Montague Grammar Induction

TL;DR

This paper introduces a flexible computational framework for inducing combinatory categorial grammars from behavioral data, allowing detailed control over grammatical assumptions, demonstrated through a distributional analysis of English verb syntax.

Contribution

It presents a novel, customizable framework for grammar induction that incorporates semantic and syntactic constraints, advancing the modeling of natural language syntax.

Findings

Successfully induced grammars consistent with semantic theories

Analyzed English verb syntactic distributions using the framework

Demonstrated control over primitive types and combinators in grammar induction

Abstract

We propose a computational modeling framework for inducing combinatory categorial grammars from arbitrary behavioral data. This framework provides the analyst fine-grained control over the assumptions that the induced grammar should conform to: (i) what the primitive types are; (ii) how complex types are constructed; (iii) what set of combinators can be used to combine types; and (iv) whether (and to what) the types of some lexical items should be fixed. In a proof-of-concept experiment, we deploy our framework for use in distributional analysis. We focus on the relationship between s(emantic)-selection and c(ategory)-selection, using as input a lexicon-scale acceptability judgment dataset focused on English verbs' syntactic distribution (the MegaAcceptability dataset) and enforcing standard assumptions from the semantics literature on the induced grammar.