Testing Distributional Properties of Context-Free Grammars
; Proceedings of The 13th International Conference on Grammatical Inference, PMLR 57:42-53, 2017.
Recent algorithms for distributional learning of context-free grammars can learn all languages defined by grammars that have certain distributional properties: the finite kernel property (FKP) and the finite context property (FCP). In this paper we present some algorithms for approximately determining whether a given grammar has one of these properties. We then present the results of some experiments that indicate that with randomly generated context-free grammars in Chomsky normal form, which generate infinite languages and are derivationally sparse, nearly all grammars have the finite kernel property, whereas the finite context property is much less common.