|Home | About | Journals | Submit | Contact Us | Français|
Using the responses of territory owners to playback to infer the territorial function of acoustic signals is common practice. However, difficulties with interpreting the results of such experiments have obscured our understanding of territorial signalling. For instance, a stronger response to playback is often interpreted as more aggressive, but there is no consensus as to whether this should be in response to the least or most threatening simulated intruder. Rather than following a gradual increase or decrease, the relationship between signal intensity and response strength may instead describe a peaked curve. We manipulated banded wren (Thryophilus pleurostictus) songs to simulate low-, median-, and high-performance singers and used these songs as stimuli in playback experiments. Banded wrens were less likely to approach the high-performance stimulus compared with the low- and median-performance stimuli. However, the birds that did approach the high-performance stimulus sang more than those that approached the low-performance stimulus. In addition, birds were more likely to match the songs when exposed to the median- and high-performance stimuli compared with the low-performance stimuli, and song matching predicted approach behavior. These results are in accordance with theoretical models of aggressive encounters in which low-performance opponents are challenged without further assessment. Median- and high-performance opponents, however, may require further assessment, and the latter may be perceived as too intimidating for approach.
Secondary sexually selected traits, such as push-up displays in lizards (Martins 1993), drumming in spiders (Elias et al. 2005), or song in songbirds, evolve in response to selection pressures exerted by potential mates and/or competitors (Darwin 1871; Andersson 1994). In intersexual selection, mating preference is usually positively related to signaller quality because the choosing sex benefits from mating with superior individuals. Competitors, however, must assess the fighting ability of their opponents and decide whether they should approach and possibly risk injury by engaging in a fight or retreat. A competitor of similar fighting abilities is expected to evoke the most aggressive behavior (Thornhill 1984; Briffa and Elwood 2001). An opponent with lower fighting prowess might elicit less aggressive behavior because it represents a relatively low threat, whereas an opponent with higher fighting prowess may elicit less aggressive behavior because it intimidates the receiver. The relationship between signal performance level and response intensity in intrasexual selection is therefore predicted to follow a peaked curve (Collins 2004).
Although intersexual selection is thought to drive song elaboration in songbirds (Vehrencamp 2000; Collins 2004; Searcy and Nowicki 2005), song is also used as an intrasexual threat signal to defend a territory against competitors in many species. The signal value of song in competitive contexts is often tested with on-territory playback experiments that expose a territory owner to 2 or more categories of song stimuli. The difference in response to different types of stimuli is used to interpret the signal value of each type. Some studies concluded that the strongest response is elicited by the stimulus simulating the more aggressive intruder (Slabbekoorn and ten Cate 1997; Leitão and Riebel 2003; Illes et al. 2006), whereas others concluded that the most threatening playback stimulus elicits the least intense response (Järvi et al. 1980; Cramer and Price 2007; Hardouin et al. 2007). These contradictory interpretations, combined with the notion that these types of experiments test the response of territory owners to intruders rather than the response of potential intruders to territory owners, led some to conclude that the function of song in territorial defense cannot be tested with on-territory playback (Searcy and Nowicki 2000).
However, the apparent discrepancies in the interpretations of these studies could also be explained by the peaked curve relating response strength to threat intensity (Collins 2004). If the experimental stimuli range from low to median intensity (stimulus set A, Figure 1), representing the upward slope of the peaked curve, one expects a stronger response to the higher intensity signal. If the stimuli range from average to high intensity (stimulus set B, Figure 1), representing the downward slope, then the strongest response should be elicited by the lower intensity signal. However, the peaked curve describing the relationship between response intensity and signal intensity is likely to remain a theoretical one. The peaked curve hypothesis, as well as many experimental studies, assumes that response intensity should be interpreted as a one-dimensional parameter. Empirically, however, it is difficult to show the difference between the reduced response to low- and high-performance signals using one behavioral parameter only, such as for instance approach. Therefore, we believe that it is necessary to measure different types of behavioral responses independently, rather than combined, in order to assess whether subjects differentiate between low-, average-, and high-intensity signals.
