Coercion and copredication as flexible frame composition

• Applicants: Professorin Dr. Laura Kallmeyer; Dr. Rainer Osswald
• Subject Area: General and Comparative Linguistics, Experimental Linguistics, Typology, Non-European Languages ; Applied Linguistics, Computational Linguistics
• Term: since 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 440934416

This project investigates the challenges posed by the dynamic semantic behavior of words in syntagmatic relationships, focusing on two phenomena: facet selection and coercion. Facet selection involves the varied meaning contributions of inherently polysemous words, particularly complex type nouns, based on the predications in which they are used. This includes copredication contexts where conflicting predicates are applied to the same noun. Coercion, on the other hand, entails a systematic enrichment of semantic meaning when there are mismatches between a predicate's selectional restrictions and the argument's semantic contribution. The project's objectives are threefold: (A) Develop a formal model for these phenomena at the lexical representation and integration levels in coercive and copredication contexts. (B) Create neural data-driven models for identifying target facets in predications, detecting coercion, and identifying complex type nouns. (C) Construct a copredication-coercion annotation environment and corpus resource. The project adopts an approach where insights from formal modeling (A) inform neural models (B), and the latter generate new data for the former. The copredication-coercion annotation environment (C) leverages theoretical insights from (A) and utilizes models from (B) for candidate selection. The generated data from (C) subsequently contributes to refining (A) and (B). The central questions revolve around (i) modeling the availability of specific meaning facets in different contexts and (ii) representing and triggering coercion. The hypothesis posits that constraint mismatches in composition block semantic default components in both cases. Part (A) of the project explores default frame logics for modeling this, while Part (B) develops probabilistic models driven by data.

• DFG Programme: Research Grants
• International Connection: France, Italy, Spain
• Cooperation Partners: Dr. Timothée Bernard; Professor Dr. Benoit Crabbé; Professorin Dr. Elisabetta Jezek; Dr. Peter Sutton, Ph.D.