Unsupervised Cross-lingual Representation Learning at Scale

TL;DR

This paper demonstrates that large-scale pretraining of multilingual language models significantly improves cross-lingual transfer performance across many languages and tasks, especially benefiting low-resource languages.

Contribution

It introduces XLM-R, a Transformer-based multilingual model trained on over two terabytes of data, achieving state-of-the-art results without sacrificing per-language performance.

Findings

XLM-R outperforms mBERT on multiple benchmarks.

Significant gains for low-resource languages.

Multilingual modeling can match monolingual performance.

Abstract

This paper shows that pretraining multilingual language models at scale leads to significant performance gains for a wide range of cross-lingual transfer tasks. We train a Transformer-based masked language model on one hundred languages, using more than two terabytes of filtered CommonCrawl data. Our model, dubbed XLM-R, significantly outperforms multilingual BERT (mBERT) on a variety of cross-lingual benchmarks, including +14.6% average accuracy on XNLI, +13% average F1 score on MLQA, and +2.4% F1 score on NER. XLM-R performs particularly well on low-resource languages, improving 15.7% in XNLI accuracy for Swahili and 11.4% for Urdu over previous XLM models. We also present a detailed empirical analysis of the key factors that are required to achieve these gains, including the trade-offs between (1) positive transfer and capacity dilution and (2) the performance of high and low resource languages at scale. Finally, we show, for the first time, the possibility of multilingual modeling without sacrificing per-language performance; XLM-R is very competitive with strong monolingual models on the GLUE and XNLI benchmarks. We will make our code, data and models publicly available.