Character is a very ambiguous word (Colless, 1985), but it is used here in the traditional sense of both a characteristic that can be used to differentiate taxa and the instantiation of this character in a particular individual. See Character representation.

The difference between taxonomic and identification characters is well known but not always applied: taxonomic splitters create taxa because they can be differentiated (identified). (This is about as absurd as pretending that races exist in Homo sapiens just because some groups of H. sapiens specimens can be identified.) At the opposite, persons interested in identification sometimes complain that some new taxa are proposed without adequate accompanying descriptions showing tendencies that characterize the taxa. Actually, taxa should be proposed for taxonomic (phylogenetic) reasons. Whether they can be identified by traditional means (naked eye or microscopic examination of morphological-anatomical characteristics) is irrelevant.

This being said, it remains that most species can be identified from morphological-anatomical characters. Even with sibling species undistinguishable from each other by traditional means, the group they form (super-species) can be identified from the other species in the genus. Therefore, it is necessary to address the topic of morphological-anatomical identification characters.

The Quest for the Perfect Identification Character

Obviously, for a character to be used for identification of a taxon it must be present in the members of the taxon and absent outside this taxon. There must be a "gap" between the range of character states observed in one taxon and the range observed for the same character in another taxon.

Like Thiele (1993) and other authors, we do not consider that there is a difference in nature between quantitative and qualitative characters, as most (all?) qualitative characters can be expressed quantitatively ‹ by ratios, as suggested by Thiele, transforms (for outlines), numbers (absent = 0, present = 1 or more), etc.‹ and that, conversely, any quantitative value can be stated in a qualitative way ‹ as in small, numerous, etc.

Beyond this well known "gap" requisite, we make a difference between "easy" identification characters and the characters that are not so easy to observe/measure in specimens of the taxon.

Primary Identification Characters

The concept of Primary Identification Characters was proposed in Fortuner 1989 and Fortuner 1993 (See publications). The basic idea is that identification should rely mostly on characters that are easy to observe in the unknown, which means that they present a minimum risk of errors.

Using our representation of characters, a Primary Identification Character must:

Describe a reasonably visible structure; obviously, the length of a clearly visible organ is more to be trusted than the length of a structure that has indistinct outlines, is not visible by conventional means, or blends into its background (similar color or texture);

represent a basic property that is not ambiguous in the taxon considered; for example, "presence" of seeds in raisins is very ambiguous as "seedless" raisins do have seeds, but very small ones.

have low variability in the taxon, or at least its variability should be well described and clearly delimited and there must be a "gap" between the range of character states/values in the specimens that belong to the taxon and the specimens that belong to other taxa.

Selection of Primary Identification Characters

Primary Identification Characters are taxon-specific, as visibility of structures, ambiguity of properties, and range of states/values vary from taxon to taxon. In Genisys, structure visibility and basic property ambiguity are entered as metadata. Range of states/values can be computed from the database.

Use of Primary Identification Characters and Secondary Characters

Different identification tasks may use different characters. For example, elimination is a very drastic procedure, one that is most susceptible to the adverse effects of errors. Pure dichotomous methods have absolutely no graceful degradation and one wrong choice in a couplet of a key eliminates the correct answer. It is therefore important that elimination relies on primary identification characters only. The other characters (secondary identification characters, those that failed in at least one of the conditions above) should not be used at all for elimination purposes.

Comparison by, e.g., similarity coefficient is a less risky procedure as the taxa are ranked but none are eliminated. Also, a coefficient is computed on several characters taken simultaneously, which makes for built-in graceful degradation. For example, if a coefficient is computed from 20 characters, an error made on one character would result in a 95% correct coefficient. Accuracy can be improved even further by attributing a higher weight to primary identification characters. Secondary characters can be used but a lower weight should be attributed to them. The coefficient of endorsement could be used to weight characters.

_____________________
Colless, DH (1985). On "character" and related terms. Syst. Zool. 34: 229-233.

Thiele, K. (1993). The holy grail of the perfect character: the cladistic treatment of morphometric data. Cladistics 9: 275-304.