BEHAVIOUR, 2005, 70, 507–515 doi:10.1016/j.anbehav.2004.11.023

Mixed-species aggregations in : , Zenaida aurita, respond to the alarm calls of carib grackles, Quiscalus lugubris

ANDREA S. GRIFFIN*,RAHULS.SAVANI†, KRISTINA HAUSMANIS* &LOUISLEFEBVRE* *Department of Biology, McGill University, Montreal yDepartment of Mathematics, University of London

(Received 12 May 2004; initial acceptance 27 July 2004; final acceptance 5 November 2004; published online 9 August 2005; MS. number: A9882R)

Aggregating with heterospecifics may be particularly beneficial for a species that is able to exploit the antipredator behaviour of another. Territorial zenaida doves vigorously exclude conspecific intruders from their territory, but forage with, and acquire novel foraging techniques from, carib grackles. Given that doves associate with no other conspecific than their mate and they have no vocal alarm signals of their own, they might benefit from attending to the conspicuous alarm calls of carib grackles. In the present study, we found that zenaida doves suppressed foraging both in response to a model predator and in response to the sound of grackle alarm vocalizations. Although doves’ responses to the predator model also involved moving away from the immediate vicinity, their responses to grackle alarm vocalizations consisted of remaining alert and tail flicking. Together, these results strongly suggest that doves attend to the antipredator behaviour of carib grackles. These findings extend earlier work demonstrating that doves obtain foraging benefits from their association with grackles, to show that they may also obtain predator avoidance benefits. Ó 2005 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

Mixed-species foraging aggregations confer advantages that territories against conspecifics (Lefebvre et al. 1996). In can be greater than those of monospecific groupings. For contrast, they show little or no aggression towards carib example, species with different diets may obtain predator grackles, despite considerable dietary overlap in areas avoidance benefits while avoiding the costs of food com- where food scraps left by humans are an important source petition (Morse 1977; Terborgh 1990). In case of dietary of food (Dolman et al. 1996). As a consequence, grackles, overlap, however, individuals may nevertheless join heter- which forage in small, mobile flocks, often feed in close ospecifics because it increases their probability of finding proximity to doves. The preferential association of doves food and/or because it allows them to acquire information and grackles, rather than of doves and conspecifics, has about novel foods and novel foraging techniques (Webster been found to drive the direction of social learning about & Lefebvre 2001). Predator avoidance benefits may also food. Zenaida doves from territorial populations acquire outweigh the costs of food competition (Morse 1977; a novel foraging technique from a carib grackle demon- Terborgh 1990) if a species with poor antipredator detection strator more readily than from a conspecific dove (Dolman skills is able to exploit an efficient antipredator behaviour of et al. 1996). In addition, grackles often land at an another species (Gaddis 1980; Munn 1986; Rasa 1990). For experimental food source before doves, raising the possi- example, a species that does not produce any alarm call may bility that doves, which are relatively neophobic, may use benefit substantially from associating with one that does. grackles to detect and investigate novel food patches Information transfer about food and predators are not (Webster & Lefebvre 2001; D. J. White, unpublished data). mutually exclusive and there are most likely to be cases in Another potential benefit that doves may obtain from which both operate (Wolters & Zuberbu¨hler 2003). their close association with grackles is heightened preda- The carib grackle and the are two urban- tor avoidance. Carib grackles travel in small flocks and ized avian species that are common and sympatric on give high-amplitude, broadband pulsatile chuck vocal- several islands of the Lesser Antilles. In most areas of izations, which are strongly associated with the presence Barbados, zenaida dove pairs aggressively defend year-round of predators (Jaramillo & Burke 1999). In free-living grackles, high rates of chuck calls are typically evoked by Correspondence and present address: A. Griffin, School of Behavioural mongooses, Herpestes auropunctatus, cats, Felis catus, vervet Sciences, The University of Newcastle, Callaghan, NSW 2308, monkeys, Chlorocebus aethiops, and dogs, Canis familiaris, Australia (email: andrea.griffi[email protected]). and are sometimes accompanied by mobbing, particularly 507 0003–3472/04/$30.00/0 Ó 2005 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved. 508 ANIMAL BEHAVIOUR, 70, 3

