On the representation of observational data used for
classification and identification of natural objects

Jacques LE RENARD, Noël CONRUYT

Muséum National d'Histoire Naturelle, CNRS D-0699 Laboratoire de Biologie des Invertébrés Marins et Malacologie, CNRS D-0699
55, rue de Buffon, 75005 Paris France.

Abstract:Starting from an analysis of the qualities of descriptions and of the observational mechanisms, this paper illustrates the interest of using a "naturally" structured representation of observational data. A particular attention is given to the formalisation of descriptive concepts and their corresponding representation from a biological point of view. The main goal is to build classification (class definition) and identification systems that take into account diversity, inter-dependancy and variability of observed characters and to handle as well as possible incomplete information.

1.

The central role of descriptions in natural sciences

The so called observational sciences have to deal with the ability to analyse the reality of things, in a word to describe. The description activity is so straightforward that we may wonder why we are analysing it. Yet every one knows that there exists some good descriptions and some bad ones, and that their use raises a lot of problems.

Why do we need descriptions? What are their qualities ? Here are questions for which we should bring clear answers, before searching which computer solutions are likely to facilitate and improve the activity of describing that occur for example when classifying or identifying living creatures and other natural objects.

1.1

Goals of a description

The description of different entities that compose our world appeared in the early antiquity as the fundamental way to increase knowledge. To "learn" what is an animal, a plant, a rock, etc., one needs to observe, but in addition to make a cognitive representation (for oneself) or a written representation (for others). The transmission of knowledge implies the notion of description.

A scientific description is an objective abstraction. It is an abstraction because it allows to free oneself from real observations that gave substance to it; and objective because it does not admit interpretation. Ideally, there is no distortion but a simple transcription "as identical" of concrete features of the observed subject into characters or characteristics that are represented. Traditionally the representation is made under textual form, using pictures to illustrate it; today, we can also use a lot of media that allow more power and flexibility.

One proceeds to describe, in the first place, in order to increase the number of particular (individual) descriptions, and next to learn about Nature at a more general level and better understand it. Qualities expected from descriptions are derived from this double objective.

1.2

Qualities of a description

We saw that the essential quality of a description is its objectivity, a perfect description is at the same time true and complete. Any method that is aimed at describing more easily must therefore allow to cover all observable features and to express them exactly, without ambiguity. In this way, the information content of the description is maximised. Ideally, a