Accommodating surprise in taxonomic tasks : a psychological and computational investigation
This thesis reports psychological and computational research conducted in a scientific domain: plant taxonomy. The purpose of the investigation was to explore how expert scientists cope with puzzling observations. Firstly, a psychological study of category induction was conducted to investigate the procedures used by professional botanists as they tried to identify a botanical category from a set of positive and negative instances. The target category in the study was defined by a feature which was unusual and inconsistent with subjects' botanical knowledge; consequently, situations of uncertainty and puzzlement were generated. The verbal reports elicited by the subjects, as they were working through the categorisation tasks, were recorded and analysed. The analysis of the protocols indicated that, in order to cope with conflicting evidence, the botanists heavily relied on their expert knowledge, which intricately influenced their inductive procedures (i.e. observation and comparison of instances). These results are consistent with previous categorisation studies which looked at the effects of background knowledge in concept induction. However, the role of background knowledge had not been sufficiently stressed in earlier investigations of scientists' reactions to surprise. The psychological results were simulated in a computer programme, Proto-ReTAX, which reproduced the categorisation behaviour of some of the subjects during the study. Some of the mechanisms implemented in this programme were subsequently used in ReTAX, an artificial intelligence system for taxonomy revision. Using a set of consistency checking rules and a set of refinement operators, ReTAX succeeded in replicating taxonomic revisions which had been suggested by professional botanists for the botanical family Ericaceae.