We tested whether there is a difference in response to gradually increasing threat signal intensity in banded wrens (Thryophilus pleurostictus, formerly Thryothorus pleurostictus, Mann et al. 2006). Like many avian species, banded wrens produce a rapid repetition of short sound pulses in their song, referred to as trills. Trill rate (number of trill notes per time unit) is constrained by a bounded relationship with the frequency bandwidth of the trill notes. This pattern has been described as a performance constraint in several species (Podos 1997), including the banded wren (Illes et al. 2006). Trills that are close to the performance constraint may be difficult to produce (Hartley and Suthers 1990), and they probably reflect more competitive singers. Correspondingly, this was the preferred interpretation in several studies testing for a difference in response by males and females to songs with different performance levels in several species (Drgnoiu et al. 2002; Ballentine et al. 2004; Illes et al. 2006; Schmidt et al. 2006, 2008; Cramer and Price 2007).
We exposed territorial banded wrens to 3 renditions of the same song that differed in frequency bandwidth, and therefore performance level, simulating low-, median-, and high-performance intruders. If performance level has a function in agonistic interactions, then we expect the subjects to discern all 3 stimuli. The response to signals that differ in intensity is affected by the resource holding potential (RHP, relative quality, motivation, and status) of the subject (Mennill et al. 2002; Schmidt et al. 2006, 2008; Osiejuk et al. 2007). For instance, male nightingales (Luscinia megarhynchos) successful in obtaining a mate and thus probably with high RHP responded stronger to songs containing broadband trills than unsuccessful males with most likely a lower RHP (Schmidt et al. 2008). In order to minimize the effect of variation in relative RHP between the subjects on the response to playback, we used stimuli that range in performance from just below the population minimum to just above the population maximum. The frequency range of the manipulated stimuli remained within the general frequency range produced by banded wrens, which prevents the stimuli from being treated as heterospecific. However, the performance values of the high- and low-quality stimuli ascertained that the simulated intruders signalled a higher or lower quality, respectively, than all the subjects.
The banded wren is a territorial, insectivorous, nonmigratory neotropical oscine. Banded wrens produce a repertoire of approximately 20 song types, with each song consisting of a relatively low-amplitude complex introduction and a louder terminal trill (see, e.g., Figure 2). Each individual is able to produce several distinct trill types. Experiments were conducted between 8 July and 15 August 2006 in Santa Rosa National Park in Costa Rica (10°51′N, 85°38′W). For a detailed description of the habitat type and population, see Molles and Vehrencamp (1999). Subjects were individually color-banded male territory owners that were in various stages of the breeding cycle.
We assessed territory boundaries by monitoring the subjects’ movements and song posts and recording the coordinates (Garmin 12XL GPS Personal Navigator) on a map of the study area. In the center of each territory, we marked a lure area of 10 by 10 m using colored flagging tape. A speaker (RadioShack mini amplifier speaker) was placed in the center of the lure area, connected to an iPod-mini with a 25-m long cable. At 25 m from the lure speaker, a second speaker (Anchor Mini-Vox model pb-25) was placed well within the subject's territory and connected to an iPod with a 25-m cable. The second speaker formed the center of an experimental circle with a radius of 15 m, also marked with flagging tape. Speaker volume was adjusted to 90 dB, measured at 1 m from the source with a RadioShack Realistic Sound Pressure Level Meter (33-2050), which approximates natural banded wren singing levels.
Before a trial, the subject was lured into the lure area in order to standardize the distance of the subject to the experimental speaker at the onset of playback. To attract the subjects into the lure area, we played male calls and female banded wren song. If the subject did not come into the lure area within 900 s, it was considered not responsive and the trial was aborted. A second and last attempt with the same individual was made at least 2 days later. We continued recruiting new subjects until 30 individuals responded by coming into the lure area.
