitation in first language learners. Journal of Speech andHearing Research, 13, 193-202.

Bull, D., Eilers, R., & Oiler, D. (1984). Infants' discrimination of inten-sity variation in multisyllabic stimuli. Journal of the Acoustical Soci-ety of America, 76, 13-17.

Cairns, G. R, & Butterfield, E. C. (1975). Assessing infants'auditoryfunctioning. In B. Z. Fiedlander, G. M. Sterrit, & G. Kirk (Eds.), Theexceptional infant: Assessment and intervention (Vol. 3). New %rk:Brunner/Mazel.

Christophe, A., Dupoux, E., Bertoncini, J., & Mehler, J. (1994). Doinfants perceive word boundaries? An empirical study of the boot-strapping of lexical acquisition. Journal of the Acoustical Society ofAmerica, 95, 1570-1580.

Colombo, L. (1992). Lexical stress effect and its interaction with fre-quency in word pronunciation. Journal of Experimental Psychology:Human Perception and Performance, 18, 987-1003.

Cooper, R. P., & Aslin, R. N. (1990). Preference for infant directedspeech in the first month after birth. Child Development, 61, 1584-1595.

Cowan, N., Suomi, K., & Morse, P. A. (1982). Echoic storage in infantperception. Child Development, 53, 984-990.

Cutler, A., & Butterfield, S. (1992). Rhythmic cues to speech segmenta-tion: Evidence from juncture misperception. Journal of Memory andLanguage, 31, 218-236.

DeCasper, A. J., & Fifer, W. P. (1980). Of human bonding: Newbornsprefer their mother's voices. Science, 208, 1174-1176.

Echols, C. H. (1993). A perceptually-based model of children's earliestproductions. Cognition, 46, 245-296.

Echols, C. H., & Newport, E. L. (1992). The role of stress and positionin determining first words. Language Acquisition, 2, 189-220.

Eilers, R. E., Bull, D. H., Oiler, D. K., & Lewis, D. C. (1984). Thediscrimination of vowel duration by infants. Journal of the AcousticalSociety of America, 75, 1213-1218.

Fernald, A. (1985). Four-month-old infants prefer to listen to motherese.Infant Behavior and Development, 8, 181-195.

Floccia, G, Christophe, A., & Bertoncini, J. (in press). HAS and new-borns: Further understanding for underlying mechanisms. Journal ofExperimental Child Psychology.

Gerken, L. A., Landau, B., & Remez, R. E. (1990). Function morphemesin young children's speech perception and production. DevelopmentalPsychology, 26, 204-216.

Giovanelli, G. (1991). Ritmi biologici, funzione del linguaggio, e iden-tita. Rivista di Psicologia, 3, 55-64.

Giovanelli, G., Callegati, I., Fameti, A., Tuozzi, G., Zucchini, E., &Colucci, C. (1990, August). Differential cardiac response to unfamil-iar/maternal voice in the newborn. Paper presented at the IVth Euro-pean Conference on Developmental Psychology, Stirling, Scotland.

Gleitman, L. R., & Wanner, E. (1982). Language acquisition: The stateof the state of the art. In E. Wanner & L. R. Gleitman (Eds.), Lan-guage acquisition: The state of the art (pp. 3-48) . New \brk: Cam-bridge University Press.

Goodsitt, J. V., Morse, P. A., Ver Hoeve, J.N., & Cowan, N. (1984).Infant speech recognition in multisyllabic contexts. Child Develop-ment, 55, 903-910.

Halle, P., Boysson-Bardies, de B., & Vihman, M. (1991), Beginnings ofprosodic organization: Intonation and duration patterns of disyllablesproduced by Japanese and French infants. Language and Speech, 34,299-318.

Hirsh-Pasek, K., Kemler Nelson, D. G., Jusczyk, P. W., Wright Cassidy,K., Druss, B., & Kennedy, L. (1987). Clauses are perceptual unitsfor young infants. Cognition, 26, 269-286.

Jusczyk, P. W. (1995). Language acquisition: Speech sounds and the

beginnings of phonology. In J. L. Miller & P. D. Eimas(Eds.),S(peec/i,language and communication (pp. 263-301). New "York: AcademicPress.

Jusczyk, P. W., Bertoncini, J., Bijeljac-Babic, R., Kennedy, L. J., &Mehler, J. (1990). The role of attention in speech perception by younginfants. Cognitive Development, 5, 265-286.

Jusczyk, P. W., Cutler, A., & Redanz, N. (1993). Preference for thepredominant stress patterns of English words. Child Development,64, 675-687.

Jusczyk, P. W., Friederici, A. D., Wessels, J., Svenkerud, V. Y., & Jusc-zyk, A. M. (1993). Infants' sensitivity to the sound patterns of nativelanguage words. Journal of Memory and Language, 32, 402-420.

Jusczyk, P. W., Hirsh-Pasek, K., Kemler Nelson, D. G., Kennedy, L. J.,Woodward, A., & Piwoz, J. (1992). Perception of acoustic correlatesto major phrasal units by young infants. Cognitive Psychology, 24,252-293.

Jusczyk, P. W., Pisoni, D. B., &. Mullenix, J. (1992). Some consequencesof stimulus variability on speech processing by two-month-old infants.Cognition, 43, 253-291.

Jusczyk, P. W, & Thompson, E. (1978). Perception of a phonetic con-trast in multisyllabic utterances by 2-month-old-infants. Perceptionand Psychophysics, 23, 105-109.

