The Ascent of Babel: An Exploration of Language, Mind, and Understanding
Page 3
On babies' sensitivity to rhythm and intonation
Intonation (variation in pitch) and rhythm are properties of a language which may differ depending on the precise language being spoken. They are also properties of the language to which babies are exposed very early, primarily in the form of the mother's voice, which travels through bone and tissue to the uterus and, although still muffled, is louder than any speech sounds coming in from the outside. The variation in the physical signal that gives rise to the perception of varying intonation and rhythm is generally referred to as prosodic variation-the variation in pitch, amplitude (which is perceived as loudness), and duration (how long the sounds are, how quickly they vary, and so on). In fact, prosodic variation is just about the only variation in the language that the baby has exposure to before it is born.
In the mid-1980s, a research group in Paris headed by Jacques Mehler demonstrated that very young babies (just four days old) already know enough about their own language to be able to tell it apart from another language. At such an early age, one might expect that the baby is doing all it: can to distinguish speech sounds from other sounds (whether doors shutting, dogs barking, dishwashers washing, and so on). To distinguish between different languages is surely overkill. Why should the baby want to distinguish between languages? The answer, of course, is that it does not. If babies are sufficiently sensitive to changes in, for instance, the prosodic characteristics of the speech they hear, it follows that they should be able to distinguish between speech exhibiting different prosodic characteristics. And if the speech from two languages differs with respect to prosody, then babies should be able to distinguish between the languages.
Nlehler used the same non-nutritive sucking technique described briefly in Chapter 1. Speech from one language (French) was played to four-day-old babies each time they sucked on their teat until they habituated-that is, until the novelty wore off and they started to suck less. Once their sucking rate fell below some pre-determined amount, speech from another language (Russian) was played to them each time they sucked (the speech was recorded from a single French-Russian bilingual speaker so that the voice characteristics remained the same). Mehler found that the babies, all of whom had French-speaking parents, sucked more when the different language was played to them (because of the novelty of this new stimulus, the infants had dishabituated). It did not matter whether the babies were first given French or Russian; if the language was changed after the babies had habituated, they sucked faster. If the language did not change, they sucked the same.
In a related experiment, Mehler and his colleagues wanted to be sure that the babies were not simply discriminating between the languages on the basis of the different sounds that they use. So they filtered the speech so as to remove the higher frequencies (recall that one cannot make out the different sounds in a language without those higher frequencies). In effect, they artificially simulated the kinds of speech that the babies would have heard in utero, and gave the babies just the prosodic content of each language. But still the babies could distinguish between them.
Finally, if babies were given language pairs that they had had no experience of (for instance, playing English and Italian to the French babies, or French and Russian to babies who were neither French nor Russian), the habituation/dishabituation pattern was not found. The babies were therefore discriminating between the languages on the basis of prior familiarity.
Prenatal learning
So infants as young as just four days are sensitive to the kinds of prosodic variation that is normal for their maternal language. It is unlikely that in this short time they hear enough of their own language to develop such a high degree of sensitivity; it is more likely that some learning has taken place prenatally. But although Mehler and his collaborators' studies are suggestive, they do not actually prove that the learning took place in utero. However, there are a number of findings, notably by Anthony DeCasper and his colleagues at the University of North Carolina, which can only be explained in terms of prenatal learning.
In one study, DeCasper and his associates arranged for a group of pregnant women to read aloud the same short story each day for the last six weeks of their pregnancy. After the babies were born, DeCasper used a non-nutritive sucking technique to establish whether the babies would prefer to hear the same story that they had, in effect, already heard, or whether they would prefer a new story. In fact, they preferred the story they had already heard (although all they could have heard of it, in utero, was the prosody with which it was spoken).
DeCasper used a slightly different technique to the one used by Mehler's team. He took advantage of the fact that babies will quickly learn to modify their sucking rate in order to receive some pleasurable stimulation. They will suck faster than normal, or slower than normal, depending on which way ensures they get what they want. So to find out whether babies prefer to hear one thing or another, you set things up so that they have to suck one way to hear one thing, and the other way to hear the other. And whichever way they suck (faster than normal, or slower than normal) will indicate which thing they preferred. In DeCasper's experiment, they sucked whichever way ensured that they would hear the familiar story, as opposed to a new story.
The babies in this experiment preferred the familiar story even if the recordings used to establish the preference were made by someone other than their mother. But several studies have shown that newborn babies prefer to hear their mother's voice when given the choice. So this last finding, coupled with the DeCasper result, suggests that babies manage to do two things when listening prenatally; they can learn about the specific prosodic characteristics of the voice that they hear most of in utero, and they can learn about the general prosodic characteristics of a sequence of sounds (making up a story, for instance) uttered by that voice.
