Bilingual Synchronization: Restoring Translational Relationships with Editing Operations

TL;DR

This paper introduces a flexible edit-based approach for bilingual synchronization, transforming initial target sequences into accurate translations, and demonstrates its effectiveness across various practical settings compared to specialized systems.

Contribution

It presents a novel, generic edit-based system for bilingual synchronization that can match or outperform dedicated translation systems after fine-tuning.

Findings

Single edit-based system performs well across tasks

Fine-tuning improves translation accuracy

Outperforms dedicated systems in experiments

Abstract

Machine Translation (MT) is usually viewed as a one-shot process that generates the target language equivalent of some source text from scratch. We consider here a more general setting which assumes an initial target sequence, that must be transformed into a valid translation of the source, thereby restoring parallelism between source and target. For this bilingual synchronization task, we consider several architectures (both autoregressive and non-autoregressive) and training regimes, and experiment with multiple practical settings such as simulated interactive MT, translating with Translation Memory (TM) and TM cleaning. Our results suggest that one single generic edit-based system, once fine-tuned, can compare with, or even outperform, dedicated systems specifically trained for these tasks.