Investigating Domain Information as Dynamic Support
for the Learner during Spoken Conversations
I researched my PhD for three years at the University of Birmingham, graduating to "Dr." in 2005 with guidance from supervisors Professor Mike Sharples and Doctor Chris Baber. The original idea for the PhD came from seeing Professor Sharples' idea of a resource for lifelong learning, "HandLeR" (Hand-held Learning Resource). This device would be with you, in one form or another, throughout your life, helping with everything that you learned. So for example, my father taught me to play chess when I was about eight. Had I been using a HandLeR, it would have kept a record of the chess openings that I learned. So now, if I have trouble recalling one of those openings, I could go to my current HandLeR and revise them.
Now, suppose I converse with you, and you start to tell me your favourite chess opening. Your HandLeR would know that opening (because it was there when you learned it), and it could 'converse' with my HandLeR to discover that I had learned the same opening. Now, the two HandLeRs as a system are in a good position to help. My HandLeR can show me the opening as I learned it (with my own annotations, for example) and your HandLeR can tell you that I already know that opening, but call it by a different name.
Thus the PhD was born. Of course, some of this is not yet technically feasible, so it was necessary to adjust the situation a little. The concept of "your HandLeR" was changed to a definition of the syllabus you used during learning, held as an expert-created concept map (but hidden from the learner). My HandLeR did not take into account my previous learning, and so offered more assistance than often was necessary. None-the-less, a useful approximation of the original idea was created, with interesting results.
So, the aim of the final PhD was to help the learner during telephone conversations between an adult learner and a relative expert. The assistance comes in the form of small amounts of immediately-relevant domain information, taken from a hidden concept map. I wrote a concept map editor as a tool for the experimental work. Early experiments asked the learner to create a cognitive map as the conversation progressed. The concept map editor was modified to select small parts of the expert's map and list those concepts at the top of the screen. You can try out this software here. (Later, the concept map editor became a stand-alone application for pocket PCs, winning an award in an international JAVA development competition (hosted in 2002 by the then Insignia) and contributing to a paper and book chapter).
For the main experiment, speech recognition was used to estimate continually the learner’s area of interest within the hidden concept map of the domain (freeing the learner from the necessity to create a cognitive map). This concept, plus those semantically related (i.e. linked) concepts, were again made visible for the learner. The experiments showed that this targeted information assisted the learner to expand their conversation into areas as yet unlearned, producing a richer interaction with the ‘expert’ and improved understanding by the learner.
From a theoretical standpoint, the work examined feedback within the learning process, drawing a connection between spoken communication (e.g. Clark) and learning (taking a constructivist approach based on Pask and Laurillard). The result is a shedding of new light on the importance of context to the learning process, and the potential of further research into conversation, cognition, and support for adult learning. I am in the process of completing a paper related to this work.
The idea behind this work is currently being showcased on the Futurelab web site.