Mimus mockingbirds are renowned for their large vocal repertoires and highly versatile singing style (Brewer and MacKay 2001). Their songs are made of a finite number of short (ca. 0.3–1s), acoustically distinct sounds known in the Northern Mockingbird (M. polyglottos) as song types (Derrickson 1987) or syllable types (Wildenthal 1965). We refer to these units as syllables because this term is consistent with the terminology used in most other studies on songbirds (see Read and Weary 1992; Catchpole and Slater 1995).

Song versatility, continuity of singing, and repetitiveness of sequential syllables have been shown to vary contextually in the Northern Mockingbird (Burnett 1978; Derrickson 1988; Derrickson and Breitwisch 1992). In the spring, males typically sing each syllable type several times in a bout before switching to a different one, show long recurrence intervals before singing the same syllable type, and make brief or no substantial pauses between consecutive bouts (i.e., are continuous singers). Fall song, on the other hand, is significantly less repetitive and contains a larger proportion of temporally discrete multiple-syllable phrases lacking any repetition (Burnett 1978). Aggressive response is stronger to playback of spring song than of fall song during both seasons (Logan and Fulk 1984). During the breeding season, syllable versatility is highest and bout length is shortest for unmated and courting males, compared to those singing during the incubation and nestling stages (Derrickson 1988). In the context of territorial patrolling and countersinging with a rival male, singing is even more repetitive and less versatile, brought about by an increase in bout length, a decrease in recurrence intervals, and a reduction in presentation of rare syllable types (Derrickson 1988). Although these studies find a consistent pattern of greater syllable repetition and lower versatility associated with more aggressive contexts, a systematic investigation of the independent effects of repetitiveness (e.g. bout length) and vocal versatility on behavioral responses has not been undertaken. Moreover, such a study would be difficult to achieve in the Northern Mockingbird because the two singing patterns are confounded.

The Tropical Mockingbird is closely related to the Northern Mockingbird and strongly resembles it in both appearance and behavior (the two species can hybridize in sympatry, Howell and Webb 1995). Although males in both species sing syllables with similar general structure, consistency of delivery, and duration (Fig. 1), Tropical Mockingbirds throughout most of their range are not continuous singers but rather sing temporally discrete “songs” like those of most other songbirds. The songs of the Tropical Mockingbird are typically composed of less than ten syllables and sometimes include three or more different syllable types (i.e., a singing style that resembles the structure of the Northern Mockingbird’s fall song). Although males at our study site are primarily immediate variety singers (Fig. 1a), individuals conspicuously vary their vocal versatility (Fig. 1b) and may alter the repetitiveness of their syllable sequences during countersinging.

Fig. 1 Variation in the song versatility of the Tropical Mockingbird. The typical song sequences in our population have high within-song versatility (song complexity) and high between-song versatility (switching rate) (a). Occasionally, males become more repetitive (more ...)

We conducted a playback experiment to determine the possible signal value of different syllable presentation patterns during simulated male intrusions in the Tropical Mockingbird. Based on the patterns observed in the Northern Mockingbird (Derrickson 1988) and the fact that the typical, non-aggressive singing style of Tropical Mockingbirds at our study site is highly versatile (Fig. 1a), we test the hypothesis that more repetitive singing represents a stronger threat and generates a stronger aggressive approach. Because of their discontinuous singing style, Tropical Mockingbirds can become more repetitive by singing songs composed of a single syllable type, by repeating the same song several times, or both. We make no a priori hypothesis about the specific mechanism through which repetitiveness is achieved but rather test for the effects of within-song versatility (i.e., song complexity) and between-song versatility (i.e., switching rate) on male response through an experiment with a factorial design. In our playback sequences we controlled for other potential acoustic signals in male-male interactions such as repertoire size, duty cycle (i.e., the fraction of the total playback time in which the recorded bird actually sings), song length, song quality, song rate, and song amplitude.

Fig. 2 Examples of complex songs obtained from recordings of male Tropical Mockingbirds in Bonaire (a, c) and the repetitive songs we derived from them (b, d). Note the similarity between the derived songs and the natural examples of repetitive songs shown in (more ...)

