Cladistic, stemmatics and the role of humanities.

I recently read (and discussed with its author) Philipp Maas' Computer Aided Stemmatics –The Case of Fifty-Two Text Versions of Carakasaṃhitā Vimānasthāna 8.67-157 (to be published (to be published in WZKS 2009-2010). This paper is the first clear introduction to cladistic as applied to the problem of building a stemma I had the pleasure to read. It is (almost) exhaustive, precise, and also easy enough to be understood by an outsider.
Cladistic is a device developed within biology and the study of palaeontology. It links characters within a tree (called cladogram) and in this way traces today's extant individuals to their common ancestors. Cladistic analysis has been used in the last 20 years also to detect common ancestors (hyparchetyp) among the manuscripts of a lost original text.
Maas' article states at the outset one of the oddest problem of the use of cladistics:
«The question to deal with is whether the variants used by the computer program to establish the branching of the tree really reveal the genealogical relationship of manuscripts.»

In other words: how can we trust the computer to build a stemma based on variants whose different significance it does not know? WE know that a lacuna is much more important, as a conjunctive variant, than a typo. The computer does not!

Hence, explains Maas, a critical judgement of variants in inevitable:

«My approach here is similar to that of Salemans (2000) in so far as we both combine a cladistic analysis of variants with their philological judgement. An important difference is, however, that Salemans decides a priori which kinds of variants reveal relationships and only then analyzes the mechanically selected variants with cladistic software, whereas I start with a cladistic analysis of the complete set of variants and decide upon the quality of variants only a posteriori».

In sum, the Salemans-type editor (if I am understanding correctly Maas) decides not to feed the cladistic software with non significant variants (such as typos, phonetic variants, minor variants which can occur more than once in the history of the transmission…). On the other hand, Maas fills the software with almost all possible variants (he also excludes some minor graphic ones) and then emends the cladogram according to their value.

This is not a minor difference, insofar as Maas' approach gives at first (almost) full confidence to the software and only later modifies the output, whereas the Salemans-like editor directs the software towards what SHE decides to be the right interpretation.

Why does Maas include also minor variants? Because, he explains, «Mistakes that can be easily corrected do not reveal the genealogical relationship of manuscripts in their own right, If, however, these variants occur frequently within a genealogically closely related group of witnesses, they add credibility to the stemmatical hypothesis.» That is, even minor variants may have a role. One cannot exclude them all a priori, it cannot be a coincidence if, say, manuscripts A and B share a huge amount of these "minor" variants.