Multilingual Denoising Pre-training for Neural Machine Translation

TL;DR

This paper introduces mBART, a multilingual denoising auto-encoder pre-training method that significantly improves machine translation performance across various tasks by pre-training a full sequence-to-sequence model on large monolingual corpora.

Contribution

It presents the first complete sequence-to-sequence pre-training approach for multiple languages, enabling direct fine-tuning for diverse MT tasks without task-specific modifications.

Findings

Up to 12 BLEU points improvement in low-resource MT

Over 5 BLEU points gain in document-level and unsupervised models

Enables transfer to language pairs without bi-text data

Abstract

This paper demonstrates that multilingual denoising pre-training produces significant performance gains across a wide variety of machine translation (MT) tasks. We present mBART -- a sequence-to-sequence denoising auto-encoder pre-trained on large-scale monolingual corpora in many languages using the BART objective. mBART is one of the first methods for pre-training a complete sequence-to-sequence model by denoising full texts in multiple languages, while previous approaches have focused only on the encoder, decoder, or reconstructing parts of the text. Pre-training a complete model allows it to be directly fine tuned for supervised (both sentence-level and document-level) and unsupervised machine translation, with no task-specific modifications. We demonstrate that adding mBART initialization produces performance gains in all but the highest-resource settings, including up to 12 BLEU points for low resource MT and over 5 BLEU points for many document-level and unsupervised models. We also show it also enables new types of transfer to language pairs with no bi-text or that were not in the pre-training corpus, and present extensive analysis of which factors contribute the most to effective pre-training.