around nesting roosts (L. Lefebvre, personal observation; can be reliably resighted, so their responses to a predator A. S. Griffin, personal observation). In addition, experi- stimulus can be compared with those evoked by non- mental playback of chuck calls cause receivers to take predator control stimuli using a repeated measures design, flight and to begin calling (A. S. Griffin, unpublished which removes the effects of interindividual variation on data). These observations, together with published reports responses (Griffin et al. 2001, 2002; Griffin & Evans 2003). of predator–chuck call associations in carib grackles (Jaramillo & Burke 1999), strongly suggest that chuck vocalizations function as antipredator calls. In contrast, Methods zenaida doves do not produce vocal alarm signals and territory owners typically forage alone or with their mate. Subjects Doves may benefit from relying on grackles to look out for The dove population at Bellairs Research Institute and predators and responding to their antipredator calls for Folkstone Park, St-James, Barbados, has been the subject of two reasons. First, an approaching predator may be more several earlier studies (Lefebvre 1996; Webster & Lefebvre likely to be detected by a grackle than a dove because 2001). All the individuals that took part in the present grackles are more abundant, more mobile, and spend study were colour-banded and readily identifiable through more time perched in trees. Second, relying on grackles binoculars. This dove population lives in proximity to might allow doves to increase their food intake by re- humans, and the birds are accustomed to being fed. Each ducing the time they allocate to antipredator vigilance. morning and evening for 10 days prior to the start of the Given the ability of zenaida doves to acquire information experiment, we walked a transect around Bellairs and about food from carib grackles and the behavioural and Folkstone Park and placed food piles at various locations. ecological differences between the two species, zenaida We identified the territory owner(s) by noting which birds doves and carib grackles form a particularly interesting defended the food against intruders. Doves were con- avian system in which to explore cross-species antipred- sidered mates if they defended the food patch against ator call recognition (McLean & Rhodes 1991). conspecifics, but foraged in close proximity of each other There have been no previous studies of the antipredator with little or no aggression. At the end of the 10-day behaviour of zenaida doves. The aims of the present study period, we selected 11 territory owners with the highest were hence two-fold. In experiment 1, we determined resighting rates as subjects for experiment 1. These how zenaida doves respond to predators by quantifying individuals were distributed across 11 different territories. their responses to a model of a prototypical mammalian Zenaida doves are difficult to sex in the field, so we did not predator. In experiment 2, we tested whether zenaida take this factor into consideration when selecting subjects. doves give antipredator responses to the sound of grackle During each test, the focal was provided with antipredator calls. approximately three handfuls of cooked rice.