As soon as the subject entered the lure area, playback of the stimulus was started at the experimental speaker. The stimulus duration was 130 s, followed by a 230-s postplayback observation period. Each subject received 3 trials with different stimulus treatments on different days and in random order. Subsequent trials with different subjects on the same day were conducted out of acoustic range of each other to avoid interference from previous trials. All trials were conducted between 0600 and 1000 h, after the end of the dawn chorus and before the midday decline in activity.
Two experimenters observed each trial. One recorded the time of specific behaviors of the subject on a Palm handheld computer using custom-made software (Sturnus: John Burt, University of Washington, Seattle). The second experimenter made acoustic recordings to which he added spoken comments of the trial, using a digital recorder (MicroTrack 24/96, M-audio) and a directional microphone (Sennheiser ME67). The observers were blind with respect to treatment, and the acoustic differences in treatment were not detectable to the human ear.
During a trial, we collected the following data for each subject: 1) the time spent within the 15-m radius of the experimental speaker, 2) the latency from the start of playback to the first crossing by the subject of the experimental 15-m circle boundary, 3) the number of songs sung during the trial, 4) the number of calls produced during the trial, and 5) the number of songs that matched the stimulus song. Song type matching is an aggressive signal in banded wrens (Molles and Vehrencamp 2001a; Vehrencamp et al. 2007).
We created sets of 3 stimuli that differed in the frequency bandwidth, and therefore the performance level, of their trills. First, we assessed the natural range in frequency bandwidth for 12 song types with trill notes that consisted of a steep-up- or down-sweep in frequency. We analyzed 10 renditions for each song type from 10 individuals using spectrograms in Avisoft (Fast Fourier Transformation = 512 Hz, Hamming window, frequency resolution = 94 Hz, time resolution = 2.67 ms, Avisoft SASLab Pro 4.39, R. Specht, Berlin, Germany, http://avisoft.com).
From a separate set of recordings, we then chose 30 songs, each from a different individual representing the 12 different song types (each song type was represented on average 2.7 times in the final stimulus set). These recordings were made with a Sennheiser ME67 microphone and a Marantz PMD 690 or MicroTrack 24/96 M-audio digital recorder at a sample rate of 48 kHz during the breeding season (April to August) of 2005 and 2006. These songs were recorded from individuals that were not part of the group of subject individuals or their neighbors in order to minimize the chance that subjects were familiar with the stimuli.
We used the “pitch bender” function in Adobe Audition (version 1.0) software to manipulate the frequency bandwidth of the trill notes. The pitch bender function allows resampling at gradually varying rates across a selected time window of a sound recording. The sample rate can be adjusted such that it increases or decreases linearly between the beginning and the end of a selection. After the manipulation, the selection is reinserted into the original recording, where the different sample rate of the selection compared with the original recording results in a stretched or compressed frequency bandwidth for an increase or decrease in sample rate, respectively.
Because we only wanted to manipulate the high frequency part of the trill notes, we selected the first 30 ms for notes with decreasing frequency and the last 30 ms of the notes that increased in frequency (Figure 2). The selection comprised approximately 25% of the total duration of the note. The result was that the bottom end of the frequency spectrum of the selection had an unmanipulated sampling rate equal to the remaining part of the note outside of the selection, whereas the top end had an increased or decreased sampling rate. Each trill note was first manipulated to the median value of the population for trill notes of that song type (see Table 1) resulting in a median-performance stimulus (hereafter referred to as Median). The upper frequency of the low-performance stimulus (hereafter referred to as Low) was compressed resulting in a 1-kHz reduction in the maximum frequency, and therefore in frequency bandwidth, whereas the upper frequency of the high-performance stimulus (hereafter referred to as High) was stretched resulting in a 1-kHz increase in the maximum frequency and in bandwidth with respect to the median value. The differences in frequency range between the stimuli types are likely to be perceived by the birds (Dooling et al. 2000). The larger (75%) and lower frequency part of each manipulated trill note remained unaltered (see Figure 3).
