MEDCOD: A Medically-Accurate, Emotive, Diverse, and Controllable Dialog System

We present MEDCOD, a Medically-Accurate, Emotive, Diverse, and Controllable\nDialog system with a unique approach to the natural language generator module.\nMEDCOD has been developed and evaluated specifically for the history taking\ntask. It integrates the advantage of a traditional modular approach to\nincorporate (medical) domain knowledge with modern deep learning techniques to\ngenerate flexible, human-like natural language expressions. Two key aspects of\nMEDCOD's natural language output are described in detail. First, the generated\nsentences are emotive and empathetic, similar to how a doctor would communicate\nto the patient. Second, the generated sentence structures and phrasings are\nvaried and diverse while maintaining medical consistency with the desired\nmedical concept (provided by the dialogue manager module of MEDCOD).\nExperimental results demonstrate the effectiveness of our approach in creating\na human-like medical dialogue system. Relevant code is available at\nhttps://github.com/curai/curai-research/tree/main/MEDCOD\n