Abstract
Classification, and particularly taxonomic revision, have not been
generally addressed by computational models of scientific discovery.
In this paper we present a framework for the automation of taxonomic
revision in biological domains. This framework views taxonomy
formation as an interaction between: a) observation; b) creation and
structuring of a taxonomic hierarchy; c) identification of relevant
taxonomic descriptors; and d) use of background knowledge. We
describe a prototype system for taxonomic revision, ReTAX, which
implements relevant aspects of such a framework. ReTAX receives as
input a pre-established taxonomy, and is presented with new items
which contradict in some way the original classification. Using a set
of consistency criteria, ReTAX identifies the inconsistencies between
the new information and the taxonomy. The system then applies a set
of refinement operators to modify the taxonomy and resolve the
inconsistencies. ReTAX has been tested on a botanical domain,
replicating taxonomic revisions which had been suggested by
professional botanists for the family Ericaceae. Finally, we propose
extensions to ReTAX, which we hope will enable us to further develop
the framework, and subsequently create an aid which taxonomists can
use to revise existing taxonomies.
full paper