THE CONTROL OF SINGING IN VARIED THRUSHES by CARL LINN WHITNEY B.S., Iowa State University, 1970 M.SCi, The University of British Columbia, 1973 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Department of Zoology) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA April, 1979... (c) Carl Linn Whitney, 1979
In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by. the Head of my Department or by his representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. n 4. * Zoology Department of _ The University of British Columbia 2075 Wesbrook Place Vancouver, Canada V6T 1W5 22 April 1979
i i ABSTRACT Male varied thrushes (Zoothera naevius) have repertoires of three to seven different songs. A typical song is a pure tone, about two seconds long and modulated in amplitude; i t can be described quantitatively by its dominant frequency (Hz) and period of modulation;.. Beduced to these two dimensions, the songs within repertoires are more dissimilar to each other than would be expected i f they were drawn at random from a l l the songs in the population. This pattern is consistent with a theory that explains the adaptive significance of song and song repertoires. According to the theory, song helps territorial males to repel other males and to attract and stimulate females, and repertoires enhance these effects by reducing the rate at which other birds habituate. This being so, i t is to the advantage of males to have repertoires of dissimilar songs, since other birds will generalize less between the songs and will therefore habituate at a slower rate. The relationship between generalization and similarity of songs forms the basis of a theory of how singing is controlled by the brain of a varied thrush. The theory assumes that a control center feeds motivational impulses to units in charge of the different songsi When the motivational state of a unit reaches a threshold, the unit produces a song and then inhibits i itself, dropping abruptly to a lower motivational state. At the same time, i t inhibits each of the other units an amount
that is directly proportional to the similarity of the songs produced by the two units. Some random variability in inhibition is not explained by this relationship. k simulation model based on this theory accounts for the following patterns found in sequences of songs produced by varied thrushes: 1) Once a song has been sung, i t is not usually repeated immediatelyi 2) Songs are not sung with equal frequency. 3) The most common song in a repertoire is more likely than any other to be repeated immediately. 4) There is a negative correlation between the lengths of successive recurrence intervals (where a recurrence interval is defined as one plus the number of other songs between successive repetitions of a given song). 5) The two most dissimilar songs in a repertoire are more likely to be sung in succession than would be expected i f they occurred independently of each other. 6) When several of the songs in a repertoire are dropped and replaced by others, the ones that remain occur in different relative frequencies;, 7) Two very similar songs in a repertoire can occur with quite different frequencies. 8) Intervals between successive songs are variable but are usually at least several seconds long. 9) The average lengths of the intervals following different songs are positively correlated with the relative frequencies of the songs;. 10) The absolute variability of the intervals between successive songs is positively correlated with the rate of singingi The theory can be elaborated to account for patterns of singing in other species.
iv TABLE OF CONTENTS Page ABSTRACT..... i i TABLE OF CONTENTS...... ^... iv LIST OF TABLES.... :... 1... :... vi LIST OF FIGURES... '.... v i i i ACKNOWLEDGEMENTS...... x i i I.. INTRODUCTION........... 1 THE BIRDS AND THEIR SONGS 3 ADAPTIVE SIGNIFICANCE OF SONG AND SONG REPERTOIRES 13 GENERAL METHODS... 17, II. SIMILARITY OF SONGS IN INDIVIDUAL REPERTOIRES 18 METHODS,.. -i-.'-i 24 RESULTS i U. 30 DISCUSSION 30 III.. CONTROL OF SINGING... i...,i i... ^... 52 INTERNAL AND EXTERNAL CONTROL... i i i........... 53 A FIRST MODEL.... i 58 INHIBITION BETWEEN SONG UNITS...i 69 FREQUENCIES OF DIFFERENT SONG TYPES... 86 INTERVALS BETWEEN SONGS, 97 SYNTHESIS... I... 1 117 IV. GENERAL DISCUSSION.... i l............. i. 122 EVOLUTIONARY SIGNIFICANCE OF TEMPORAL PATTERNS OF SINGING......i 122
V CONTROL OF SINGING,*... 125 Similarity of Songs in Individual Repertoires '.. 125 Control of Singing in Selected Species... 128
vi LIST OF TABLES Table Page I; Frequencies (and proportions) of different songs sung by the thrush named Mick in response to playback and in control sequence... 56 II. Frequencies (and proportions) of different songs sung by the thrush named Bingo in response to playback and in control sequence..,, 57 III. Sequences of songs recorded from the thrushes named Buddy, Chuck, and Bod...,...;..,.:..:... 59 IV. Matrix of inhibition values....,.... 66 Vi. Part of a sequence simulated using'the values of inhibition from Table IV 67 VI. Belative frequencies of songs repeated immediately., 68 VII. Spearman rank correlation between successive 1 recurrence intervals.. 72 V i l l i Transition frequencies!between the songs of the thrush named Otis* ;. >.,:........ i........... 74 IX.. Transition frequencies between the songs of the thrush named John... i........ 75 X. Transition frequencies between songs for two ' simulation runs of 500 songs each...i...i... 76 XI. Matrix of inhibition values for repertoire shown in Figure 19... '.. 83 X l l i Observed/expected probability of the two most dissimilar songs in a repertoire being sung consecutively 87
