Jason Riggle just spoke at the linguistics colloquium of the University of Southern California, San Diego, on Generative models of variation within and between languages, the abstract of which follows:
In this talk, I evaluate three proposed mechanisms for generative
models of variation within language: (i) sampling from partially
specified grammars, (ii) grammars with noisy parameters, and (iii) probabilistic grammars. After describing qualitative differences among the types of variation they allow, I assess the ability of these three models to fit observed patterns of variation and ask whether there is (or could be) an empirical basis for selecting one of them. Finally, I ask the same questions in terms of the models’ ability to account for typological variation (i.e., the frequency with which various patterns are attested typologically).