XLM-E: Cross-lingual Language Model Pre-training via ELECTRA

TL;DR

XLM-E introduces ELECTRA-style pre-training tasks for cross-lingual models, achieving better transferability and performance on understanding tasks with reduced computational costs.

Contribution

The paper proposes novel ELECTRA-style pre-training tasks for cross-lingual models, enhancing transferability and efficiency over previous methods.

Findings

Outperforms baseline models on cross-lingual tasks

Requires less computation than existing models

Shows improved cross-lingual transferability

Abstract

In this paper, we introduce ELECTRA-style tasks to cross-lingual language model pre-training. Specifically, we present two pre-training tasks, namely multilingual replaced token detection, and translation replaced token detection. Besides, we pretrain the model, named as XLM-E, on both multilingual and parallel corpora. Our model outperforms the baseline models on various cross-lingual understanding tasks with much less computation cost. Moreover, analysis shows that XLM-E tends to obtain better cross-lingual transferability.