We welcome you to the next Natural Language Processing and Vision (NLPV) seminar at the University of Exeter.
Scheduled: Thursday 25 Sep 2025 at 15:00 to 16:00, GMT+1 Location: https://Universityofexeter.zoom.us/j/92525879857?pwd=goytDAcZ4zTpM0OIogW9YbI... (Meeting ID: 925 2587 9857 Password: 485835)
Title: From Framing to False Premises: A Two-Axis View of Robust LLM Reasoning
Abstract: Large language models often appear to “reason”, yet their answers can be steered by how we ask and by what assumptions slip into the question. This talk advances a two-axis view of question robustness grounded in two benchmarks. On the surface, logically equivalent phrasings (e.g., more vs less) can directionally bias conclusions, a systematic framing effect, captured by More or Less Wrong. In the depths, multi-hop questions can embed false premises; when such premises are present, models often answer anyway instead of detecting and rejecting them, behavior exposed by MultiHoax. Together, these benchmarks set non-negotiable standards for robust reasoning: framing invariance and premise integrity.
Speaker's bio: Dr Nafise Sadat Moosavi is a Lecturer in Natural Language Processing at the University of Sheffield. Her research tackles the nuanced challenges of large language models, including reasoning, efficiency, robustness and fairness. She earned her PhD in Computational Linguistics from the University of Heidelberg. In the research community, she has served as Senior Area Chair for ACL, NAACL, EMNLP, and COLING, and co-initiated and co-organized the SustaiNLP workshop series. At Sheffield, she is Deputy Director of EDI, leading initiatives that foster equity, inclusion and a positive research culture.
We will update future talks on the website: https://sites.google.com/view/neurocognit-lang-viz-group/seminars
Joining our *Google group* for future seminar and research information: https://groups.google.com/g/neurocognition-language-and-vision-processing-gr...