Language Distances are Practical for Equitable Cross-Lingual Transfer

2026-09-26

AI Summary

This study addresses the challenges of source language selection and fairness in low-resource scenarios within cross-lingual transfer. It presents the first systematic evaluation of language distance-based source language ranking paradigms from a fairness perspective. By integrating multilingual models, training-free composite distance metrics, and supervised learning-to-rank algorithms, this work thoroughly investigates the trade-off between task performance and resource inequality. The findings validate the effectiveness of language distance as a criterion for source language selection and demonstrate that trained rankers offer significant advantages in mitigating inequality. Accordingly, we recommend prioritizing trained rankers when evaluation data are available, and resorting to composite distance metrics otherwise, thereby achieving an optimal balance between performance and fairness in cross-lingual transfer.

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.

Problem

• cross-lingual transfer
• language distances
• source language selection
• equity
• low-resource languages

Innovation

• Cross-lingual Transfer
• Language Distances
• Equity-focused Evaluation
• Composite Distances
• Trained Rankers