EvenVLM: A Text and Multimodal Dataset for the Even Language with Vision-Language Fine-Tuning
Fine-tuning lifted visual question-answering accuracy in Even to 65%, from zero for the base model.
Even is an endangered Northern Tungusic language spoken by approximately 6,800 people in Siberia. The researchers assembled 50,521 Even–Russian dictionary and phrasebook entries, then generated 107,543 synthetic text training examples. Expanding the tokenizer cut average tokens per word by 27%, while the best text configuration reached 11.4% exact match. The authors release the datasets, evaluation benchmarks and training pipeline.
HF Daily Papers' note
Even is an endangered Northern Tungusic language spoken by approximately 6,800 people in Siberia. The researchers assembled 50,521 Even–Russian dictionary and phrasebook entries, then generated 107,543 synthetic text training examples. Expanding the tokenizer cut average tokens per word by 27%, while the best text configuration reached 11.4% exact match. The authors release the datasets, evaluation benchmarks and training pipeline.
HF Daily Papers' note
score 4