First Komi-Yazva–Russian Parallel Corpus for Low-Resource LLM Translation

arXiv·low signal

Researchers release the first Komi-Yazva–Russian parallel corpus — 457 aligned sentence pairs from 74 narrative texts — with documented provenance, sentence-level alignment, and story IDs enabling leakage-aware evaluation of modern LLMs in an endangered, extremely low-resource setting. It establishes an explicit evaluation protocol for zero- and few-shot translation. A small but methodologically careful benchmark for the low-resource MT community.