![[2511.10453] Reasoning about Intent for Ambiguous Requests](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2511.10453] Reasoning about Intent for Ambiguous Requests
Summary
This paper explores how large language models can better handle ambiguous requests by generating multiple interpretation-answer pairs, enhancing user experience and safety.
Why It Matters
As AI systems become more integrated into daily tasks, understanding user intent is crucial. This research addresses the challenges of ambiguity in user requests, aiming to improve interaction quality and reduce misunderstandings, which can lead to safety risks.
Key Takeaways
- Proposes a method for generating multiple interpretations for ambiguous requests.
- Utilizes reinforcement learning and customized reward functions for training.
- Demonstrates higher coverage of valid answers compared to baseline methods.
- Enhances transparency and efficiency in AI responses.
- Supports downstream applications with structured output formats.
Computer Science > Computation and Language arXiv:2511.10453 (cs) [Submitted on 13 Nov 2025 (v1), last revised 13 Feb 2026 (this version, v2)] Title:Reasoning about Intent for Ambiguous Requests Authors:Irina Saparina, Mirella Lapata View a PDF of the paper titled Reasoning about Intent for Ambiguous Requests, by Irina Saparina and Mirella Lapata View PDF HTML (experimental) Abstract:Large language models often respond to ambiguous requests by implicitly committing to one interpretation. Intent misunderstandings can frustrate users and create safety risks. To address this, we propose generating multiple interpretation-answer pairs in a single structured response to ambiguous requests. Our models are trained with reinforcement learning and customized reward functions using multiple valid answers as supervision. Experiments on conversational question answering and semantic parsing demonstrate that our method achieves higher coverage of valid answers than baseline approaches. Human evaluation confirms that predicted interpretations are highly aligned with their answers. Our approach promotes transparency with explicit interpretations, achieves efficiency by requiring only one generation step, and supports downstream applications through its structured output format. Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) Cite as: arXiv:2511.10453 [cs.CL] (or arXiv:2511.10453v2 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2511.10453 Focus to...
