Scott Thornbury’s latest Sunday post gave what I thought was a very poor account of the poverty of the stimulus argument and of objections to it. While Scott was quite measured in his original remarks, his post showed a spectacular disregard for logic, and the wave of enthusiastic messages of support which flooded in from a frightening array of dimwits and cranks seemed to unhinge our normally restrained hero, provoking him to ever more outrageous and fanciful claims. I and a couple of other sensitive souls did our modest best to keep him on the rails, but we failed, the wheels came off, and last time I looked, the whole crazy bunch of them were swapping quotes from Derrida, counting backwards from 666, trying to communicate with each other without switching their brains on, and using impoverished input devices like the Microsoft keyboard. Since they’ve all shown themselves to be useless at marshalling a case against Chomsky for themselves, I thought I’d offer a helping hand. I’m all heart, really. So here’s the case against Chomsky as argued by two of his leading critics: Geoffrey Sampson and Elizabeth Bates.
Before we start on Sampson, let’s quickly state the poverty of the stimulus argument. It says: since children know things about language that they’ve never been exposed to, that knowledge must be innate. Vivian Cook puts it like this:
Step A. A native speaker of a particular language knows a particular aspect of syntax.
Step B. This aspect of syntax could not have been acquired from language input. This involves considering all possible sources of evidence in the language the child hears and in the processes of interaction with parents.
Step C. This aspect of syntax is not learnt from outside. If all the types of evidence considered in Step B can be eliminated, the logical inference is that the source of this knowledge is not outside the child’s mind.
Step D. This aspect of syntax is built-in to the mind (Cook, 1991).
The UG argument is that all natural languages share the same underlying structure, and the knowledge of this structure is innate.
Sampson says that Chomsky’s claims about the linguistic data available to the child are “untrue”, and he takes Chomsky’s example (used at the famous 1975 conference at Royaumont, where Piaget, Chomsky, Fodor, and others gathered to discuss the limitations of the genetic contribution to culture) of two different hypotheses about the grammar of yes/no questions in English. Turning an English statement into the corresponding yes/no question involves operating on a finite verb in the statement. Either the verb itself is moved to the left (if the verb is a form of be, do, have, or a modal verb such as will) – thus ‘The man is tall’ becomes ‘Is the man tall?’; or, in all other cases the verb is put into the infinitive and an inflected form of do is placed to the left – thus ‘The man swims well’ becomes ‘Does the man swim well?’ (Sampson, 1997: 40).
Chomsky says there are two hypotheses that the child learning English might try: 1. operate on the first finite verb; 2. operate on the finite verb of the main clause. Hypothesis 1 violates the structure dependence universal and is false (applied to the sentence “The man who is tall is sad.”, it would give: “Is the man who tall is sad?”). Hypothesis 2 is correct. Yet both hypotheses work in all questions except those formed from statements containing a subordinate clause which precedes the main verb. The child cannot decide by observation whether one or the other hypothesis is true, because cases of statements containing a subordinate clause which precedes the main verb are extremely rare. Therefore, the child decides on the basis of innate knowledge. In reply to this Sampson says that many examples actually exist, including the well-known line from Blake’s The Tyger “Did he who made the Lamb make thee?” Sampson goes on to give a number of other examples from a children’s corpus, and concludes:
Since Chomsky has never backed up his arguments from poverty of the child’s data with detailed empirical studies, we are entitled to reject them on the ground that the data available to the child are far richer than Chomsky supposes. (Sampson, 1997: 42)
Sampson then attacks Chomsky’s “question-begging idealizations”. Chomsky distinguishes between competence (a certain type of knowledge which is the phenomenon that he wants to explain), and performance (data, much of which he judges to be irrelevant). To examine competence, Chomsky argues that it’s necessary to make various simplifying assumptions, but Sampson claims that Chomsky’s use of simplifications distorts the substantial point at issue. Each of the counterfactual simplifying assumptions about human language which Chomsky makes “eliminates a plausible alternative from consideration through what is presented as a harmless, uncontroversial assumption” (Sampson, 1997: 51). Sampson gives the example of the assumption that language acquisition is an instantaneous process. This, says Chomsky, is “a harmless assumption, for if it mattered then we would expect to find substantial differences in the result of language learning depending on such factors as order of presentation of data, time of presentation, and so on. But we do not find this” (Chomsky, cited in Sampson, 1997: 51-52). Sampson replies that language acquisition is not an instantaneous process (as Chomsky elsewhere admits), and it is not a harmless simplification to say that it is. As Sampson says:
To claim that it is harmless to pretend that language acquisition is instantaneous is, in effect, to assume that language acquisition does not work in a Popperian fashion, without going to the trouble of arguing the point. (Sampson, 1997: 52)
Chomsky acknowledges that children do not move from ignorance to mastery of language instantaneously, but he insists that “fairly early in life” a child’s linguistic competence reaches a “steady state”, after which there are no significant changes. Sampson points out, however, that this “steady state” idea is contested by Bloomfield and Whitney (both of whom see language learning as a lifelong process), and is also completely at odds with the Popperian approach to learning, which brings us to Sampson’s alternative explanation of language acquisition.
