A Cognitive Assistant that Uses Small Talk in Tutoring Conversation

Abstract

This paper presents a cognitive conversational agent for use in teaching and learning processes named THOTH (Training by Highly Ontology-oriented Tutoring Host) that is capable of formulating and enunciating a well-defined set of small talk segments in a Q&A (Question and Answer) interaction. The small talk structures are placed within the tutoring conversation by an agent designed as a cognitive assistant, in order to make communication smoother and less formal, presenting a more “concerned” behavior. Twelve small talk segments are suggested, included in conversation stages such as opening and closing the conversation, maintaining the rhythm and managing learning. We also explore some branches of the theoretical assumptions and concepts grounding THOTH, such as Dennett’s intentional stance, Bloom’s taxonomy and microlearning theory. In order to measure the perception and effects of using THOTH, we performed a quantitative and qualitative study with a group of students from a course in Applied Artificial Intelligence over one semester. The outcomes are classified into two main categories of analysis – interactivity and intentionality – informing the discussion on the potential uses of a small talk agent as a valuable resource in tutoring interaction, and also raising some points for improvement. In addition to this study, we also drew a small talk profile for this group of students revealing what structures and topics they use the most, as well as a partial performance analysis that allows identifying some effects on learning.