Abstract
Recent work in cross-topic argument mining attempts to learn models that generalise across topics rather than merely relying on withintopic spurious correlations. We examine the effectiveness of this approach by analysing the
output of single-task and multi-task models for cross-topic argument mining through a combination of linear approximations of their decision boundaries, manual feature grouping, challenge examples, and ablations across the
input vocabulary. Surprisingly, we show that cross-topic models still rely mostly on spurious correlations and only generalise within closely related topics, e.g., a model trained only on closed-class words and a few common open-class words outperforms a state-of-theart cross-topic model on distant target topics.
output of single-task and multi-task models for cross-topic argument mining through a combination of linear approximations of their decision boundaries, manual feature grouping, challenge examples, and ablations across the
input vocabulary. Surprisingly, we show that cross-topic models still rely mostly on spurious correlations and only generalise within closely related topics, e.g., a model trained only on closed-class words and a few common open-class words outperforms a state-of-theart cross-topic model on distant target topics.
Originalsprog | Engelsk |
---|---|
Titel | Proceedings of *SEM 2021 : The Tenth Joint Conference on Lexical and Computational Semantics |
Antal sider | 9 |
Forlag | Association for Computational Linguistics |
Publikationsdato | 2021 |
Sider | 263-277 |
DOI | |
Status | Udgivet - 2021 |
Emneord
- cross-topic argument mining
- spurious correlations
- multi-task models
- linear approximations
- input vocabulary ablations