Those things, that all languages have in common, or that are necessary to every language, are treated of in a science, which some have called Universal or Philosophical grammar
- James Beattie, 1788
- James Beattie, 1788
Every language in the world has a finite number of words in its vocabulary - it can't help it, since the speakers have to know them all, and they only live so long. But every language makes possible an infinite number of sentences, thanks to the recursive way those words can be combined, and so allows humans to express an infinite number of thoughts. We don't know of any animals that can do this. There was a chimpanzee called Nim Chimpsky who was taught about a hundred words of sign language - but he didn't have the infinitely creative ability that we have, because he had no notion of syntax. He could use words on their own and even sometim
es in pairs, but his communications had to remain at that basic level. We're having similar trouble teaching language to computers. Some researchers believe that computers will never have human levels of competence - partly because we don't really know what we have to teach them. Even humans find it very hard to learn if they tackle the task as adults - second languages come with effort, and natives almost always identify a foreigner.So if language is so complicated and so hard to learn for animals, computers and adult humans, why are children so good at it? Except in a few cases of pathological disability, they do it without fail, and incredibly fast at that. And they don't need explicit teaching: in some cultures, it's considered inappropriate to address children directly until they can speak like adults, but they learn anyway from overhearing the conversations of others.
Noam Chomsky (the chimp's namesake) suggested that this could be because we're born already knowing how language works. I have an uncle who told me this idea was ridiculous: whoever heard of people inheriting ideas in their genetic code? But if you remember that the knowledge is supposed to be subconscious, I think the idea becomes more plausible. Babies are born knowing all sorts of things: they have an idea of number constancy, for example. If a single ball disappears for a moment behind a screen and, a moment later, re-emerges on the other side, babies remain pretty relaxed about life. You can tell this by putting a dummy in their mouths that has sensors in it to measure how much sucking they're doing - sucking levels stay fairly similar when the ball comes out from behind the screen. But if one ball goes in and two balls come out, their eyes go wide and their sucking gets harder as they try and deal with this affront to their world-view.
So we're born with simple number theory in our heads. But are we really born with simple grammatical theory too? Well, my favourite piece of evidence that we are is the question of the tall man: in 1975, Chomsky imagined a Martian learning English. At the moment we come into the story, the Martian has the hang of statements, but is trying to work out how to turn them into questions. He makes some observations and finds that the statement "The man is tall" can become the question "Is the man tall?" Fine, thinks the Martian. The rule must be to move "is" to the beginning. At first this seems to generalise well: the Martian comes across the sentence "My hovercraft is full of eels" and, although he's never seen the sentence before, his rule serves him well and he finds he can correctly form the question "Is my hovercraft full of eels?" But here's where the fun starts. Suppose he's faced with "The man who is tall is in the room", and he wants to turn it into a question. Which "is" should he move to the front? He might come up with any number of theories for which one to move, all of them logically possible - perhaps the rule is: when you have two choices, always pick the first one. Then he'd mistakenly produce "Is the man who tall is in the room?" Now, this is the clincher. Children never make that kind of mistake. They know without having to be told that all rules depend on the structure of the sentence, not on any of the other logical possibilities like the order of the words. I think they're born knowing.
11 comments:
The arguments from stimulus poverty are IMHO severely damaged by research done at the last decade.
This is Pullum on auxiliary fronting, and he had papers (2002, 2005 if I recall correctly) taking some other similar arguments to the test.
http://www.ecs.soton.ac.uk/~harnad/Papers/Py104/pullum.learn.html
With a finite sentence length, and a finite vocabulary, I fail to see how this leads to an infinite number of sentences rather than "a very, very, large number of sentences"...
Hillel: thanks for the link. I'll have a look.
Freshvictims: are you suggesting that sentences have a maximum length? What is it?
Hillel, that's an interesting paper: it shows Chomsky sticking his neck out too far and getting his head chopped off. I wish he hadn't, because you really don't need to go round claiming these constructions are once-in-a-lifetime occurrences for the argument to be convincing. All you need to say is that they're rare enough that children will sometimes need to generate them before they've come across them. As is shown in the Crain and Nakayama experiment that Pullum cites, children never make this kind of mistake.
Lay that alongside the fact that there are many kinds of mistake that children do fall into when they're forming questions. Examples from another Crain experiment include 'Why you have your vest on?' and 'What do you think what pigs eat?' What's the difference between these mistakes and the one in the article? The answer is that these ones are linguistically possible - they're attested in Romance languages (in the vest example) and some Germanic ones (for the pigs). The structure-independence mistake in the article is logically possible but linguistically impossible.
