This project investigates whether the differences between various sources of disagreement (e.g., noise, ambiguity, subjective bias) can be detected using statistical models, particularly in structured data, and how to use such insight to guide the development of approaches for training and evaluating NLP models with datasets containing disagreements.
Most research on disagreement in text classification focuses on disagreement on atomic judgements (e.g., whether a post is offensive or not). But many types of observed disagreements have to do with complex judgments, many of which involve tree structures - e.g., about the structure of a discourse, or the syntactic analysis of a sentence.
This PhD project is investigating disagreements of this type, starting with disagreements on the RST analysis of a text.
References:
Debopam Das, Manfred Stede, and Maite Taboada. 2017. The Good, the Bad, and the Disagreement: Complex ground truth in rhetorical structure analysis. In Proceedings of the 6th Workshop on Recent Advances in RST and Related Formalisms, pages 11–19, Santiago de Compostela, Spain. Association for Computational Linguistics.
Héctor Martínez Alonso, Barbara Plank, Arne Skjærholt, and Anders Søgaard. 2015. Learning to parse with IAA-weighted loss. In Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1357–1361, Denver, Colorado. Association for Computational Linguistics.
Yang Janet Liu, Tatsuya Aoyama, and Amir Zeldes. 2023. What’s Hard in English RST Parsing? Predictive Models for Error Analysis. In Proceedings of the 24th Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 31–42, Prague, Czechia. Association for Computational Linguistics.