Compositional Generalization for Kinship Prediction through Data Augmentation

Kangda Wei, Sayan Ghosh, Shashank Srivastava


Abstract
Transformer-based models have shown promising performance in numerous NLP tasks. However, recent work has shown the limitation of such models in showing compositional generalization, which requires models to generalize to novel compositions of known concepts. In this work, we explore two strategies for compositional generalization on the task of kinship prediction from stories, (1) data augmentation and (2) predicting and using intermediate structured representation (in form of kinship graphs). Our experiments show that data augmentation boosts generalization performance by around 20% on average relative to a baseline model from prior work not using these strategies. However, predicting and using intermediate kinship graphs leads to a deterioration in the generalization of kinship prediction by around 50% on average relative to models that only leverage data augmentation.
Anthology ID:
2022.wnu-1.2
Volume:
Proceedings of the 4th Workshop of Narrative Understanding (WNU2022)
Month:
July
Year:
2022
Address:
Seattle, United States
Venue:
WNU
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
13–19
Language:
URL:
https://aclanthology.org/2022.wnu-1.2
DOI:
10.18653/v1/2022.wnu-1.2
Bibkey:
Cite (ACL):
Kangda Wei, Sayan Ghosh, and Shashank Srivastava. 2022. Compositional Generalization for Kinship Prediction through Data Augmentation. In Proceedings of the 4th Workshop of Narrative Understanding (WNU2022), pages 13–19, Seattle, United States. Association for Computational Linguistics.
Cite (Informal):
Compositional Generalization for Kinship Prediction through Data Augmentation (Wei et al., WNU 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.wnu-1.2.pdf
Video:
 https://aclanthology.org/2022.wnu-1.2.mp4