Neural Machine Translation Leveraging Phrase-based Models in a Hybrid Search

TL;DR

This paper presents a hybrid search method for neural machine translation that integrates phrase-based models to improve translation quality, achieving up to 2.3% BLEU score gains over standard NMT systems.

Contribution

It introduces a novel hybrid search approach combining NMT and phrase-based SMT models to enhance translation accuracy.

Findings

Up to 2.3% BLEU improvement on German-English and English-Russian tasks.

Effective integration of phrase-based features into neural search.

Demonstrates benefits of hybrid models in NMT quality.

Abstract

In this paper, we introduce a hybrid search for attention-based neural machine translation (NMT). A target phrase learned with statistical MT models extends a hypothesis in the NMT beam search when the attention of the NMT model focuses on the source words translated by this phrase. Phrases added in this way are scored with the NMT model, but also with SMT features including phrase-level translation probabilities and a target language model. Experimental results on German->English news domain and English->Russian e-commerce domain translation tasks show that using phrase-based models in NMT search improves MT quality by up to 2.3% BLEU absolute as compared to a strong NMT baseline.