New Approach for Conversational Agent Definition by Non-Programmers: A Visual Domain-Specific Language

Abstract

Intelligent tutors and conversational agents (CAs) have proven to be useful learning tools. They have potential not only as stand-alone devices but also as integrable components to enrich and complement other educational resources. For this, new authoring approaches and platforms are required. They should be accessible to non-programmers (such as most teachers) and they should be integrable into current web-based educational platforms. This paper proposes a new approach to define such agents through a visual domain-specific language based on Google Blockly (a scratch-like language). It also develops a web-based integrable authoring platform to serve as a prototype, describing the requirements and architecture. To evaluate whether this novel approach is effective, a multi-stage experiment was conducted. First, participants learned to use the prototype authoring platform through an interactive tutorial. Second, they created a CA with a specific domain model. Times and performance were measured. Finally, they answered a standardized usability questionnaire (UMUX) and a purpose-specific survey. Results show that participants were able to learn to use the domain-specific language in a short time. Moreover, the purpose-specific survey indicates that their perception of the approach (and its potential) is positive. The standardized questionnaire indicates that even in its prototype stage, its usability is satisfactory.