Language distances improve fair cross-lingual model transfers
Language Distances are Practical for Equitable Cross-Lingual Transfer
Computation and Language
Summary
Choosing the best language to train a computer model for translating or understanding another language can be very hard, especially for languages with little data. The authors studied ways to measure how 'close' two languages are and how helpful that is for picking good training languages. They found that combining several distance measures or using learned rankings makes transfers fairer and more reliable across different tasks and languages. This helps technology work better for less common languages without needing lots of extra testing.
What this means in practice
- •For machine learning engineers: Select better source languages when building multilingual models to improve accuracy and fairness for low-resource target languages.
- •For natural language processing developers: Use composite language distances to rank training languages without needing task-specific performance data, speeding up development for diverse languages.
Authors
York Hay Ng, Razan Ahsan Rifandi, Aditya Khan, En-Shiun Annie Lee
Abstract
Cross-lingual transfer is strongly conditional on how the source language is chosen, but it is impractical to determine the best candidate source for every target language, especially for low-resource target languages. Language distances are widely used to rank candidate sources due to their correlation with transfer efficacy and applicability in resource-sparse settings. However, the reliability of distance-based rankers across tasks and resource levels remains underexplored. We therefore present the first equity-focused evaluation of paradigms for ranking source languages, studying resource-level inequality and task inequality across ten cross-lingual tasks and two multilingual models. While both inequalities are most pronounced for individual language distances and an English-always baseline, they are substantially reduced by training-free composite distances, and nearly eliminated by trained rankers. We further demonstrate the reliability of rankers using language distances compared to rankers using language model internals. Overall, we find that language distances provide a practical basis for equitable and performant transfer language selection. We recommend using trained rankers when task-specific transfer evaluations are available, and composite distances otherwise.