BEAR Dataset

Dataset to evaluate common factual knowledge in language models. This dataset was created as part of the paper "BEAR: A Unified Framework for Evaluating Relational Knowledge in Causal and Masked Language Models".

For more information visit the LM Pub Quiz website.

Dataset Structure

Data Instances

For each instance, there is a template string, a subject to be filled in, and a list of object strings (one of which is correct), as well as a template used to create the specific multiple choice item.

Data Fields

• composite_id: A unique string (each part of the dataset) identifying the item within the dataset. The string is made up from relation, item, and template_index.
• relation: An string identifiying the wikidata relation (between the subject and the answer options).
• item: The indext of the item within the relation.
• template_index: The index of the template used for this instance.
• template: The template used to create the text_options.
• subject: The subject of the relation.
• answer_options: Possible objects in the relation.
• correct: The index of the correct object (index of the answer in answer_options / correct statement in text_options).
• text_options: Alternative textual statements one of which is correct (identified by the index in correct).

Citation

When using the dataset or library, please cite the following paper:

@inproceedings{wiland-etal-2024-bear,
    title = "{BEAR}: A Unified Framework for Evaluating Relational Knowledge in Causal and Masked Language Models",
    author = "Wiland, Jacek  and
      Ploner, Max  and
      Akbik, Alan",
    editor = "Duh, Kevin  and
      Gomez, Helena  and
      Bethard, Steven",
    booktitle = "Findings of the Association for Computational Linguistics: NAACL 2024",
    month = jun,
    year = "2024",
    address = "Mexico City, Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-naacl.155/",
    doi = "10.18653/v1/2024.findings-naacl.155",
    pages = "2393--2411"
}