perfect description should permit to rebuild exactly the primitive object; practically, a description is satisfactory when it gives a "good estimate" of this object, especially concerning its specificities. This implies to take into account not only the descriptive characters, but also the different links (topologic, relational, of dependancy etc.) that exist between these characters, because these links carry information too.

Aditional qualities can be mentioned such as clarity and concision, as for all scientific writings. Some authors give attention to the elegance of the text. It is rare to mention understandability as a quality that makes description more easily understood by someone who is not a specialist of the domain. This means that one should use a less technical vocabulary, with the possible counterpart of loosing accuracy and conciseness : so there is some compromise to be found, still waiting for a solution that allows to adapt the "level" of the description to the user. However, it is not sufficient for a description to be excellent in itself : it needs moreover to allow comparisons with other descriptions.

1.3

Qualities of descriptions

With the classification or identification goal in mind, our main concern is to compare description one another. When these descriptions have been written by the same author, they follow generally a common framework, and it is easier to compare them because homologous characters are located in corresponding parts of the texts. But when the authors are different, they could follow heterogeneous observation "methods"; comparisons between descriptions are then more difficult to achieve.

The notion of homology is very important; it allows to ensure that only comparable characters are compared. It is based on the fact that every biological object has an organization plan which is found identical in the other objects of the same kind. Recognizing and taking into account this general constitution plan (bauplanin German) allows a natural structuration of descriptions, following what we will call a descriptive model.

Remark: The above considerations virtually concern all natural subject descriptions. Although, both for classification and identification, each specialist restricts his studies to a more particular domain, like a zoological or botanical group, and/or a geographic area, and/or an ecosystem etc. In the following paper, it is a delimited domain instead of a "universal system" still not reachable at the present time that we have in mind.

2.

Representing descriptive data

We start from the point that we can understand only what we can model, and that it is better to adjust the model to the reality than the reverse. We will study in more details what are the elements that constitute a description, and how they are arranged together by the descriptor (i.e. the person, generally the specialist, who makes the description, and not a described character which should be called ... a descriptum). We will deduce from this how descriptive models must be conceived, according to the above quality constraints.

2.1

Natural Structuration

As an example, let us take a particular domain : farm animals. In such a domain, anyone is a "specialist". Let us see how the specialist will classify and identify these animals.

First observation : all these animals have four limbs, two anterior, two posterior. The anterior limbs are either legs, or wings for poultries. So we learn that there are two principal categories, that the specialist will immediately name Mammals and Birds (with capital letters because we are in a scientific area). Next, among the Mammals, the cat and the dog of the farm are distinguished from others because they eat meat. Here are two other categories : Carnivorous for them, Herbivorous for the others. Among the Carnivorous, there is the Cat