Is anisotropy really the cause of BERT embeddings not being semantic?

Alejandro Fuster Baggetto, Victor Fresno


Abstract
In this paper we conduct a set of experiments aimed to improve our understanding of the lack of semantic isometry in BERT, i.e. the lack of correspondence between the embedding and meaning spaces of its contextualized word representations. Our empirical results show that, contrary to popular belief, the anisotropy is not the root cause of the poor performance of these contextual models’ embeddings in semantic tasks. What does affect both the anisotropy and semantic isometry is a set of known biases: frequency, subword, punctuation, and case. For each one of them, we measure its magnitude and the effect of its removal, showing that these biases contribute but do not completely explain the phenomenon of anisotropy and lack of semantic isometry of these contextual language models.
Anthology ID:
2022.findings-emnlp.314
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:
4271–4281
Language:
URL:
https://aclanthology.org/2022.findings-emnlp.314
DOI:
10.18653/v1/2022.findings-emnlp.314
Bibkey:
Cite (ACL):
Alejandro Fuster Baggetto and Victor Fresno. 2022. Is anisotropy really the cause of BERT embeddings not being semantic?. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 4271–4281, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
Cite (Informal):
Is anisotropy really the cause of BERT embeddings not being semantic? (Fuster Baggetto & Fresno, Findings 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.findings-emnlp.314.pdf
Video:
 https://aclanthology.org/2022.findings-emnlp.314.mp4