Does Pullum make the case for the position that examples of these kinds of constructions are so common that children's total lack of mistakes is unsurprising? I don't think he does. He's very careful not to assume a priori that language learners know what a clause is, but he does assume that they know what an auxiliary verb is, and that they know evidence gathered about one auxiliary verb's behaviour can be safely applied in the context of another. It would be more impressive if he was able to show lots of situations where the embedded auxiliary was the same as the main one, and more impressive still if he was able to show lots for each one - including 'is'. How much knowledge do you need to gather - without the help of any biological endowment - before you can see the relevance here of the question ‘How fundamental are the changes these events portend’?
Hi again. In the 1996 paper Pullum suggested one way for children to be able to infer the rule based on their given input. Since then, there have many papers on similar problems, the ones I find most promising are from the field of probabilistic linguistics. (You can check out this one: crl.ucsd.edu/~elman/Papers/BU2001.pdf , for a neural network approach to the problem of auxiliary fronting).
Since we have no way of knowing how the child actually gets to know the rule, we can only choose the explanation that sounds best (or come up with one of our own). If saying that all (or some) grammar rules are innate sounds reasonable to you, that by all means go that way (and if you do, crain is a good researcher to read, no doubt). The point I'm trying to make is that recent years provided some promising alternative approaches for the ones uncomfortable with that assumption.
I think it'd be obvious to the Martian, or a child, that the 'is' you move is the one that's the verb of the sentence, even when they're too young/Martian to be able to express it that way.
I don't see that it would be obvious to the Martian, given the fact that other logical possibilities exist.
I can't see a logical explanation for "Is the man who tall is in the room". You know, if you know English, that the 'is' you moved is to do with being tall, not to do with being in the room. But yeah, maybe this just seems obvious to me because I was "born knowing".
Do children really never make that kind of mistake? I remember "Whose is that's?" (for "Whose is that?") being one of my pet hates early in school.
"You know, if you know English" - you're begging the question! And even assuming you do know the grammatical role of each instance of is, how do you know that's relevant to question formation?
There are tons of logical possibilities - one can picture a language which forms questions by simply reversing the order of the words in the sentence, but of course no human language does that, because it violates the principle that all rules must be structure-dependent.
Children really never make that kind of mistake. Unfortunately I don't think the relevant paper is freely available on the web. It appeared in the journal Language in 1987: Stephen Crain and his co-author got children making various errors, but never ones like that.
"Whose is that's" is interesting. I wonder whether the 's is a possession marker or a copy of the moved is.
Language nature or nurture
Children may not been born knowing grammar or number theory. The may simply learn. The fact that there are some grammar mistakes which never happen could just as well be explained because language is passed on by word of mouth. Indeed this seems to me the most likely explanation. So there’s no particular need to be “born knowing” except to the extent that the brain has the capacity to lay down the connections in the learning phase – and probably the chimp doesn’t.
Take your number theory example. Try a thought experiment. Imagine taking a new born baby and popping him into a world where, whenever something disappeared, there would be two when it/they reappeared. So Mummy dips below the cot and up pop two Mummies. (To make life less impossible, where there are two of something in this world they could reappear as one – when the two Mummies leave the room only one returns.) How would we expect the baby to respond to the ball experiment now? One ball turning into two would leave him unimpressed. But one ball emerging still as only one could be expected to create a ferment of dummy sucking. Because his world-view had been affronted. But not because he was born with number theory – because he had learned it.
Now what about language? You mention that children never make certain types of mistake. OK, let’s do another thought experiment. Devise a new language “Margram” which deliberately employs some/lots of these “mistakes” in its grammar. So they are not mistakes in Margram – they are correct grammar. Have a team of researchers become fluent in Margram. Now take a new born identical twin. Bring him up in a Margram-only environment.
What would we expect? Surely we would expect him to acquire Margram complete with Margram’s unique grammar. We might expect him to acquire it just as quickly as his twin brother acquires English (or whatever language he is brought up in). And if we later send them for a two year period aged say 6 to each other’s homes we might expect them both to become equally bilingual as well.
So what you suggest we are born with, could just as well be learned. Indeed I think that is much more likely.
So how come certain grammatical possibilities, such as selected for Margram, do not occur in any natural language? If they don’t exist now, they are unlikely to arise because each child is brought up learning a language where they do not. His grammar has been passed to him by word of mouth. So going back in time you can see that they must not have existed in ancestral generations either. I cannot argue that we are definitely not born with grammar or number theory. But learning seems just as good a possibility for the examples you cite. And it may be a simpler one too.
If babies can hear (and learn to recognise) their mother's voice inside the womb, couldn't they also pick up some grammatical rules?
Post a Comment