An Operation Sequence Model for Explainable Neural Machine Translation

TL;DR

This paper introduces an explainable neural machine translation model that generates translations by walking through the source sentence with explicit word alignment, improving interpretability and maintaining competitive translation quality.

Contribution

It proposes a novel operation sequence model for NMT that provides explicit alignments and explainability, outperforming standard models on several language pairs.

Findings

Outperforms plain text systems in BLEU score on Japanese-English and Portuguese-English.

Maintains near state-of-the-art BLEU scores on Spanish-English.

Provides explicit word alignment information for better explainability.

Abstract

We propose to achieve explainable neural machine translation (NMT) by changing the output representation to explain itself. We present a novel approach to NMT which generates the target sentence by monotonically walking through the source sentence. Word reordering is modeled by operations which allow setting markers in the target sentence and move a target-side write head between those markers. In contrast to many modern neural models, our system emits explicit word alignment information which is often crucial to practical machine translation as it improves explainability. Our technique can outperform a plain text system in terms of BLEU score under the recent Transformer architecture on Japanese-English and Portuguese-English, and is within 0.5 BLEU difference on Spanish-English.