metadata

license: mit
language:
  - am
  - en
  - nso
  - ha
  - sw
  - yo
  - zu
size_categories:
  - 1K<n<10K
multilinguality:
  - multilingual
pretty_name: Uhura-Arc-Easy
language_details: am, en, ha, nso, sw, yo, zu
tags:
  - uhura
  - arc-easy
  - arc
task_categories:
  - multiple-choice
  - question-answering
task_ids:
  - multiple-choice-qa
configs:
  - config_name: am_multiple_choice
    data_files:
      - split: train
        path: am_train.json
      - split: test
        path: am_test.json
      - split: validation
        path: am_dev.json
  - config_name: en_multiple_choice
    data_files:
      - split: train
        path: en_train.json
      - split: test
        path: en_test.json
      - split: validation
        path: en_dev.json
  - config_name: ha_multiple_choice
    data_files:
      - split: train
        path: ha_train.json
      - split: test
        path: ha_test.json
      - split: validation
        path: ha_dev.json
  - config_name: nso_multiple_choice
    data_files:
      - split: train
        path: nso_train.json
      - split: test
        path: nso_test.json
      - split: validation
        path: nso_dev.json
  - config_name: nso_multiple_choice_unmatched
    data_files:
      - split: train
        path: nso_train_unmatched.json
      - split: test
        path: nso_test_unmatched.json
  - config_name: sw_multiple_choice
    data_files:
      - split: train
        path: sw_train.json
      - split: test
        path: sw_test.json
      - split: validation
        path: sw_dev.json
  - config_name: yo_multiple_choice
    data_files:
      - split: train
        path: yo_train.json
      - split: test
        path: yo_test.json
      - split: validation
        path: yo_dev.json
  - config_name: zu_multiple_choice
    data_files:
      - split: train
        path: zu_train.json
      - split: test
        path: zu_test.json

Dataset Card for Uhura-Arc-Easy

Dataset Summary

Uhura-ARC-Easy is a widely recognized scientific question answering benchmark composed of multiple-choice science questions derived from grade-school examinations that test various styles of knowledge and reasoning.

The original English version of the benchmark originates from Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge (Clark et al., 2018) and is divided into "Challenge" and "Easy" subsets, with 2,590 and 5,197 questions, respectively.

We translated a subset of Arc-Easy into 6 low-resource African languages using professional human translators. Relying on human translators for this evaluation increases confidence in the accuracy of the translations.

You can find more details about the dataset in our paper Uhura: A Benchmark for Evaluating Scientific Question Answering and Truthfulness in Low-Resource African Languages.

Languages

Uhura includes six widely spoken Sub-Saharan African languages, representing millions of speakers across the continent: Amharic, Hausa, Northern Sotho (Sepedi), Yoruba, and Zulu.

Dataset Structure

Data Instances

For the multiple_choice configuration, each instance contains a question and multiple-choice answer choices with corresponding labels and an answer key as well as an id.

{
    "id": "Mercury_7072328",
    "question": "Ìdí ago ẹnu ọ̀nà ní pàtó ni láti sọ agbára iná ẹ̀lẹ̀tírìkì di?",
    "choices": { 
        "label": [ "A", "B", "C", "D" ],
        "text": [ "Ohùn", "Ìrìn", "Agbára Iná", "Agbára Kẹ́míkà" ] 
    },
    "answerKey": "A",
}

Data Fields

id: a string feature.
question: a string feature.
choices: a dictionary feature containing:
- text: a string feature.
- label: a string feature.
answerKey: a string feature.

Data Splits

name	train	dev	test
am	656	92	491
ha	655	93	452
nso	440	3	509
sw	650	90	491
yo	659	93	494
zu	909	0	300

Note: Numbers vary across languages due to differences in the number of questions that can be translated for each language.

Dataset Creation

You can find more details about the dataset creation in our paper Uhura: A Benchmark for Evaluating Scientific Question Answering and Truthfulness in Low-Resource African Languages.

Curation Rationale

From the paper:

[Needs More Information]

Source Data

Initial Data Collection and Normalization

[Needs More Information]

Who are the source language producers?

[Needs More Information]

Annotations

Annotation process

[Needs More Information]

Who are the annotators?

[Needs More Information]

Personal and Sensitive Information

[Needs More Information]

Considerations for Using the Data

Social Impact of Dataset

[Needs More Information]

Discussion of Biases

[Needs More Information]

Other Known Limitations

[Needs More Information]

Additional Information

Dataset Curators

[Needs More Information]

Licensing Information

The Uhura-Arc-Easy dataset is licensed under the MIT License.

Citation

To cite Uhura, please use the following BibTeX entry:

@article{bayes2024uhurabenchmarkevaluatingscientific,
      title={Uhura: A Benchmark for Evaluating Scientific Question Answering and Truthfulness in Low-Resource African Languages}, 
      author={Edward Bayes and Israel Abebe Azime and Jesujoba O. Alabi and Jonas Kgomo and Tyna Eloundou and Elizabeth Proehl and Kai Chen and Imaan Khadir and Naome A. Etori and Shamsuddeen Hassan Muhammad and Choice Mpanza and Igneciah Pocia Thete and Dietrich Klakow and David Ifeoluwa Adelani},
      year={2024},
      eprint={2412.00948},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2412.00948}, 
}

Acknowledgements

This work was supported by OpenAI. We also want to thank our translators, whose contributions made this work possible.