Visual stimuli EXPERIMENT 1 To measure responses to a prototypical predator stimu- lus, we used a taxidermic mount of an ermine, Mustela As in many islands, the mammalian predator community erminea. Although ermines are unfamiliar predators to in Barbados is entirely established by humans. Dogs are zenaida doves, their morphology is convergent with that thought to have been brought by the Amerindians, the of many mammalian predators, such as mongooses, original colonizers of Barbados, prior to the 1500s. Cats, which are abundant on the island of Barbados. To com- Indian mongooses and vervet monkeys were introduced pare responses evoked by a mammalian predator with more recently by European settlers. Except for dogs, there those evoked by a nonpredator-like control object of is circumstantial evidence that each of these species similar volume, we presented the birds with a cardboard represents some degree of threat to either doves or their box painted black (Maloney & McLean 1995; McLean offspring. Vervet monkeys are common nest predators et al. 1996; Go¨th 2001; Wisenden & Harter 2001). Visual (A. S. Griffin, personal observation), cats have been observed stimuli were attached to a cart (0.45 ! 0.65 ! 0.1 m) that stalking doves (A. S. Griffin, personal observation), and could be wheeled in and out of a hide. Finally, to detect mongooses are able to kill and consume vulnerable adults spontaneous changes in behaviour over time, we con- (J. Morand-Ferron, personal communication). There have ducted a blank control, in which the hide and the test been no systematic attempts, however, to quantify the device were set up on the subject’s territory, but no responses of doves to predators. We considered it neces- stimulus was presented. Each of the three visual treat- sary therefore to address this question before we tested the ments was conducted once on each territory in an order responses of doves to grackle alarm calls. that was balanced across subjects. Quantifying antipredator behaviour during natural in- stances of predation is often impossible. Consequently, we elected to stage predator encounters by presenting a model Test procedure to free-living zenaida doves. This approach has the further We conducted all trials in the early morning between advantage that attributes of the encounter that might 0545 and 0900 hours. These are the times at which influence response levels substantially, such as the pred- zenaida doves are most likely to be found foraging on ator’s speed, gait and direction of approach, can be their territories and when tests were the least likely to be controlled across subjects. Furthermore, territorial doves disturbed by visitors to Folkstone Park and Bellairs. We ran GRIFFIN ET AL.: ANTIPREDATOR RESPONSES OF ZENAIDA DOVES 509 three to five stimulus presentations per day on widely lowered towards an approaching conspecific and often spaced territories, and tests on the same territory were began a territorial display or fight. If the bird exited the separated by 2–4 days. circle, but no intruder could be identified, we scored the We began by placing a small pile of rice on the focal entire time spent outside the circle as ‘away’. subject’s territory. A hide (0.5 ! 0.70 ! 0.45 m) contain- We defined pecking as a clear downward movement of ing the cart and the visual stimulus was then placed 3 m the beak towards the ground regardless of whether the away from the food source. The cart could be pulled in beak actually contacted food or not. We only scored and out of the hide via a quiet pulley system operated by pecking if it occurred within the 2-m circle area centred the experimenter who sat approximately 7 m away. Doves on the food source because we wanted to test whether the in the Bellairs population often forage within considerably birds were willing to forage in proximity to the area where closer distances of humans. Correspondingly, we observed the visual stimulus was presented. In practice, foraging no behaviour that indicated that the focal subject was behaviour was very rare outside the defined zone, pre- disturbed by the experimenter. We waited for a maximum sumably because the experimental food was far more of 20 min for the territory owner to appear and begin abundant than any naturally occurring food. foraging on the experimental food source. If the territory For each bird, we counted the number of pecks and tail owner had not appeared after 20 min, the experiment was flicks produced during the 30-s prestimulus baseline and postponed until the next scheduled test time. each 60-s interval after the appearance of the stimulus. For The stimulus was only presented if the subject had been each behaviour, we then calculated the difference between foraging at the experimental food source for 30 s without the rate of occurrence during baseline and the rate of being interrupted by a conspecific intruder. This con- occurrence during each 60-s interval after stimulus onset. trolled for baseline behaviour and distance to the hide at For each behaviour, we examined effects of stimulus on stimulus onset across all tests. To present the stimulus, the the mean change in rate from baseline using a two-way experimenter slowly moved the cart out of the hide repeated measures ANOVA with factors for stimulus (pre- (approximately 0.5 m/s) to approximately 2 m away from dator, box, blank control) and time (three 60-s intervals). the food source where it remained immobile for 60 s. At Significant main effects of stimulus in ANOVA models of the end of the 60-s presentation period, the experimenter pecking and tail flicking were investigated further using moved the stimulus slowly back into the hide. paired t tests. We calculated the mean percentage of time spent away and chasing per minute between the appearance of the Data analysis stimulus and 120 s after the stimulus had disappeared. We We videorecorded the doves for 30 s immediately prior did not calculate changes in these behaviours from to stimulus presentation (baseline), 60 s during stimulus prestimulus baseline because we had ensured experimen- presentation, 120 s after the stimulus had disappeared. To tally that ‘away’ and ‘chase’ did not occur before the quantify responses to the stimulus, we measured (1) stimulus appeared (see test procedure). We used non- changes in pecking rates from prestimulus baseline, (2) parametric statistics to analyse these data because they the tendency of the birds to move away from the stimulus were not normally distributed. We tested for an effect of and (3) changes in rate of tail flicking, a flight intention stimulus on the mean percentage of time spent away per movement typically shown by alarmed zenaida doves. Tail minute using a nonparametric Friedman test. To ensure flicking consists of a rapid upwards, then downwards, that any effect of stimulus on pecking and tail-flicking movement of the whole body, accompanied by scanning, rates was not a consequence of differential levels of neck elongation and a characteristic jerking motion of the territorial defence, we also tested for an effect of stimulus tail; the posture is often followed by flight if the source of on the mean percentage of time spent chasing per minute alarm keeps approaching. Although tail flicking has not using a nonparametric Friedman test. Significant effects of been studied in zenaida doves, there is evidence from stimulus were investigated further using Wilcoxon other avian species that this behaviour is strongly associ- matched-pairs tests. ated with antipredator behaviours, such as alarm calling, Significance levels were set at 0.05, except for post hoc mobbing and wing flicking (Curio et al. 1978; Vieth et al. multiple comparisons, in which we used Bonferroni- 1980). corrected P values adjusted for three successive compar- Because zenaida doves actively defend their territories isons. Significance levels for post hoc tests were hence set against conspecifics, focal doves sometimes stopped for- at 0.017. All analyses were carried out on untransformed aging and left the experimental food source to chase away data using Statview 5.2 (SAS Institute 1998). an intruder. To separate locomotion away from the food that was caused by conspecific intrusion from locomotion away from the food that was not, we defined a 2-m Results and Discussion diameter circle centred on the food source. In the field, we placed natural objects, such as stones, on the borders of Zenaida doves suppressed pecking more in response to the circle so that we could later identify this area on video a predator stimulus than in response to a similar-sized recordings. In cases where the focal bird exited the circle control object (box), or during blank control trials (Fig. 1a). to chase away an intruder, the entire time spent outside This was reflected by both a significant ANOVA main the circle was scored as ‘chasing’. Chases were unambig- effect of stimulus on peck rate (F2,20 Z 9.667, P Z 0.001), uous to score; the focal bird moved rapidly with its head and the results of post hoc pairwise comparisons (paired 510 ANIMAL BEHAVIOUR, 70, 3

