The University of Cambridge's Machine Translation Systems for WMT18

TL;DR

The paper describes the University of Cambridge's approach to machine translation for WMT18, combining various neural models and phrase-based systems to improve translation quality across multiple language pairs.

Contribution

It introduces a hybrid translation system that integrates diverse neural models with phrase-based SMT, achieving incremental improvements over existing Transformer ensembles.

Findings

Consistent gains over strong Transformer ensembles

Effective combination of recurrent, convolutional, and self-attention models

Successful integration of neural and phrase-based systems

Abstract

The University of Cambridge submission to the WMT18 news translation task focuses on the combination of diverse models of translation. We compare recurrent, convolutional, and self-attention-based neural models on German-English, English-German, and Chinese-English. Our final system combines all neural models together with a phrase-based SMT system in an MBR-based scheme. We report small but consistent gains on top of strong Transformer ensembles.