Academic journal article Educational Technology & Society

Designing for Automatic Affect Inference in Learning Environments

Academic journal article Educational Technology & Society

Designing for Automatic Affect Inference in Learning Environments

Article excerpt

Introduction

Computer-based learning now encompasses a wide array of innovative learning technologies including adaptive hypermedia systems to sophisticated tutoring environments, educational games, virtual environments and online tutorials. These continue to enrich the learning process in numerous ways. Keen to emulate the effectiveness of human tutors in the design and functioning of learning technologies, researchers have continually looked at the strategies of expert human teachers for motivation and are making directed efforts to make this machine-learner interaction more natural and instinctive. Detection of learners' affective states can give better insight into a learners' overall experience which can be helpful in adapting the tutorial interaction and strategy. Such a responsive interface can also alleviate fears of isolation in learners and facilitate learning at an optimal level. To enhance the motivational quality and engagement value of instructional content, affect recognition needs to be considered in light of its implications to learning technologies.

Effective tutoring by humans is an interactive yet guided process where learner engagement is constantly monitored to provide remedial feedback and to maximise the motivation to learn (Merill, Reiser, Trafton, & Ranney, 1992). Indeed, formative assessment and feedback is an important aspect of effectively designed learning environments and should occur continuously and unobtrusively as an integral part of the instruction (Bransford, Brown, & Cocking, 1999). In naturalistic settings, the availability of several channels of communication facilitates the constant monitoring necessary for such an interactive and flexible learning experience (Picard et al., 2004; de Vicente & Pain, 1998). One of the biggest challenges for computer tutors then is to achieve the mentoring capability of expert human teachers (van Vuuren, 2006). To give such a capability to a machine tutor entails giving it the ability to infer affect.

Learning has a strong affective quality that impacts overall performance, memory, attention, decision-making and attitude. Recent research provides compelling evidence to support the multiplicity and functional relevance of emotions for the situational and ontogenetic development of learners' interest, motivation, volition, and effort (Pekrun, 2005). It reflects the growing understanding of the centrality of emotion in the teaching-learning process and the fact that as yet this crucial link has not been addressed in machine-learner interactions (O'Regan, 2003).

Despite this recognition of affect as a vital component of learning processes and a context for cognition, computer- based learning environments have long ignored this aspect and have concentrated mostly in modelling the behaviour of a learner in response to a particular instructional strategy (Picard et al., 2004; du Boulay & Luckin, 2001). This relative bias towards the cognitive dimension of learning is now being criticised and the inextricable linkage between affective and cognitive functions is being stressed. This comes at a time when advances in the field of affective computing have opened the possibility of envisioning integrated architectures by allowing for formal representation, detection, and analysis of affective phenomena. This increasing interest in building affect-sensitive human-computer interactions thus finds an important application in learning technologies (Cowie et al., 2001).

Building on a discussion of recent studies highlighting the relevance of emotions in learning, this paper describes different techniques for measuring emotions and efforts in automatic recognition and/or prediction of affect in learning contexts before proposing a parallel emotion inference system. This is not an exhaustive survey of the past work but a selected discussion of recent works highlighting the concern and those attempting to address it. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.