Karzon, R. G. (1985). Discrimination of polysyllabic sequences by one-to-four month-old-infants. Journal of Experimental Child Psychology,39, 326-342.

Landau, B., & Gleitman, R. L. (1984). The language of perception inblind children. Cambridge, MA: Harvard University JVess.

Lehiste, I. (1970). Suprasegmentals. Cambridge, MA: MIT Press.Levitt, A., & Wang, Q. (1991). Evidence for language-specific rhythmic

influences in the reduplicative babbling of French- and English-learn-ing infants. Language and Speech, 34, 235-249.

Mehler, J., Bertoncini, J., Barriere, M., & Jassik-Gerschenfeld, D.(1978). Infant recognition of mother's voice. Perception, 7, 491 —497.

Mehler, J., Jusczyk, P. W., Lambertz, G., Halsted, N., Bertoncini, J., &Amiel-Tison, C. (1988). A precursor of language acquisition in younginfants. Cognition, 29, 143-178.

Moon, C , Bever, T. G., & Fifer, W. P. (1992). Canonical and non-canoni-cal syllable discrimination by two-day-old infants. Journal of ChildLanguage, 19, 1-17.

Moon, C , Cooper, R. P., & Fifer, W. P. (1993). Two-day-olds prefertheir native language. Infant Behavior and Development, 16, 495-500.

Morgan, J. L. (1986). From simple input to complex grammar. Cam-bridge, MA: MIT Press.

Morgan, J. L., Meier, R. P., & Newport, E. L. (1987). Structural packag-ing in the input to language learning: Contributions of prosodic andmorphological marking of phrases to the acquisition of language. Cog-nitive Psychology, 19, 498-550.

Nespor, M., & Vogel, I. (1986). Prosodic phonology. Dordrecht, TheNetherlands: Foris.

Pegg, J. E., Werker, J. F., & McLeod, P. J. (1992). Preference for infant-directed over adult-directed speech: Evidence from 7-week-old in-fants. Infant Behavior and Development, 15, 325-345.

Peters, A. M. (1983). The units of language acquisition. Cambridge,England: Cambridge University Press.

Sansavini, A. (1994). Percezione della prosodia del linguaggio neiprimi giorni di vita [Perception of the prosody of speech in the firstdays of life]. Unpublished doctoral dissertation, Bologna University,Bologna, Italy.

Sansavini, A., Bertoncini, J., & Giovanelli, G. (1994). Newborns dis-criminate stress patterns in phonetically complex Italian words. InfantBehavior and Development, 17, 924.

NEONATAL DISCRIMINATION OF STRESSED WORDS 11

Siqueland, E. R., & De Lucia, C. A. (1969). Visual reinforcement ofnonnutritive sucking in human infants. Science, 165, 1144-1146.

Spring, D. R., & Dale, P. S. (1977). Discrimination of linguistic stressin early infancy. Journal of Speech and Hearing Research, 20, 224-232.

Trehub, S. E. (1976). Infants discrimination of two syllable stimuli: Therole of temporal factors. Paper presented at the annual meeting of theAmerican Speech and Hearing Association, Houston, TX.

Irehub, S. E., & Chang, H. W. (1977). Speech as reinforcing stimulationfor infants. Developmental Psychology, 13, 170-171.

Weiss, M. J., & Zelazo, P. R. (1991). A taxonomy of newborn attention.

In M. J. Weiss & P. R. Zelazo (Eds.), Newborn attention: Biologicalconstraints and the influence of experience (pp. 466-511). Norwood,NJ: Ablex.

Werker, J. F , & McLeod, P. J. (1989). Infant preference for both maleand female infant-directed talk: A developmental study of attentionaland affective responsiveness. Canadian Journal of Psychology, 43,230-246.

Williams, J. A., & Golenski, J. (1978). Infant speech sound discrimina-tion: The effects of contingent versus non contingent stimulus presen-tation. Child Development, 49, 213-217.

Appendix

Stimuli Used in Experiments 1, 2, 3

'mamama'ma

'tacalata'cala

'dag a'nata'mara'baga'lama'taca'rama'gaba

Meanda'gana'tama'raba'gala'mata'cara'maga'ba

Mean

1st syll.

356.0250.0

310.0181.0

432.0413.0360.0390.0391.0354.0423.0422.0

398.1238.0230.0217.0230.0246.0201.0240.0313.0

239.4

Duration

2nd syll.

313.0376.0

180.0313.0

330.0348.0358.0387.0363.0350.0349.0398.0

360.4358.0385.0368.0403.0397.0365.0404.0407.0

385.9

(ms)

3rd syll. Total

Experiment 1

— 669.0— 626.0

Experiment 2

296.0 786.0322.0 816.0

Experiment 3

— 762.0— 761.0— 718.0— 777.0— 754.0— 704.0— 772.0— 820.0

— 758.5— 596.0— 615.0— 585.0— 633.0— 643.0— 566.0— 644.0— 720.0

— 625.2

Fundamental frequency (H2)

1st syll.

236.2231.7

251.2310.8

240.2247.5247.3247.2250.3246.2257.8257.7

249.3207.6254.6137.0191.2213.1239.8244.6259.5

218.4

2nd syll.

241.2249.5

250.1321.8

257.7246.5252.8260.0253.8243.5257.9253.5

253.2227.5235.3232.6241.5235.1241.7241.7253.2

238.6

3rd syll.

—

254.0325.8

————————

————————

_

Note. Stress mark precedes stress syllable. Syll. = syllable.

Received November 9, 1994Revision received January 11, 1996

Accepted February 11, 1996