DialogueTrainer

'DialogueTrainer' develops conversation models that are used to teach how to conduct certain conversations, such as assessing performance, delivering bad news or giving feedback. The conversations are conducted with a virtual character in a virtual environment (also called DialogueTrainer). The conversation models are used to develop scenarios in which a learner takes different steps through a conversation with the virtual character. At each step, the learner is given a choice between different options, and the emotion and reaction of the virtual character depends on this choice.

The past two years have seen strong developments in machine learning technologies relevant to training communication skills using applications such as DialogueTrainer. First, the availability of large language models, such as the model used for ChatGPT, and dialog corpora on particular topics has a strong impact on all human language-related tasks. Second, the automatic recognition of emotions from voice, facial expressions, posture, distance, gaze direction or pauses is greatly improved.

Purpose
The goal of this project is twofold. On the one hand, the researchers want to use large language models and dialogue corpora to analyze and improve the quality of dialogue scenarios. On the other hand, they want to extend DialogueTrainer so that it facilitates and uses multimodal input in interventions to practice communication skills.

In particular, they will investigate how and under what conditions communication skills interventions are effective, and how we can assess and improve the quality of these interventions.

120,000 will be used as a PPP program grant.

Earlier we spoke to Johan Jeuring, lead applicant and professor at Utrecht University. Read here the article.