Towards Conditioning Clinical Text Generation for User Control
This work addresses the challenge of deploying reliable natural language generation systems in clinical settings to reduce human oversight, though it appears incremental as it builds on existing LLM methods.
The paper tackled the problem of hallucinations and factual inconsistencies in clinical text generation by exploring automated dataset augmentation using LLMs to condition models for clinician control, achieving a 9% to 34% relative improvement on a shared task.
Deploying natural language generation systems in clinical settings remains challenging despite advances in Large Language Models (LLMs), which continue to exhibit hallucinations and factual inconsistencies, necessitating human oversight. This paper explores automated dataset augmentation using LLMs as human proxies to condition LLMs for clinician control without increasing cognitive workload. On the BioNLP ACL'24 Discharge Me! Shared Task, we achieve new state-of-the-art results with simpler methods than prior submissions through more efficient training, yielding a 9\% relative improvement without augmented training and up to 34\% with dataset augmentation. Preliminary human evaluation further supports the effectiveness of our approach, highlighting the potential of augmenting clinical text generation for control to enhance relevance, accuracy, and factual consistency.