Odia ARC-Challenge: Parallel English–Odia Science Questions for Model Evaluation

Odia ARC-Challenge is a translation of the AI2 ARC-Challenge dataset, containing 2,590 multiple-choice science questions. The dataset, created by 'tripathysagar' and part of the OdiaBench collection, provides parallel English and Odia text for evaluating language models. It was last updated on Hugging Face on 2026-05-22.

Use Cases

• Benchmarking Odia language model performance on science question-answering tasks based on the parallel English-Odia text.
• Training or fine-tuning models for multilingual reasoning based on the multiple-choice science question format.
• Studying cross-lingual transfer learning for low-resource languages using the translated question stems and answer blocks.

Strengths

• Contains 2,590 total rows of data, split into test (1,172 rows), train (1,119 rows), and validation (299 rows) sets.
• Provides parallel English and Odia text for each question, enabling direct cross-lingual evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Translation of the AI2 ARC-Challenge dataset.

Collection Method

Likely involves machine or human translation of the original English questions and answer choices into Odia.

Freshness

Last updated 2026-05-22 07:23:34; freshness should be verified.