Emerging Cross-lingual Structure in Pretrained Language Models

TL;DR

This paper investigates why multilingual masked language models excel at cross-lingual transfer, revealing that shared parameters and latent symmetries in embedding spaces enable effective transfer even without shared vocabularies or similar domains.

Contribution

The study demonstrates that cross-lingual transfer is possible without shared vocabularies or domains, due to shared top-layer parameters and universal embedding symmetries learned during training.

Findings

Transfer occurs without shared vocabularies.

Shared top-layer parameters are crucial.

Universal embedding symmetries are discovered during training.

Abstract

We study the problem of multilingual masked language modeling, i.e. the training of a single model on concatenated text from multiple languages, and present a detailed study of several factors that influence why these models are so effective for cross-lingual transfer. We show, contrary to what was previously hypothesized, that transfer is possible even when there is no shared vocabulary across the monolingual corpora and also when the text comes from very different domains. The only requirement is that there are some shared parameters in the top layers of the multi-lingual encoder. To better understand this result, we also show that representations from independently trained models in different languages can be aligned post-hoc quite effectively, strongly suggesting that, much like for non-contextual word embeddings, there are universal latent symmetries in the learned embedding spaces. For multilingual masked language modeling, these symmetries seem to be automatically discovered and aligned during the joint training process.