AMR Dependency Parsing with a Typed Semantic Algebra

TL;DR

This paper introduces a semantic parser for AMR that leverages a typed semantic algebra and neural techniques, achieving state-of-the-art accuracy in dependency parsing of AMR graphs.

Contribution

It proposes a novel typed semantic algebra framework for AMR parsing, integrating neural supertagging and dependency parsing with principled type constraints.

Findings

Achieves state-of-the-art accuracy in AMR dependency parsing

Develops two approximative decoding algorithms

Outperforms strong baseline models

Abstract

We present a semantic parser for Abstract Meaning Representations which learns to parse strings into tree representations of the compositional structure of an AMR graph. This allows us to use standard neural techniques for supertagging and dependency tree parsing, constrained by a linguistically principled type system. We present two approximative decoding algorithms, which achieve state-of-the-art accuracy and outperform strong baselines.