Footnote:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Description:
This interdisciplinary thesis reviews and refines the concepts of generalisation and generalising statements (e.g. all ravens are black) as a phenomenon of human cognition and language. The main parts of the thesis target researchers from linguistics, (computational) literary studies and natural language processing. The first part establishes a formal framework that explains the interplay between belief worlds, logical inference and the truth value judgements for generalising statements. Hereby, it incorporates findings from psycholinguistics and provides a unifying analysis for various types of quantification, including universal, generic and numerical quantification. The second part investigates how generalisation is encoded and used in language and narration. To achieve this, a comprehensive corpus of narrative fiction is constructed, containing annotations of generalising statements. A newly developed natural language processing pipeline is used to extract linguistic features from both generalising and non-generalising statements, and statistical and qualitative analyses are performed on the results. The third part develops tools for the automatic identification of generalising statements in texts. These tools range from simple rule-based taggers to complex deep learning systems. The best approaches achieve performances of 63% F1 and 68% F1 on the narrative corpus (in-domain) and a mixed-genre corpus (off-domain), respectively. Analyses of the neural systems provide further understanding of the extent to which artificial intelligence can learn the concept of generalisation. ; 2024-05-23