Specimen Identification and Key Construction With Time‐sharing Computers

Abstract

Current methods of specimen identification are surveyed, and algorithms are given for constructing dichotomous keys and for identifying specimens directly with a polythetic polyclave procedure. Although usable without computers, these techniques have been automated as Fortran programs on a time-sharing system. Taxon/character data are encoded as a hierarchy of data matrices for use with the various programs. Largescale taxonomic information systems are also briefly discussed.

Regional and Local Genetic Variation in Japanese Stiltgrass (Microstegium vimineum)

“…The process of determining an I-Rank is therefore polythetic (cf. Morse, 1971), drawing upon an overall fact pattern that does not require any prespecified set of individual questions to be addressed. Answering the more heavily weighted and/or more easily addressed questions facilitates determination of the I-Rank, which is a skill gained through experience.…”

Theoretical Aspects of Numerical Identification

“…Much has been written about identification keys (10,12,13,15), and some common themes can be discerned. The most important of these themes appear to be the following.…”

Character Selection During Interactive Taxonomic Identification: “Best Characters”

“…A polyclave can be difficult to use correctly, such as by novices who make frequent errors, or for polythetic taxa which are distinguished by possessing a majority of a set of character states, rather than by required states (Morse 1971(Morse , 1975). These problems can be treated with polythetic polyclaves that provide an "error-tolerance" or "mismatch threshold" setting, and by weighting character states and taxa (specifying which character states or taxa are rare).…”

“…The coefficients have names such as "Best character" (Dallwitz et al 1998(Dallwitz et al /2012, "Separation coefficient" (ETI Bioinformatics undated), "dichotomizing value" (Morse, 1971), or "Best" (Brach and Song 2005;Pankhurst, 1978;Lucidecentral.org 2010). Other software provides a choice of coefficients (Ung et al 2010), such as variants of a Jaccard coefficient, or Simple Matching coefficient (also called "Sokal & Michener coefficient", Sneath and Sokal 1973).…”

“…The separation coefficient is the separation number divided by the number of possible pairs, so it is normalized to the Table 1, petal color = white exhibits overlap). Morse (1971) extended the separation number to count the extent of overlap, but only for binary characters. The Simple Matching ("Sokal & Michener") coefficient, the ratio of character-state matches to character-state pairs, and the Average Pairwise Jaccard Distance (which excludes "negative" , also called "absence" matches, i.e., shared notapplicable states) are similar, relatively simple, calculations.…”