CAP: A Source-Grounded Proposition Scaffold for Faithful Clinical Dialogue-to-Note Generation

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research, Clinical Care & Medical Practice · Depth: Expert, quick

Summary

Clinical Atomic Propositions (CAPs) introduce a novel dialogue-aware intermediate representation designed to enhance faithful clinical note generation from noisy, distributed, and often revised dialogue evidence. Addressing issues like omissions and unsupported fill-in, CAPs extract source-grounded clinical assertions, meticulously preserving critical modifiers such as verification status, temporality, speaker/source, and action type. The system also explores an optional event consolidation layer that groups CAPs into problem-oriented care bundles before note rendering. Evaluated on a 197-case ACI-Bench cohort, CAP was compared against a transcript-only baseline and prompt-based reimplementations of Cluster2Sent and MEDSUM-ENT, alongside CAP+Event. Using GPT-R/P/F1 and semCAP-R/P/F1 metrics, results demonstrate that CAP significantly improves the preservation of transcript-grounded clinical propositions while maintaining competitive concept-level GPT scores. While CAP+Event offers organizational benefits, it can introduce omissions. All associated code and evaluation artifacts are publicly available.

Key takeaway

For NLP Engineers developing clinical dialogue-to-note systems, you should integrate source-grounded intermediate representations like Clinical Atomic Propositions (CAPs) to significantly enhance faithfulness and detail preservation. This approach directly addresses common issues such as omissions and unsupported fill-in, ensuring clinical assertions retain critical context. Evaluate your systems using proposition-grounded metrics to verify source-grounded faithfulness, and carefully assess the trade-offs of event consolidation for organizational benefits versus potential data loss.

Key insights

CAPs use a source-grounded intermediate representation to improve faithfulness and detail preservation in clinical dialogue-to-note generation.

Principles

• Clinical note generation benefits from dialogue-aware intermediate representations.
• Preserving modifiers (e.g., temporality, speaker) is crucial for faithfulness.
• Event consolidation can improve organization but risks omissions.

Method

The method involves extracting Clinical Atomic Propositions (CAPs) from dialogue, preserving modifiers, and optionally grouping them into problem-oriented care bundles before note rendering.

In practice

• Use CAPs to extract granular, source-grounded clinical assertions.
• Consider event consolidation for problem-oriented note organization.
• Audit note generation systems with proposition-grounded faithfulness metrics.

Topics

• Clinical Note Generation
• Clinical Atomic Propositions
• Source-Grounded Faithfulness
• Intermediate Representations
• Dialogue Summarization
• Medical NLP

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• Possible or Definite? A Benchmark for Evaluating Diagnostic Uncertainty Preservation in Clinical Text — LLMs systematically distort diagnostic uncertainty in clinical text, often converting nuanced statements into definite assertions.
• CorPipe at CRAC 2026: Empty Nodes and Cross-Lingual Transfer in Multilingual Coreference Resolution — CorPipe 26 integrates empty node prediction with mention and coreference resolution for improved multilingual performance.
• Faithful by Construction: Claim-Anchored Attribution for Multi-Document Summarization — A modular, claim-anchored summarization framework inherently builds fine-grained attribution and faithfulness into the generation process.
• EDEN: A Large-Scale Corpus of Clinical Notes for Italian — EDEN provides the largest Italian clinical note corpus and a CRF-filling benchmark for LLM-based information extraction.
• A Proactive EMR Assistant for Doctor-Patient Dialogue: Streaming ASR, Belief Stabilization, and Preliminary Controlled Evaluation — Proactive EMR assistants require streaming ASR, belief stabilization, and objectified retrieval for real-time consultation support.
• EHRNote-ChatQA: A Benchmark for Evidence-Grounded Multi-Turn Clinical Question Answering over Longitudinal Discharge Summaries — EHRNote-ChatQA benchmarks LLMs on evidence-grounded, multi-turn clinical QA over longitudinal discharge summaries, revealing current limitations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.