v i i XIII. Frequencies (and proportions) of songs that were in the repertoire of the thrush named Fats both before and after he altered the repertoire... 98 XIV.. Frequencies (and proportions) of songs that were in the repertoire of the thrush named Simon both before and after he altered the repertoire... 99 XV. Spearman rank correlation between the relative frequencies of different songs within individual repertoires and the lenqths of the intervals following them...... 107 XVI. Product-moment correlations between 1) mean lengths of intervals between successive songs and standard deviations of interval lengths, and 2) mean lengths of intervals between successive songs and coefficients of variation of interval lengths;,,.. j,........................ 114 XVII. Matrix of inhibition values that will produce a fixed sequence: ABCDE ABCDE 12 9
v i i i LIST OF FIGURES Figure Page 1.. Sonagram of a typical varied thrush song;.;...:.. 4 2: Sonograms of typical varied thrush songs, showing differences in frequency (Hz) and period of modulation... 7 3,i Sonagrams of varied thrush songs more complex than those shown in Figures 1 and 2i,. 9 4.. Sonagrams of aberrant songs sung by a male varied thrush...... 11 5- Dominant frequencies and periods of modulation of songs sung by the thrush named Chubby... 19 6. Dominant frequencies and periods of modulation of a l l 150 songs sung by the 30 male thrushes... 21 7,. Distribution of varied thrush songs along the dimension of dominant frequency: observed distribution (solid lines) and theoretical distribution of best f i t (dashed lines)... 26 8. Distribution of varied thrush songs along the dimension of period of modulation: observed distribution (solid lines) and theoretical distribution of best f i t (dashed lines)... 28 9. Observed (shaded) and expected distributions of similarity to nearest song along the dimension of dominant frequency in varied thrush songs 31 10. Observed (shaded) and expected distributions of similarity to nearest song along the dimension of period of modulation in varied thrush songs... 33
ix 11. a. Sonagram of a rufous-sided townee song, showing introductory phrase and t r i l l of single, repeated, syllable., b*, Sonagrams of. ruf pus-sided towhee songs, showing variations in syllable length... 36 12. Observed and expected distributions of similarity to nearest song along the dimension of syllable length in rufous-sided towhee songs...... 3 8 13. Habituation of male rufous-sided towhees'to 1 playback of songs on their territories;,;..,:... 40 14. Response of rufous-sided towhees to playback of song A after having habituated to one of four songs. A, B, C, or D (see Figure 11b).,..,...,... 43 15. Responsiveness and rate of habituation, of rufous^ sided towhees to songs A, B, C, and D... 45 16,i A first model of the control of singing in varied thrushes....... 61 17. Behavior of f i r s t model of control... 63 18i Eelationship between the lengths of successive recurrence intervals for the most common song in a seguence produced by the f i r s t model... 70 19. Banks of dominant frequency and period of modulation of songs sung by the thrush named Fats* 78 20. Behavior of model in which song units not only inhibit themselves but also inhibit each other... 80 21. Relationship of the observed/expected probability of two songs occuring in succession and the similarity of the songs in a simulated sequence based on the repertoire shown, in Figure 19,.... 84
X 22. Relationship of the relative frequency of a song and the total amount of inhibition its controlling unit receives from other units in a simulated sequence based on the repertoire shown in Figure 19... 89 23. Changes in the repertoire of the thrush named Fats*,... i i ;.. 91 24. Changes in the repertoire of the thrush named Simons..... A. A 93 25. Relationship of the relative frequency of a song and the amount its controlling unit inhibits itself in a simulated sequence based on the matrix of inhibition values shown in in Table IV.. 95 26. Lengths of intervals between successive songs in sequence sung by the thrush named Simon;... A... 100 27. Lengths of intervals between successive songs in a simulated sequence based on the repertoire shown in Figure 1 9,... 102 28. Intervals between successive songs in a sequence, sung by the thrush named John. ;. 105 29. Asymmetry of inhibition between song units.in models having a stochastic relationship between the similarity of songs and the amount songs inhibit each other... 109 30. Sonagrams of two very similar songs in the repertoire of the thrush named Fats. 111 31. Relationship of the coefficient of variation of interval length to the average interval length in simulated sequences*... 115 32. Final model of the control.of singing in varied thrushes..118
xi 33* Elaborations of the varied thrush model to account for the control of singing in other species. 133 34. Intervals between successive songs in a sequence sung by a rufous-sided towhee...;... 136
