| Research in semantic parsing has focused on developing computational systems capable of simultaneously performing syntactic, i.e. structural, and semantic, i.e., meaning-based, analyses of given sentences. We present an implementation of a semantic parsing system using a constraint-based grammatical formalism called Lexicalized Well-Founded Grammars (LWFGs). LWFGs are a type of Definite Clause Grammars, and use an ontology-based framework to represent syntactico-semantic information in the form of compositional and interpretation constraints. What makes LWFGs particularly interesting is the fact that these are the only constraint-based grammars that are provably learnable. Furthermore, there exist tractable learning algorithms for LWFGs, which make these especially useful in resource-poor language settings.;In this thesis, we present a revised parsing implementation for Lexicalized Well-Founded Grammars. Previous work implemented semantic parsers using Prolog, a declarative language, which is slow and does not allow for an easy extension to a stochastic parsing framework. Our implementation utilizes Python's Natural Language Toolkit which not only allows us to easily interface our work with the natural language processing community, but also allows for a future possibility of extending the parser to support broad-coverage and stochastic parsing. |
