Empirical Demonstration of LLM Failure Modes in Pronoun Ambiguity Resolution
Abstract
This project documents empirical testing of large language model behaviour on a structured set of six pronoun ambiguity tasks, using structurally identical sentences across controlled context conditions. Five systematic failure modes are identified and demonstrated: premature ambiguity collapse, contradictory resolution rules applied to structurally identical sentences, content-dependent mode switching, post-hoc rationalization of arbitrary choices, and failure to preserve ambiguity when both referents are contextually supported.
Fields of Study & Subjects
Natural Language Processing Techniques
Topic Modeling
Syntax
Semantics
Linguistic Variation
Publication details
- Journal / source
- OSF Preprints (OSF Preprints)
- Publication date
- 28 February 2026
- Language
- EN
- Document type
- Preprint