The Ascent of Babel: An Exploration of Language, Mind, and Understanding
Page 6
The idea that the way in which we learn words has something in common with the associative learning exhibited by rats, pigeons, and, come to that, Pavlov's dogs has led to considerable controversy. There is certainly a superficial resemblance between associating the sound of a bell with the arrival of food and associating the sound of a word with the object to which that word refers. But did the sound of the bell have meaning for Pavlov's dogs in the same way that the word 'dinner' had meaning for Pavlov himself? Probably not, but this does not mean that the mechanism that underlies the dogs' learning plays no part at all in our own learning. Obviously, children are more sophisticated than other animals in terms of the kinds of things they can learn, and the circumstances in which they can learn them. This sophistication is nowhere more apparent than in our ability to communicate our beliefs and intentions, and to communicate not just the physical world around us, but also the mental world we inhabit. And whilst other animals (most notably the primates) do have intentions and beliefs, their ability to communicate these aspects of their mental world is severely limited. Although there is evidence that chimpanzees can communicate some of their intentions, and can be aware of another chimp's beliefs, there is little evidence that the chimp can communicate another chimp's intentions, or another chimp's beliefs.
The distinction between the mental world and the observable world is particularly important when we consider that we use language not simply to refer to things that we can see but to refer to things and events that we cannot see, or that happened in the past, or may happen in the future. Language is not simply a way of directing the attention of the hearer or reader to some part of his or her immediate environment. If it were, as more than one psycholinguist has observed, we could just point and grunt. So the task of mapping words onto their meanings is going to be pretty difficult if those meanings are not there. How is the child to know that the meaning of `drink' in `Drink your milk' refers precisely to the thing that the child is not doing? There are lots of things that a child is not doing at any one time, so which specific one is it supposed to be?
Analyses of the earliest speech spoken to infants and young children suggests that only the nouns used in that earliest speech tend to refer to things in the here-and-now. The verbs do not necessarily do so. This explains, in part, why nouns are acquired earlier than verbs. But there is another reason too. Verbs are very much more complex in their meanings. Nouns refer to things, whereas verbs refer to events that often involve more than one of those things. The meaning of 'drink' presumes that someone is doing the drinking and something is being drunk. Interpreting a verb requires an appreciation of the things that take part in the event that the verb refers to. Interpreting a noun requires an appreciation only of the single thing that it refers to.
Several theorists believe that the meanings of verbs can be acquired only once a core body of nouns, to which those verbs will apply (the something and the someone in the drink example), have already been learned. Experiments with young children show that if they see a video of a rabbit feeding a duck, and either hear `The rabbit is zorking the duck' or `The duck is zorking', they will interpret `zorking' to mean `feeding' (as in `supplying food') in the first case, and `eating' in the second. Why? Because the video shows two events; it shows the rabbit doing something to the duck (feeding it), and the duck doing something itself (eating). When the child hears the sentence `The rabbit is zorking the duck' it knows that it should map this sentence onto the event that involves both the rabbit and the duck (hence the importance of knowing the meanings of `rabbit' and `duck'). When it hears `The duck is zorking', it knows it should map this sentence onto the event that involves only the duck. Equivalent effects have been found by Letty Naigles at Yale University in children as young as two years. So what? It sounds obvious, so why care? There are at least two reasons why this is a very important result.
The first reason is that these children interpret the meaning of a novel verb-word on the basis of two things: the structure of the sentence they hear, and the structure in the world they see. They do not ignore the surrounding words and simply focus in on `zorking', and map this onto a feeding event that is somehow isolated from who is doing the feeding and who is being fed. They use the structure of the events they see in the world to aid in their interpretation of the structure of the sentences they hear. Conversely, they use the structure of the sentence they hear to guide what they should be attending to in the world they see. It is a two-way process. This view of what is going on, developed by Lila Gleitman and colleagues at the University of Pennsylvania, is important because it means that children are not attempting to map individual words, somehow excised from the rest of the utterance, onto individual meanings somehow excised from the rest of the world. And yet this was the classical view of what word-learning involved.
The second reason why that Naigles result is important is that it means that verbs are learned in conjunction with a rudimentary knowledge of grammar. The meaning of a verb involves knowledge about the ways in which different things take part in the event described by that verb. For `eat': something does the eating, and something gets eaten. For `give': someone does the giving, someone does the receiving, and something goes from one to the other. Adults, and indeed children, rarely confuse which things are doing what. When children hear that `Daddy is eating the peas', they rarely assume (if they have been brought up in the English language) that the peas are doing the eating. Even children who have not progressed beyond uttering single words will be aware that Daddy tickling Mummy means something different from Mummy tickling Daddy. The point, quite simply, is that when a child knows what a verb means, it apparently knows who-does-what-to-whom.
