1
2
3
4
5
6
7
8
9
6. Discussion
IKBS has been implemented in Java language and is fully operational on (http://www.univ-reunion.fr/~ikbs/). Experts unaccustomed to computers are able to model and describe, and any non-specialists interested in the field can describe and identify new observations. IKBS is used directly by experts for creating descriptive models and filling cases without any help from a computer scientist. They find the interface very pleasant and enjoy the effectiveness of the tool.
In our methodology, it is important that the case base contains descriptions of specimens made by biologists other than the expert. This, in order to counterbalance his interpretation of observations (inter-observer variation) when consulting the knowledge base. The results of the identification process are more dependable when we mix descriptions of different users for the same specimens (shown in Table 1). As they are labeled with the correct identification name from the expert, we can integrate the noise due to misinterpretations from end-users directly into the case base.
Similarly, because of the intra-specific variability, the number of described specimens by species must be increased. Insofar as Pocilloporidæ is concerned, this family is one of the sixteen families of corals containing the greatest intra-variability, and its complex diversity was covered with detailed precision.
The difficulty arises due to the number of attributes applicable to each case. Thus, the building of an exhaustive knowledge base is time-consuming for describers: updating a case with the latest descriptive model on Pocilloporidæ requires nearly a whole day's work!
7. Related work
In other domains such as botany and zoology, some researchers have come up with solutions for coding descriptions [5]. Their programs enable to compare descriptions and facilitate identification process from databases [10], [15].
In Case-Based Reasoning methodology, IKBS can be compared with AcknoSoft's KATE, Isoft's RECALL and TecInno's CBR-Works. These decision support systems have been designed to cope with industrial fields and very large databases [3]. In the life sciences, our objective is to deal with more complex descriptions and less data (cases) by class.
8. Conclusions and future work
In collaboration with three experts, we are presently experimenting with IKBS on three other families of corals of the Mascarene archipelago (Fungiidæ, Poritidæ, Thamnasteriidæ). The meticulous choice of terms, drawings and images seems decisive for generating a dependable knowledge base and managing the complexity of natural objects.
This is why we are designing IKBS to build cooperative knowledge bases. The aim is to encourage experts to draw up a common thesaurus of vocabulary and illustrations (i.e. the questionnaire) on the same Family.
Collections of specimens, like experts, are distributed around the world. Thanks to satellite high-speed broadband networks, we have been able to demonstrate Telesystematics using video-conferencing and IKBS. At ATM Developments'98, experts were able to share their interpretations of observations of specimens under a microscopic examination synchronously between La Reunion (South-West of Indian Ocean) and Rennes (France).
Nowadays, expertise in natural sciences is precious (it becomes very rare). It is therefore urgent to develop tools that will ensure that expertise be collected and safeguarded for transmission to future generations. If this is not done, we will be left only with monographic descriptions and museum collections.