This last distinction is important. When I am describing a Cat, if I say it has no tail, then this information leads to the fact that it is a member of the Manx race (cats without tail of Man Island), unless it is an accident and I know that it "had" a tail. On the other hand, if I do not mention the tail, I bring no information; the "value «unknown»" which is often invoked in this case is non-sense, or worse an artificial way to treat as an information what is not. The most convenient way to treat unknown facts in a description is to leave them blank.

2.1.2

"Point of view" logic

It often happens that a description of a natural object might be done at different levels. For example, it will focus on morphology, anatomy, cytology, or again on biochemistry or the genetic map. This is true anyway for each of the observational parts. The information attached to the different points of view are linked by the existing structural relations between these various observational levels.

Practically, the "point of view" logic is very similar to the composition logic. However, it doesn't have as rich a semantic; the fact that one level of analysis cannot be accessible for a given part of the description doesn't imply that this level remains inaccessible when describing its subparts. Another difference can be found when processing classification : a missing subpart will be taken into account whereas a missing point of view is devoid of classificatory signification.

One of the major interests of defining a descriptive model is to preserve the homology of characters even between different levels of observation. Thus, the descriptive model is a way to index knowledge and position it in order to compare it to others; it corresponds somehow to the relational and/or hierarchical structures in data bases.

2.1.3

Specialization logic

Let us come back to the farm animals, supposing that we can make use of a classification about different kinds of breeding farms. If we know nothing about "our" farm, the general model of bred animals contains four limbs, but if we know that it is specialised in aviculture, we can start from a more precise model, animals with two wings, two legs, a beak, feathers, or on the contrary without horns, teeth, etc.

The fact that a more precise concept of our farm is available, at an already abstract level, allows us to restrict the area of domain knowledge, and gives information in advance (without real observation) about some local descriptions. This mechanism, called specialization, is so general that it can appear in a lot of descriptions written by naturalists, in place of true local descriptions. Thus, simply stating that our farm breeds aquatic birds (ducks for instance) partly replaces a description about legs (that are always web-footed) or feathers (always watertight).

The specialisation is a convenient short cut : it allows to fill in "by default" (by inheritance) whole or part of a real local description by a conceptual one. Of course, there is a risk to be imprecise, or moreover to be incorrect. It is thus necessary to complement "manually" the deduced information.

2.1.4

Logic of exceptions

Whereas specialisation is the process of restricting the observable domain, exception is conversely a way to enlarge the current domain in order to handle particular cases. Suppose we learn that our farm does aquaculture; thus no more animals with four limbs, but fishes ("pisciculture") or even oysters ("ostreiculture") The descriptions will have to take into account characters about scales, fins, or shells. If those characters were not present in the general model of farm animals, it is needed for this particular case to be extended.

This process is complementary to the specialisation one even if it appears as a complication (like some "patches" that take place in computer programs). It seems better to follow this