Simpler Context-Dependent Logical Forms via Model Projections

TL;DR

This paper introduces a method for learning context-dependent logical forms by projecting complex models onto simpler, faster models, and demonstrates its effectiveness with new datasets and a novel parser.

Contribution

It proposes a novel projection approach to simplify context-dependent semantic parsing, enabling faster learning and inference, and provides new datasets and a parser for this task.

Findings

Simpler models are faster and surprisingly effective.

Projections can bootstrap the full model.

New datasets and a left-to-right parser are introduced.

Abstract

We consider the task of learning a context-dependent mapping from utterances to denotations. With only denotations at training time, we must search over a combinatorially large space of logical forms, which is even larger with context-dependent utterances. To cope with this challenge, we perform successive projections of the full model onto simpler models that operate over equivalence classes of logical forms. Though less expressive, we find that these simpler models are much faster and can be surprisingly effective. Moreover, they can be used to bootstrap the full model. Finally, we collected three new context-dependent semantic parsing datasets, and develop a new left-to-right parser.