Towards Better Characterization of Paraphrases

Timothy Liu, De Wen Soh


Abstract
To effectively characterize the nature of paraphrase pairs without expert human annotation, we proposes two new metrics: word position deviation (WPD) and lexical deviation (LD). WPD measures the degree of structural alteration, while LD measures the difference in vocabulary used. We apply these metrics to better understand the commonly-used MRPC dataset and study how it differs from PAWS, another paraphrase identification dataset. We also perform a detailed study on MRPC and propose improvements to the dataset, showing that it improves generalizability of models trained on the dataset. Lastly, we apply our metrics to filter the output of a paraphrase generation model and show how it can be used to generate specific forms of paraphrases for data augmentation or robustness testing of NLP models.
Anthology ID:
2022.acl-long.588
Volume:
Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
Month:
May
Year:
2022
Address:
Dublin, Ireland
Venue:
ACL
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
8592–8601
Language:
URL:
https://aclanthology.org/2022.acl-long.588
DOI:
10.18653/v1/2022.acl-long.588
Bibkey:
Cite (ACL):
Timothy Liu and De Wen Soh. 2022. Towards Better Characterization of Paraphrases. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 8592–8601, Dublin, Ireland. Association for Computational Linguistics.
Cite (Informal):
Towards Better Characterization of Paraphrases (Liu & Soh, ACL 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.acl-long.588.pdf
Software:
 2022.acl-long.588.software.zip
Code
 tlkh/paraphrase-metrics
Data
GLUEMRPCPAWS