t tests: predator versus box, t10 Z 2.886, P Z 0.016; that the effects of the predator on peck rate were not Z Z predator versus blank, t10 4.080, P 0.002; box ver- attributable to differences in levels of territorial defence 2 sus blank, t10 Z 0.268, P Z 0.760). There was also a main between stimulus treatments (Friedman test: c2 Z 2.800, effect of time on peck rate (ANOVA: F2,20 Z 6.772, N Z 11, P Z 0.247; Fig. 2b, left panel). P Z 0.006), but no stimulus by time interaction To the extent that our prototypical model predator (F4,40 Z 0.109, P Z 0.979). In contrast, we found no effect was a representative exemplar of the greater category of of stimulus presentation on tail-flick rates (ANOVA: main mammalian predators, decreases in peck rates and in time Z Z effect stimulus, F2,20 2.005, P 0.161; main effect spent at the food patch are consistent with the idea that Z Z ! time, F2,20 1.050, P 0.369; stimulus time interac- doves respond more cautiously to predator stimuli than to tion, F4,40 Z 0.934, P Z 0.454; Fig. 1b). nonpredator stimuli. Doves spent significantly more time away from the experimental food patch during trials in which we presented a predator than during trials in which we EXPERIMENT 2 presented the control stimulus or no stimulus (Fig. 2a, left panel). This was reflected by both a significant effect of Grackles may be a more common source of information 2 about predators for zenaida doves than conspecifics. First, stimulus (Friedman test: c2 Z 19.579, N Z 11, P ! 0.001), and the results of post hoc pairwise comparisons (Wil- unlike doves, which forage alone or with their mate, coxon matched-pairs tests: predator versus box, T Z 66, grackles forage in small flocks. Consequently, there are N Z 11; P Z 0.003; predator versus blank, T Z 66, many grackle eyes looking out for predators. Second, N Z 11, P Z 0.003; box versus blank, T Z 10, N Z 11, doves are seed eaters and forage primarily on the ground P Z 0.500). In contrast, there was no effect of stimulus on within their own territory, whereas grackles are very time spent chasing conspecific intruders, demonstrating mobile omnivores, spending much of their time in the air or perched in trees. Therefore, grackles may be more likely to detect approaching danger, such as a troop of (a) vervet monkeys. Third, territorial zenaida doves are found 0 more often in close association with grackles than with conspecifics, so doves may be more likely to observe the –5 alarm responses of grackles than those of a conspecific. Finally, unlike doves, which do not vocalize in the presence of predators, grackles give conspicuous alarm –10 vocalizations, which can be heard tens of metres away. Responding to the calls of grackles would be a straightfor- –15 ward means for zenaida doves to avoid predators. The aim of experiment 2 was hence to determine whether zenaida (pecks/min) –20 doves respond to the sound of grackle alarm calls.

–25 Mean change from baseline Methods –30 Subjects (b) We selected 14 territory owner zenaida doves distributed 1.5 across 14 different territories in the same way as we had Predator 1.25 done for experiment 1. Six of these individuals had taken Box part in the previous experiment. 1 Blank 0.75 Acoustic stimuli 0.5 We measured the effects of grackle chuck calls on the behaviour of zenaida doves by comparing them to 0.25 responses evoked by a playback of grackle song. To create (tail flicks/min) the playback sequences of grackle chuck vocalizations, we 0 evoked these calls in captive grackles by holding them in

Mean change from baseline –0.25 a cage and standing beside them while staring at them. To obtain playbacks of song, we recorded free-living grackles. –0.5 1 2 3 Both kinds of acoustic stimuli were recorded using a Sony Time (60-s intervals) dynamic F-V620 microphone connected to a G3 iBook computer (Amadeus sound software, stereo recording, Figure 1. Behavioural responses of zenaida doves to a model predator, a box of similar volume, and during a blank control trial. sample rate 44.1 kHz, 16-bit amplitude encoding). Mean C SE (N Z 11) changes in peck rates (a) and tail-flick rates Folkstone Park and Bellairs Research Institute are located (b) from prestimulus baseline are shown for three successive 60-s close to roads, so they are relatively noisy environments intervals from the appearance of the stimulus to 2 min after the in which to record sound. Using Cool Edit Pro 1.2a stimulus had disappeared. (Syntrillium Software 2000), we edited each stimulus GRIFFIN ET AL.: ANTIPREDATOR RESPONSES OF ZENAIDA DOVES 511

(a) Visual Acoustic 100 100

80 80 * * 60 60

40 40 Mean time away from food/min (%) 20 20

0 0 Predator Box Blank AlarmSong Blank

(b) 100 100

80 80

60 60

40 40

Mean time allocated 20 20 to territorial defence/min (%) 0 0 Predator Box Blank AlarmSong Blank Figure 2. Mean percentage of time spent (a) away from the food source and (b) in territorial defence in response to three visual (left panel) and three acoustic (right panel) stimuli. For visual stimuli, mean C SD (N Z 11) percentage of time per minute was averaged across the 1-min presentation period and two 1-min postpresentation intervals. For acoustic stimuli, mean C SD (N Z 14) percentage of time was averaged across the 30-s stimulus presentation period and two 30-s poststimulus presentation intervals.*P ! 0.017. recording to reduce background noise. First, we used (frequency response 0.07–18 kHz) located 3 m away from a sample of the stimulus where the birds were not the food source, and separated by approximately 1 m. We vocalizing to calculate a noise reduction algorithm. We conducted one blank control trial and presented one set the level of noise reduction such that the bird calls randomly selected exemplar of each of the two acoustic were not altered to our ear. Second, we removed a narrow treatments (alarm, song) on each territory, such that each frequency range of the remaining background noise, exemplar was only used three or four times. Treatment which consisted mainly of traffic sound, using parame- order was balanced across territories. tric equalization (Cool Edit Pro 1.2a, Syntrillium Software 2000). Finally, we edited the recordings to make four Test procedure distinct 30-s exemplars of alarm call sequences (Fig. 3) and As in experiment 1, we conducted all trials in the early four distinct exemplars of song sequences (Fig. 3) in order morning between 0545 and 0900 hours. As in experiment 1, to sample natural variation in the acoustic structure of we conducted three to five stimulus presentations per day these vocalizations. Each exemplar began with a 2-s fade- on widely separated territories, and tests on the same in and ended with a 2-s fade-out to avoid startling the territory were separated by 2–4 days. birds. We began by placing a pile of rice on the focal subject’s To examine the differential effect of alarm call versus territory. We then placed the loudspeakers 3 m away from song per se, we chose to match the volume of both stimuli the food and connected them to an iBook laptop com- to the natural volume of alarm calls. We adjusted the puter held by the experimenter, who sat approximately amplitude of all stimuli at the output using a digital sound 8 m away from the food source. As in experiment 1, we level meter (Radioshack, model no. 33-2085) and played waited for a maximum of 20 min for the territory owner to back the stimuli at a mean amplitude of 82 dB (A appear and begin foraging at the food source. If the weighting; peak; G1 dB measured 3 m in front of the territory owner had not appeared after 20 min, we post- speakers). This is roughly equivalent to the birds’ own poned the experiment until the next scheduled test time. alarm call output volume measured at a distance of 2 m. As in experiment 1, the stimulus was only presented if the We played back the stimuli using an iBook G3 Apple subject had been foraging at the experimental food source computer laptop through two Altec Lansing 220 speakers without interruption for 30 s. 512 ANIMAL BEHAVIOUR, 70, 3

