Toward a Phylogenetic Classification of the Recent Birds of the World (class Aves)

--This paper proposes that avian classifications hould express the content of natural groups, that is, taxa that are our best estimates of genealogically related groups of species. The information content of classifications consists only of the hierarchical arrangement of groups within groups, and phylogenetic classifications can store and retrieve such information precisely because genealogical relationships are hierarchical in their logical structure. Present avian classifications, to the extent that they are based on overall (phenetic) similarity or on elevating morphologically distinct taxa to ranks higher than those of their close relatives, do not allow information about natural groups to be stored and retrieved unambiguously. This paper is a first attempt at a phylogenetic classification of the higher taxa of Recent birds. Most currently recognized orders and families are probably monophyletic (with the exception of some family-rank taxa in the Passeriformes), but present classifications do not adequately express the interrelationships of these groups. Twenty orders are here classified into nine Divisions: Division 1 contains the Sphenisciformes, Gaviiformes (including grebes), Procellariiformes, and Pelecaniformes; Division 2 the Palaeognathiformes; Division 3 the Ciconiiformes and Falconiformes (including owls); Division 4 the Anseriformes and Galliformes; Division 5 the Gruiformes, Charadriiformes, and Columbiformes; Division 6 the Psittaciformes; Division 7 the Cuculiformes; Division 8 the Caprimulgiformes and Apodiformes; and Division 9 the Piciformes, Coliiformes, Coraciiformes, and Passeriformes. The families of each order are arranged into taxa that seem best to express our current knowledge of their phylogenetic relationships. Received 26 November 1980, accepted 12 May 1981. HISTORICALLY, biological classifications have existed to convey knowledge about the structure of nature. With few exceptions natural historians have recognized that the knowledge to be specified in classifications concerns the identity of what are called natural groups. Indeed, it is the search for natural groups that has stoked the passions of countless biologists since the time of Aristotle. There have been, and continue to be, differences of opinion as to what exactly is meant by naturalness, but a general consensus has emerged, one that can be said to apply to preas well as to post-evolutionary times: a classification is natural to the extent that it conveys knowledge about the hierarchical structure of nature. This conception of natural classification represents the basic historical theme of biologists' attempts to classify; in modern systematic biology arguments exist over alternative conceptions of the hierarchical structure to be expressed in classifications, not over whether the structure of nature is hierarchical. Thus, knowledge about the heirarchical structure of nature can be specified either by a branching diagram or by a Linnaean classification: it does not really matter which, because each depicts the set membership of the included taxa. This paper adopts a view of natural classification that has existed, to one degree or the other, within biology for nearly 200 yr: the hierarchy of nature consists of groups within groups, produced as a result of genealogical descent. The purpose of a natural classification, therefore, is to express our current state of knowledge about the genealogical relationships of taxa. 681 The Auk 98: 681-714. October 1981 682 JOEL CRACRAFT [Auk, Vol. 98 THE LOGICAL STRUCTURE AND INFORMATION CONTENT OF CLASSIFICATIONS As implied above, the knowledge, or information, contained in branching diagrams or in Linnaean classifications i the same; it is the arrangement of groups within groups. This isomorphy in information content between branching diagrams and classifications has not always been recognized or accepted by systematists, and this has led to extended, unnecessary discussion and debate in the systematic literature. Because it has an obvious bearing on the scientific content of avian classifications, and because it will be illustrated repeatedly later, this relationship needs further amplification (see also Nelson 1972, 1973; Cracraft 1974a; Farris 1977; Wiley 1979; Eldredge and Cracraft 1980; Nelson and Platnick 1981). Linnaean classifications consist of taxa (species or groups of species) placed in various taxonomic categories (genera, families, etc.). The logical structure of Linnaean classifications is such that taxa can be classified by subordination, that is, inclusion in another taxon of higher categorical rank, or they can be listed in a sequence with one or more taxa of equal rank (see Nelson 1973, Eldredge and Cracraft 1980: 165-171). Given this, a Linnaean classification can specify only the set-membership of its included taxa. In other words, Linnaean classifications do not necessarily contain intrinsic information about phylogenetic relationships or about genetic or morphological similarity, only about the hierarchical structure of groups within groups. So if biologists are to speak of the information content of classification, then the latter must be the set-membership of taxa. Systematists compare taxa, and, using some similarity relationship, they cluster these taxa into sets that are then classified according to some criterion. There are alternative methods for comparing and clustering taxa, the most well-known being embodied in the systematic philosophies of numerical taxonomy (phenetics), evolutionary systematics, and phylogenetic systematics (cladistics). If the sole information content of a classification is its set-membership, then it is logical to conclude that the maximum amount of information for a given classification is obtained if the setmembership retrieved from that classification precisely reflects that of the clustering methods that were originally used to form the classification. Cladistic hypotheses can be interpreted as hypotheses of genealogical relationships. Taxa are clustered on the basis of synapomorphies (Hennig 1966), and the taxa so formed are postulated to be strictly monophyletic. Cladistic hypotheses (cladograms) are hierachical in structure, and, whereas an analysis of derived similarity is used to form these hypotheses, the latter do not, in themselves, contain any exact information about the nature of those similarities, only that a pattern of nested synapomorphy is assumed to be present (see Eldredge and Cracraft 1980, Nelson and Platnick 1981). What cladograms do contain is a hypothesis of set-membership (presumably describing the genealogical hierarchy), and this can be classified within a Linnaean system. More important, if classified so as to preserve that set-membership, then the original cladisfic hypothesis can be retrieved precisely from the classification: there is an isomorphy between the original genealogical hypothesis and the classification in terms of the sets of taxa that are recognized. In this sense, then, phylogenetic classifications constitute hypotheses about the structure of nature: they are repositories of our knowledge about the membership of natural (genealog-