How does BERT process disfluency?

Ye Tian, Tim Nieradzik, Sepehr Jalali, Da-shan Shiu


Abstract
Natural conversations are filled with disfluencies. This study investigates if and how BERT understands disfluency with three experiments: (1) a behavioural study using a downstream task, (2) an analysis of sentence embeddings and (3) an analysis of the attention mechanism on disfluency. The behavioural study shows that without fine-tuning on disfluent data, BERT does not suffer significant performance loss when presented disfluent compared to fluent inputs (exp1). Analysis on sentence embeddings of disfluent and fluent sentence pairs reveals that the deeper the layer, the more similar their representation (exp2). This indicates that deep layers of BERT become relatively invariant to disfluency. We pinpoint attention as a potential mechanism that could explain this phenomenon (exp3). Overall, the study suggests that BERT has knowledge of disfluency structure. We emphasise the potential of using BERT to understand natural utterances without disfluency removal.
Anthology ID:
2021.sigdial-1.22
Volume:
Proceedings of the 22nd Annual Meeting of the Special Interest Group on Discourse and Dialogue
Month:
July
Year:
2021
Address:
Singapore and Online
Editors:
Haizhou Li, Gina-Anne Levow, Zhou Yu, Chitralekha Gupta, Berrak Sisman, Siqi Cai, David Vandyke, Nina Dethlefs, Yan Wu, Junyi Jessy Li
Venue:
SIGDIAL
SIG:
SIGDIAL
Publisher:
Association for Computational Linguistics
Note:
Pages:
208–217
Language:
URL:
https://aclanthology.org/2021.sigdial-1.22
DOI:
10.18653/v1/2021.sigdial-1.22
Bibkey:
Cite (ACL):
Ye Tian, Tim Nieradzik, Sepehr Jalali, and Da-shan Shiu. 2021. How does BERT process disfluency?. In Proceedings of the 22nd Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 208–217, Singapore and Online. Association for Computational Linguistics.
Cite (Informal):
How does BERT process disfluency? (Tian et al., SIGDIAL 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.sigdial-1.22.pdf
Video:
 https://www.youtube.com/watch?v=bQypRvHeOR0
Data
SNLI