Understanding Cross-Lingual Alignment -- A Survey

TL;DR

This survey comprehensively reviews techniques for cross-lingual alignment in multilingual models, categorizing methods, summarizing key insights, and discussing applications across different model architectures.

Contribution

It provides a taxonomy of cross-lingual alignment methods, summarizes extensive research findings, and discusses future directions for various model types.

Findings

Effective trade-off between language-neutral and language-specific info is crucial.

Various techniques have been developed with differing strengths and limitations.

Insights can be extended beyond encoder models to other architectures.

Abstract

Cross-lingual alignment, the meaningful similarity of representations across languages in multilingual language models, has been an active field of research in recent years. We survey the literature of techniques to improve cross-lingual alignment, providing a taxonomy of methods and summarising insights from throughout the field. We present different understandings of cross-lingual alignment and their limitations. We provide a qualitative summary of results from a large number of surveyed papers. Finally, we discuss how these insights may be applied not only to encoder models, where this topic has been heavily studied, but also to encoder-decoder or even decoder-only models, and argue that an effective trade-off between language-neutral and language-specific information is key.