license: cc-by-4.0
language:
- en
tags:
- world knowledge
- plausibility
- world modeling
pretty_name: EWoK-core-1.0
size_categories:
- 1K<n<10K
extra_gated_heading: Acknowledge LICENSE and TERMS OF USE to access the data repository
extra_gated_button_content: I have read and agree to the EWoK LICENSE and TERMS OF USE.
extra_gated_prompt: >-
⚠️ PLEASE DO NOT distribute any of the EWoK materials or derivatives publicly
in plain-text! Any materials should appear in password-protected ZIP files or
behind gated authentication mechanisms such as Huggingface datasets.
⚠️ Any use of EWoK materials in pretraining/training of models requires
EXPLICIT ACKNOWLEDGMENT!
[Full Terms of Use
(TOU)](https://github.com/ewok-core/ewok-paper/blob/main/TERMS_OF_USE.txt)
extra_gated_fields:
I have read and agree to the terms of use of EWoK-core-1-0 written in the TERMS OF USE file: checkbox
I agree to NOT share unzipped plain-text contents of the dataset or any derivative of it on the internet: checkbox
I agree to explicitly ACKNOWLEDGE and ATTRIBUTE any use of this dataset or any derivative in the training of language models (we suggest mentioning it in your README, the MODEL CARD, the GitHub repo, and any paper): checkbox
EWoK-core-1.0
The repository hosts data from the paper Elements of World Knowledge (EWoK): A cognition-inspired framework for evaluating basic world knowledge in language models
What is EWoK?
The ability to build and leverage world models is essential for a general-purpose AI agent. Testing such capabilities can be hard, in part because the building blocks of world models are ill-defined. Here, we present Elements of World Knowledge (EWOK), a framework for evaluating world modeling in language models by testing their ability to use knowledge of a concept to match a target text with a plausible/implausible context. EWOK targets specific concepts from multiple knowledge domains, known to be vital for world modeling in humans. Domains range from social interactions (help/hinder) to spatial relations (left/right). Both contexts and targets are minimal pairs. Objects, agents and locations in the items can be flexibly filled in, enabling easy generation of multiple controlled datasets. We then introduce EWOK-CORE-1.0, a dataset of 4,374 items covering 11 world knowledge domains. We evaluate the performance of 20 open-weights large language models (ranging from 1.3B-70B parameters) across a battery of evaluation paradigms along with a human norming study comprising 12,480 measurements. The overall performance of all tested models is worse than human performance, with results varying drastically across domains. These data highlight simple cases in which even large models fail and present rich avenues for targeted research on LLM world modeling capabilities.
Usage
⚠️ PLEASE DO NOT distribute any of the EWoK materials or derivatives publicly in plain-text! Any materials should appear in password-protected ZIP files or behind gated authentication mechanisms such as Huggingface datasets.
⚠️ Any use of EWoK materials in pretraining/training of models requires EXPLICIT ACKNOWLEDGMENT!
Authors and Citation
Anna A. Ivanova*, Aalok Sathe*, Benjamin Lipkin*, Unnathi Kumar, Setayesh Radkani, Thomas H. Clark, Carina Kauf, Jenn Hu, Pramod R.T., Gabe Grand, Vivian Paulun, Maria Ryskina, Ekin Akyurek, Ethan Wilcox, Nafisa Rashid, Leshem Choshen, Roger Levy, Evelina Fedorenko, Josh Tenenbaum, and Jacob Andreas.
@article{ivanova2024elements,
title={Elements of World Knowledge (EWOK): A cognition-inspired framework for evaluating basic world knowledge in language models},
author={Anna A. Ivanova and Aalok Sathe and Benjamin Lipkin and Unnathi Kumar and Setayesh Radkani and Thomas H. Clark and Carina Kauf and Jennifer Hu and R. T. Pramod and Gabriel Grand and Vivian Paulun and Maria Ryskina and Ekin Akyurek and Ethan Wilcox and Nafisa Rashid and Leshem Choshen and Roger Levy and Evelina Fedorenko and Joshua Tenenbaum and Jacob Andreas},
journal={arXiv preprint arXiv:2405.09605},
year={2024},
url={https://arxiv.org/abs/2405.09605}
}