Cooperative Learning of Disjoint Syntax and Semantics

TL;DR

This paper introduces a recursive, cooperative model that learns syntax and semantics jointly without supervision, achieving high accuracy on mathematical expressions and competitive results on natural language tasks.

Contribution

The proposed model uniquely combines separate syntax and semantics modules trained cooperatively, enabling out-of-domain generalization without linguistic supervision.

Findings

Near-perfect accuracy on mathematical expression parsing

Competitive performance on natural language inference and sentiment analysis

Effective out-of-domain generalization with little performance loss

Abstract

There has been considerable attention devoted to models that learn to jointly infer an expression's syntactic structure and its semantics. Yet, \citet{NangiaB18} has recently shown that the current best systems fail to learn the correct parsing strategy on mathematical expressions generated from a simple context-free grammar. In this work, we present a recursive model inspired by \newcite{ChoiYL18} that reaches near perfect accuracy on this task. Our model is composed of two separated modules for syntax and semantics. They are cooperatively trained with standard continuous and discrete optimization schemes. Our model does not require any linguistic structure for supervision and its recursive nature allows for out-of-domain generalization with little loss in performance. Additionally, our approach performs competitively on several natural language tasks, such as Natural Language Inference or Sentiment Analysis.