This paper shows that infants who are just beginning to learn words do not use the phonetic perceptual abilities that we know they must have. The authors show that even though 14-month-old infants are able to distinguish two minimally contrasting words, they "ignore" this level of phonetic detail when associating the words with objects. This result is perhaps surprising only to those who are not aware of the history of the field.
For years it has been known that children, from the point of learning their first words up to ages as high as 10 years, have trouble distinguishing words that are phonetically very similar. This is first obvious in their speech production (see, for example, Ferguson and Farwell 1975), but also extends to their speech perception. That is, a number of studies (see Barton 1978 for a review) reported that children could not reliably discriminate two words that differ in only a single segment (a minimal pair). The conditions reported in these experiments varied considerably -- the subjects' ages, the use of real or nonsense words, the familiarity of the subjects with the real words, and the amount of practice the subjects were given with the words, among other factors -- but the error rates reported were as high as 30% (Barton 1978). The general trend that children perform worst at this task when they begin learning words, and gradually improve with age, naturally led to what Ingram (1989, p. 84) calls the "perceptual learning theory". According to this theory, children cannot distinguish these words because they cannot perceive the distinction between the phones that contrast them. Furthermore, according to the theory, infants are born not being able to perceive any phonetic contrasts at all, and gradually acquire the ability to distinguish them throughout childhood. This theory seemed to account for all the linguistic data that was and could easily become available, as it was not obvious how to gather linguistic data on infants who could not speak or understand any words.
However, the perceptual learning theory was instantly disconfirmed in 1971 by the work of Eimas et al. Using the high-amplitude sucking paradigm, these workers showed that infants between 1 and 4 months of age, raised in English-speaking households, can readily discriminate between minimally contrasting syllables in English (e.g. /pa/ vs. /ba/). The paper sparked a flurry of research on the perceptual abilities of infants. The ability was soon shown to extend all the way back to newborn infants, and to a wide range of English phonemic contrasts. More interestingly, however, they could also discriminate a wide range of contrasts found in languages other than English. Since then, it has been shown that all infants can discriminate many, perhaps all, of the contrasts used phonemically in all the world's languages.
This presented what seemed to be a strange state of affairs. Newborn infants are able to perceive non-native contrasts, contrasts that not even the average adult can perceive without training. Children of intermediate age, meanwhile, not only seem unable to perceive these contrasts at all, they seem to have difficulty with even native contrasts!
The next question, then, was to discover at what age children lose their discriminatory abilities, and the task of answering it was taken up by many researchers, but most notably by researchers in Werker's laboratory. In retrospect, it is interesting to note that they decided to concentrate on the non-native contrasts at this point, even though it is the loss of native contrasts that is most puzzling.
Noting that children can learn a second language natively up to the onset of puberty, Werker's laboratory hypothesized that the ability to discriminate non-native contrasts might be lost at that age. However, they found that children 12 years of age could not discriminate the contrasts. So they tested progressively younger and younger children (Werker and Tees 1983, 1984a; Kuhl et al. 1992) until it was established that the loss of discrimination occurs between 6 and 12 months of age. Thus, infants' ability to perceive phonetic distinctions is tuned very rapidly after birth to detect distinctions found only in the environmental language.
Importantly, however, this loss of discriminatory ability cannot be a "loss" in the sense of complete nonexistence and unrecoverability. Several lines of evidence point to this conclusion. First, there is the fact, previously noted, that children much older than infants can be placed in a new linguistic environment and learn another language without an accent (with varying degrees of success). Second, the Werker and Tees experiments showed that after age 4, children very gradually regain their non-native discriminatory ability, resulting in varying levels of proficiency in adulthood. Third, adults can be trained to relearn non-native distinctions. Finally, and most interestingly, Werker and Tees (1984b) showed that if the non-native phones were presented in a nonlinguistic context (by stripping off the rest of the syllable), adult subjects could discriminate the sounds easily.
If the ability to discriminate non-native contrasts is not "lost", then, where has it gone? The evidence suggests that the ability is simply "ignored" when auditory input is processed linguistically; this is what Stager and Werker mean by "functional reorganization". One model of this is as follows. All acoustic input is processed by a general auditory processor of the brain; this processor is able to discriminate, among other nonlinguistic sounds, any pair of phones used as a phonemic minimal pair in any of the world's languages. Evidence from other animals -- parrots obviously, but also chinchillas and others -- suggests that this discriminatory ability may be common to higher animals. In humans older than 6 months, however, the output of this processor that is perceived to be linguistic in nature is obligatorily routed into a second processor, the phonological processor. This processor lumps together noncontrastive phones into single phonemes, and its output is what is "perceived". It is the phonological processor that is being developed during the critical period between 6 and 12 months, and the ability of a listener to prevent the output of the auditory processor from entering the phonological processor -- the ability to perceive sounds nonlinguistically -- will later be developed over time, aided occasionally by practice and the properties of the input.
With all of this as background, Stager and Werker have now turned their attention to the problem of the loss of perception of native contrasts that formed the original puzzle. As with the non-native contrasts, we can ask two questions: (1) At what age does the loss of perception occur? (2) Is the "loss" complete, or is there something else going on? Stager and Werker's work is important because it goes a long way toward answering both questions.