x i i ACKNOWLEDGEMENTS I am grateful to my supervisor, Jamie. Smith, for his support and encouragement during the five years of this study, and most especially for his patience during the final months, when I was in short supply of this virtue myselfi The members of my research committee, Lee Gass, Charley Krebs, Robin Liley, Judy Myers, and Jamie Smith, helpfully criticized earlier drafts of this thesis. Susan Harrison and B i l l Webb backed me in my battles with the computer* Mary Perkins, a welcome companion on recording trips, helped me to lug equipment over the steep terrain of Seymour and Hollyburn Mountains. John Spence was a kindred spirit throughout my career as a Ph.D. student; whatever understanding I have of the philosophical issues that underlie this research can be traced to discussions we have had during the past five years. Together, John and Debbie Spence contributed to my mental health while I was writing this thesis by indulging with Joan Miller and me in weekly bouts of beer drinking. Joan Miller helped by drawing the figures and reading a number of different versions of the thesis* More importantly, she enthusiastically discussed my research with me and contributed much to the development and clarification of my ideas about how singing is controlled in birds*
1 CHAPTEB I INTRODUCTION One of the things that behavioral biologists attempt to explain about behavior is how i t is controlled by the nervous system; Observing that behavior is organized, that i t follows definite patterns, we ask: What are the mechanisms underlying these patterns? Dawkins (1976) has recently pointed out that, contrary to what many of us may think, a satisfactory understanding of control mechanisms does not follow as a natural consequence of detailed neuroanatomical and neurophysiological descriptions. This is especially true for behavioral patterns controlled by complex nervous systems. What is necessary, Dawkins argues, is not a complete (and hopelessly detailed) account of the activities of neurons and synapses during behavior but instead a "distillation of general principles at a higher level*" In other words, to summarize an analogy he makes between nervous systems and computers, a software theory based on general principles is a more appropriate goal than is a complete description of hardware: a set of programming instructions is a better explanation of control than is a wiring diagram. Dawkins (1976) believes that software principles can be inferred from behavioral evidence alone* This is not an entirely new idea. The ethological literature abounds with studies of the temporal patterning of behavior, many of them done under the assumption that the results would elucidate the
2 underlying control mechanisms (Slater 1973). An important part of Dawkins' contribution, however, has been to emphasize something that has not always been appreciated in the past: a detailed description of behavior may be no more enlightening than a detailed description of hardware unless i t is done in a theoretical context* In this thesis, I develop a software theory which ties together a number of patterns observed in the singing behavior of individual varied thrushes (Zoothera - naeyius). The remainder of this chapter and the whole of Chapter II build up to the theory, which is developed in Chapter III. In Chapter I, after introducing the birds and their songs, I develop an evolutionary argument that the songs within the repertoires of individuals should be more dissimilar to each other than i f they were drawn at random from a l l the songs in the population. This prediction is confirmed in Chapter II- In Chapter III, I begin with a very simple theory of control and elaborate i t, step by step, until i t accounts not only for the dissimilarity of songs within individual repertoires but also for patterns based on 1) the relative frequencies with which different songs are sung, 2) the order in which songs are sung, and 3) the lengths of intervals between successive songs*. Finally, in Chapter IV, I discuss the possible evolutionary significance of several patterns described in Chapter III, and I consider the control of singing in other species.
3 THE BIRDS AND THEIR SONGS la some parts of their range, varied thrushes are known as "tin whistle birds." The name refers to their song, which Louis Agassiz Fuertes, the wildlife artist; described as "a single long-drawn note, uttered in several different keys, some of the high-pitched ones with a strong vibrant t r i l l.. Each note grows out of nothing, swells to a f u l l tone, and then fades away to nothing until one is carried away with the mysterious song" (quoted by Bailey 1924). More prosaic but just as fitting is the description in Nature West Coast by Smith et al. (1973): "a single monotone note that varies each time, sounding like someone whistling and humming at the same time. " A sonagram of a typical song is shown in Figure 1. The song is a pure tone, about two seconds long and modulated in amplitude. It can be described quantitatively by i t s dominant frequency (Hz) and period of modulation, both measured from the sonagram. The songs of varied thrushes are familiar to hikers in the Coast Mountains around Vancouver, but the birds themselves are often overlooked. Slightly smaller than robins, which they resemble in general appearance, they have orange wingbars and an orange eye stripe* Males have a black breast band, females a gray breast band and duller orange feathers. Only the males sing. I studied the singing behavior of varied thrushes during three summers, 1974-76, at two areas about 1000 m in elevation
4 Figure 1. Sonagram of a typical varied thrush song* Left set of axes shows a time versus frequency (Hz) plot of the beginning of the song; the remainder of the song is similar. Period of modulation is defined as the interval (measured in seconds) between successive pulses. Right set of axes shows a plot of relative intensity versus frequency taken at a single point of time during the song: Dominant frequency is defined as the frequency (Hz) at which intensity is greatest.
