BEHAVIOURAL SIMULATOR FOR PROFESSIONAL TRAINING BASED ON NATURAL LANGUAGE INTERACTION

The virtual patient is an online simulation system designed to train and assess relational and clinical abilities in a realistic interactive problem-based learning scenario, where users (medical students) can interact and communicate with characters specifically designed to challenge their clinical and relational skills and facilitate the generation of learning objectives.

In this paper we will present an enhancement to the system by simulating a normal interview with real users through natural language, thus enabling users to behave more naturally without keyboards or other input devices. We evaluated the system with a sample of users and found that the new voice-based interaction is user-friendly and facilitates user acceptance.

The Virtual Patient is based on artificial intelligence and pre-recorded movies, in which the learner plays the role of a physician who confronts a simulated patient. The patient in the scenario is played by a professional actor who is trained to simulate variable moods, attitudes and emotional responses through verbal and non-verbal communication. The interview can be paused at any point to give time for discussion, generation of learning issues and interaction with the tutor. Virtual Patients are simulators designed to improve users' effectiveness in the areas of anamnesis, diagnosis, treatment and follow-up. Also, the focus is on the process of establishing a trusted relation with the patient. In medical education, they can also be used in a classroom, to trigger the discussion around some learning objectives, under the guidance of a tutor. Dialogue interactions between the simulator and the user occur through selection from a set of closed answers, that we call "canned texts", using the mouse. At the end of each visit, the system provides a feedback, given directly through the patient's comments. The feedback is based on the decisions and the communication strategy that the user applied during the last visit or visits.

To allow a natural voice interaction with the Virtual Patient, we built a matching module for the system that maps the speech input into a particular piece of canned text.

This was performed in two steps:
(1) ranking all available canned texts regarding their similarity to the speech input;
(2) checking whether the highest ranked canned text is a suitable match for the input.

We conducted a preliminary evaluation with a small number of medical students to assess the learnability, discoverability, ease of use, satisfaction, and user friendliness. The participants' reaction towards the proposed approach was positive, in spite of some problems
still unsolved in the current version of the voice based Virtual Patient: non-matching situations mainly due to questions asked by the students that were no part of the set of the canned texts, not-correct matching due to the presence of similar phases but with a different meaning, and the difficulty, in some cases, to know what question to ask. Although these problems might represent an obstacle to the wide diffusion of the voice based Virtual Patient, we believe that the new version of the Virtual Patient is a step ahead toward a more natural simulation based medical education practice. We are working to solve the problems encountered in the evaluation and to make the Virtual Patient closer to the needs and to the practice of medical training.