The range of possible answers to the first question were narrow even before the experiment was done: the child apparently "loses" her ability to discriminate native contrasts even when she speaks her first words, around the age of a year and a half. Stager and Werker's experiment pushes back the time of "loss" even further, to a period between 8 and 14 months.
But, in answer to the second question, the "loss" is not really a loss, as the experiment also shows. While the 14-month-old infants were not able to distinguish between 'bih' and 'dih' in the context of a word-object association task (experiments 1 and 2), they were able to distinguish them outside of that context (experiment 4). This shows that the ability to distinguish the phones is still there.
The key word here is context. The infants fail to discriminate the native phones in the word-learning context, but are able to outside the context. This is completely analogous to the evidence that adults fail to discriminate non-native phones in a linguistic context, but are able to outside that context. This parallel suggests a parallel mechanism; this is what Stager and Werker mean when they refer to the second "functional reorganization". In terms of the model outlined above, perhaps what is happening at this stage is that the output of the phonological processor is now being fed into a third processor, which filters the input even more. Given that this occurs in a word-learning context, we might suppose that it is the lexical processor, which has the task of assigning to the perceived word a meaning, a minimal phonological representation, and a place in memory. If the auditory input is perceived as linguistic but not lexical in nature, the output of the phonological processor will not be processed by the lexical processor, and the child will be able to discriminate the native phones easily.
What I have outlined above is an account of when and how these reorganizations of perceptual abilities take place. A perhaps more fundamental question is why they occur. The answer seems clear with the first reorganization: the discrimination of non-native phonemic contrasts is probably of little use to the child (unless he is moved into a new linguistic environment, in which case his discriminatory ability is still available to him). In fact, such discrimination interferes with phonological processing, since it focuses on irrelevant detail, and thus should be removed quickly.
But what the second reorganization? Why should the child put in all the effort to learn which contrasts are native and which are non-native during his first year of life, only to mostly ignore this information in the second year? Here again (in this paper and in Werker 1995), Werker seems to be on the right track with her suggestion, which again parallels the first reorganization. She suggests that the level of phonetic detail would still be too much to handle for the task of associating words with objects. The child thus stores the word in lexical memory with just enough phonological detail to discriminate it from all the other words in his small vocabulary. The phonetic detail is filled in only when it is needed to distinguish words that would otherwise have identical phonological structure.
If true, this suggestion puts the contrast between Experiments 1 and 2 in a new perspective. In the first experiment, the infant is given two word-object pairs to memorize; in the second, there is only one pair. Stager and Werker imply that the second experiment should be easier for the infants, presumably because there is only one word to learn. However, if it is true that the phonetic detail is added only when needed, one might expect the first experiment to be easier for the young word learner. Only in the first experiment is she given two words which minimally contrast to learn; only then would the phonetic detail be forced to be present in the lexical entries. Thus, only in the first experiment would the lexical entries be rich enough to allow discrimination in the test phase.
In summary, this is good work, and it fills a important gap in the research of the field. Given the nice parallel between the loss of native discrimination in word-learning studied here, and the loss of non-native discrimination in linguistic perception studied earlier, the work is unlikely to be highly controversial to those knowledgeable of the field, although there are certain to be differences of opinion. Rather, it is more likely to have confirmed many researchers' suspicions.
In follow-up work, Werker and her colleagues should confirm that the ability shown in 14-month-olds to discriminate phonemes outside the word-learning task is also present in older children, as has been suggested. Ironically, this is probably easier to test with infants than with older children. With infants, it is simple to place the syllables in a non-word-learning environment, by presenting them as disembodied sound, unassociated with any object. With older children, disassociated syllables may still be perceived as words and thus still processed by the lexical processor. This is why the children in Barton's work, and in all the studies cited by Barton, were having trouble discriminating minimal pairs. In all these experiments, the minimal pairs were presented as word-learning tasks. It does not seem to have occurred to these researchers to try to present them as sounds alone!
Barton, David. (1978) The role of perception in the acquisition of phonology. Ph.D. thesis, Stanford University. Distributed by the Indiana University Linguistics Club.
Eimas, P.D., E. R. Siqueland, P. Jusczyk, and J. Vigorito. (1971) Speech perception in infants. Science 171:303-6.
Ferguson, Charles A. and Carol B. Farwell. (1975) Words and sounds in early language acquisition. Language 51(2):419-39.
Ingram, David. (1989) First Language Acquisition. Cambridge: Cambridge University Press.
Kuhl, Patricia K., Karen A. Williams, Francisco Lacerda, Kenneth N. Stevens, and Bjorn Lindblom. (1992) Linguistic experience alters phonetic perception in infants by 6 months of age. Science 255:606-8.
Werker, Janet. (1995) Exploring developmental changes in cross-language speech perception. In L. R. Gleitman and M. Liberman (Eds.), Invitation to cognitive science: Language, pp. 87-106. Cambridge, MA: MIT Press.
Werker, J. F., and R. C. Tees. (1983) Developmental change across childhood in the perception of non-native speech sounds. Canadian Journal of Psychology 37:278-86.
Werker, J. F., and R. C. Tees. (1984a). Cross-language speech perception: evidence for perceptual reorganization during the first year of life. Infant Behavior and Development 7:49-63.
Werker, J. F., and R. C. Tees. (1984b). Phonemic and phonetic factors in adult cross- language speech perception. Journal of the Acoustical Society of America 75:1866-78.