Generalization in NLI: Ways (Not) To Go Beyond Simple Heuristics
Abstract
Much of recent progress in NLU was shown to be due to models' learning dataset-specific heuristics. We conduct a case study of generalization in NLI (from MNLI to the adversarially constructed HANS dataset) in a range of BERT-based architectures (adapters, Siamese Transformers, HEX debiasing), as well as with subsampling the data and increasing the model size. We report 2 successful and 3 unsuccessful strategies, all providing insights into how Transformer-based models learn to generalize.
- Publication:
-
arXiv e-prints
- Pub Date:
- October 2021
- DOI:
- 10.48550/arXiv.2110.01518
- arXiv:
- arXiv:2110.01518
- Bibcode:
- 2021arXiv211001518B
- Keywords:
-
- Computer Science - Computation and Language
- E-Print:
- Workshop on Insights from Negative Results (EMNLP 2021)