.3.4 increasing intensity
6 on Seymour and Hollyburn Mountains near Vancouver. Thrushes breed here in the subalpine forest. They arrive in April when snow s t i l l covers the ground, and males establish territories on which they can be heard singing until late July. Perched typically in the tops of conifer trees, they sing one song after another, the songs separated by a few seconds of silence. The songs are usually loud, and can be heard as far away as 1000 m. Each male has a repertoire of three to seven songs, which differ in dominant frequency and period of modulation (Fig. 2). Having sung a particular song, a male does not usually return to i t until he has sung one or more other songs* One individual, for example, which had a repertoire of five songs (A,B,C,D,E), sang the following sequence: BACDBCDCBDACD BCEDBCADCBEDBCAC. Some songs are slightly more complex than those shown in Figure 2, but even these (Fig. 3) can be reduced to the two dimensions, dominant frequency and period of modulation* With only a single exception, the 31 males I studied had songs that could be described along these two dimensions. To appreciate the suitability of varied thrush songs for such precise description, one has only to compare them to the much more complex songs of most birds (see, for example, the sonagrams in Birds of North America by Bobbins et al. 1966),* The songs of the aberrant male (Fig. : 4) resemble songs normally sung under quite different circumstances from those described above. I have observed other males singing songs like the ones shown in Figure 4 while hopping about on the ground feeding, while "warming up" to typical loud singing,
Figure 2* Sonagrams of typical varied thrush song, showing differences in frequency (Hz) and period of modulation. Note that the f i r s t song is not modulated at a l l. As in Figure 1, only part of each song is shown. 7
Figure 3. Sonagrams of varied thrush songs more complex than those shown in Figures 1 and 2, 9
Figure 4. Sonagrams of aberrant songs sung by a male varied thrush. 11
(ziw) Aoiianbay 12
13 during disputes with other males, and in response to playback. In a l l these situations, however, the songs are uttered quietly and almost without pause. This thesis is concerned only with the loud singing that males do from.high perches. It describes the behavior of the 30 males that sang normal loud songs. The ideas in this thesis grew out of an observation I made in.. the r spring of 1974 while hiking through, a subalpine forest where several varied thrushes were singing. The observation was simply, that..the songs in the repertoires of individual birds were quite distinct, the birds seeming to obey a rule that says: sing only songs that are not very similar to each other. This rule, which is the basis of the theory developed later in this thesis, also f i t s neatly into a theoretical framework that explains the adaptive significance of song and song repertoires. ADAPTIVE SIGNIFICANCE OF SONG AND SONG EEPEBTOIRES Like many other songbirds, varied thrushes defend territories during the breeding season and sing within the boundaries of their territories. Territorial song is commonly thought to have two general functions: defense.of the territory against other males, and attraction and stimulation of females. More than 100 years ago, Bernard Altum (1868) suggested that song is a long-range warning to conspecific males, repelling them from a distance. There have been three recent attempts to test this hypothesis. In support of i t ; Peek
14 (1972) found that surgically muted red-winged blackbirds (Agelaius phoeniceus) suffered more trespassing by other males and were more likely to lose their territories; But, in a similar experiment with the same species* Smith (1976) found no difference in the ability of muted and control males to hold their territories, and i t is not clear whether these contradictory results reflect differences in experimental procedure or differences in the role of song between the two populations of red-wings (Smith 1976). Krebs et al. (1978), taking a different approach in a study of great tits (Parus major), found that after the removal of territorial pairs, new birds colonized empty control areas more quickly than they did areas occupied only by loudspeakers playing back song. Therefore, in great t i t s song does repel other males. There is also evidence, though equally scanty, that song attracts and stimulates females. Although i t is not known whether females locate mates by their song, female chaffinches (Fringilla coelebs) are attracted to sources of recorded song (Marler 1956). Vocalizations of male budgerigars (Melopsittacus undulatus) and ring doves (Streptopelia risoria) stimulate gonadal activity in females (e.g. Brockway 1969; Lott and Brody 1966). This may also be true for canaries (Serinus canarius); females build nests more actively when exposed to song (Hinde and Steel 1976), a pattern that presumably has a hormonal basis* Varied thrushes are also similar to many other song birds in having repertoires of several different song types. In a few species, males sing different songs in different contexts.
15 Chestnut-sided warblers (Dendrojca gensylvaniea), for example, sing certain songs in the centers of their territories when no other males are nearby and other songs near the boundaries during encounters with other males (Lein 1978). In most species, however, males seem to sing a l l their songs in the same contexts (Krebs 1977b).. There is perhaps no better example of this than varied thrushes, which sing one song type after another ACDBCEDBCAD from the same perch. But why do birds sing more than one type of song i f the songs a l l occur in the same contexts and presumably therefore carry the same message? A likely answer is that repertoires enhance one or both of the general effects of song discussed above* Thus, a male that has a repertoire might be better able to maintain his territory against intruders. This seems to be the case for great t i t s. Intruders (birds that have either no territories or inferior ones) are quicker to take over empty territories when resident males have been replaced by recordings of single song types than when they have been replaced by repertoires of several song types (Krebs et al. 1978). The reason for this difference is not known, but one possibility is that potential intruders habituate -that i s, cease to be repelled sooner to playback of single songs (Krebs 1977b). This difference in the rate of habituation to single songs and repertoires does, in fact, occur in another context, that of resident males responding aggressively to playback of song on their territories (Krebs et al. 1978). A male that has a repertoire might also be more attractive
16 and stimulating to females,. In canaries, which sing complex songs of 30 to 40 different syllables, females exposed to normal songs build nests sooner and lay more eggs than do those exposed to songs having an a r t i f i c i a l l y small number of syllables (Kroodsma 1976). These differences might also be due to differences in rate of habituation (Kroodsma 1976). Whether i t is important for male songbirds to prevent females and potential male intruders from habituating to their song is s t i l l an open question. S t i l l, a predictable result in studies of habituation is that after a response has waned, i t can be restored by a different stimulus (Hinde 1970b; Thompson et al. 1973). It would not be surprising therefore i f repertoires did reduce habituation* The amount that habituation is reduced should depend on the similarity of the songs in the repertoire. A pattern often observed in studies of habituation is that the response to one stimulus is smaller after the animal has habituated to another stimulus. This is called "stimulus generalization", a term also applied to the situation when a conditioned response is made to a stimulus that has not been associated with reinforcement* In either case, the amount of generalization normally increases with increasing similarity of the stimuli (Thompson et al. 1973; MacKintosh 1974). Thus, i f varied thrushes benefit by singing in ways that reduce the habituation of other birds, they might be expected to have repertoires of dissimilar songs. This prediction is put to a test in the next chapter.
