In this talk, I discuss a recent controversy in the philosophies of AI and cognitive science. The issue of ‘impossible grammars’ has some pedigree in theoretical linguistics as an offshoot of both the Universal Grammar postulate and the Principles and Parameters framework of the early 1980’s. Recently, the idea that human language is constrained by particular kinds of rules has been used to challenge the emerging optimism concerning the role of language models in cognitive science.

I offer some caution to both sides of an emerging debate, arguing the language models are neither irrelevant due to their alleged unconstrained learning nor necessarily aligned when they show some human-like learning biases.