(a) Grackle alarm 16 14 12 10 8 6 4 2 0 0 5101520 25 30 Time (s)

(b) Grackle song 16 Frequency (kHz) 14 12 10 8 6 4 2 0 0 5101520 25 30 Time (s) Figure 3. Sonagrams and spectrograms of one of the four exemplars of acoustic stimuli used to study recognition of grackle alarm calls by zenaida doves. Sampling rate Z 44.1 kHz, delta frequency Z 172.3 Hz, delta time Z 5.8 ms, fast Fourier transform Z 256, Hanning window, cutoff values: low: 40 dB, high: 6 dB.

Data analysis using multiple paired t tests and Wilcoxon matched-pairs We videorecorded the doves for 30 s immediately prior tests. Significance levels were set as in experiment 2. to stimulus presentation (baseline), 30 s during stimulus presentation, and 1 min after the end of the stimulus. As Results and Discussion in experiment 1, we scored and analysed changes in pecking rate and tail-flicking rate from prestimulus base- The sound of grackle alarm calls evoked a sustained line, the percentage of time spent away from the food decrease in peck rates in zenaida doves from stimulus patch, and the percentage of time allocated to chasing onset to 1 min afterward (ANOVA: stimulus main effect, away intruders. F2,26 Z 23.523, P ! 0.001; time main effect, F2,26 Z 0.587, Changes in pecking and tail-flicking rates from baseline P Z 0.563; stimulus ! time interaction, F4,52 Z 1.578, were quantified by calculating difference scores between P Z 0.194; Fig. 4a). The decrease was significantly greater the 30-s prestimulus baseline and each of the three 30-s than that evoked by either grackle song or blank control intervals after the onset of the acoustic stimulus. We trials (paired t tests: alarm versus song, t13 Z 4.254, ! Z ! calculated the mean percentage of time allocated to P 0.001; alarm versus blank, t13 7.436, P 0.001; chasing and away from the food per minute between song versus blank, t13 Z 1.405, P Z 0.184). Zenaida doves the onset of the acoustic stimulus and 60 s after the end of also tail-flicked significantly more both during presenta- the playback. tions of grackle alarm vocalizations and for 1 min afterward As in experiment 1, we tested for a differential effect of than in response to grackle song (Fig. 4b), as indicated by stimulus on changes in pecking rate and tail-flicking rate a main effect of stimulus on tail-flick rate (ANOVA: from prestimulus baseline using two-way repeated meas- F2,26 Z 4.040, P Z 0.030), a nonsignificant main effect Z Z ures ANOVAs with factors for stimulus (alarm, song, blank of time (F2,26 0.811, P 0.455) and a nonsignificant control) and time (three 30-s intervals). We tested for stimulus by time interaction (F4,52 Z 0.390, P Z 0.815). a differential effect of stimulus on the mean percentage of This effect was, however, small and was not apparent in Z time away per minute and the mean percentage of time Bonferroni-corrected t tests (alarm versus song, t13 2.110, allocated to chasing per minute using nonparametric P Z 0.055; alarm versus blank, t13 Z 2.580, P Z 0.230; Friedman tests. Significant effects were examined further song versus blank, t13 Z 0.906, P Z 0.382). GRIFFIN ET AL.: ANTIPREDATOR RESPONSES OF ZENAIDA DOVES 513

(a) was absent from this reduced data set, but there was 5 a significant stimulus by time interaction, revealing that alarm stimuli continued to evoke significantly more tail 0 flicking, but only during stimulus presentation (Fig. 4b). Insofar as our exemplars of grackle alarm calls and song –5 were representative of the greater category of these vocal- –10 izations, our results are consistent with the idea that zenaida doves adjust their levels of foraging in response to –15 the vocal alarm behaviour of grackles. First, changes in

