Servicenavigation

Project Details

Back

Projekt Print View

Computer-aided Analysis of Unreliability and Truth in Fiction - Interconnecting and Operationalizing Narratology (CAUTION)

Subject Area German Literary and Cultural Studies (Modern German Literature)
General and Comparative Linguistics, Experimental Linguistics, Typology, Non-European Languages
Term from 2020 to 2025
Project identifier Deutsche Forschungsgemeinschaft (DFG) - Project number 449444411
 
Final Report Year 2025

Final Report Abstract

The project focused on a multi-tracked, computer-aided investigation of the narrative phenomenon of unreliable narration. Unreliable narration in the variant investigated here refers to narrators in fictional literary texts that make incorrect claims about the fictional world of the narration. In literary research, this phenomenon is classified as strongly interpretation-dependent, which is why it is assumed that the assessment of the reliability of a narrative instance can (legitimately) vary depending on the interpreter. The justifications provided for this assumption vary in the research literature, and no empirical verification has yet taken place. At the same time, research has identified various textual features, so-called indicators, which can point to a narrator being unreliable. A discussion of the question which indicators are connected to which variant of unreliable narration as well as (theoretical or empirical) investigations concerning the indicative power of said indicators have not yet been addressed. In CAUTION, the aforementioned research desiderata were addressed within the framework of a multi-tracked experimental computer-assisted approach. First, a corpus of nine short fictional narratives (reliable and unreliable) from the period between the 19th and 21st centuries was created. Potentially relevant textual indicators (or indicative characteristics of narrators) for unreliable narration were then selected (emotional agitation, intention to appear certain, uncertainty, addressee-orientation, distraction and intention to distract). On the one hand, these indicators were to be determined as automatically as possible in the corpus using available NLP models (shallow track). On the other hand, the indicators were identified using manual annotation, as well as occurrences of unreliably narrated passages (middle track). Finally, the decisions underlying the annotation decisions were made visible and put up for discussion (deep track). The data resulting from the three approaches can be evaluated and compared in order to generate different (initially tentative due to the small size of the corpus) findings: (1) Identification of indicators: Do the automatically identified indicators match the manually identified indicators? Can the manual annotations be used to optimize the NLP models? (2) Indicative power: What is the actual indicative power of the indicators? Which indicators actually correlate sufficiently often with unreliable narration, and in which way? (3) Interpretation dependence: How much do interpreters' assessments of unreliable narration actually differ? What effect do the visualization and discussion of the underlying decisions have? The results will be reflected upon in terms of literary theory and made available for subsequent use.

Publications

 
 

Additional Information

© 2026 DFG Disclaimer / Copyright / Privacy Policy

Textvergrößerung und Kontrastanpassung