Disagreements and their limits in discourse structure annotation: From insights to machine learning (C11*)

• Subject Area: General and Comparative Linguistics, Experimental Linguistics, Typology, Non-European Languages
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 317633480

Project C11 investigates Human Label Variation (HLV) in pragmatic annotation, challenging the traditional “single ground truth” assumption, and examining its potential as a meaningful signal for discourse structure. The project will identify causes of both inter- and intra-annotator disagreement, to inform new HLV-enriched discourse models. The project will develop input representations and evaluation schemes for machine learning under HLV, which ultimately may benefit downstream tasks. By addressing the hidden limits of discourse annotation variability, C11 contributes to CRC Cluster C's focus on Hidden Variability.

• DFG Programme: Collaborative Research Centres
• Subproject of: SFB 1287: Limits of Variability in Language: Cognitive, Computational, and Grammatical Aspects
• Applicant Institution: Universität Potsdam
• Project Heads: Professorin Dr. Tatjana Scheffler; Professor Dr. Manfred Stede