Cross-lingual representations have the potential to make NLP techniques available to the vast majority of languages in the world. However, they currently require large pretraining corpora or access to typologically similar languages. In this work, we address these obstacles by removing language identity signals from multilingual embeddings. We examine three approaches for this: (i) re-aligning the vector spaces of target languages (all together) to a pivot source language; (ii) removing language-specific means and variances, which yields better discriminativeness of embeddings as a by-product; and (iii) increasing input similarity across languages by removing morphological contractions and sentence reordering. We evaluate on XNLI and reference-free MT across 19 typologically diverse languages. Our findings expose the limitations of these approaches-unlike vector normalization, vector space re-alignment and text normalization do not achieve consistent gains across encoders and languages. Due to the approaches' additive effects, their combination decreases the cross-lingual transfer gap by 8.9 points (m-BERT) and 18.2 points (XLM-R) on average across all tasks and languages, however. Our code and models are publicly available.
|Title of host publication||*SEM 2021 - 10th Conference on Lexical and Computational Semantics, Proceedings of the Conference|
|Editors||Lun-Wei Ku, Vivi Nastase, Ivan Vulic|
|Number of pages||12|
|Publisher||Association for Computational Linguistics, ACL Anthology|
|Publication status||Published - 2021|
|Event||10th Conference on Lexical and Computational Semantics, *SEM 2021 - Virtual, Bangkok, Thailand|
Duration: 5 Aug 2021 → 6 Aug 2021
|Conference||10th Conference on Lexical and Computational Semantics, *SEM 2021|
|Period||05/08/2021 → 06/08/2021|
|Sponsor||ACL Special Interest Group on the Lexicon, SIGLEX|
|Series||*SEM 2021 - 10th Conference on Lexical and Computational Semantics, Proceedings of the Conference|
Bibliographical noteFunding Information:
We thank the anonymous reviewers for their insightful comments and suggestions, which greatly improved the final version of the paper. This work has been supported by the German Research Foundation as part of the Research Training Group Adaptive Preparation of Information from Heterogeneous Sources (AIPHES) at the Technische Uni-versität Darmstadt under grant No. GRK 1994/1, as well as by the Swedish Research Council under grant agreement No 2019-04129.
© 2021 Lexical and Computational Semantics