Exploiting Semantics In Neural Machine Translation With Graph Convolutional Networks

ARXIV  ·  2018  ·  Impact 5

Diego Marcheggiani; Jasmijn Bastings; Ivan Titov;

HIGHLIGHT

In this work, we are the first to incorporate information about predicate-argument structure of source sentences (namely, semantic-role representations) into neural machine translation.

RELATED

Papers      Patents     Grants     Venues     Experts    Cite    Share    Save

ABSTRACT

Semantic representations have long been argued as potentially useful for enforcing meaning preservation and improving generalization performance of machine translation methods. In this work, we are the first to incorporate information about predicate-argument structure of source sentences (namely, semantic-role representations) into neural machine translation. We use Graph Convolutional Networks (GCNs) to inject a semantic bias into sentence encoders and achieve improvements in BLEU scores over the linguistic-agnostic and syntax-aware versions on the English--German language pair.

 Paraphrase     Questionfy     Wikify     Simplify     Summarize

Download