Do Language Models Understand Measurements?

Sungjin Park, Seungwoo Ryu, Edward Choi


Abstract
Recent success of pre-trained language models (PLMs) has stimulated interest in their ability to understand and work with numbers. Yet, the numerical reasoning over measurements has not been formally studied despite their importance. In this study, we show that PLMs lack the capability required for reasoning over measurements. Furthermore, we find that a language model trained on a measurement-rich corpus shows better performance on understanding measurements. We propose a simple embedding strategy to better distinguish between numbers and units, which leads to a significant improvement in the probing tasks.
Anthology ID:
2022.findings-emnlp.128
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2022
Month:
December
Year:
2022
Address:
Abu Dhabi, United Arab Emirates
Editors:
Yoav Goldberg, Zornitsa Kozareva, Yue Zhang
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
1782–1792
Language:
URL:
https://aclanthology.org/2022.findings-emnlp.128
DOI:
10.18653/v1/2022.findings-emnlp.128
Bibkey:
Cite (ACL):
Sungjin Park, Seungwoo Ryu, and Edward Choi. 2022. Do Language Models Understand Measurements?. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 1782–1792, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
Cite (Informal):
Do Language Models Understand Measurements? (Park et al., Findings 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.findings-emnlp.128.pdf
Video:
 https://aclanthology.org/2022.findings-emnlp.128.mp4