Disentangling Linguistic Relatedness from Task Alignment in Cross-Lingual Transfer

Document Type

Conference Proceeding

Role

Author

Published In

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics

Publisher

Association for Computational Linguistics

First Page

683

Last Page

700

Publication Date

7-2026

Abstract

We study cross-lingual transfer by fine-tuning seven large language models (4B–671B parameters) on Arabic and evaluating zero-shot reading comprehension on Semitic languages and non-Semitic controls. Across dense and Mixture-of-Experts architectures, we find no evidence of Semitic-specific transfer: models with weak baselines improve dramatically across all languages, while strong-baseline models show only marginal gains regardless of language family. A chain-of-thought ablation reinforces this finding: the same models that benefit most from fine-tuning benefit equally from inference-time reasoning, suggesting both mechanisms address task-format alignment rather than cross-lingual knowledge transfer.

Share

COinS