Kathleen Currie Hall 9 May 2005 Notes on: Kuhl, Patricia K., & Meltzoff, Andrew N. (1996). Infant vocalizations in response to speech: Vocal imitation and developmental change. Journal of the Acoustical Society of America, 100(4), 2425-2438. Summary: In this paper, Kuhl and Meltzoff present the results from a tightly controlled infant vowel production study. They show that infant vowel productions vary along the same dimensions that adult productions do; that even between 12 and 20 weeks, infants tighten up their vowel spaces; and that infants show imitation of ambient input at these ages. Details: 1. While it is generally accepted that there are basically 5 universal stages of child language development (reflexive phonation, cooing, expansion, canonical babbling, and meaningful speech), the progression from one to the next has not been very well studied and is not fully understood. This paper tries to explore at least some of these issues. The two main hypotheses are: anatomical change and vocal learning. 2. Infants of three ages, 12-, 16-, and 20-week-olds, participated in the experiment. Each child heard and saw 5 minutes of a woman's face producing either [a], [i], or [u]. The child was recorded and transcribed, and acoustic measurements were taken. 3. Unlike other imitation studies, the stimuli were tightly controlled -- the same for all infants, presented at the same level and for the same duration, etc. 4. The three questions: a. How do infant productions compare to adult? -- infants produce vowels that do not use the full range of formant values that adults' do, but which are still concomitant with the adult vowel range. They differentiate vowels along the same dimensions (F1, F2, etc.) that adults do. b. Are there developmental changes between 12 and 20 weeks? -- yes, there are; these seem to be along the lines of tightening up the vowel space encompassed by each vowel. Kuhl and Meltzoff propose that this is due to the formation of memory representations of clusters of vowels encountered in the ambient language, which in turn helps shape the motor gestures and the output produced by the child. c. Do infants imitate? -- Yes, there was a significant amount of infant vowel productions that matched adult stimuli, which was above that expected if the infants were just randomly cooing. This imitation presumably helps with the memory representations needed to produce speech closer and closer to that of the ambient language. The question of normalization (how the child knows that his [a] is [a]-like when the frequencies are in fact different from those of the adult, etc.) is still unresolved. Questions: 1. The infants each came in for 3 sessions. Did they hear the same vowel each time? Why were there 3 sessions? Was there any effect of session? E.g., did children imitate more during the third session? (Especially c.f. comments on p. 2435 about 15 minutes of exposure being "sufficient to influence vocalizations" -- how do we know that it was the 15 minutes that was sufficient? Some infants did not imitate the stimuli. Maybe those who did were predisposed to anyway, regardless of the 15 minutes . . . . 2. What was done with the 14% inter-transcriber disagreement? Did they come to a consensus or throw out the data? (The latter seemed to be the case in one of the analyses, but not all?) 3. Why do they say that grave/acute and compact/diffuse values are "not frequency specific"? I would think that without some sort of standardization, it still wouldn't be fair to simply compare infant and child productions directly (especially with the grave/acute measure). From: LAURA SLOCUM 1) "Homo sapiens is the only mammal that displays vocal learning, the tendency to acquire the species-typical vocal repertoire by hearing the vocalizations of adults and mimicking them." (p. 2426)...I question the accuracy of this statement on a couple of grounds: A) When does "listening" begin? We've talked before about perception in utero. Why do researchers consistently fail to take this into consideration? B) I hardly think that we have studies enough on non human mammalian species to be able to make this claim as fact. Just because we don't know that they are doing it certainly does not mean that they aren't doing it! (There is some interest ing research going on at OSU's primate center looking at vocalization and semanticity in chimpanzees.) 2) 86% transcriptional agreement just goes to show how difficult (and subjective) a task transcription is! 3) I was glad to see that they took potential subject preferences ("/a/" "/i/" and "/u/" infants) into consideration. But I don't understand this analysis. Can you please clarify? From: Junko Davis - What is "raspberries" in "... a wide variety of new sounds such as yells, screams, whispers, and raspberries"? (p.2425) - I wonder whether there were any instances in which the produced vowels sounded somewhere in between the two vowels, or /e/-like or /o/-like. If there were su ch cases, I wonder how the transcriber made the decisions. - I don't know what "the compact-diffuse (CD) feature" and "the grave-acute (GA) feature" are. - In Figure 5 (p.2433), it is true that the /i/ utterances occurred the most in response to the stimulus /i/. However, if we look at how the infants reacted to the stimulus /i/, the /a/ utterances occurred the most and not the /i/ utterances. - It seems that the infants have the predominant vocalization pattern. What could be the source of this variability? From: Fangfang Li This paper examines infants' vocalization in response to vowel stimuli, namely a, i and u. both perceptual analysis and instrumental analysis suggest that infants produce corresponding vowel-like vocalization with respect to the different vowels presented. this is interpreted as infants' ability to imitate adults' productions of vowel as early as 12 weeks, and then more precise productions as the developmental pattern for 16 and 20-week's babies. questions: 1. can we go over figure 7 in class? i didn't get it, especially about the cell numbers. 2. what kind of "target" that babies have already stored in their mind as early as 12 weeks? apparently from the discussion of the paper that this "target" is a normalized one, invariant to the difference between adults and children, male and female. if this is the case, does that suggest that infants learn to normalize well before 12 weeks of life? that sounds incredible. or this is something that is born with? 3. i don't think 15-minutes in all can give infants enough time to do the auditory-articulatory mapping for a novel sound if they did not try it before. it is very possible that for those infants who successfully coo back the right vowel, they should already have the "target" and the mapping in their mind, and for those who failed to coo back, part of the reason is, again, possiblly that they don't have the target and the mapping stored in their mind or didn't get enough practice before. From: HELENA RIHA p. 2425 ?