So the acquisition of meaning is a complex process that depends, in part, on what kinds of words are being learned. Nouns are relatively easier to learn than verbs. Verbs require an appreciation of the fact that events in the world are structured-things cause those events to happen to other things. Consequently, the child, when learning the meanings of verbs, must also appreciate that the utterances it hears are also structured, so as to reflect the structure `out there'.
The idea that the child learns about language through observing the way in which language maps onto the world (and vice versa) is certainly compelling. But it does present some problems for the child. For example, time, in the world, runs in just one direction, yet we can say things like `Sam moved his bicycle after his mother cleaned it' even though the event that is mentioned first in the sentence happened last in the world. The temporal order of the descriptions of events does not necessarily map onto the temporal order in which those events happened. And, not surprisingly, this causes problems-it takes children until they are around six years old before they stop assuming that the first mentioned event happened first. Attempting to map structure in the language onto structure in the world is not without its problems.
One way or another, structure in the language is important. But when do children start to reflect it in their own language? When do they start producing grammatical sentences?
Learning grammar
Infants start combining words once they have an expressive vocabulary in the range of 50-100 words, and the more words they can produce, the more combinations they produce also. But infants need to know more than just a whole lot of words in order to produce multi-word utterances. They rarely produce random combinations. Instead, they somehow learn which combinations are meaningful. They also learn that certain combinations require particular inflections on certain words (e.g. `I walk' but `he walks'). This is particularly difficult in a language like English because its inflections are not the most salient, or noticeable, parts of words. In some languages they are more noticeable (Turkish, for example), and children brought up in languages like these acquire the use of inflections sooner. But somehow, children do learn what is required of them, and this requires the acquisition of knowledge that goes well beyond the meanings of individual words, and which is therefore quite different fro
m the kind that can be learned simply by seeing a cow and hearing the appropriate word.
Since the late 1950s when Noam Chomsky first revolutionized the field of linguistics (he managed to revolutionize it several more times in subsequent years), a major concern for linguists has been to determine how children learn about word order and inflection, and specifically, the correct word order and the correct inflections for their language. Learning about grammar, and specifically, the correct grammar for a particular language, is no easy task. Chomsky had proposed that adults use rules to generate meaningful sentences, and that the task for the child is to somehow acquire these rules. For instance, there are specific rules for transforming active sentences such as `Children learn languages' into their passive form `Languages are learned by children', and there are specific rules for creating the past tense of a verb. How children might acquire such rules is the subject of considerable controversy.
Children are not taught explicitly which rules apply to their language, and so they must presumably learn by example-another psycholinguistic truth that is hardly surprising. But what is surprising is that no one tells their children that certain sentences are ungrammatical, and they certainly do not give their children examples of ungrammatical sentences. Yet the result of this learning process is an adult who knows that certain combinations of words are indeed ungrammatical. How can this be? When children do make errors, and produce ungrammatical sentences, the evidence suggests that, generally, no one corrects them; when children tell untruths, then they are corrected. Roger Brown, one of the pioneers of research into language acquisition, pointed out, slightly tongue in cheek, that this leads to an interesting paradox; if children are corrected for producing untruths, and are not corrected for producing bad grammar, how is it that the result of this regime is an adult who is adept at telling untruths but whose sentences are perfectly grammatical?
To summarize the problem: in order to learn a language one needs to be able to generate grammatical sentences, and to avoid generating ungrammatical ones. The fact that children learn to avoid ungrammatical sentences cannot simply be the result of never hearing any; what is so remarkable about language is that users of the language can generate sentences which they have never heard before. So never having heard an example of a particular sentence does not prevent it from being produced.
The concept of grammaticality is one that is defined not in terms of how to combine or inflect actual words, but in terms of how to combine or inflect types of words. A sentence such as `The girl knew the language was beautiful' can be described as a sequence of syntactic categories: `determiner noun verb determiner noun verb adjective'. Any word in a grammatical sentence can be replaced by any other word of the same syntactic category and the sentence will still be grammatical. However, the task for the child is not simply to learn about which sequences are grammatical, but to learn also about the internal structure of the sentence. It is the breakdown of a sentence into its internal (or constituent) structure that allows the sentence to be broken down into subjects, verbs, and objects. And in order to figure out who did what to whom (or the equivalent), we need to be able to identify the who, the whom, and the what in the internal structure of the sentence. In `The girl knew the language was beautiful', what was beautiful was the language and not necessarily the girl. What was known by the girl was not necessarily the language itself, but that it was beautiful. So the internal structure of the sentence (shown below), broken down into its constituent parts ('the girl', `the language', `the language was beautiful', `knew the language was beautiful', and so on), only licenses certain inferences to be made. That is what grammar is all about. (We shall come back to grarnmar, and how it is used, as opposed to learned, in Chapters 7 and 8.)
