Abstract
We evaluate two popular neural cognate generation models’ robustness to several types of human-plausible noise (deletion, duplication, swapping, and keyboard errors, as well as a new type of error, phonological errors). We find that duplication and phonological substitution is least harmful, while the other types of errors are harmful. We present an in-depth analysis of the models’ results with respect to each error type to explain how and why these models perform as they do.- Anthology ID:
- 2022.lrec-1.458
- Volume:
- Proceedings of the Thirteenth Language Resources and Evaluation Conference
- Month:
- June
- Year:
- 2022
- Address:
- Marseille, France
- Editors:
- Nicoletta Calzolari, Frédéric Béchet, Philippe Blache, Khalid Choukri, Christopher Cieri, Thierry Declerck, Sara Goggi, Hitoshi Isahara, Bente Maegaard, Joseph Mariani, Hélène Mazo, Jan Odijk, Stelios Piperidis
- Venue:
- LREC
- SIG:
- Publisher:
- European Language Resources Association
- Note:
- Pages:
- 4299–4305
- Language:
- URL:
- https://aclanthology.org/2022.lrec-1.458
- DOI:
- Cite (ACL):
- Winston Wu and David Yarowsky. 2022. On the Robustness of Cognate Generation Models. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 4299–4305, Marseille, France. European Language Resources Association.
- Cite (Informal):
- On the Robustness of Cognate Generation Models (Wu & Yarowsky, LREC 2022)
- PDF:
- https://preview.aclanthology.org/nschneid-patch-1/2022.lrec-1.458.pdf