(pecks/min) foraging were evoked by grackle alarm calls, but not by –20 a control stimulus of similar amplitude (grackle song), indicating that changes in dove behaviour do not simply Mean change from baseline –25 reflect a response to loud sounds played back from nearby loudspeakers. Second, changes in behaviour are unlikely to –30 reflect a startle response to broadband noises. We gradually (b) increased the volume of all playbacks over a 2-s period until 2 the stimulus reached maximum volume, thereby reducing Alarm the likelihood of a startle reaction. In addition, the doves’ 1.5 Song responses were long-lasting, unlike startle responses; sub- Blank jects did not resume foraging after the end of the playback, 1 but rather remained vigilant for at least 60 s afterward. However, we cannot exclude the possibility that zenaida doves are, in general, more fearful of broadband noises that 0.5 are more similar in frequency composition to grackle alarm calls than to grackle song. Such responsiveness would be 0 (tail flicks/min) sufficient to allow doves to use grackle alarm calls as a signal for danger. Heterospecific alarm calls are usually considered –0.5 Mean change from baseline to be recognized as alarm calls per se, rather than inherently aversive sounds, if they trigger responses that are qualita- –1 1 2 3 tively similar to those evoked by conspecific alarm calls, Time (30-s intervals) such as alarm calling and mobbing (Evans 1972; Krams & Krama 2002). Zenaida doves do not display these responses, Figure 4. Behavioural responses of zenaida doves to a playback of so we have to rely on more general changes in behaviour to grackle alarm calls, grackle song, and during a blank control trial. Mean C SE (N Z 14) changes in peck rates (a) and tail-flick rates quantify recognition. In the present experiment, subjects (b) from prestimulus baseline are shown for three successive 30-s did not move away from the origin of the sound, the intervals from the appearance of the stimulus to 1 min after the end loudspeakers. Rather, they adopted an alert erect posture of the stimulus. and tail-flicked, as if scanning their surroundings, and maintained this behavioural state for 60 s after the end of the alarm stimulus. In contrast, in experiment 1, the Contrary to the results obtained during presentations of presence of the predator model caused the birds to move a model predator, there was no main effect of stimulus on away from the food, suggesting that the model was in- 2Z time away from the food source (Friedman test: c2 2.649, herently aversive. We suggest therefore that grackle alarm N Z 14, P Z 0.266; Fig. 2a, right panel). There was also no calls may not be inherently frightening to doves, but rather effect of stimulus on time allocated to chasing, confirming that they are recognized as a sound associated with danger. that, as in experiment 1, effects of stimulus on peck rate and tail-flick rate were not attributable to differential 2 GENERAL DISCUSSION levels of territorial defence (Friedman test: c2 Z 2.632, N Z 14, P Z 0.268, Fig. 2b; right panel). This study represents the first attempt to characterize the In four of 14 alarm call presentation trials, the playback responses of zenaida doves to predators, and to determine stimulus attracted nearby grackles and triggered alarm whether doves respond with antipredator behaviour to calling in these individuals. The observation that our the alarm calls of a closely associated species, the carib playback stimulus facilitated alarm calling in free-ranging grackle. We found that zenaida doves suppressed foraging grackles provides excellent support that the acoustic both in response to a model predator and in response to stimulus reliably reproduced the sound of grackle alarm the sound of grackle alarm vocalizations. Responses to the calls. However, we cannot exclude the possibility that predator model also involved moving away from the during these trials the doves responded to the alarm immediate vicinity, whereas responses to grackle alarm behaviour of the live grackles, rather than to the playback calls consisted of remaining alert and tail flicking. Together, stimulus. We ensured therefore that removing these trials these results strongly suggest that doves respond adap- from the data set did not change the outcome of our tively to the antipredator behaviour of carib grackles. analyses. Main effects of stimulus on pecking rate re- Empirical demonstrations of heterospecific alarm call mained significant. The effect of stimulus on tail flicking recognition that employ experimental playback to tease 514 ANIMAL BEHAVIOUR, 70, 3

