0

Bridging the Data Gap between Training and Inference for Unsupervised Neural Machine Translation

An online self-training approach for unsupervised neural machine translation reduces style and content gaps by using pseudo parallel data to mimic the inference scenario, outperforming strong baselines.

Year
2022
Venue
ACL 2022 5
Authors
5
Hosting
Abstract onlyARXIV-DEFAULT

Cite

Notes

Only stored in your browser.

Attribution

Abstract & full text
arxiv.org/abs/2203.08394v4ARXIV-DEFAULT
TL;DR
Semantic Scholar
Attribution policy →

Abstract

Back-translation is a critical component of Unsupervised Neural Machine Translation (UNMT), which generates pseudo parallel data from target monolingual data. A UNMT model is trained on the pseudo parallel data with translated source, and translates natural source sentences in inference. The source discrepancy between training and inference hinders the translation performance of UNMT models. By carefully designing experiments, we identify two representative characteristics of the data gap in source: (1) style gap (i.e., translated vs. natural text style) that leads to poor generalization capability; (2) content gap that induces the model to produce hallucination content biased towards the target language. To narrow the data gap, we propose an online self-training approach, which simultaneously uses the pseudo parallel data {natural source, translated target} to mimic the inference scenario. Experimental results on several widely-used language pairs show that our approach outperforms two strong baselines (XLM and MASS) by remedying the style and content gaps.

Authors

5