README.md

---

Dataset Card for KAMEL: Knowledge Analysis with Multitoken Entities in Language Models

Table of Contents

• Dataset Description
  • Dataset Summary
  • Supported Tasks and Leaderboards
  • Languages
• Dataset Structure
  • Data Instances
  • Data Fields
  • Data Splits
• Dataset Creation
  • Curation Rationale
  • Source Data
  • Annotations
  • Personal and Sensitive Information
• Considerations for Using the Data
  • Social Impact of Dataset
  • Discussion of Biases
  • Other Known Limitations
• Additional Information
  • Dataset Curators
  • Licensing Information
  • Citation Information
  • Contributions

Dataset Description

• Homepage: https://github.com/JanKalo/KAMEL
• Repository: https://github.com/JanKalo/KAMEL
• Paper: @inproceedings{kalo2022kamel, title={KAMEL: Knowledge Analysis with Multitoken Entities in Language Models}, author={Kalo, Jan-Christoph and Fichtel, Leandra}, booktitle={Automated Knowledge Base Construction}, year={2022} }

Dataset Summary

This dataset provides the data for KAMEL, a probing dataset for language models that contains factual knowledge from Wikidata and Wikipedia.

See the paper for more details. For more information, also see: https://github.com/JanKalo/KAMEL

Languages

en

Dataset Structure

Data Instances

Data Fields

KAMEL has the following fields:

• index: the id
• sub_label: a label for the subject
• obj_uri: Wikidata uri for the object
• obj_labels: multiple labels for the object
• chosen_label: the preferred label
• rel_uri: Wikidata uri for the relation
• rel_label: a label for the relation

Data Splits

The dataset is split into a training, validation, and test dataset. It contains 234 Wikidata relations. For each relation there exist 200 training, 100 validation, and 100 test instances.

Dataset Creation

Curation Rationale

This dataset was gathered and created to explore what knowledge graph facts are memorized by large language models.

Source Data

Initial Data Collection and Normalization

See the reaserch paper and website for more detail. The dataset was created from Wikidata and Wikipedia.

Annotations

Annotation process

There is no human annotation, but only automatic linking from Wikidata facts to Wikipedia articles. The details about the process can be found in the paper.

Who are the annotators?

Machine Annotations

Personal and Sensitive Information

Unkown, but likely information about famous people mentioned in the English Wikipedia.

Considerations for Using the Data

Social Impact of Dataset

The goal for the work is to probe the understanding of language models.

Discussion of Biases

Since the data is created from Wikipedia and Wikidata, the existing biases from these two data sources may also be reflected in KAMEL.

Additional Information

Dataset Curators

The authors of KAMEL at Vrije Universiteit Amsterdam and Technische Universität Braunschweig.

Licensing Information

The Creative Commons Attribution-Noncommercial 4.0 International License. see https://github.com/facebookresearch/LAMA/blob/master/LICENSE

Citation Information

@inproceedings{kalo2022kamel, title={KAMEL: Knowledge Analysis with Multitoken Entities in Language Models}, author={Kalo, Jan-Christoph and Fichtel, Leandra}, booktitle={Automated Knowledge Base Construction}, year={2022} }