Modeling Future Conversation Turns to Teach LLMs to Ask Clarifying Questions

Large language models (LLMs) must often respond to highly ambiguous user requests. In such cases, the LLM's best response may be to ask a clarifying question to elicit more information. We observe existing LLMs often respond by presupposing a single interpretation of such ambiguous requests, frustrating users who intended a different interpretation. We speculate this is caused by current preference data labeling practice, where LLM responses are evaluated only on their prior contexts. To address this, we propose to assign preference labels by simulating their expected outcomes in the future turns. This allows LLMs to learn to ask clarifying questions when it can generate responses that are tailored to each user interpretation in future turns. In experiments on open-domain QA, we compare systems that trained using our proposed preference labeling methods against standard methods, which assign preferences based on only prior context. We evaluate systems based on their ability to ask clarifying questions that can recover each user's interpretation and expected answer, and find that our training with our proposed method trains LLMs to ask clarifying questions with a 5% improvement in F1 measured against the answer set from different interpretations of each query. Ambiguity is a hallmark of natural language that enables concise communication by allowing speakers to exclude details that are inferable from the context (e.g., conversational, temporal, geographical) (Piantadosi et al., 2012). At times, however, the speaker's intent is unclear despite the context, and further interaction is necessary to clarify their intent. Asking clarifying questions is particularly important for large language models (LLMs), which are tasked with serving a wide audience, often without access to the personalized context available in human interactions. In this work, we develop LLMs that can ask clarifying questions to resolve ambiguity in their users' requests. State-of-the-art LLMs (OpenAI, 2023; Gemini Team, 2024) often do not ask clarifying questions when presented with an ambiguous request, and instead respond directly by assuming the user's intent (see an example in Figure 1).