Clarification Question Generation

Clarification question generation studies how AI systems detect underspecified requests and ask questions that reveal the user’s intended meaning. Work spans conversational question answering, text-to-SQL, coding, recommendations, and tool-using agents, using methods such as uncertainty estimates, information gain, value-of-information reasoning, and reinforcement learning. Evaluation increasingly uses multi-turn benchmarks and human or simulated users to assess whether questions resolve intent, improve task outcomes, and avoid unnecessary turns. A central challenge is balancing the risks of guessing against the cost of interrupting users: models may recognize ambiguity yet fail to ask, or ask questions that do not change the outcome.