Graph Transformer for Graph-to-Sequence Learning

TL;DR

This paper introduces Graph Transformer, a novel model for graph-to-sequence learning that uses explicit relation encoding to enable direct communication between distant nodes, improving performance on text generation and translation tasks.

Contribution

The paper presents a new Graph Transformer model that enhances global graph structure modeling by allowing direct node communication, outperforming existing graph neural network approaches.

Findings

Achieved 27.4 BLEU on AMR-to-text generation, surpassing previous results.

Set new state-of-the-art BLEU scores on syntax-based translation tasks.

Demonstrated improved performance over existing models in multiple graph-to-sequence applications.

Abstract

The dominant graph-to-sequence transduction models employ graph neural networks for graph representation learning, where the structural information is reflected by the receptive field of neurons. Unlike graph neural networks that restrict the information exchange between immediate neighborhood, we propose a new model, known as Graph Transformer, that uses explicit relation encoding and allows direct communication between two distant nodes. It provides a more efficient way for global graph structure modeling. Experiments on the applications of text generation from Abstract Meaning Representation (AMR) and syntax-based neural machine translation show the superiority of our proposed model. Specifically, our model achieves 27.4 BLEU on LDC2015E86 and 29.7 BLEU on LDC2017T10 for AMR-to-text generation, outperforming the state-of-the-art results by up to 2.2 points. On the syntax-based translation tasks, our model establishes new single-model state-of-the-art BLEU scores, 21.3 for English-to-German and 14.1 for English-to-Czech, improving over the existing best results, including ensembles, by over 1 BLEU.