Multilingual Alignment of Contextual Word Representations

TL;DR

This paper introduces procedures for evaluating and improving multilingual contextual embedding alignment, significantly enhancing zero-shot transfer performance of multilingual BERT across languages and analyzing its systematic deficiencies.

Contribution

It presents novel alignment procedures for multilingual BERT, demonstrating improved zero-shot transfer and a new contextual word retrieval task for measuring alignment.

Findings

Improved zero-shot performance on XNLI after alignment

Alignment correlates with downstream transfer effectiveness

Identified systematic deficiencies in BERT's multilingual alignment

Abstract

We propose procedures for evaluating and strengthening contextual embedding alignment and show that they are useful in analyzing and improving multilingual BERT. In particular, after our proposed alignment procedure, BERT exhibits significantly improved zero-shot performance on XNLI compared to the base model, remarkably matching pseudo-fully-supervised translate-train models for Bulgarian and Greek. Further, to measure the degree of alignment, we introduce a contextual version of word retrieval and show that it correlates well with downstream zero-shot transfer. Using this word retrieval task, we also analyze BERT and find that it exhibits systematic deficiencies, e.g. worse alignment for open-class parts-of-speech and word pairs written in different scripts, that are corrected by the alignment procedure. These results support contextual alignment as a useful concept for understanding large multilingual pre-trained models.