Exploiting Semantics in Neural Machine Translation with Graph Convolutional Networks
This work addresses the challenge of meaning preservation in machine translation for NLP researchers, though it is incremental as it builds on existing methods with a novel semantic integration.
The paper tackled the problem of improving neural machine translation by incorporating semantic-role representations to enforce meaning preservation, achieving improvements in BLEU scores over linguistic-agnostic and syntax-aware versions on the English-German language pair.
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.