Sampson argues that the essential feature of languages is their hierarchical structure. Children start with relatively crude systems of verbal communication, and gradually extended syntactic structures in a pragmatic way so as to allow them to express more ideas in a more sophisticated way. The way they build up the syntax is piecemeal; they concentrate on assembling a particular part of the system from individual components, and then put together the subassemblies. This gives them low level structures which are then combined, with modifications on the basis of input, into higher level structures, and so on.
Sampson uses the Watchmaker parable, first made by Herbert Simon (see Sampson, 1997:111-113), to explain linguistic development. I won’t go into it here, but Sampson says that Simon’s parable shows that “complex entities produced by any process of unplanned evolution, such as the Darwinian process of biological evolution, will have tree-structuring as a matter of statistical necessity” (Sampson, 1997: 113). Furthermore, in Sampson’s view, “the development of knowledge, as Popper describes it, is a clear case of the type of evolutionary process to which Simon’s argument applies, and can be applied to syntactic structures”. Sampson describes how the communication system of our ancestors gradually became more complex as language learners made longer sentences, which would enter the language if they made a significant enough contribution to transmitting information more economically, or if they were semantically innovative. Similarly, a child acquires language by composing sub-assemblies from individual components, and then putting together the sub-assemblies.
Only a general learning theory is involved in Sampson’s explanation, which adopts a decidedly Popperian approach. The child tests various hypotheses about grammaticality against input, and slowly builds up the right hierarchically structured language by following a Popperian programme of conjectures and refutations. This supposes, of course, that the child is exposed to adequate input. Sampson’s argument has two main strands: first, following Simon, gradual evolutionary processes have a strong tendency to produce tree structures; and second, following Popper, knowledge develops in a conjectures-and-refutations evolutionary way. Sampson claims that these two strands are enough to explain language acquisition.
Perhaps Sampson’s criticism of one of Chomsky’s most central assumptions can serve to highlight the differences between them. Chomsky says that
Linguistic theory is concerned primarily with an ideal speaker-listener, in a completely homogenous speech community, who knows its language perfectly (Chomsky, cited in Sampson, 1997: 53).
This assumption, which Chomsky describes as being of “critical importance” for his theory, excludes Sampson’s Popperian approach without even considering it. For Sampson, learning is a “non-terminating process”, and language has no independent existence over and above the representations of the language in the minds of the various individuals belonging to the speech community that uses it.
What the language learner is trying to bring his tacit linguistic theory into correspondence with is not some simple, consistent grammar inhering in a collective national psyche…. Rather, he is trying to reconstruct a system underlying the usage of the various speakers to whom he is exposed; and these speakers will almost certainly be working at any given time with non-identical tacit theories of their own – so that there will not be any wholly coherent and unrefutable grammar available to be formulated. The notion of a speaker-listener knowing the language of his community “perfectly” is doubly inapplicable – both because there is no particular grammar, achievement of which would count as “perfect” mastery of the language, and because even if there were such a grammar, there is no procedure by which a learner could discover it. (Sampson, 1997: 53-54)
From Sampson’s Popperian perspective, even if language learners were “ideal” they would not attain “perfect” mastery of the language of the community. As Sampson says:
Popperian learning is not an algorithm which, if followed without deviation, leads to a successful conclusion. Therefore, to assume that it makes sense to describe an “ideal” speaker-listener as inhabiting a perfectly homogenous speech community and as knowing its language perfectly amounts, once again, to surreptitiously ruling the Popperian view of acquisition out of consideration. (Sampson, 1997: 55)
I personally don’t find Sampson’s arguments persuasive, and I’ll explain why after I’ve presented Bates’ case against Chomsky in the next post.
Cook, V. J. (1991) The poverty-of-the-stimulus argument and multi-competence. Second Language Research, 7,2, 103-117
Sampson, G. (1999) Educating Eve: the `language instinct’ debate. London: Cassell.