Properties and Challenges of LLM-Generated Explanations
2024 (English)In: Proceedings of the Third Workshop on Bridging Human-Computer Interaction and Natural Language Processing / [ed] Su Lin Blodgett, Amanda Cercas Curry, Sunipa Dey, Michael Madaio, Ani Nenkova, Diyi Yang, Ziang Xiao, ASSOC COMPUTATIONAL LINGUISTICS-ACL , 2024, p. 13-27Conference paper, Published paper (Refereed)
Abstract [en]
The self-rationalising capabilities of large language models (LLMs) have been explored in restricted settings, using task-specific data sets.However, current LLMs do not (only) rely on specifically annotated data; nonetheless, they frequently explain their outputs.The properties of the generated explanations are influenced by the pre-training corpus and by the target data used for instruction fine-tuning.As the pre-training corpus includes a large amount of human-written explanations “in the wild”, we hypothesise that LLMs adopt common properties of human explanations.By analysing the outputs for a multi-domain instruction fine-tuning data set, we find that generated explanations show selectivity and contain illustrative elements, but less frequently are subjective or misleading.We discuss reasons and consequences of the properties’ presence or absence. In particular, we outline positive and negative implications depending on the goals and user groups of the self-rationalising system.
Place, publisher, year, edition, pages
ASSOC COMPUTATIONAL LINGUISTICS-ACL , 2024. p. 13-27
Keywords [en]
Natural Language Processing, Large Language Models, Explainability, Human-AI Interaction
National Category
Natural Language Processing
Identifiers
URN: urn:nbn:se:liu:diva-207112DOI: 10.18653/v1/2024.hcinlp-1.2ISI: 001596065400002ISBN: 9798891761117 (print)OAI: oai:DiVA.org:liu-207112DiVA, id: diva2:1894037
Conference
Third Workshop on Bridging Human-Computer Interaction and Natural Language Processing at NAACL 2024, MEXICO, JUN 21, 2024
Note
Funding Agencies|Wallenberg AI, Autonomous Systems and Software Program (WASP) - Knut and Alice Wallenberg Foundation; European Commission [101135671]
2024-09-022024-09-022025-12-18