apart the effects of the vocalization from other environ- groupings (Rasa 1990), particularly if the ecology of the mental events, such as the behaviour of social compan- eavesdropping species puts it at a disadvantage that is ions, remain relatively rare in birds compared with the compensated for by the antipredator behaviour of the huge literature on intraspecific communication (Frings producer species. For example, Sullivan (1985) found that et al. 1955; Nuechterlein 1981; Sullivan 1984, 1985; Go¨th downy woodpeckers, Picoides pubescens, which often for- 2001; Krams & Krama 2002). This is rather surprising since age alone, respond to the alarm vocalizations of black- heterospecific communication has been suggested to play capped chickadees, Parus atricapillus, which associate with a role in the evolution of alarm call structure (Marler conspecifics and have a high propensity to call. Western 1957). The paucity of data in birds stands in stark contrast grebes, Aechmophorus occidentalis, have reduced flying to the large literature on heterospecific chemical alarm capabilities during the breeding season and use the aerial substance recognition in fish, the mechanisms of which alarm calls of Forster’s terns, Sterna forsteri, to detect are now well understood. Heterospecific alarm substances overhead danger, thereby perhaps reducing their vulner- that are structurally similar to those released by conspe- ability to predators (Nuechterlein 1981). The alarm calls of cifics evoke responses that are not dependent upon prior antbirds and tanagers in Amazonian polyspecific flocks are experience, suggesting that these chemicals are recognized exploited by a variety of gleaner and prober species, which by simple generalization from conspecific alarm signal find food by scanning the ground with their bills close to recognition (Chivers & Smith 1998; Magurran 1999). In the substrate (Munn 1986). Similarly, in the grackle–dove contrast, heterospecific signals that bear little structural system, an approaching predator is more likely to be similarity to those released by conspecifics elicit responses detected by a grackle than a dove because of the differ- that arise as a consequence of associative learning (Chivers ences in these species’ social systems, their use of habitat et al. 1995; Mirza & Chivers 2001, 2003). and their alarm behaviour. Grackles are hence an ideal Zenaida doves do not produce any vocal alarm signals of source of predatory information for doves. Earlier work their own, so generalization from conspecific alarm call has shown that doves can parasitze information from recognition can be ruled out. They may, however, respond grackles about food and novel foraging techniques. The to broadband pulsatile grackle alarm calls through a gen- benefits grackles may provide for doves, both in terms of eral sensitivity to loud broadband sounds. We consider antipredator vigilance and information about food avail- this unlikely, however, because our subjects did not move ability (Webster & Lefebvre 2001) and novel foraging away from the source of the sound during the entire 30-s techniques, may explain why doves forage with grackles playback, suggesting that they did not find the acoustic so readily, despite considerable dietary overlap in areas stimulus inherently aversive. We suggest, therefore, that where anthropogenic food is an important resource (Dol- responses to grackle alarm calls may be acquired through man et al. 1996). More specifically, relying on grackles to associative learning in which grackle alarm behaviours signal danger may outweigh the costs of enhanced become associated with the presence of predators. Our foraging competition, particularly if searching for food thinking is analogous to that proposed to explain the or intruders is dependent upon the use of search images or results of cross-species learning about food in the grackle– diverts the dove’s attention from other environmental zenaida dove system. Dolman et al. (1996) found that stimuli, such as predators (Dukas & Kamil 2000, 2001). territorial doves acquire a novel foraging technique more Finally, our results raise the possibility that grackle alarm readily from a demonstrator grackle than from a conspe- responses may facilitate avoidance learning of novel cific dove, whereas doves from a group-feeding harbour predators in doves. This question awaits future research. population learn more readily from a conspecific. Territo- rial doves caught at a site adjacent to the harbour, which Acknowledgments experience both group-feeding at the harbour and aggres- sive defence at their site of residence, learn equally well A. S. Griffin is supported by the Swiss National Foundation from doves and grackles (Carlier & Lefebvre 1997). Given for Scientific Research. The research was supported by the lack of isolation between dove populations with dif- grants from the Natural Sciences and Engineering Re- ferent tutor preferences, the rapid changes in defence search Council of Canada and Fonds que´becois pour la brought about by manipulations of food distribution in recherche en nature et technologie (FQRNT) (Que´bec) to the field (Goldberg et al. 2001) and the lack of differences Louis Lefebvre. We thank the staff of the Bellairs Research between group-feeding and territorial doves kept for long Institute for technical assistance and Evan Balaban for periods under identical conditions in aviaries (B. Livoreil, helping with sound analyses and depictions. All proce- unpublished data), learning is the most plausible mecha- dures were approved by the McGill University Animal nism for the differences in response to feeding informa- Ethics Committee (protocol no. 4660). tion. During ontogeny, territorial doves learn to associate food with the presence of grackles and to associate conspecifics with territorial aggression. The idea that References doves may learn to behave adaptively to the alarm calls Carlier, P. & Lefebvre, L. 1997. Ecological differences in social of carib grackles is supported by evidence that associative learning between adjacent, mixing populations of zenaida doves. learning can mediate the acquisition of antipredator Ethology, 103, 772–784. responses to novel acoustic stimuli (Shriner 1999). Chivers, D. P. & Smith, R. J. F. 1998. Chemical alarm signalling in Information parasitism of one species’ alarm system by aquatic predator–prey systems: a review and prospectus. E´co- another is likely to be an important benefit of polyspecific science, 5, 338–352. GRIFFIN ET AL.: ANTIPREDATOR RESPONSES OF ZENAIDA DOVES 515