(In addition to anatomical change), a second factor responsible for change in infants? vocalizations ? is vocal learning.? ?Infants acquire the specific inventory (of) a particular language in part through imitation.? Kuhl and Meltzoff?s experiments indicate that infants derive representations of vowels from the speech sounds they hear and that these representations serve as targets for their production. One question not addressed in the paper is infants? motivation for this process. What drives infants to react to speakers of the ambient language in this way? What compels them to make sense of speech in the way they do? Is the formation of mental representations and imitation a part of our innate capacity for language? p. 2426 ?One question is whether vocal learning begins only after the first or second year ? or whether there is learning taking place earlier.? This statement reminded me of our discussion of covert contrasts. It is possible that infants? earliest vocal learning is not perceived by researchers just as covert contrasts are not audible to listeners and have to be identified through acoustic analysis. p. 2428 ?The perceptual analyses (of the 224 utterances) were done by having a (trained) individual listen to each of the utterances and code it as one of eight vowels (i, I, E, @, a, ^, U, u).? Isn?t this kind of like a forced-choice experiment? What if the infants were not producing one of the eight vowels? Categorizing infants? productions as one eight vowels also ignores the question of how speech sounds are differentiated and represented in the infant?s mind. p. 2431 ?Perhaps infants are listening to the vowels produced by adult speakers? and are striving to produce vowels themselves that perceptually resemble those they hear adults produce.? This is interesting to consider from the point of view of general learning strategies. As noted in Edwards (2000), future research should consider the effects of general learning principles in phonological acquisition. To what degree is imitation an important principle in learning generally? How does vocal imitation resemble imitation in other kinds of learning? p. 2433 ?The vowel stimulus strongly affected the type of vowel infants produced in response.? In response to my first question above, maybe infants engage in vocal imitation unconsciously at first, and as they mature, they do so consciously, as well. Perhaps at the canonical babbling stage (5-10 months), their vocal imitations become more conscious in an effort to approximate adult targets more accurately. From: Kirk Baker Kuhl et al. Developmental study: 12-20 wks infants the vowel space becomes more separated; also evidence for vocal imitation. -why is /i/ so less frequent for the younger groups? -why 20 wks fewer vocalizations? -why doesn't imitation % increase with age? From: ASIMINA SYRIKA Questions on the Kuhl and Meltzoff article: -How much input is necessary for an infant to form memory representations of speech that will later on guide his/her own sound productions? -What counts as the ambient language when the child is exposed to more than one language or dialect from birth? How does bilingualism affect vocal learning? -On what basis were the three vowels /-a/, /-i/ and /-u/ chosen for the experiment? How did they account for dialectal variations? How was the dialect of the recorded speaker similar to the one the infants had already been exposed to? -What counts as a good vowel (vowel 'prototype') given the large variability in vowel production? -We are not usually exposed to isolated vowels in everyday speech. If the vowels were played in 'familiar' words or CV sequences would that also result in vocal imitation? What was the rationale behind using vowels in isolation? -How do you know that the infants were actually imitating the sound rather than looking at the visual stimuli (face) or practising vowels they prefer or are more familiar with? In Greek, for example, infants are probably more exposed to /a-like/ vowels in isolation (since they appear in exclamations)than they are to /u-like/ vowels. How do you know that frequency of exposure doesn't affect infants' vocal imitation? -Where was the boundary set for the transcription of infants' vocalisations as /a-like/, /i-like/ and /u-like/? What about categories falling outside and in-between these categories? What if the infant started with an a-like vowel and later on turned it into an e or i-like vowel? How would this production be transcribed? -If an infant is exposed to both accented and unaccented speech (native-like) what kind of representations would he/she form? At what age can infants tell that a sound is 'approximating' the one in their native language? It seems that the ambient language of each child can vary if you consider dialectal variations too. How did they know that the infants used in the experiment were exposed to the same ambient language and had similar amount of input? -Do infants perceive their own productions as ambient-like (close to 'target') or not? In the case of twins that hear themselves and each other almost all the time would you predict that they will differentiate baby language from adult language at a young age? Why would the adult input necessarily serve as their model for forming memory representations? From: nagar@ling.ohio-state.edu The paper talks about the role that imitation plays in the development of speech in 12, 16 and 20 month olds. The authors discuss three different vowels /a/, /i/, /u/ and talk about the way imitation affects the way speech is learned. The authors examine the speech of children at ages 12, 16 and 20 months. They give a stimulus to the children which is both visual and audio and the vocalizations are analyzed perceptually and instrumentally. Their results show 1) that the vowel space that infants use is spread more than the vowel space that adults use, 2) utterance produced by children cluster in certain acoustic spaces, 3) vowel category and age of infant affect vocalization 4)stimulus given to the children has an affect on their vocalizations 5) infants can recognize the constancy in vowels in spite of the variation in vowel and their frequency ranges among adults and in their own productions 6) exposure to speech is an important factor which influences vocalization. From: ejkong@ling.ohio-state.edu - The study claims that infants' production changes over early weeks of their lives not only because of the vocal tract development but also because of thier imitation of adults' speech. - The infants' imitating speech input indicates that they've been preparing for the speech even before actually starting producing meaningful words by experiencing and recognizing the acoustic characteristics of the ambient language and their articulatory(motor) connection. - Infants' production is language-specific (the production magnet effect) because it is led by the linguistic representations which are established by the perceptual experience and the auditory-articulatory connection of the ambient language.