Learning compositional structures for semantic graph parsing

TL;DR

This paper introduces a neural latent-variable model for AM dependency parsing that learns compositional structures directly from graphs, reducing manual effort and maintaining high accuracy.

Contribution

It proposes a novel neural latent-variable approach for training AM dependency parsers without explicit tree annotations, simplifying the process for new semantic graph datasets.

Findings

Model learns linguistic phenomena autonomously.

Achieves comparable accuracy to supervised methods.

Reduces manual heuristic requirements.

Abstract

AM dependency parsing is a method for neural semantic graph parsing that exploits the principle of compositionality. While AM dependency parsers have been shown to be fast and accurate across several graphbanks, they require explicit annotations of the compositional tree structures for training. In the past, these were obtained using complex graphbank-specific heuristics written by experts. Here we show how they can instead be trained directly on the graphs with a neural latent-variable model, drastically reducing the amount and complexity of manual heuristics. We demonstrate that our model picks up on several linguistic phenomena on its own and achieves comparable accuracy to supervised training, greatly facilitating the use of AM dependency parsing for new sembanks.