AMR Parsing as Graph Prediction with Latent Alignment

TL;DR

This paper presents a neural AMR parser that models alignments as latent variables within a joint probabilistic framework, achieving state-of-the-art results by using variational auto-encoding and continuous relaxation techniques.

Contribution

It introduces a novel joint modeling approach for AMR parsing that integrates alignment prediction directly into the parsing process, improving accuracy.

Findings

Achieves 74.4% F1 score on the standard benchmark.

Joint modeling outperforms pipeline approaches.

Uses variational auto-encoding with continuous relaxation for inference.

Abstract

Abstract meaning representations (AMRs) are broad-coverage sentence-level semantic representations. AMRs represent sentences as rooted labeled directed acyclic graphs. AMR parsing is challenging partly due to the lack of annotated alignments between nodes in the graphs and words in the corresponding sentences. We introduce a neural parser which treats alignments as latent variables within a joint probabilistic model of concepts, relations and alignments. As exact inference requires marginalizing over alignments and is infeasible, we use the variational auto-encoding framework and a continuous relaxation of the discrete alignments. We show that joint modeling is preferable to using a pipeline of align and parse. The parser achieves the best reported results on the standard benchmark (74.4% on LDC2016E25).