The increase or decrease in sample rate of the selected time window resulted in a decrease of 3 ms or increase of 5 ms in note duration, respectively. It is unlikely that the birds can perceive this difference in duration between individual trill notes in stimuli presented on different days (Maier and Klump 1990; Dooling et al. 2000). We compensated for the increase or decrease in trill note duration by inserting or deleting the same amount of silence in-between trill notes. This procedure was repeated for all trill notes in the song, resulting in a stimulus with similar trill note rate and identical total song duration. Each manipulated song was then copied with an interval of 6.5 s, creating a sound file that consisted of 20 identical songs. The end results were 30 sets consisting of 3 stimuli of 130 s each, with a single perceptual difference between the stimuli within a set, namely, the frequency bandwidth of the trill notes (see, e.g., Figure 2). The stimuli were saved on the iPod as uncompressed files in WAV format.
Response variables such as time spent within the experimental circle and number of songs produced were analyzed using a repeated measures model. For normally distributed data, we applied a mixed model with treatment as a repeated factor with unstructured covariance type and order of presentation as a fixed factor. Differences between pairs of stimuli were analyzed with least significant difference post hoc comparison of the estimated marginal means. Nonnormally distributed data were transformed (Sokal and Rohlf 1998) to achieve normal distribution. Call rate could not be normalized; therefore, we analyzed the data with the Wilcoxon signed-rank test using a sequential Bonferonni correction for multiple tests (Rice 1989). When the dependent variable showed a Poisson distribution, we tested for a difference in response using a repeated GLIMMIX procedure with a correction for Poisson distribution using SAS.
Approach latency was measured as the time between the start of the playback and the moment the subject entered the experimental circle. However, a number of subjects did not enter the experimental circle within the duration of a trial. In order to avoid having to discard these subjects as missing values, we calculated Kaplan–Meier survival estimates for approach latency for each treatment (Botero and Vehrencamp 2007). Kaplan–Meier survival estimate is one of several statistical techniques for analyzing the interval between 2 events when information is incomplete. This technique was developed for the medical sciences when the event of interest may not occur for all subjects during the period in which they are observed (Norušis 2004). In our case, the entering of the subjects into the experimental circle is the event of interest, but in the medical sciences, the event of interest may be death or the lack thereof, hence the name of the technique. These survival estimates were then compared with a log-rank pairwise comparison while controlling for the order of presentation. Throughout we report mean ± standard error of the mean. All analyses were conducted with SPSS version 13.0, except when stated otherwise.
We successfully conducted the first trial with 30 subjects although 7 subjects were not responsive on the first trial. Six more subjects appeared not responsive on subsequent trials, and their results were discarded. In 4 out of these 6 cases, the lack of response coincided with the construction of a new nest in a different corner of the territory in-between trials. The interval between successive trials was on average 4.3 ± 0.32 days.
There was neither an order by treatment interaction (F4,63 = 1.19, P = 0.33) nor an effect of order of presentation of the stimuli (F2,63 = 1.93, P = 0.15) on the time spent near the speaker. There was an effect of treatment on the time spent within the experimental circle (F2,63 = 5.60, P < 0.01). Post hoc analysis showed that the subjects spent less time within the experimental circle in response to High than in response to Low (t63 = 3.01, P < 0.01) and Median (t63 = 2.90, P < 0.01). There was no difference in response to Median and Low (t63 = 0.12, P < 0.91; Figure 4, top panel).
Males approached the speaker more rapidly in response to both Median (χ2 = 3.98, P = 0.048) and Low (χ2 = 5.48, P = 0.02) than to High. There was no difference in the survival estimates of the response to Low and Median (χ2 < 0.01). Thus, the approach latency to High was greater than to Low and Median (Figure 5).