We generated each playback set by first creating high and low between-song switching rate treatments with complex songs extracted from high quality recordings (i.e., recorded at close range, good signal to noise ratio, no overlapping sounds and little to no detectable reverberation). We then obtained the corresponding two other treatments by replacing each complex song with a digitally derived repetitive song (see below). High switching rate treatments consisted of a one-minute sequence of seven different songs by a single male repeated four times (i.e., 1, 2, 3, 4, 5, 6, 7, 1, 2, etc…). Low between-song switching rate treatments consisted of the same 28 songs and inter-song intervals but each song type was presented four times before introducing the next one (i.e., 1, 1, 1, 1, 2, 2, 2, 2, 3, etc…). To create repetitive songs we took a single syllable from the complex songs and repeated it several times until achieving similar song duration (Fig. 2; mean duration of songs = 2.06s; SD = 0.71s; min = 0.88s; max 3.73s). In this process we deliberately used syllables that would produce repetitive songs that resembled those we had previously heard in the wild (note the duration of our repetitive songs is well within the range of natural repetitive songs, e.g. Fig. 1b). We chose to generate repetitive songs out of complex ones (as opposed to using recorded repetitive songs from the same male) because this procedure allowed us to produce playback sets with equivalent recording quality, song quality, duty cycle, song rate and song duration. To test whether generating repetitive songs out of a single syllable had any influence over the outcome of our experiment, we also generated heterospecific controls by repeating single syllables of sympatric troupials (Icterus icterus, Figs. 2e and 2f). In natural situations troupials sing repetitive sequences of a single syllable type that do not evoke aggressive responses in mockingbirds.

Each trial consisted of five min of pre-playback observation, four min of playback of one of the five possible playback treatments, and six min of post-playback observation. We scored the subjects’ responses with a personal digital assistant (Palm™ m105) with custom event recording software (courtesy of J. M. Burt at University of Washington, Seattle). Data collected in real time included the times when the bird sang or crossed one of the distance markers (specifying whether he was going toward or away from the speaker). From these data we computed the following seven response variables: (1) number of songs, (2) time spent within 15 m from the speaker, (3) closest distance bin of approach (1 = 0 – 15 m, 2 = 15 – 25 m, 3 = 25 –50 m, and 4 = >50 m), (4) distance bin after first approach (1 = 0 – 15 m… 4 = >50 m), (5) whether the subject approached or not within 15 m from the speaker (0 = no, 1 = yes), (6) latency to sing, and (7) latency to approach within 50 m of the speaker. We computed all response variables based on the combined playback and post-playback observations because responses were never restricted to the duration of the playback.

Fig. 4 Kaplan-Meier cumulative plot of the latency to approach during trials. A vertical dotted line marks the end of the playback period. Convention: HSC = High Song Complexity; LSwR = Low Switching Rate.

The remaining response variables were collapsed into three composite factors using factor analysis with varimax rotation. Note that the assumptions regarding the distributions of variables included in a factor analysis are not in force when this technique is only used descriptively (Tabachnick and Fidell 2001). We assessed the effects of within and between-song versatility and the interaction between them on these composite factors with general linear mixed models (GLMM) in SAS (SAS Institute, Cary, North Carolina) using the containment method for computing the denominator degrees of freedom for the tests of fixed effects. To meet the normality assumptions for GLMM, we used cubic values of factor two and log-transformed values of factor three (data transformations followed Osborne 2002). We corrected for multiple sampling of the different subjects by including the identity of the test subject as a random factor in the GLMM and a categorical variable in the survival analyses. Besides applying a different order of presentation of treatments to the different individuals, we statistically corrected for order effects in our survival and GLMM models by including trial order as an ordinal variable in our models. We also corrected for the subjects’ behavior prior to the playback by including the following three variables: (1) number of songs during pre-playback, (2) number of calls during pre-playback and (3) closest bin of approach in pre-playback (1=0–15m; 4=50+m). We removed covariates and the interaction term from our final models (in that order) when their p-values were above 0.40.

Conspecific playback treatments produced fast and evident spatial responses. Males (n=13) typically interrupted their current activities, approached the speaker to varying degrees and visibly inspected the surroundings after we had played only a few songs. Although most responses were quick, subjects were able to sample the between-song switching rate of the treatments before their first response. Males heard at least four playback songs before initiating approach in 72 out of the 85 trials (and heard at least two in 82 trials).

