Empirically-motivated generalizations of CCG semantic parsing learning algorithms

Abstract

Learning algorithms for semantic parsing have improved drastically over the past decade, as steady improvements on benchmark datasets have shown. In this paper we investigate whether they can generalize to a novel biomedical dataset that differs in important respects from the traditional geography and air travel benchmark datasets. Empirical results for two state-of-the-Art PCCG semantic parsers indicates that learning algorithms are sensitive to the kinds of semantic and syntactic constructions used in a domain. In response, we develop a novel learning algorithm that can produce an effective semantic parser for geography, as well as a much better semantic parser for the biomedical dataset. © 2014 Association for Computational Linguistics.