What Can a Generative Language Model Answer About a Passage?

Douglas Summers-Stay, Claire Bonial, Clare Voss


Abstract
Generative language models trained on large, diverse corpora can answer questions about a passage by generating the most likely continuation of the passage followed by a question/answer pair. However, accuracy rates vary depending on the type of question asked. In this paper we keep the passage fixed, and test with a wide variety of question types, exploring the strengths and weaknesses of the GPT-3 language model. We provide the passage and test questions as a challenge set for other language models.
Anthology ID:
2021.mrqa-1.7
Volume:
Proceedings of the 3rd Workshop on Machine Reading for Question Answering
Month:
November
Year:
2021
Address:
Punta Cana, Dominican Republic
Editors:
Adam Fisch, Alon Talmor, Danqi Chen, Eunsol Choi, Minjoon Seo, Patrick Lewis, Robin Jia, Sewon Min
Venue:
MRQA
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
73–81
Language:
URL:
https://aclanthology.org/2021.mrqa-1.7
DOI:
10.18653/v1/2021.mrqa-1.7
Bibkey:
Cite (ACL):
Douglas Summers-Stay, Claire Bonial, and Clare Voss. 2021. What Can a Generative Language Model Answer About a Passage?. In Proceedings of the 3rd Workshop on Machine Reading for Question Answering, pages 73–81, Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
What Can a Generative Language Model Answer About a Passage? (Summers-Stay et al., MRQA 2021)
Copy Citation:
PDF:
https://preview.aclanthology.org/emnlp22-frontmatter/2021.mrqa-1.7.pdf
Data
GLUE