Zhenhai Pan

The Hong Kong Polytechnic University, Hong Kong Abstract Most automated electronic medical record (EMR) pipelines remain output-oriented: they transcribe, extract, and summarize after the consultation, but they do not explicitly model what is already known, what is still missing, which uncertainty matters most, or what question or recommendation should come next. We formulate doctor-patient dialogue as a proactive knowledge-inquiry problem under partial observability. The proposed framework combines stateful extraction, sequential belief updating, gap-aware state modeling, hybrid retrieval over objectified medical knowledge, and a POMDP-lite action planner. Instead of treating the EMR as the only target artifact, the framework treats documentation as the structured projection of an ongoing inquiry loop. To make the formulation concrete, we report a controlled pilot evaluation on ten standardized multi-turn dialogues together with a 300-query retrieval benchmark aggregated across dialogues. On this pilot protocol, the full framework reaches 83.3% coverage, 80.0% risk recall, 81.4% structural completeness, and lower redundancy than the chunk-only and template-heavy interactive baselines. These pilot results do not establish clinical generalization; rather, they suggest that proactive inquiry may be methodologically interesting under tightly controlled conditions and can be viewed as a conceptually appealing formulation worth further investigation for dialogue-based EMR generation. This work should be read as a pilot concept demonstration under a controlled simulated setting rather than as evidence of clinical deployment readiness. No implication of clinical deployment readiness, clinical safety, or real-world clinical utility should be inferred from this pilot protocol. 1 Introduction Clinical consultations generate rich evidence: symptom descriptions, onset patterns, aggravating and relieving factors, prior history, test results, risk signals, and physician reasoning. Yet physicians must often conduct the consultation and document it at the same time. Recent progress in clinical note generation and dialogue summarization has substantially improved post hoc documentation support [1, 2, 3]. However, most existing systems still follow a linear workflow: transcribe speech, extract information, and generate a final note after the fact. That workflow is useful for documentation assistance, but it does not explicitly reason about what is known so far, what remains missing, or what action would best close the most important gap while the consultation is still unfolding. A clinician does not wait until the end of the encounter to decide what matters. The clinician maintains a running case model, identifies unresolved uncertainties, weighs risk, and chooses the next question or recommendation accordingly. We argue that a useful AI assistant should support exactly this process. In that sense, multi-turn doctor-patient dialogue is more naturally framed as a proactive knowledge-inquiry process than as a passive note-generation pipeline. This paper proposes a unified framework for proactive knowledge inquiry in doctor-patient dialogue. The framework represents dialogue evidence as stateful events, maintains a structured CurrentState, compares it against a target-oriented GoalState, derives typed GapSignals, retrieves supportive objects and reasoning paths from an objectified knowledge base, and selects the next action using a tractable POMDP-lite controller. Within this formulation, the EMR is not the control loop itself; it is a structured output view of the evolving internal state. Our contributions are fourfold. First, we formulate dialogue-based EMR generation as a proactive inquiry problem under partial observability rather than a purely linear speech-to-record task. Second, we define a stateful extraction representation that distinguishes observed results, historical findings, recommendations, pending verification, negations, and related clinically meaningful evidence states. Third, we specify a unified inference-and-control framework that combines sequential belief updating, typed gap modeling, hybrid retrieval over objectified knowledge, and path-aware action planning. Fourth, we provide a controlled pilot protocol with raw-count denominators and stable audit units, together with a companion systems manuscript that instantiates the framework as an online prototype [16]. SpeechTextFieldsEMRLinear pipeline