tripathysagar/odia-arc

Odia ARC-Challenge

Odia translation of AI2 ARC-Challenge. Multiple-choice science questions with English choices and gold labels, plus Odia translations of the question stem and formatted answer block.

Part of **OdiaBench** — parallel English–Odia benchmark translations for evaluating Odia-capable language models.

Splits

Total rows: 2,590

Schema

Translation methodology

OdiaBench translations were produced via Cursor subagents (Composer 2.5-fast) using the v3 translation prompt in `cursor_translate/odia_trans`:

1. 1.Download upstream split and shard into 30-row JSONL batches
2. 2.Translate question and answer into odia_question / odia_answer
3. 3.Apply structural fixers, grammar audit (7 rules), and quality evaluation
4. 4.Merge batches, dedupe by id, and export parquet splits

Translation rules preserve numbers as digits, keep benchmark-specific structural tokens (e.g. A:–D: for ARC, _ blank for Winogrande, <<calc>> / #### N for GSM8K), and transliterate proper nouns into Odia script.

Usage

from datasets import load_dataset

ds = load_dataset("tripathysagar/odia-arc")
print(ds)
print(ds["validation"][0]["odia_question"])

Source dataset

• —Original (English): allenai/ai2_arc

Citation

If you use this dataset, please cite OdiaBench and the upstream source:

@misc{odiabench2026,
  title={OdiaBench: English-Odia Benchmark Translations},
  author={OdiaBench Contributors},
  year={2026},
  howpublished={\url{https://github.com/tripathysagar/OdiaBench}}
}

Upstream source:

@article{clark2018think,
  title={Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge},
  author={Peter Clark and Isaac Cowhey and Oren Etzioni and others},
  journal={arXiv:1803.05457},
  year={2018}
}