Ensemble Self-Training for Unsupervised Machine Translation

TL;DR

This paper introduces an ensemble self-training framework for unsupervised neural machine translation that leverages model diversity and ensemble decoding to generate synthetic data, significantly improving translation quality.

Contribution

It proposes a novel ensemble-driven self-training method for UNMT that enhances performance by using structured model diversity and ensemble decoding for synthetic data generation.

Findings

Achieved 1.7 chrF improvement in English translation

Achieved 0.67 chrF improvement into English

Significant performance gains over single-model UNMT baselines

Abstract

We present an ensemble-driven self-training framework for unsupervised neural machine translation (UNMT). Starting from a primary language pair, we train multiple UNMT models that share the same translation task but differ in an auxiliary language, inducing structured diversity across models. We then generate pseudo-translations for the primary pair using token-level ensemble decoding, averaging model predictions in both directions. These ensemble outputs are used as synthetic parallel data to further train each model, allowing the models to improve via shared supervision. At deployment time, we select a single model by validation performance, preserving single-model inference cost. Experiments show statistically significant improvements over single-model UNMT baselines, with mean gains of 1.7 chrF when translating from English and 0.67 chrF when translating into English.