In contrast to the spatial responses, vocal responses to conspecific playback were generally weak. Test subjects sang fewer than four songs during the four minutes of playback in 57 trials (67%). Furthermore, in 51 trials (60%) they sang fewer than ten songs in the ten minutes of playback plus post-playback observation (a very low song rate when compared to courtship standards). Subjects that sang during trials usually did so softly during playback and loudly later on.

Table 1 Principal Components Factor Analysis after varimax rotation.

The proximity response score, factor one (Fig. 3), was significantly higher in trials with high versus low between-song switching rate (GLMM: F_1,66 = 6.43, P = 0.014) and marginally higher in trials with low versus high song complexity (GLMM: F_1,66 = 3.92, P = 0.052; Trial order and number of calls before playback were removed from final model). Higher scores for this factor translate to closer approaches to the speaker (Table 1). Factor two, the temporal response score, was not significantly affected by song complexity or switching rate (GLMM: Song complexity: F_1,68 = 0.47, P = 0.497; Switching rate: F_1,68 = 0.22, P = 0.638; Number of calls before playback, closest approach before playback and the interaction term were removed from final model), and neither was factor three, the vocal response score (GLMM: Song complexity: F_1,67 = 0.43, P = 0.515; Switching rate: F_1,67 = 0.04, P = 0.85; Interaction term: F_1,67 = 1.37, P = 0.245; Number of songs before playback, closest approach before playback were removed from final model).

Fig. 3 Mean proximity response scores during playback trials with low (white bars) and high (gray bars) within-song versatility at low and high between-song switching rates. Error bars reflect one SE. Higher score values correspond to closer approach.

The latency to sing after the initiation of playback was not significantly affected by the song complexity (Survival Analysis: n = 85, 26 right-censored values; Z = 0.01, P = 0.990) or switching rate of the stimulus (Z = 0.18, P = 0.858; Number of calls before playback and interaction term were removed from final model). The analysis of the subset of trials in which the subject’s first song occurred after four or more playback songs revealed similar results (Survival analysis: n = 78, 26 right-censored values; Song complexity: Z = −0.87, P = 0.383; Switching rate: Z = −0,01, P = 0.989; Number of calls before playback and the interaction term were removed from final model).

The response of male Tropical Mockingbirds to conspecific intruders is affected by the vocal versatility of their rivals. Both vocal versatility parameters manipulated in this experiment, song complexity and switching rate, influenced how fast and how close males approached the speaker during trials. As predicted, more repetitive singing at the within-song level was associated with stronger responses. However, this was not true at the between-song level where higher switching rates elicited more aggression. Similar opposing trends in within-song and between-song versatility have been reported for male Northern Cardinals (Cardinalis cardinalis), which reduce short-term versatility (by repeating the notes in a song a few extra times) and increase song switching rate in aggressive situations (Ritchison 1988). It is possible that senders increase the conspicuousness of their signals during agonistic encounters because of the striking contrast of low within-song versatility with high between-song versatility. Nevertheless, the lack of a significant interaction in our statistical models and the similar differences between levels of one parameter at the different levels of the other (Figs. 3 and 4) suggest that within- and between-song versatility serve independent purposes and/or transmit different information.

Four hypotheses could explain the responses of male Tropical Mockingbirds to variation in the vocal versatility of intruding males. First, vocal versatility may signal aggressive intentions in territorial contexts as observed in other oscines (reviewed in Vehrencamp 2000). Under this scenario, extremely repetitive songs may signal aggressiveness (because they are very different from the songs typically used in non-agonistic contexts, Vehrencamp 2000) while high song switching rates may be used to reduce receiver habituation (Falls and D'Agingcourt 1982). Given the qualitative differences of the effects of variation of within- and between-song versatility in this experiment, another possibility is that switching rate indicates a singer’s aggressiveness (switching rate had a stronger effect on proximity score than song complexity) while song complexity indicates his willingness to attack (song complexity had a stronger effect on approach latency than switching rate).