17 GENERAL METHODS Three general methods were used in this study: recording of songs onto magnetic tape, analysis of recorded songs, by a sound spectrograph, and playing back of songs to birds through a loudspeaker* Songs were recorded with a Stellayox Professional Recorder Type Sp7 and a Sennheiser MKH 815 condenser microphone. The recording speed was 19 cm/s for songs to be analyzed on the sound spectrograph and 9.5 cm/s for long sequences of songs. During the three years of the study (1974-76), I recorded birds only between 0500 and 1100 hours PDT from 20 May to 27 July. Not having banded the birds, I could not identify individuals with certainty. But i f two recordings made in the same area at different times contained identical songs, I assumed that they were of the same individual. The sound spectrograph was a Kay Electric Missilizer 675. The Flat Shape and Wide Band settings were used for a l l sonagrams and the Flat Shape and Narrow Band settings for plots of frequency (Hz) versus intensity. For playback experiments, tape loops 57 cm long were played at 9.5 cm/s on a Oher 4000 Report-L tape recorder* The playback rate of 10 songs/min was within the normal range of singing rates for varied thrushes* The songs were played through a Nagra DH portable amplifier/loudspeaker, placed near the middle of a male's territory. Just before playback I made a rough map of the territory by observing the male's movements for 30 to 60 minutes*
18 CHAPTEE II SIMILARITY OF SONGS IN INDIVIDUAL REPERTOIRES Eeduced to two dimensions, as shown in Figure 1, varied thrush songs can be plotted as points on a graph that has dominant frequency represented on one axis and period of modulation on another. The repertoire of one bird is graphed this way in Figure 5, and a l l 150 songs recorded from the 30 males are shown in Figure 6. The general question asked in this chapter is whether the songs in individual repertoires, for example the one shown in Figure 5, are more dissimilar to each other than would be expected i f they were drawn at random from a l l the songs in the population, of which those in Figure 6 are assumed to be a random sample. In other words, the question is whether individual repertoires are organized to reduce.generalization in birds listening to the songs. In the analysis, I assume that birds generalize along the dimensions of dominant frequency and period of modulation^ and that they generalize according to a rule which I gleaned from published studies of generalization. The most detailed of these describe generalization of conditioned responses (e.g. Mostofsky 1965; MacKintosh 1974}, not generalization of habituation, but I assume that the rule applies in both situations. According to this literature, birds should generalize along each dimension, not according to the absolute difference between two values, but according to their ratio.
Figure 5. Dominant frequencies and periods of modulation of songs sung by the thrush named Chubby,. 19
40 co E c o is D "D O 30 A 20i g 10-0 2.5 3.0 3.5 4.0 4.5 Dominant Frequency (khz)
Figure 6. Dominant frequencies and periods of modulation of a l l 150 songs sung by the 30 male thrushes,. Numbers correspond to the number of songs. Explanation for the logarithmic scaling of the axes is given in the text. 21
22 70.7 58.9-1 49.0-J 1 2 1 40.8 33.9-2 11 12 ST", 23.5i 1 16.3. 1»H o 11.3 9.4-1 6.5-1 5.4-1 1 1 1 1 1 2 1 2 2 1 1 2 1 2 1 2 3 :3 4 2 3 11 1 2 1 2 2 5: T I I I I i I 7 1 1 3 3 2 2 4 1 1 2 5 1 4 4 2 1 2 3 2 2 3 1 1 1 3 2 2810 3146..3522 3943 1 4414 ' 4942 ' 5533 2656 2973 3329 3727 4172 4671 5229 Dominant. Frequency (Hz) 1 1 1 1
23 For example, i f a thrush generalizes a certain amount from a song having a period of modulation of 6 ms to one having a period of 12 ms, i t should generalize an equal amount from the latter song to one having a period of 24 ms. I was aisle to find only one study of generalization along a dimension (the interval between pairs of clicks) that corresponds to period of modulation, and in that study (Wilkinson and Howse.1975) a bullfinch (Pvrrhula purrhula) behaved according to the rule of equal generalization to equal ratios* Many studies (several of which are. summarized in Mostofsky 1965 and MacKintosh 1974)- show that the same rule applies to generalization along the dimension of frequency (Hz) i If the axes of Figure 5 are transformed logarithmically (as in Figure 6), a given change along one axis from any point on the axis represents a constant amount of generalization. There is s t i l l the problem, however, that two songs may differ in both dominant frequency and period of modulation, in which case the overall amount of generalization is some function of the generalization along each axis. Although several "combination" rules have been considered in studies of generalization along two dimensions, no single rule seems to apply generally. About a l l we know at present i s that i f a certain amount of generalization occurs between two stimuli that differ along one dimension, less will occur i f they differ along two (Blough 1972),. Therefore, in this analysis, no combination rule was assumed, and instead a conservative approach was adopted in which each dimension was considered
24 independently. The general procedure was to calculate an index of generalization by measuring the distance (on a graph such as Figure 6) from each song to the nearest song in the same repertoire;. The data for a l l 150 songs were combined into an observed distribution of "similarity to nearest song", which was compared to the distribution expected i f individual birds acquired their songs at random from those in the.population. METHODS To simplify the analysis, each song was reduced from a dominant frequency and a period of modulation to a rank based on each of these dimensions* Although the exact procedure used to establish these ranks differed slightly between the two dimensions, the goal in each case was to put the data in a form where a given difference (in number of ranks) implies a constant amount of generalization. Thus i f a thrush generalizes a certain amount' from a song in rank 2 of dominant freguency to a song in rank 5, i t should generalize an equal amount from the latter song to one in rank 8. Fifteen ranks, based on the logarithms of the original data, were established along each dimension. For dominant frequency, the minimum value of the f i r s t rank was the logarithm of the lowest dominant frequency (2510 Hz) observed for a l l 150 songs, and the maximum value of the 15th rank was the logarithm of the highest frequency (5854 Hz). The procedure was not quite so simple for period of modulation, since a few songs (3% in my