Song rate varied with treatment (F2,24 = 4.3, P = 0.025), with post hoc analysis showing that Median elicited a significantly higher song rate than both Low (P = 0.014) and High (P = 0.04). There was no order of presentation by treatment interaction (F4,46 = 1.65, P = 0.18) nor was there a main effect of order of stimulus presentation (F2,43 < 1).
The subset of males that approached to within 15 m of the speaker sang at lower rates in response to Low (Figure 4, middle panel). We found an effect of treatment on song rate (F2,18 = 10.74, P < 0.01), and post hoc comparison showed that Low elicited a decreased song rate compared with High (P < 0.01) and Median (P < 0.01) for those individuals that approached the speaker to within 15 m. There was no difference in song rate in response to High and Median (P = 0.29).
Males produced more calls in response to High than Low (Figure 4, bottom panel; Z = −2.17, P = 0.03). There was no difference in response to Low and Median (Z = −0.54, P = 0.59) or to High and Median (Z = −0.72, P = 0.57).
Although there was no evidence of whole song type matching, there was a numerical difference in trill type matching for the different treatments. Eleven subjects matched at least one trill with Median, whereas only 5 and 3 subjects, respectively, matched at least one trill with High and Low. Subjects that matched Median stimuli approached faster than those that did not match (Kaplan–Meier survival estimate: χ2 = 7.24, P < 0.007); this is consistent with the interpretation that matching is an aggressive signal (Molles and Vehrencamp 2001b; Gil and Gahr 2002). In addition, all subjects that matched Median, as well as all 5 that matched High, entered the experimental circle.
Banded wrens responded differently to song stimuli that were manipulated to vary in frequency bandwidth and therefore performance level. Physical approach responses were stronger to Median- and Low-performance treatments, whereas vocal responses were higher in response to Median- and High-performance treatments. Therefore, subjects responded differentially to the 3 stimulus types that were presumed to differ in their threat level. These results are in agreement with an earlier study that showed that banded wrens responded differently to playback songs that differed in performance level (Illes et al. 2006).
Changing frequency parameters in bird song may lead to reduced responses when broadcasted to subjects on their territories when the manipulated songs are not recognized as conspecific (Brémond 1968; Dabelsteen and Pedersen 1985; Nelson 1988; Slabbekoorn and Ten Cate 1998). Although the current experiment cannot conclusively exclude this as an explanation for the obtained results, we do not believe this to be correct for several reasons. First, it is imperative to note that subjects’ vocal response to High was equal compared with Median. The vocal response to Low was reduced compared with the other 2 stimulus categories. The approach, however, was reduced in response to High compared with Low and Median. This response pattern suggests that the birds perceived all stimuli as coming from a conspecific intruder because both Low and High were treated like Median in at least one response parameter, but each elicited a reduced response in another response parameter. Second, the frequency range of the manipulated notes fell well within the frequency range used by the species (see Figure 2), suggesting that it did not extend outside of the frequency sensitivity range of the species. Third, the larger part, including the fundamental frequency, of the manipulated syllables remained unaltered in our study (Figure 3) in contrast to the studies mentioned above in which entire syllables were shifted in frequency. We believe that shifting an entire syllable in frequency, as opposed to only increasing or decreasing the higher part of a syllable, is likely to have a larger effect on the perception of the song and is therefore more likely to be perceived as irrelevant by the birds. Finally, if animals only respond to the average or most common value of a signal component, then it would not allow signals to evolve in a directional way as generally assumed by theories of signal evolution in the context of sexual selection. Stabilizing selection on the most common or median value of parameters cannot lead to the extreme values observed in traits subject to sexual selection. Also, it is well known that animals respond to signals that extend beyond the species-specific range, a phenomenon referred to as supernormal stimuli. Therefore, we believe that the difference in response observed in the present study should be interpreted as the subjects perceiving the 3 stimulus categories as differing in level of threat intensity.