The hypothesis that variation in vocal versatility is used to mediate territorial conflicts generates several testable predictions. First, it predicts that birds singing with low within-song versatility and high switching rate will approach a boundary and engage in territorial expansion or boundary disputes more often that birds singing in a different way. Given that Tropical Mockingbirds occasionally live in social groups in which several males defend a common territory (Morton et al. 2004, E. Laurent unpubl. data), this hypothesis also predicts that males living in polyandrous groups should sing repetitive songs and variable song sequences in response to potential intruders (i.e., neighbors or strangers) but not to fellow group members. Finally, since male mockingbirds sing mostly during courtship (Logan 1983) and since most natural countersinging events in the Tropical Mockingbird occur when at least one of the rivals is courting a female (Botero, personal observation), this hypothesis predicts that territorial conflicts are most prevalent while courting. Such a temporal bias in the frequency of territorial conflicts could occur if the area defended by males varies throughout the breeding cycle (see Biedenweg 1983). Nevertheless, the fact that most of our focal males did not sing back in response to playback (which seems to be also the case in the Northern Mockingbird, see Logan and Fulk 1984) does not support the idea that they use variation in vocal versatility as a mechanism for mediating territorial disputes.

An alternative explanation is that the song of male Tropical Mockingbirds serves primarily a mate attraction function (as seems to be the case in the Northern Mockingbird, Derrickson and Breitwisch 1992) and that intruders singing with low within-song versatility and high-between song versatility represent a strong threat to the paternity of resident males. This hypothesis predicts that certain singing styles are more likely to yield extra-pair copulations than others and that resident males are able to block cuckolding attempts by responding aggressively to nearby singers (see Westneat and Stewart 2003; Kokko and Morrell 2005). In this case, immediate repetition of syllable types could for example show off a singer’s ability to produce each song type consistently (Lambrechts and Dhondt 1988) while high switching rates could show off the size of his repertoire or help him reduce receiver habituation (Falls and D'Agingcourt 1982).

A third hypothesis is that variation in vocal versatility is a conventional signal of aggressive intentions (as in the first hypothesis), but that it is emitted in the context of competition for females and not for territory. Although conventional signals like the ones described here are not likely to influence female choice directly, they have the potential to do so indirectly if females keep track of the outcome of competitive interactions between males (West et al. 1981; Berglund et al. 1996; Mennill et al. 2002; Collins 2004). Female mockingbirds are known to constantly re-assess their mates (Logan 1997), and divorce, mate switching, facultative monogamy and extra-pair copulations also occur in this group (Derrickson 1989; Logan 1991; Logan 1997; Morton et al. 2004). Intense countersinging interactions between neighboring male mockingbirds seem to be associated with female mate assessment (Logan 1997; Botero personal obs.). The hypothesis of competition over the access to reproductive females explains why countersinging occurs primarily when at least one of the males is courting and predicts that males singing with low within-song versatility and high between-song versatility will tend to interact aggressively with their rivals. In contrast to the first hypothesis, it predicts that males living in polyandrous groups will also use this singing style in response to adult males of the same territory if both of them are unrelated to the breeding female.

A fourth alternative is that the song of male Tropical Mockingbirds serves both mate-attraction and territorial functions and that variation at the two levels of vocal versatility is intended for receivers of different sexes. For example, it is possible that high switching rates are employed to attract females (e.g. Catchpole et al. 1984; Buchanan and Catchpole 1997; Langmore 1997; Collins 1999) and that repetitive songs are used to show aggression in male-male interactions (Vehrencamp 2000). Under this scenario, resident males would be most threatened by male-aggressive female-attractive singers, somewhat less threatened by male-non-aggressive female-attractive singers and much less threatened by male-non-aggressive female-non-attractive singers.

At this point we can only conclude that male Tropical Mockingbirds pay attention to the vocal versatility of other males and that they perceive intruders singing repetitive songs or variable song sequences as a stronger threat than those that sing complex songs or repetitive song sequences. We are currently pursuing definitive tests of these alternatives.

We would like to thank S. Despa and F. Vermeylen (Cornell Statistical Consulting Office) for their advice on statistical procedures. Many thanks as well to R. Boy (at Kunuku Warahama), and G. van Hoorn and P. Montanus (at Dienst Ruimtelijke Ordening en Beheer Bonaire) for their logistic support while in the field, to N. Bloch for her assistance in the collection of data for this project and to A. Kock, N. Bouten and their families for their incredible hospitality. This manuscript greatly benefited from the comments and suggestions of three anonymous reviewers. Funding was provided by NIH grant R01MH60461. SYRINX software is available at www.syrinxpc.com.