Project Details
Projekt Print View

From distributions to roots - Towards a linguistically grounded theory of the conceptual underpinnings of verb meaning

Applicant Dr. Tillmann Pross
Subject Area General and Comparative Linguistics, Experimental Linguistics, Typology, Non-European Languages
Term from 2018 to 2021
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 406004556
 
Final Report Year 2021

Final Report Abstract

I investigated whether and how neural word embeddings can be understood to encode not only idiosyncratic aspects of word meaning but also the kind of general and abstract concepts that are central to theoretical approaches of lexical semantics. To this end, I computed the difference between general-purpose embeddings of intransitive verbs, and task-specific embeddings of the same verbs that capture their similarity according to the unaccusative hypothesis. The project showed that the difference that retraining makes is neither trivial nor random but captures surprisingly well the cues for unergativity and unaccusativity that have been proposed in the theoretical literature. The project thus suggests that word embeddings may provide a novel and empirically grounded perspective on the conceptual underpinnings of verb meaning.

Publications

  • (2019): Towards a correlation of form, use and meaning of ge-prefixed predicative participles in German (mit Antje Roßdeutscher). Glossa – a Journal of General Linguistics 4(1):93. 1–31
    Pross, Tillmann & Roßdeutscher, Antje
    (See online at https://doi.org/10.5334/gjgl.753)
  • (2020): Dispositions and the verbal description of their manifestations. A case study on Emission Verbs. Linguistics & Philosophy 43:149—191
    Pross, Tillmann
    (See online at https://doi.org/10.1007/s10988-019-09268-5)
  • (2020): Distributional semantics and the conceptual foundations of verb meaning: how neural word embeddings memorize the unaccusative hypothesis Empirical Issues in Syntax and Semantics 13:81–107
    Pross, Tillmann
 
 

Additional Information

Textvergrößerung und Kontrastanpassung