Portal de Periódicos da CAPES

Nelson’s arrested development

2012; Wiley; Volume: 28; Issue: 6 Linguagem: Inglês

10.1111/j.1096-0031.2012.00404.x

1096-0031

James S. Farris,

Botanical Studies and Applications

Sir, As Williams and Knapp (2010, p. xi), presented it, the cause of Nelson’s (2004)“arrested development” was unidentified: Nelson’s title [“Cladistics: its arrested development”] implies... that there is still a way to go, progress being halted, unnaturally so, and the cause of the arrest obscure and unidentified. Wheeler (2008, p. 5) instead traced the cause to neglect of morphology: Unless we continue to discover, clarify, corroborate and map distributions of morphological characters, much of the rationale for doing phylogenies—the need to interpret special similarities in their historical context—is gone. I agree with those who suggest that the Hennigian revolution is unfinished (e.g. Nelson, 2004). To Nelson (2004, p. 127) himself, however, the cause was perfectly clear: The subsequent development of cladistics has been arrested, too, by, computer implementations of character optimization [parsimony] and by the ideology of total evidence, which reflects a phenetic rather than cladistic objective: the overall similarity of synapomorphy. Why, then, did Nelson’s supporters not say that? Perhaps because they found his arguments embarrassing. As I will point out here, Wheeler’s (2008) and Mooi and Gill’s (2010) appeals to Nelson are actually undercut by Nelson’s own attempts to rationalize his position, and the same attempts also reveal an important flaw in his defence of 3ta (three-taxon analysis of Nelson and Platnick, 1991). Consider first Nelson’s (2004, p. 139) comments on total evidence: How do we test the secondary homologies?... According to one viewpoint, sloganized as total evidence, the only way is to add more data to the matrix... Recall phenetics and its hypothesis of the matches asymptote... the wishful thought... that beyond a certain number, adding more characters to the matrix does not change the result... With total evidence, need no such asymptote be expected? If it were, we might witness the reemergence, even the vindication, of phenetics as the overall similarity of synapomorphy. One would think from Nelson’s remarks that only pheneticists hope for stable results, whereas in fact anyone would. Advocates of total evidence, similarly, are scarcely alone in using new data to test phylogenetic conclusions, and Nelson suggested no other test, so that connecting phenetics with total evidence was much like calling Hitler a libertarian on the grounds that he ate tomatoes. Nelson’s fancied vindication, further, was patent nonsense. The overall similarity used in phenetics is similarity in all states—plesiomorphic as well as apomorphic—whereas synapomorphy is naturally restricted to apomorphic similarity, so that “the overall similarity of synapomorphy” is simply self-contradictory. I suspect that Nelson’s misconception actually arose from an elementary confusion. Total evidence—properly simultaneous analysis sensuNixon and Carpenter (1996)—requires including all available characters (transformation series) in an analysis. Phenetics requires including all states in overall similarity, but states have often been called “characters” too, so that “all characters” can be ambiguous if taken out of context, and of course confusing characters with character-states is most common among older systematists.1 Another kind of confusion cropped up in Nelson’s (2004, p. 137) discussion of parsimony. He managed to forget the subject of Platnick et al.’s (1996) paper—a paper that he had co-authored: These few matrices show that the [parsimony] programs fail to find nodes where they should and they find nodes where they should not, so that the theory of optimization, implemented in the programs, is not perfect (Platnick et al., [1996]). Platnick et al.’s argument had nothing to do with Nelson’s (2004) matrices, instead proposing a different excuse for adopting 3ta (Platnick et al., 1996, p. 247)2: The three-taxon approach to phylogenetic analysis implements interpretation 2 of Nelson and Platnick (1980), whereas the standard approach [parsimony] implements only interpretation 1. That assertion involved some confusion in itself, as it turned out that Platnick et al.’s own examples contradicted their claim that parsimony amounted to interpretation 1 (see Farris, 1997, 2011)—perhaps that accounts for Nelson’s loss of memory. As for Nelson’s (2004) matrices, they fell into two series, the smallest members of which are illustrated in Fig. 1, the larger members having more terminals and/or characters while showing similar patterns of 0s and 1s. According to Nelson (2004, p. 137): Hypothetical matrices used in Nelson’s (2004) arguments. Matrix 1 is redrawn from Nelson’s (2004) fig. 6.4. Matrix 4 is redrawn from Nelson’s (2004) fig. 6.7. In both matrices, the outgroup O is