Learning Algebraic Recombination for Compositional Generalization

TL;DR

This paper introduces LeAR, a neural model that learns algebraic recombination to improve compositional generalization in semantic parsing by modeling the task as a homomorphism between syntactic and semantic algebras.

Contribution

LeAR is the first end-to-end neural model to explicitly learn algebraic recombination for semantic parsing, bridging syntax and semantics through a homomorphic approach.

Findings

LeAR outperforms existing models on two compositional generalization benchmarks.

The model effectively learns latent syntax and semantic operations.

Results demonstrate improved generalization to complex compositional structures.

Abstract

Neural sequence models exhibit limited compositional generalization ability in semantic parsing tasks. Compositional generalization requires algebraic recombination, i.e., dynamically recombining structured expressions in a recursive manner. However, most previous studies mainly concentrate on recombining lexical units, which is an important but not sufficient part of algebraic recombination. In this paper, we propose LeAR, an end-to-end neural model to learn algebraic recombination for compositional generalization. The key insight is to model the semantic parsing task as a homomorphism between a latent syntactic algebra and a semantic algebra, thus encouraging algebraic recombination. Specifically, we learn two modules jointly: a Composer for producing latent syntax, and an Interpreter for assigning semantic operations. Experiments on two realistic and comprehensive compositional generalization benchmarks demonstrate the effectiveness of our model. The source code is publicly available at https://github.com/microsoft/ContextualSP.