A Study on Self-attention Mechanism for AMR-to-text Generation

Abstract

Introduced by Vaswani et al., transformer architecture, with the effective use of self-attention mechanism, has shown outstanding performance in translating sequence of text from one language to another. In this paper, we conduct experiments using the self-attention in converting an abstract meaning representation (AMR) graph, a semantic representation, into a natural language sentence, also known as the task of AMR-to-text generation. On the benchmark dataset for this task, we obtain promising results comparing to existing deep learning methods in the literature.