Song Features as Species Discriminants in American Warblers (Parulidae)

Abstract

-Using multivariate discriminant analysis, we examined 337 songs of 19 species of wood warblers sympatric in New Brunswick, Canada. We divided the warblers into five overlapping groups of species based on habits and songs. Our hypothesis was that song features would be the most reliable at high noise levels or under conditions of poor transmission. Hence, we predicted that within most of these groups the songs would segregate highly on the basis of song features alone, as opposed to features of individual sounds or phones. In four of the groups the analysis correctly classified 84 to 95% of the songs on the basis of song features alone. In Group 5 (Yellow, Chestnut-sided, Redstart, Magnolia warblers), only 68% were correctly classified on the same basis. The addition of phone features to discrimination for this group increased the correct classification to 85%. In some groups the frequency modulation patterns of the phones are so simple that they contribute little to improved discriminations. The relative contributions of phones may be functionally related to the importance of possible competitors, the distance of the communication and other noise factors. and to the relative development of repertoires. An important function ascribed to bird song is the conveyance of information by which members of different species recognize their own kind. This role has been demonstrated in pairs, or, at most, small groups of species of wood warblers (Family Parulidae; Stein 1962, Ficken and Ficken 1967, 1969, Gill and Murray 1972) and other birds (Thielcke 196 1, Stein 1963, Catchpole 1973, Becker 1976). In such studies the songs were compared either subjectively by visual analysis of sonograms or by measuring a limited number of variables. Such comparisons are less satisfactory, however, when considering many species interacting in a natural environment. In that case, an objective method to compare quantifiable attributes of song is needed. In this paper, we examine relationships of songs among 19 sympatric species of warblers, using multiple discriminant analysis to objectively weigh differences among the songs. This multivariate technique is used both descriptively and to test an initial set of predictions on the structure of bird song necessary to achieve functional species recognition in species-rich habitats. To our knowledge, this approach has been used to only a limited extent (Sparling and Williams 1978, Brown and Lemon 1979). Two general types of variables characterize the structure of bird song. Songs of most species are generated from a limited set of basic sounds, continuous in time, which we call “phones.” A song may be simply the clustered repetition of a single phone, or it may have combinations and sub-groupings of phones. We use these two levels of features-phone and song-to define variables that are useful in any comparison of songs. Playback experiments on species recognition have shown that particular details of phones are sometimes essential to evoke a maximum degree of response (Falls 1963, Emlen 1972, Shiovitz 1975, Marler and Peters 1977, Shiovitz and Lemon 1980). Response to such detail is often limited even to members of the particular vocal community or dialect (Lemon 1969, Milligan and Verner 197 1). Sounds are often distorted, however, through masking, attenuation and related phenomena, depending on the presence of other species’ songs, distance, variable densities of air, etc. (Wiley and Richards 1978, Lemon et al. 198 1, Richards 198 1). Consequently, overall features of song, as opposed to precise details of phones, should be expected also to contribute strongly to species recognition in natural environments. For example, among certain warblers, the species differ by song characteristics such as range and mode of frequency, as well as by groupings of phones (Ficken and Ficken 1962, Lemon et al. 198 1). Because the problems of distortion should apply to details of phones more than to more general features of timing and sequence, we