Abstract
While improving neural dialogue agents’ factual accuracy is the object of much research, another important aspect of communication, less studied in the setting of neural dialogue, is transparency about ignorance. In this work, we analyze to what extent state-of-the-art chit-chat models are linguistically calibrated in the sense that their verbalized expression of doubt (or confidence) matches the likelihood that the model’s responses are factually incorrect (or correct). We find that these models are poorly calibrated, yet we show that likelihood of correctness can accurately be predicted. By incorporating such metacognitive features into the training of a controllable generation model, we obtain a dialogue agent with greatly improved linguistic calibration.- Anthology ID:
- 2022.tacl-1.50
- Volume:
- Transactions of the Association for Computational Linguistics, Volume 10
- Month:
- Year:
- 2022
- Address:
- Cambridge, MA
- Venue:
- TACL
- SIG:
- Publisher:
- MIT Press
- Note:
- Pages:
- 857–872
- Language:
- URL:
- https://aclanthology.org/2022.tacl-1.50
- DOI:
- 10.1162/tacl_a_00494
- Cite (ACL):
- Sabrina J. Mielke, Arthur Szlam, Emily Dinan, and Y-Lan Boureau. 2022. Reducing Conversational Agents’ Overconfidence Through Linguistic Calibration. Transactions of the Association for Computational Linguistics, 10:857–872.
- Cite (Informal):
- Reducing Conversational Agents’ Overconfidence Through Linguistic Calibration (Mielke et al., TACL 2022)
- PDF:
- https://preview.aclanthology.org/remove-xml-comments/2022.tacl-1.50.pdf