Talk at Puddle • Computational and statistical learning theory for language.

I am speaking at Puddle on Friday of seventh week (28 November) at 3 pm in Lecture Theatre B of computer science. (Enter via the door to Lecture Theatres A and B, or go to the reception.)

Puddle is a seminar arranged by doctorants in computer science. As far as I can tell, it nearly exclusively hosts (1) category theorists and (2) token philosophers. (Plausibly this talk has more to do with computer science than, er, modal potentialism.)

Consider the following roughly Chomskyan argument. (1) Without innate knowledge of language, first language acquisition would be so hard as to be impracticable. (2) But we generally succeed in acquiring our first language. (3) So we must have innate knowledge of language. The argument can only be assessed with respect to some understanding of hardness. Some linguists, cognitive scientists, and philosophers appeal to learning theory: hardness is (sample or computational) complexity. I discuss two difficulties for this approach. First, the applicability of these learning-theoretic results to cases of language acquisition is unclear. Unlearnability results are régime-dependent, and it’s not obvious that régimes where unlearnability results hold are any more plausible than other régimes. Second, many learning theory-inspired accounts of what innate knowledge actually is do not seem to explain learnability. I propose an explanation. The problem is the assumption that language learning should scale asymptotically, rather than simply on plausible inputs. I suggest this assumption arises from a confusion between two different forms of computationalism about cognition; one justifies the assumption, but only the other is well-motivated. If I have time, I will explain how similar considerations apply to arguments against linguistic competence in language models appealing to learning or complexity theory. I shall assume no substantial prior knowledge of philosophy of language, Chomsky, computational complexity, or philosophy of mind. For an informal exposition of some of these ideas, see a short note on computational complexity and philosophy; and a sketch of the thesis from work on which I shall draw the material in the talk.

À propos.

Étiquettes.

philo, computational complexity, learning theory, Chomsky, linguistics, talks.

Mises à jour.

• J.P. Loo (24 novembre 2025): New Puddle abstract.
• J.P. Loo (14 novembre 2025): Puddle.