Cross-Lingual Transfer for Machine Translation in Turkic Languages

arXiv:2607.29355v1 Announce Type: cross Abstract: Cross-lingual transfer is central to low-resource machine translation, but its behavior within closely related language families remains insufficiently characterized. We study transfer among five Turkic languages; Turkish, Azerbaijani, Uzbek, Kazakh, and Kyrgyz; using pairwise transfer matrices. In this setting, each model is fine-tuned with one transfer source and evaluated on a different transfer target while the translation target remains the same. Across mT5 experiments, we find that transfer is strongest between closely related Turkic pairs, especially Turkish-Azerbaijani and Kazakh-Kyrgyz. We also show that transfer direction matters, and that the same transfer source-transfer target pair can behave differently when the translation target changes. Latinization improves BLEU and chrF in several script-mismatched settings, but its effect is not uniform across metrics. Additional analyses show that transfer sources are mostly stable across different datasets and model settings.

Related

• Why Low-Resource NLP Needs More Than Cross-Lingual Transfer: Lessons Learned from Luxembourgish
• Extracting Small Translation Specialists from LLMs by Aggressively Pruning Experts
• Should We be Pedantic About Reasoning Errors in Machine Translation?
• Shared Doubt: Zero-shot Cross-Lingual Confidence Estimation for Language Models

Source: arXiv cs.AI | 2026-08-03