Testing babies in utero
The studies so far have found postnatal responses to prenatal experience. In a further study, DeCasper and his colleagues wanted to go one step further, and find a prenatal response to prenatal experience. They arranged for pregnant mothers to read aloud a short story every day between the 34th and 38th week of pregnancy. At the end of the 38th week, that same story and a different story were played to the fetus, through the mother's abdomen. The mother was not aware of which story was being played when, so that any response in the fetus could not be due to some sensitivity to the mother's response to the story. Fetal response was judged on the basis of changes in the fetal heart-rate. One story was played for 30 seconds, then there was a 75-second period of silence, after which the other story was played for 30 seconds. DeCasper found that fetal heart-rate decreased when the familiar story was played, irrespective of whether the familiar story was played first or second. Apparently, therefore, the fetuses distinguished between the two stories, and did so on the basis of what they had learned previously. Although further work remains to be done to corroborate these results, the evidence all converges on the view that babies are able to learn about the sounds they hear prenatally.
It was research of this kind that led to the practice, prevalent mainly in the USA, of talking to babies prenatally through the mother's abdomen (a device called a `pregaphone' was marketed for this very purpose). There is no evidence that prenatal stimulation other than that normally experienced has any beneficial effects. In fact, it is unclear whether any prenatal learning, even under normal circumstances, is necessary. For instance, children with normal hearing who are born of deaf parents in communities where sign language is the more usual form of communication do not seem to suffer from the lack of prenatal exposure to spoken language; their spoken language skills develop within normal limits so long as they have sufficient exposure to (and social interaction using) the spoken language. This does not mean that prenatal learning is of no use. It may well be useful (more of this shortly), but it may not be necessary for normal language development. A baby born without any prenatal language experience may be able to learn, subsequently, everything it needs
to know about the prosody of the language.
So the next question is simply this: how could knowing about the prosody of the language (whether from pre- or postnatal experience) possibly have any use? Being able to distinguish statements from questions, or another language from its own, is hardly something that the newborn infant has to worry about, and would be of little help when it comes to figuring out where, in all the jumble of sounds it hears, the actual words are.
The importance of being prosodic
As a step towards at least a partial answer to this question, we need to detour slightly and consider a further ability of very young infants. The problem for a newborn baby is similar to the problem faced by an adult when listening to speech uttered in an unknown foreign language. Where do words begin and end? At least the adult knows about words, and that the speech he or she can hear probably contains some. And knowing about words, even in principle, is likely to be a big help when learning to understand a new language. After all, if the task, when learning that language, is to associate meanings with words, then identifying the words is half the task. But knowing about words may also help the adult distinguish the noise coming from the speaker's mouth (or radio, or wherever else it may be coming from) from all the other noises that can be heard. How? Because spoken words have (at least) one property that is not shared with other kinds of noise.
A fundamental property of spoken words is that they are formed by opening and closing the mouth in subtly different ways, which include changing the shape of the lips, moving the tongue in certain ways, and opening or closing the velum (the flap of skin that allows air to go through the mouth or through the nose or through both). The result of this opposition between the mouth being open and the mouth being closed (perhaps only partially) is that we produce syllables.
Syllables are, of course, the fundamental building blocks of words. Some languages combine syllables together to make words even though the individual syllables themselves may have no meaning of their own (English, for instance). In other languages (Chinese, for instance) each syllable has a distinct meaning, and the meaning of a word made up of more than one syllable need not be related to the individual meanings of its component syllables. In each case, though, the syllable is a fundamental unit of spoken language. So knowing about syllables, and being able to spot them, would be a useful ability to have if one were trying to learn a language. Not only would it help in terms of an initial attempt to identify words, but it would also be useful because non-speech sounds tend not to consist of identifiable syllables.
After a number of studies, Jacques Mehler and his colleagues Josianne Bertoncini and Peter Jusczyk concluded that the ability to perceive speech in terms of syllables exists at birth. In a series of experiments with babies ranging in age from just four days to a few months, these researchers found that newborn babies could discriminate between certain sequences of sounds but not between certain others. For instance, they can distinguish between different sounds such as [p] and [t]', _link_ and not surprisingly, therefore, between different syllables such as [pat] and [tap]. What is surprising, though, is that despite this ability, they do not easily distinguish between [pst] and [tsp]. This could be because these sounds, when put together, make a noise that is very hard to tell apart. Alternatively, it could be because these sequences are not real (or legal) syllables.