Learning about constituent structure, and specifically which is the correct breakdown of a sentence into these constituents, is not easy. In order to learn the right internal structure, the child needs to learn about both the different types of syntactic category (i.e. the different types of words), and their position in the sentence. But if the child does not know about syntactic categories, how can it learn about their position? And even if the child knew about different syntactic categories and wanted to calculate their relative positions, how could it do so if it did not know which words belonged to which syntactic category?
Arguments such as these, and the fact that children hardly ever get any feedback to tell them when they have made a grammatical error, have led linguists to suggest that languages would be unlearnable if the only input provided to the child was the language itself. But it is evidently the case that languages are learnable, and that adults know what is, or is not, grammatical. So where does this notion of `grammatical' spring from?
Language in ourgenes
Noam Chomsky and, more recently, Steven Pinker, also at the Massachusetts Institute of Technology, have argued that the only viable explanation is that we are born with knowledge that aids in (and is necessary for) the task of learning language. Pinker's hypothesis is that the child is endowed, innately, with very crude knowledge about types of words (for instance, that there are types corresponding to nouns and to verbs) and about their role in the language (for instance, that there will be a word in the sentence that corresponds to its subject). This knowledge serves simply as a basis on which to determine (subconsciously, of course) which other types of words there are, and where they can be found relative to one another within the sentence. Moreover, this knowledge is supposed not to be specific to any one language, but is universal to all languages. For instance, young children apparently assume, when observing an event that is being described to them, that the thing causing the event (the rabbit who is supplying the food to the duck, for example) corresponds to the subject of the sentence. So all the child needs to figure out is which word is the subject (it varies depending on the language). If it knows the word for `rabbit', sees the rabbit feeding the duck, and hears the sentence `the rabbit is feeding the duck', it can learn that, in English at least, the first word (in fact, the first noun phrase) is the subject of the sentence. In this way, the child has learned a little bit about the grammar of English.
The fact that the speech addressed to young infants tends to be relatively uncomplicated means that the innate knowledge need only be very crude indeed. None the less, hypotheses such as these attract a high degree of controversy.
There are many abilities that are innate. As adults, for instance, we can tell which side of a room a sound might be coming from, and newborns can apparently do the same. And newly born babies will sometimes imitate facial expressions even though this requires that they somehow `know' that the image on their retina, which is an image they have never seen before, corresponds to a part of their body that they also have never seen before. So the idea that there could exist innate abilities is not itself the source of the controversy. But the innateness hypothesis attracts a number of different controversies. For instance, a distinction can be drawn between different kinds of innate knowledge. If you place your finger in a newborn's hand, the baby will grasp it. This does not mean that the baby has innate knowledge of fingers. In fact, it will grasp anything placed in its hand although, again, this does not mean that it has innate knowledge of graspable things. Whatever it is that is innate here (a reflex) is very different from knowledge about the structure of objects that the baby could encounter in its new world. In one case, the innate ability does not require knowledge of the way in which the external world is organized, and in the other case, it does.
To suggest that infants are endowed with innate knowledge of the distinction between verbs and nouns may sound as extreme as suggesting that humans are born with the innate knowledge that their food should, if possible, be heated (there would be obvious evolutionary advantages if this improved the chances of survival for the species). But perhaps the noun/verb distinction is not so extreme after all. Perhaps what we are endowed with is the ability to distinguish between objects and the changes that can happen to thos
e objects-in other words, to tell states from changes in state. Is this so very different from the distinction between nouns (which often refer to objects) and verbs (which often refer to actions, and hence, changes)? So one aspect of the controversy concerns the status of the innate knowledge that would be required in an account such as Pinker's-is the knowledge linguistic, concerning the nature of language, or is it something more general, concerning the nature of the world?
In some respects, Pinker's account is very similar to Lila Gleitman's account of how we acquire verb meanings, described earlier. There, a child would observe an event and map that event onto the sentence it was hearing. If a sentence included the names of two things that the child could see ('rabbit' and `duck', for instance), and what was seen to be happening was that one thing was doing something to the other, then the child would interpret the sentence it heard as conveying exactly that information: that one was doing something to the other. So the child would learn about subject (the thing that is doing) and object (the thing it is being done to) by taking account of both the world and the sentence, and trying to map one onto the other. The only thing that would need to be innate would be the ability to map one thing (language) onto another (the world), and vice versa. The difference between the two accounts concerns the nature of the predisposition that is assumed to be innate. In Pinker's account, it is a predisposition that is supposedly specific to language (there are nouns, there are verbs, there are subjects); in Gleitman's, it is a predisposition that is specific to things which co-vary-that is, a predisposition to establish whether one kind of input (in this case, language) correlates with, and can therefore be mapped onto, another (the observable world).