| Summary: | Conversational Intelligent Tutoring Systems aim to mimic the adaptive behaviour of human tutors by delivering tutorial content as part of a dynamic exchange of information conducted using natural language. Deciding when it is beneficial to intervene in a student’s learning process is an important skill for tutoring. Human tutors use prior knowledge about the student, discourse content and learner non-verbal behaviour to choose when intervention will help learners overcome impasse. Experienced human tutors adapt discourse and pedagogy based on recognition of comprehension and non-comprehension indicative learner behaviour. In this research non-verbal behaviour is explored as a method of computationally analysing reading comprehension so as to equip an intelligent conversational agent with the human-like ability to estimate comprehension from non-verbal behaviour as a decision making trigger for feedback, prompts or hints. This thesis presents research that combines a conversational intelligent tutoring system (CITS) with near real-time comprehension classification based on modelling of e-learner non-verbal behaviour to estimate learner comprehension during on-screen conversational tutoring and to use comprehension classifications as a trigger for intervening with hints, prompts or feedback for the learner. To improve the effectiveness of tuition in e-learning, this research aims to design, develop and demonstrate novel computational methods for modelling e-learner comprehension of on-screen information in near real-time and for adapting CITS tutorial discourse and pedagogy in response to perception of comprehension indicative behaviour. The contribution of this research is to detail the motivation for, design of, and evaluation of a system which has the human-like ability to introduce micro-adaptive feedback into tutorial discourse in response to automatic perception of e-learner reading comprehension. This research evaluates empirically whether e-learner non-verbal behaviour can be modelled to classify comprehension in near real-time and presents a near real-time comprehension classification system which achieves normalised comprehension classification accuracy of 75%. Understanding e-learner comprehension creates exciting opportunities for advanced personalisation of materials, discourse, challenge and the digital environment itself. The research suggests a benefit is gained from comprehension based adaptation in conversational intelligent tutoring systems, with a controlled trial of a comprehension based adaptive CITS called Hendrix 2.0 showing increases in tutorial assessment scores of up to 17% when comprehension based discourse adaptation is deployed to scaffold the learning experience.
|
|---|
