MIT-Structure Dependence, the Rational Learner, and Putnam’s Sane Person

Young children say many surprising and funny things – funny, often, because how they say it is not quite right in an endearing way. “My friend goed to the playground,” and “I ated two desserts” both demonstrate errors that we readily understand, sympathize with, and are confident will go away with further listening and speaking.

But there are other kinds of errors that children just don’t seem to make. In his pathbreaking work on transformational grammar, Noam Chomsky has written extensively about sentences like “The dog in the corner is hungry.” By applying a formal operation Chomsky described in detail, we can form the question “Is the dog in the corner hungry?” But confronted with “The dog that is in the corner is hungry,” we do not end up asking “Is the dog that in the corner is hungry?” Instead, we apply the transformational rule in a different, more complex way, to ask “Is the dog that is in the corner hungry?”

Chomsky draws two conclusions from close study of many such cases. First, he says, this shows that the transformational grammar rules we follow are “structure-dependent,” that is, they apply to phrases, not simply to a string of words in sequence. Second, because a person can go through life without recognizing or even encountering some structure-dependent cases – and yet make the correct choice when presented with alternatives – this aspect of grammar has deep implications for human psychology. In fact, Chomsky claims, this is an argument for the existence of invariant principles of language, a universal grammar.

Howard Lasnik cites evidence for a different interpretation: Chomsky’s “poverty of the stimulus” scenario may not be relevant. By examining a large collection of speech (drawn from the CHILDES database), and applying a Bayesian model of grammar induction – making use, in other words, of the speaker’s knowledge of prior probabilities – it is possible to show that a rational learner could in fact learn that transformational linguistic rules depend on phase structure.

Lasnik’s former student, now colleague, Juan Uriagereka, broadens the argument. Drawing on a startling range of examples – from animal behavior to protein folding, Uriagereka wonders if the structural properties of grammar are unique to human language, or extend to other forms of human cognition, including music, mathematics, and complex planning. Structure dependence may be true, it may be specific to language or at least to human thought … but how did it get there? Where does structure come from? These are the bold questions Lasnik and Uriagereka believe that contemporary linguistic cognitive science has to address.

• March 21, 2010 -- MIT-America at War? (0)
• March 21, 2010 -- MIT-Spectacular Reconstructions: The Politics of Recovery in American Urban Disasters (0)
• March 21, 2010 -- MIT-Double Restoration: Berlin after 1945 (0)
• March 21, 2010 -- MIT-Beirut, Beirut (0)
• March 21, 2010 -- MIT-Washington, DC after Conquest and Arson during the War of 1812 (0)