Enhancing Answer Attribution for Faithful Text Generation with Large\n Language Models

2024/10/22 by Juraj Vladika, Vladika, Juraj, Luca Mülln +3
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Information Retrieval (cs.IR) #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2410.17112

openalex publication_date 2024/10/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

The increasing popularity of Large Language Models (LLMs) in recent years has\nchanged the way users interact with and pose questions to AI-based\nconversational systems. An essential aspect for increasing the trustworthiness\nof generated LLM answers is the ability to trace the individual claims from\nresponses back to relevant sources that support them, the process known as\nanswer attribution. While recent work has started exploring the task of answer\nattribution in LLMs, some challenges still remain. In this work, we first\nperform a case study analyzing the effectiveness of existing answer attribution\nmethods, with a focus on subtasks of answer segmentation and evidence\nretrieval. Based on the observed shortcomings, we propose new methods for\nproducing more independent and contextualized claims for better retrieval and\nattribution. The new methods are evaluated and shown to improve the performance\nof answer attribution components. We end with a discussion and outline of\nfuture directions for the task.\n

Related

• Peering into the Mind of Language Models: An Approach for Attribution in Contextual Question Answering
• Attributed Question Answering: Evaluation and Modeling for Attributed Large Language Models
• Decomposition-Enhanced Training for Post-Hoc Attributions In Language Models
• Understanding Retrieval Augmentation for Long-Form Question Answering
• Advancing Large Language Model Attribution through Self-Improving