supposed to have all plesiomorphic states. Matrix 4 [Fig. 1] is one of a series of matrices, again purely hypothetical, that contain conflicting characters with, perhaps, no evidence of relationship. Nevertheless, the [parsimony] programs see matrices of this series as fully informative. For each matrix, the [parsimony] programs return a single resolved tree. Matrices 5 and 6 (Figure 6.8) are two more of this series, which can be extended indefinitely. There is reason to view these matrices as uninformative, but I will not argue the point here. Apparently Nelson had not bothered to check his calculations. For matrix 4, parsimony does yield a single resolved tree (O(AB)(CD))—but then so does 3ta. In fact the 3ta tree matches that produced by parsimony for each of the matrices in that series, so that such cases can hardly provide any grounds for rejecting parsimony in favour of 3ta. Turning then to the other series, for matrix 1 (Fig. 1) Nelson (2004, p. 136) advocated concluding group (BCD), which is the only informative group on the strict consensus (OA(BCD)) of 3ta trees (although he did not so identify it). Dating from Nelson (1996), the claim that matrix 1 provides definite support for (BCD) has been defended on a variety of pretexts (reviewed by Farris, 1997, in press),3 none of which has addressed the difficulty that such support would lead to a contradiction. That problem arises because 3ta assumes irreversibility, that is, does not apply reversals (Nelson and Platnick, 1991, p. 363): There may be some difficulties in implementing the three-taxon approach for unrooted algorithms, because the approach does not score apparent reversals as synapomorphic. If matrix 1 supported (BCD), then the stem-species of (BCD) would need to have at least one apomorphy among the characters of that matrix. In fact, however, a plesiomorphic 0 occurs in one of B, C, or D for each of those characters, so that under the irreversibility assumption, the stem-species of (BCD)—if it existed—must have been entirely plesiomorphic (Farris, 1997). Nelson has never addressed that contradiction directly, but has instead adopted a position under which the problem would be avoided (Nelson, 2004, p. 131): To associate biological data [states] with nodes has been a concern of cladistics from its beginning... From the relational viewpoint, however, this litany of so-called cladism reads like Matthew’s paleontology, the search for the true sequence of ancestors and descendants. By “relational,” Nelson meant non-transformational. He thought that abandoning the idea of evolutionary change would justify dispensing with the bother of associating states with nodes. In that case, the contradiction just described would presumably not apply, but that way of saving (BCD) comes at a price. By “[parsimony] programs fail to find nodes where they should,” it is now seen, Nelson actually meant “parsimony fails to find nodes that can be defended only by resorting to creationism.” Parsimony, then, has come through unscathed. Whether rationalized on creationist grounds or not, Nelson’s 3ta-motivated refusal to associate states with nodes is a problem in itself, as can be seen from Mooi and Gill’s (2010, p. 27) observation on the importance of synapomorphies: If phylogenies are based on the discovery of character homology, or synapomorphy, this evidence must be provided in support of the tree. It was once common practice that proposed synapomorphies were provided for each node so that others could then evaluate the evidence and determine if the homologies/synapomorphies/ evidence would hold up, or they could present contrasting evidence. Providing proposed synapomorphies for each node is exactly what one cannot do if states are not associated with nodes. Remarkably, Mooi and Gill cited Nelson (2004) with great approval and seem to have supposed that their views were compatible with his—apparently they had not read his paper carefully. Gill’s colleague Wheeler (2008, p. 5) appeared to have been similarly inattentive4: Unless we continue to discover, clarify, corroborate and map distributions of morphological characters, much of the rationale for doing phylogenies—the need to interpret special similarities in their historical context—is gone. I agree with those who suggest that the Hennigian revolution is unfinished (e.g. Nelson, 2004). Hennig (1966) advocated a phylogenetic system and meant it, insisted on monophyletic groups, and most definitely associated groups (nodes) with synapomorphies. Nelson (2004) wanted to be rid of that association, seemed to think that evolutionary change had nothing to do with “cladistics” or “phylogeny,” and had long advocated 3ta, which can form paraphyletic groups (Farris, 2011; Farris et al., 2001). In short, Nelson seemed much less interested in completing Hennig’s revolution than in trying to sabotage it. Fortunately for systematics, Nelson’s counterrevolution has never had much success.