Song type matching may signal an individual's overall willingness to approach an intruder (Krebs et al. 1980; Vehrencamp 2001). Indeed, all subjects that matched the trill type of the playback stimulus in response to Median and High approached within 15 m of the speaker, consistent with the notion that matching is a signal of motivation that is related to the probability of attacking. However, not all subjects that entered the experimental circle matched the playback stimulus. Especially in response to Low, only 15% of the subjects that entered the experimental circle also matched the playback stimulus, compared with 52% for Median and 45% for High. Combined with the fact that the birds that entered the experimental circle sang less in response to Low, these results suggest that the subjects did not need to assess Low further and were willing to approach without further ado. In contrast, High, and especially Median, required further assessment, and the subjects engaged in a singing contest reminiscent of a sequential assessment game in which individuals acquire additional information about their opponents’ relative fighting abilities (Enquist and Leimar 1983; Leimar and Enquist 1984; Enquist 1985).
With the High-performance stimuli, we attempted to simulate intruders that were superior to all individuals in the population and therefore should have been intimidating to all subjects. Nevertheless, High did not prevent all subjects from approaching, and those that did approach sang as much in response to High as to Median. Response intensity depends not only on the intruder's competitiveness relative to the subject (Parker 1974) but also on the subject's motivation, the subject's valuation of the resource at stake (Enquist and Leimar 1983; Enquist 1985), and the individual aggressiveness of the subject (Verbeek et al. 1996; Nowicki et al. 2002). Subjects may have varied in one or more of these characteristics, making them willing to approach even a superior intruder simulated by high-performance playback.
The results from this playback experiment strongly suggest that structural aspects of bird song, in addition to semantic information such as repertoire size (Catchpole and Slater 1995) or temporal song delivery strategy such as overlapping (Vehrencamp et al. 2007), convey information about the competitiveness of the singer (Byers 2007). Consistent with earlier studies (Drãgãnoiu et al. 2002; Ballentine et al. 2004; Illes et al. 2006; Cramer and Price 2007), this study provides additional data to support the hypothesis that trill performance level, a structural component of song (Podos 1997), affects response behavior by receivers. Some previous studies used natural songs that differed in the parameter of interest (Cramer and Price 2007; Schmidt et al. 2008). However, in these studies, the parameter of interest may be correlated with another parameter that was salient for the birds’ response. Other studies manipulated trill rate and consequently created a potential confounding parameter by also changing song duration (Illes et al. 2006). In the current study, the stimuli differed in frequency bandwidth of the trill notes. Although each of these approaches to creating stimuli introduces different, potentially confounding factors, taken together, they suggest that trill performance is the important signal rather than just trill rate or just frequency bandwidth; effects of vocal performance were found regardless of which other factors are held constant (Cardoso et al. 2007). This study therefore completes the body of work showing unambiguously that it is the bounded relationship between trill rate and frequency bandwidth that provides receivers with tools to assess the competitiveness of the singer.
Consistent with the hypothesis that bird song reflects the quality of the singer and that it may have a repellent effect on competitors, we found that territorial banded wrens approached high-performance songs less. In addition, this study shows that song and approach behavior show different response patterns depending on the threat level of the simulated intruder. These results suggest that different parameters should be combined into one multivariate response measure only with caution (McGregor 1992). Finally, this study shows the need for caution in interpreting the results of experiments in which subjects are exposed to 2 signals that differ in signal intensity. It is important to assess how the information about the relative competitiveness conveyed in the stimuli relates to that of the subjects (Schmidt et al. 2008).
National Institute of Mental Health (R01-MH60461).
We thank the Area Conservación Guanacaste for permission to work in Santa Rosa National Park and their staff for support in Costa Rica. Roger Blanco provided logistic support in the field. Thorsten Balsby advised on the statistics. Charles Eldermire, Michelle Hall, Karolina Kluk-de Kort, and Sandra Valderrama helped with fieldwork. Michelle Hall and 3 reviewers provided comments that helped us improve the manuscript.