25 sample) are not modulated at a l l. These were placed arbitrarily in the 15th rank, and the other 14 ranks were established as described for dominant frequency* The similarity to nearest song, along each dimension, was calculated for each song in each of the 30 repertoires. The procedure used to calculate the expected distributions of similarity to nearest song was the same for both dimensions: All 150 songs were placed into a frequency distribution having 15 ranks established as above, a theoretical distribution was fitted to this frequency distribution (the object being to smooth the original distribution), each of the 30 repertoires was simulated 30 times by a random selection of songs from the theoretical distribution, and the similarity to nearest song was calculated for each song. The theoretical distributions were calculated using the program FBEQ in the University of B. C: Computing Centre Library; This program calculates goodness of f i t to seven theoretical distributions: normal, Poisson, binomial, negative binomial, gamma, lognormal, and exponential* The program was applied not only to the original frequency distributions but also to two transformations of these, square root and logarithmic. Thus, a total of 21 goodness of f i t tests was done for each frequency distribution. The theoretical distributions used in the final analysis were those that gave the best f i t according to a chi-square criterion. The original distributions and the theoretical distributions they were replaced by are shown in Figures 7 and 8,:
Figure 7. Distribution of varied thrush songs along the dimension of dominant frequency: observed distribution (solid lines) and theoretical distribution of best f i t (dashed lines). Theoretical distribution is the normal distribution fitted to sguare roots of original data* 26
27
Figure 8. Distribution of varied thrush songs along the dimension of period of modulation: observed distribution (solid lines) and theoretical distribution of best f i t (dashed lines). ~ Theoretical distribution i s the negative binomial fitted to square roots of original data.: 28
29 at LO CO C\J CO CO CD in CO CNJ v. M UU I) co C\J oil c\j T r-t ^ T - ^ CNJ C\J C\J CO AouanbaJj < DC
30 RESULTS For dominant frequency, fewer songs than expected are in the same rank (Fig* 9), and according to a Kolmogorov-Smirnov one-sample test (Siegel 1956) the difference between the observed and expected distributions is significant (p<0.01). For period of modulation, almost as many songs as expected are in the same rank, but far fewer than expected differ by just one rank (Fig* 10), and again the difference between the distributions is significant (p<0.01). DISCUSSION The analysis upholds the hypothesis that repertoires are organized to reduce generalization in birds listening to the songs. The next step should be to test the assumptions that birds generalize along the dimensions of dominant frequency and period of modulation, and that they obey the rule of equal generalization to equal ratios. I attempted to test the f i r s t assumption in a field experiment. The general procedure was to play back songs on the territories of individual males, habituating them to one song, then measuring their response to another song. The prediction was that the more similar the two songs, the smaller would be the response to the. second song. This kind of experiment requires that one quantify the responses of birds. Unfortunately, varied thrushes were not cooperative* Although sometimes they responded by singing.typical loud songs and by flying back and forth over the loudspeaker, two behavioral
Figure 9. Observed (shaded) and expected distributions of similarity to nearest song along the dimension of dominant frequency in varied thrush songs; The difference is significant, p<0.01 (Kolmogorov-Smirnov one-sample test: D-Max=0.175; N=150). 31
32 8CH 7CH 60-50- 40-30- 20-104 JAM n lid 0 1 2 3 4 6 8 Distance to Nearest Song (number of ranks)
Figure 10. Observed (shaded) and expected distributions of similarity to nearest song along the dimension of period of modulation in varied thrush songs. The difference is significant, p<0.01 (Kolmogorov-Smirnov one-sample test: D-Max=0.167; N=150). 33
60H 50H cn c o CO o CD.Q =3 40 30M 20 10-0 1 8 9-15 Distance to Nearest Song (number of ranks)
35 pattern's that can be measured, more often they perched near the loudspeaker, singing quietly and continuously, or simply flew away from the vicinity of the speaker and did not return. At the same time I was studying varied thrushes, I also studied ruf ous-'sided towhees (Pipilo erythrgphthalmus), and they proved more suitable for playback experiments. In southwestern British Columbia, towhees have very simple songs. Except for a brief introductory phrase, which does not always occur, the songs are monotonous t r i l l s, composed of a single, repeated syllable (Fig. 11a). Although the detailed configuration of syllables varies between songs, the songs can be reduced to a single dimension, syllable length (Fig* 11b), which ranged from y 6 to 253 ms in the songs I recorded. The results of an analysis similar to the one done on varied thrushes suggests that rufous-sided towhees also have repertoires made up of songs less similar to each other than i f they had been taken at randpm from the songs in the population (Fig. 12). Although the results are not statistically significant (0. 20>p>0. 15), they are close enough to warrant repeating the analysis with a larger sample size. I did a field experiment, similar in design to the one^ attempted with varied thrushes, to determine whether towhees generalize along the dimension of syllable length. Males predictably sang and flew over the loudspeaker and usually habituated to the first song during 25 minutes of playback (Fig; 13). Even so, the experiment ^-an exceedingly timeconsuming one-' was in the end a failure, for reasons worthwhile to relate*
