Inducing Language-Agnostic Multilingual Representations

Abstract

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.

arXiv comment*SEM2021 Camera Ready

References in corpus (4)

• Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection
• Multilingual Alignment of Contextual Word Representations
• Concatenated Power Mean Word Embeddings as Universal Cross-Lingual Sentence Representations
• Survey on the Use of Typological Information in Natural Language Processing

Cited by in corpus (5)

• Locating Language-Specific Information in Contextualized Embeddings
• On the ability of monolingual models to learn language-agnostic representations
• Changes in European Solidarity Before and During COVID-19: Evidence from a Large Crowd- and Expert-Annotated Twitter Dataset
• Improving Pretrained Cross-Lingual Language Models via Self-Labeled Word Alignment
• A Closer Look at Few-Shot Crosslingual Transfer: The Choice of Shots Matters