On the Language Neutrality of Pre-trained Multilingual Representations
This addresses the problem of cross-lingual representation quality for multilingual NLP tasks, offering incremental improvements in language neutrality.
The paper investigates the language-neutrality of multilingual contextual embeddings, finding they are more language-neutral and informative than aligned static embeddings but only moderately so by default, and proposes two methods to improve neutrality while achieving state-of-the-art accuracy on language identification and matching statistical methods for word alignment without parallel data.
Multilingual contextual embeddings, such as multilingual BERT and XLM-RoBERTa, have proved useful for many multi-lingual tasks. Previous work probed the cross-linguality of the representations indirectly using zero-shot transfer learning on morphological and syntactic tasks. We instead investigate the language-neutrality of multilingual contextual embeddings directly and with respect to lexical semantics. Our results show that contextual embeddings are more language-neutral and, in general, more informative than aligned static word-type embeddings, which are explicitly trained for language neutrality. Contextual embeddings are still only moderately language-neutral by default, so we propose two simple methods for achieving stronger language neutrality: first, by unsupervised centering of the representation for each language and second, by fitting an explicit projection on small parallel data. Besides, we show how to reach state-of-the-art accuracy on language identification and match the performance of statistical methods for word alignment of parallel sentences without using parallel data.