Figure 11. a. Sonagram of a rufous-sided towhee song, showing introductory phrase and t r i l l of single, repeated syllable. Syllable length equals the interval (measured in seconds) from the beginning of one syllable to the beginning of the next;* Only the f i r s t part of the song is shown, but remainder is just a continuation of the t r i l l, b. Sonagrams of rufous-sided towhee songs, showing variations in syllable length. From left to right: songs A, B, C, and D. 36
Figure 12. Observed (shaded) and expected distributions of similarity to nearest song along the dimension of syllable length in rufous-sided towhee songs. The difference is not significant, 0.20>p>0.15 (Kolmogorov-Smirnov onesample test: D-Max=0.153; N=53). 38
39 CO CD c o CO CD _Q E 20-18- 16-14- 12-10 8-6- 4-2- 0 1 8 Distance to Nearest Song (number of ranks)
Figure 13. Habituation of male rufous-sided towhees to playback of songs on their territories. Playback was done for 25 minutes according to the methods described on p. 17. Results shown are averages for 48 individuals; 40
0 1 8 CO CO c o CO 6 4- «e «c * «e 2-1 13 25 Minute 24 D C CO -t ' 14 e CE o "l3 25 Minute
42 The experiment was designed so that the birds habituated to a different song in each of the four treatments (the songs are shown in Figure 11b), then responded for 5 minutes to the same song (song A in Figure 11b) in a l l treatments. Thus, in one treatment, birds habituated to the same song (song A) they later responded to, in the second treatment a different song but with the same syllable length (song B), in the third a song with a different syllable length (song C), and in the fourth a song with an even more different syllable length (song D). Superficially, the results appear to be as predicted (Fig* 14), but according to an analysis of variance there were no significant differences among the treatments (p>0.50). Also, an analysis of covariance showed that the birds did 1 not respond equally or habituate at the same rate to the four songs (Fig. 15). This could explain some of the differences in Figure 14, since i t is known from other studies (see Hinde 1970b) that an animal's responsiveness to one stimulus can affect i t s subsequent responsiveness to another stimulus. Therefore, in this experiment the response to the second song was not necessarily just a function of the amount of generalization between the two songs. This source of ambiguity could be eliminated by a more elaborate experimental design. If the f i r s t song was the same for a l l treatments, and only the second song varied, the response to the second song would be a function of the amount of generalization between the two songs and of the response the second song would have elicited i f the bird had not already habituated to another song. Any
Figure 14- Response of rufous-sided towhees to playback of song A after having habituated to one of four songs. A, B, C, or D (see Fig. 11b). Letters on the graphs refer to the four songs* Results shown are averages for 12 individuals in each treatment. Analyses of variance based on the overall means and variances for the 5 minute period showed no significant differences among the treatments for either measure of response (p>0.50 in both cases)* 43
Minute
Figure 15,. Responsiveness and rate of habituation of rufoussided towhees to songs A, B, C, and D. The regression lines were calculated for the periods during which the overall rates of response were declining (see Fig. 13). In the upper graph, the slopes of the lines are not significantly different (p>0.20) but the Y-intercepts are significantly different (p<0.0001); in the lower graph, the slopes are significantly different (p<0.02); 45
9H Minute
47 differences in the latter could be measured in separate experiments and the results used as a control. These experiments would have to be repeated several times, with different songs each time (to control for other, unmeasured differences among songs), just to establish that birds generalize along a particular dimension, a task that could take several years, since birds are normally responsive to playback only for a limited time during the breeding season (for a detailed study see Petrinovich et al. 1976). It might take much longer to determine the rules governing generalization along that dimension* One's time would surely be better spent working with birds in the laboratory where generalization could be studied in detail using procedures standard in experimental psychology (see e.g. MacKintosh 1974, chap. 10). There is s t i l l another assumption c r i t i c a l to the interpretation of the results in this chapter. This assumption follows from the fact that song has a developmental history in individual birds. Although i t is not known how varied thrushes acquire their songs, i f they are anything like chaffinches and white-crowned sparrows (Zonotrichia leucophry_s), the two species in which the development of song has been studied most intensively (see e*g. Thorpe 1958; Marler 1970), adult birds sing songs that they learned as juveniles. The general question asked at the beginning of this chapter can now be rephrased: Are the songs in individual repertoires more dissimilar to each other than would be expected i f they were acquired at random from a l l the songs that were in the
