Another model not for the learning of language

It is laudable for Yang and Piantadosi (PNAS, 2022) to tackle the important question of how language is generated and learned. Their reported success in learning a range of grammars from small amounts of data is striking, especially since some of these grammars belong to classes which are provably unlearnable under such conditions (Kearns & Valiant, 1994). However, their project is severely undermined by their evaluation method and by their conception of language.