Posted on Friday 24 October 2003

Linguistic nativism or the innateness hypothesis is the claim, advanced by Chomsky (1986) and Pinker (1994) amongst others, that human beings are endowed with an innately, presumably genetically, specified domain specific knowledge of language. This knowledge is tacit, that is to say not accessible to conscious thought, and it specifies in some detail the nature of possible human languages, including a set of syntactic categories, a set of possible phrase structure rules, constraints on admissible transformations and so on. The primary argument for this bold hypothesis is the so-called Argument from the Poverty of the Stimulus (APS), that the linguistic input or evidence available to the infant child is so impoverished and degenerate that no general, domain-independent learning algorithm could possibly learn a plausible grammar without assistance. An obvious refutation of this argument is to demonstrate the existence of an algorithm that can learn a reasonable grammar, from that amount of data. It is that issue that this thesis is intended to study. Nonetheless the algorithms presented here are I hope of general interest as pieces of computational linguistics or machine learning research.

[more (PDF)]