48 population when the birds actually learned the songs? The analysis assumes that the distribution of songs along the two dimensions was the same when the birds learned their songs as when I recorded my sample. Of course, there is no way of testing this assumption for the present analysis, but i t should not be ignored in future studies* Turning now to questions.of adaptive significance, i f the theory summarized in Chapter I is correct, males that have repertoires of dissimilar songs may be more, effective at repelling potential intruders or at attracting and stimulating females. These ideas could be tested experimentally. The idea that repertoires aid males in repelling potential intruders has been developed most fully by Krebs (1977b) in his so-called "Beau Geste" hypothesis. Given that the suitability of an area for breeding declines as the density of residents increases, i t is to the advantage of settlers to estimate the number of birds already present. According to the hypothesis they do this by listening to the song of territorial males* By the same argument, i t is to the advantage of residents to appear more common than they actually are, so that settlers will move on. Hence the evolution"of repertoires: each male, by singing several different song types, creates the impression that he i s several different individuals. The value of this hypothesis is that i t predicts several behavioral patterns that might be associated with repertoires* One is that "within repertoire variability in song types should be at least as great as between repertoire variability" (Krebs 1977b). These two sources of variability will be equal when
individuals sing a random sample of songs from the population. If resident birds and intruders are playing a deception and detection game, this would seem a good strategy for residents; there would be no fixed pattern of songs within repertoires and therefore no code for intruders to break. But Krebs (1977b) has gone on to suggest that habituation is the process by which intruders "count" the number of residents, and i f this is so a greater variability of song types within repertoires than between repertoires might be a better strategy,. This is just another way of saying that birds might be expected to have repertoires organized so as to reduce generalization by intruders; Therefore, the predicted variability of songs within repertoires compared to their variability between repertoires depends on how the hypothesis is stated, in particular on whether habituation is considered the process by which intruders estimate the density of singing males. Krebs (1977a) has found no clear pattern in his work with great t i t s. If the songs, which are relatively simple, consisting of a single phrase repeated several times, are described by the length of the phrase, the variability of songs within repertoires is greater than that between repertoires. But i f the songs are classified by the form of the notes within the phrases, the variability within repertoires is no different from that between repertoires. Thus, no conclusion can be drawn about whether the repertoires of great tits are organized so as to reduce generalization. If repertoires have evolved to deceive intruders, males should behave in ways that uphold the deception. Great t i t s
50 seem to, for example, by singing their songs in bouts of a single type (AAA.*.EBB--.) and by changing perches at the same time they change songs (Krebs et al. 1978)* Varied thrushes, on the other hand, sing one song type after another from prominent perches, a practice that would be unlikely to fool even the least perceptive intruder. It would be interesting to find out whether repertoires aid male varied thrushes in repelling intruders; i f so, the explanation would surely not be a Beau Geste deception* One final thought about our efforts to understand the adaptive significance of repertoires and the behavioral patterns associated with them. We should take Bertram's (1976) advice, proffered especially, for the study of behavioral patterns used in social interactions, and attempt to identify and measure a l l the selective pressures operating on these traits, not just the positive ones. One example will illustrate the point. In great t i t s, males are known to benefit because their repertoires aire more effective than single songs at repelling potential intruders (Krebs et al. 1978). If Krebs (1977b) is correct that repertoires reduce the rate at which potential intruders habituate to song, then the same should be true for territorial neighbors. If so, the aggressive interactions by which adjacent males establish their territories could be prolonged beyond the time'that would otherwise be necessary for territorial establishment. This would have obvious costs in time, energy, and risk of injury (and might also have some subtle benefits). It would be fatal to the Beau Geste hypothesis i f these costs outweighed the
51 benefits intruders* resulting from the more effective repulsion of
52 CHAPTER III CONTROL OF SINGING Samuel F. Rathbun, who introduced Arthur Cleveland Bent, the great monographer of North American birds, to varied thrushes (Bent 1949), studied their singing behavior early in this century. Using only his ears and a pencil and paper, he recorded sequences such as the following, in which each song is described by its pitch: "High l o w medium low -very low, this followed by a harsh note; high very high low - medium high low, then a pause as i f the bird was reflecting on its performance; high -medium medium -low..." Rathbun shared his notes with Bent* who observed that the. songs within these sequences were not delivered in a fixed order, and also that some songs were used more commonly than others. For example, in one sequence of 56 songs, the most common song was sung 21 times, the rarest only once (Bent 1949). This description, accurate insofar as i t goes, is the most detailed account I have been able to find of the singing behavior of varied thrushes* In this chapter, I elaborate this description, thanks in large part to technology and analytic techniques developed long after Rathbun made his observations. I recorded sequences of up to 110 sonqs from each of the 30 males and determined the dominant frequency and period of modulation of each song type, the order in which the songs were sung, and the lengths of the pauses between songs. These were the basic data used for