Taxonomy Revision in Botany: A Simulation of Historical Data

In this abstract we present ReTAX, a system for taxonomy revision in Botany. The function of ReTAX is to revise, and eventually modify, a taxonomic hierarchy as novel inconsistent data are provided. The system has been applied to replicate some taxonomic revisions which have taken place historically in the botanical family Ericaceae. As noted by different researchers (e.g., Lakatos, 1976; Kulkarni & Simon, 1988; Klahr, Dunbar, & Fay, 1990), focusing on unexpected observations is a powerful heuristic in scientific reasoning. Since classification is the basis of many scientific endeavours (see, e.g., Sokal 1974), studying the effect of unexpected phenomena on the revision of classificatory systems is of particular relevance for scientific discovery. Some artificial intelligence models of discovery have dealt with the creation of scientific taxonomies (e.g., Langley, Zytkow, Bradshaw & Simon, 1983; Nordhausen & Langley, 1990). Likewise some computational systems have been created to perform taxonomy related tasks such as biological identification, automation of taxonomic descriptions, database management, and various numeric analyses (see, e.g., Pankhurst, 1975; Abbot, Bisby, and Rogers, 1985; Woolley & Stone, 1987; Watson, Dallwitz, Gibbs, and Pankhurst, 1988; Domingo, 1993). However, in general, neither of these approaches has addressed explicitly taxonomy revision. On the other hand, discovery systems which have tackled the revision of faulty scientific theories to accommodate anomalous data (e.g, Rose, 1989; Rajamoney, 1990; Karp, 1993) have not normally dealt with classification. ReTAX applies a theory revision approach in the context of scientific classification. ReTAX is based on a psychological study on category induction, in the domain of plant taxonomy, that we conducted at Aberdeen. The purpose of the study was to detect the search strategies and heuristics used by expert