Learning Joint Semantic Parsers from Disjoint Data

TL;DR

This paper introduces a novel method for training semantic parsers across disjoint datasets with different formal representations by treating unobserved annotations as latent variables, leading to improved parsing performance.

Contribution

It proposes a joint learning approach that effectively handles disjoint data for semantic parsing, enhancing accuracy over existing methods.

Findings

Improved frame-semantic parsing accuracy

Enhanced semantic dependency parsing performance

Effective handling of disjoint datasets

Abstract

We present a new approach to learning semantic parsers from multiple datasets, even when the target semantic formalisms are drastically different, and the underlying corpora do not overlap. We handle such "disjoint" data by treating annotations for unobserved formalisms as latent structured variables. Building on state-of-the-art baselines, we show improvements both in frame-semantic parsing and semantic dependency parsing by modeling them jointly.