In order to test this, Mehler and his associates added a vowel to each side of the sequences, to create [upstu] and [utspu] (the [u] being pronounced as in the word `put'). Although the [pst] and [tsp] sounds remained identical, the addition of a vowel on either side meant that each sequence could now be interpreted as a sequence of two legal syllables. If [pst] and [tsp] were hard to discriminate simply because they were difficult sounds, then the addition of the extra vowels should make no difference. If, on the other hand, they had been hard to discriminate because they violated some sort of expectation of what a proper syllable should sound like, then the additional vowels should help, as they turn the sequences into proper sounding syllables. And sure enough, this is exactly what happened; the addition of the vowels did make the sounds much more discriminable.
What was so ingenious about this experiment was that the infants' sensitivity to syllables was not tested by having them discriminate between different syllables, but by having them discriminate, in effect, between legal and illegal ones. The same sounds in one context were indiscriminable ([pst] vs. [tsp]), but in another context were discriminable ([upstu] vs. [utspu]). But despite the ingenuity of the lateral thinking that went into the design of these studies, we are now faced with a new dilemma: how can infants possibly know what is a legal or illegal syllable? It is obviously a useful ability to have, for the reasons that were mentioned before, namely that being able to interpret the incoming sounds in terms of (legal) syllables is a big help when it comes to learning a language and discriminating between the sounds of that language and other non-language sounds. But where does this knowledge come from? Is it genetically programmed in our DNA? Probably not. After all, if knowledge about syllables could be preprogrammed, or innate, why not other knowledge as well? In fact, some researchers take the view that other knowledge is innate, but we shall come to that in another chapter. For the moment, we shall stick to syllables, and although there is currently no absolute proof that knowledge about syllables is not innate, there is an alternative explanation for the extraordinary abilities that the newborn infant appears to have.
It is at this point that we can return to an earlier question, namely whether knowledge about prosody could be at all useful to the newborn language learner. On first asking this question we side-tracked slightly and discovered that babies seem to organize what they hear in terms of syllables. But how can babies do this? The answer, or at least, one of a range of answers, is that babies do not organize what they hear syllabically-they organize it prosodically. Neither [pst] nor [tsp] qualify as legal syllables because syllables generally require a vowel in them somewhere. But putting a vowel in amongst a sequence of consonants necessarily changes the prosody. That is, changes are introduced which influence the pitch, rhythm, and relative intensities of the different sounds. And the reason [pst] and [tsp] may not be easily distinguished may not be because the newborn knows that there should be a vowel in there, but because they cannot recognize the prosody; it is not one of the prosodic patterns that the baby has come to expect. With [upstu] and [utspu], on the other hand, the baby can recognize the sounds as, in some sense, conforming-they each have prosodic `tunes' which are familiar on the basis of the baby's prior experience. And because they conform, the baby can make sense of them, and tell them apart.
Sensitivity to prosody might do more, though, than just help identify legal syllables. An analogy may help: many popular songs are structured so that the music accompanies the words quite precisely; each time a new word (or even syllable) is sung, a new note (or collection of notes) is played. The drummer may even provide a beat that closely accompanies the singing of each word or syllable. If someone were listening to such a song in a language they did not know, they could listen to the changing notes, and the beat, without knowing anything about the words that accompanied the music. But the changing notes and the different beats would most likely be accompanied by different words. After listening to only a little of the song, it would be possible to attend less to the notes and the beat, and more to the words that fell on the different notes and different beats. In short, being able to recognize structure in the music, would allow someone listening to the song to discern structure in the vocals. If the same words were sung in monotone, with no backing music, it would sound like a continuous jumble of uninterpretable sounds. But because the words are not sung in monotone, and have backing music which has some recognizable structure to it, the words themselves can be disentangled from one another. And although they would have no meaning (if in an unknown language), it would at least be possible to separate one from the other (or at least, one syllable from another syllab
le). The important point about this analogy is that the equivalent to the backing music and the beat in the human voice is prosody.
For the newborn, listening to the human voice is perhaps a little similar to someone listening to a song in a language they know nothing about. The newborn already knows something about the structure of the music, including the beat. The other sounds that accompany the music (that is, the actual syllables) are differentiated from one another simply because the music itself has structure. Of course, this is a very simplified account, but it gives some of the flavour of what may really be going on. The key question is this: how easy would it be for a child to acquire a language that was identical in all respects to spoken English (or any other language) except that it was spoken in monotone? Presumably, not easy at all, although we shall never know for sure-we can only conjecture. But at least we would have a solid empirical basis for that conjecture.