Saved in:
Bibliographic Details
Main Authors: Lin, Shuyuan, Duan, Lei, Hughes, Philip, Sheng, Yuxuan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.16951
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912497892589568
author Lin, Shuyuan
Duan, Lei
Hughes, Philip
Sheng, Yuxuan
author_facet Lin, Shuyuan
Duan, Lei
Hughes, Philip
Sheng, Yuxuan
contents Conversational Information Retrieval (CIR) systems, while offering intuitive access to information, face a significant challenge: reliably handling unanswerable questions to prevent the generation of misleading or hallucinated content. Traditional approaches often rely on external classifiers, which can introduce inconsistencies with the core generative Large Language Models (LLMs). This paper introduces Self-Aware LLM for Unanswerability (SALU), a novel approach that deeply integrates unanswerability detection directly within the LLM's generative process. SALU is trained using a multi-task learning framework for both standard Question Answering (QA) and explicit abstention generation for unanswerable queries. Crucially, it incorporates a confidence-score-guided reinforcement learning with human feedback (RLHF) phase, which explicitly penalizes hallucinated responses and rewards appropriate abstentions, fostering intrinsic self-awareness of knowledge boundaries. Through extensive experiments on our custom-built C-IR_Answerability dataset, SALU consistently outperforms strong baselines, including hybrid LLM-classifier systems, in overall accuracy for correctly answering or abstaining from questions. Human evaluation further confirms SALU's superior reliability, achieving high scores in factuality, appropriate abstention, and, most importantly, a dramatic reduction in hallucination, demonstrating its ability to robustly "know when to say 'I don't know'."
format Preprint
id arxiv_https___arxiv_org_abs_2507_16951
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Harnessing RLHF for Robust Unanswerability Recognition and Trustworthy Response Generation in LLMs
Lin, Shuyuan
Duan, Lei
Hughes, Philip
Sheng, Yuxuan
Computation and Language
Conversational Information Retrieval (CIR) systems, while offering intuitive access to information, face a significant challenge: reliably handling unanswerable questions to prevent the generation of misleading or hallucinated content. Traditional approaches often rely on external classifiers, which can introduce inconsistencies with the core generative Large Language Models (LLMs). This paper introduces Self-Aware LLM for Unanswerability (SALU), a novel approach that deeply integrates unanswerability detection directly within the LLM's generative process. SALU is trained using a multi-task learning framework for both standard Question Answering (QA) and explicit abstention generation for unanswerable queries. Crucially, it incorporates a confidence-score-guided reinforcement learning with human feedback (RLHF) phase, which explicitly penalizes hallucinated responses and rewards appropriate abstentions, fostering intrinsic self-awareness of knowledge boundaries. Through extensive experiments on our custom-built C-IR_Answerability dataset, SALU consistently outperforms strong baselines, including hybrid LLM-classifier systems, in overall accuracy for correctly answering or abstaining from questions. Human evaluation further confirms SALU's superior reliability, achieving high scores in factuality, appropriate abstention, and, most importantly, a dramatic reduction in hallucination, demonstrating its ability to robustly "know when to say 'I don't know'."
title Harnessing RLHF for Robust Unanswerability Recognition and Trustworthy Response Generation in LLMs
topic Computation and Language
url https://arxiv.org/abs/2507.16951