Authors
Alexander Clark
Year
2021
Abstract
This paper presents an algorithm for strong learning of probabilistic multiple context free grammars from a positive sample of strings generated by the grammars. The algorithm is shown to be a consistent estimator for a class of well-nested grammars, given by explicit structural conditions on the underlying grammar, and for grammars in this class is guaranteed to converge to a grammar which is isomorphic to the original, not just one that generates the same set of strings.