Chivers, D. P., Wisenden, B. D. & Smith, R. J. F. 1995. The role of Magurran, A. E. 1999. The causes and consequences of geographic experience in the response of fathead minnows (Pimephales variation in antipredator behavior. Perspectives from fish popula- promelas) to skin extract of Iowa darters (Etheostoma exile). tions. In: Geographic Variation in Behavior. Perspectives on Behaviour, 132, 665–674. Evolutionary Mechanisms (Ed. by S. A. Foster & J. A. Endler), pp. Curio, E., Ernst, U. & Vieth, W. 1978. The adaptive significance of 139–163. Oxford: Oxford University Press. avian mobbing. II. Cultural transmission of enemy recognition in Maloney, R. F. & McLean, I. G. 1995. Historical and experimental blackbirds: effectiveness and some constraints. Zeitschrift fu¨r learned predator recognition in free-living New Zealand robins. Tierpsychologie, 48, 184–202. Animal Behaviour, 50, 1193–1201. Dolman, C., Templeton, J. & Lefebvre, L. 1996. Mode of foraging Marler, P. 1957. Specific distinctiveness in the communication competition is related to tutor preference in Zenaida aurita. Journal signals of birds. Behaviour, 11, 13–39. of Comparative Psychology, 110, 45–54. Mirza, R. S. & Chivers, D. P. 2001. Learned recognition of Dukas, R. & Kamil, A. C. 2000. The cost of limited attention in blue heterospecific alarm signals: the importance of a mixed predator jays. Behavioral Ecology, 11, 502–506. diet. Ethology, 107, 1007–1018. Dukas, R. & Kamil, A. C. 2001. Limited attention: the constraint Mirza, R. S. & Chivers, D. P. 2003. Fathead minnows learn to underlying search image. Behavioral Ecology, 12, 192–199. recognize heterospecific alarm cues they detect in the diet of Evans, S. M. 1972. Specific distinctiveness in the calls of cordon bleus a known predator. Behaviour, 140, 1359–1369. (Uraeginthus spp.; Estrildidae). Animal Behaviour, 20, 571–579. Morse, D. H. 1977. Feeding behavior and predator avoidance in Frings, H., Frings, M., Cox, B. & Peissner, L. 1955. Auditory and heterospecific groups. BioScience, 27, 332–339. visual mechanisms in food-finding behaviour in the herring gull. Munn, C. A. 1986. Birds that ‘cry wolf’. Nature, 319, 143–145. Wilson Bulletin, 67, 155–170. Nuechterlein, G. L. 1981. ‘Information parasitism’ in mixed colonies Gaddis, P. 1980. Mixed flocks, accipiters, and antipredator of western grebes and Forster’s terns. Animal Behaviour, 29, 985–989. behavior. Condor, 82, 348–349. Rasa, O. A. E. 1990. Interspecific defense aggregations – a model for Goldberg, J. L., Grant, J. W. A. & Lefebvre, L. 2001. Effects of the evolution of sociality and kin selection. Netherlands Journal of temporal predictability and spatial clumping of food on the Zoology, 40, 711–728. intensity of competitive aggression in the zenaida dove. Behavioral SAS Institute 1998. Statview 5.0. Cary, North Carolina: SAS Ecology, 12, 490–495. Institute. Go¨th, A. 2001. Innate predator-recognition in Australian brush- Shriner, W. M. 1999. Antipredator responses to a previously neutral turkey (Alectura lathami, Megapodiidae) hatchlings. Behaviour, sound by free-living adult golden-mantled ground squirrels 138, 117–136. Spermophilus lateralis (Sciuridae). Ethology, 105, 747–757. Griffin, A. S. & Evans, C. S. 2003. Social learning of antipredator Sullivan, K. A. 1984. Information exploitation by downy wood- behaviour in a marsupial. Animal Behaviour, 66, 485–492. peckers in mixed-species flocks. Behaviour, 91, 294–311. Griffin, A. S., Evans, C. S. & Blumstein, D. T. 2001. Learning Sullivan, K. A. 1985. Selective alarm calling by downy woodpeckers specificity in acquired predator recognition. Animal Behaviour, 62, in mixed-species flocks. Auk, 102, 184–187. 577–589. Syntrillium Software 2000. Cool Edit Pro 1.2.a. Scottsdale, Arizona: Griffin, A. S., Evans, C. S. & Blumstein, D. T. 2002. Selective Syntrillium Software. learning in a marsupial. Ethology, 108, 1103–1114. Terborgh, J. 1990. Mixed flocks and polyspecific associations: costs Jaramillo, A. & Burke, P. 1999. New World Blackbirds: the Icterids. and benefits of mixed groups to birds and monkeys. American Princeton, New Jersey: Princeton University Press. Journal of Primatology, 21, 87–100. Krams, I. & Krama, T. 2002. Interspecific reciprocity explains Vieth, W., Curio, E. & Ernst, U. 1980. The adaptive significance of mobbing behaviour of the breeding chaffinches, Fringilla coelebs. avian mobbing. III. Cultural transmission of enemy recognition in Proceedings of the Royal Society of London, Series B, 269, 2345–2350. blackbirds: cross-species tutoring and properties of learning. Lefebvre, L. 1996. Raging dove. Natural History, 105, 34–37. Animal Behaviour, 28, 1217–1229. Lefebvre, L., Palameta, B. & Hatch, K. K. 1996. Is group-living Webster, S. & Lefebvre, L. 2001. Problem-solving and neophobia associated with social learning? A comparative test of a gregarious in a passeriform–columbiform assemblage in Barbados. Animal and a territorial columbid. Behaviour, 133, 241–261. Behaviour, 62, 23–32. McLean, I. G. & Rhodes, G. I. 1991. Enemy recognition and Wisenden, B. D. & Harter, K. R. 2001. Motion, not shape, facilitates response in birds. In: Current Ornithology (Ed. by D. A. Power), pp. association of predation risk with novel objects by fathead 173–211. New York: Plenum. minnows (Pimephales promelas). Ethology, 107, 357–364. McLean, I. G., Lundie-Jenkins, G. & Jarman, P. J. 1996. Teaching Wolters, S. & Zuberbu¨hler, K. 2003. Mixed-species associations of an endangered mammal to recognise predators. Biological Diana and Campbell’s monkeys: the costs and benefits of a forest Conservation, 75, 51–62. phenomenon. Behaviour, 140, 371–385.