Acoustic communication in crocodilians: information encoding and species specificity of juvenile calls

Transcription

1 Anim Cogn (2012) 15: DOI /s ORIGINAL PAPER Acoustic communication in crocodilians: information encoding and species specificity of juvenile calls Amélie L. Vergne Thierry Aubin Samuel Martin Nicolas Mathevon Received: 21 November 2011 / Revised: 2 July 2012 / Accepted: 2 July 2012 / Published online: 21 July 2012 Ó Springer-Verlag 2012 Abstract In the Crocodylia order, all species are known for their ability to produce sounds in several communication contexts. Though recent experimental studies have brought evidence of the important biological role of young crocodilian calls, especially at hatching time, the juvenile vocal repertoire still needs to be clarified in order to describe thoroughly the crocodilian acoustic communication channel. The goal of this study is to investigate the acoustic features (structure and information coding) in the contact call of juveniles from three different species (Nile crocodile Crocodylus niloticus, Black caiman, Melanosuchus niger and Spectacled caiman, Caiman crocodilus). We have shown that even though substantial structural differences exist between the calls of different species, they do not seem relevant for crocodilians. Indeed, juveniles and adults from the species studied use a similar and nonspecies-specific way of encoding information, which relies on frequency modulation parameters. Interestingly, using conditioning experiments, we demonstrated that this tolerance in responses to signals of different acoustic structures was unlikely to be related to a lack of discriminatory A. L. Vergne N. Mathevon (&) Equipe de Neuro-Ethologie Sensorielle, CNPS, Université de Lyon Saint-Etienne, CNRS UMR 8195, Saint-Etienne, France mathevon@univ-st-etienne.fr A. L. Vergne T. Aubin N. Mathevon Centre de Neurosciences Paris-Sud, Centre National de la Recherche Scientifique, UMR 8195, Paris, France T. Aubin Equipe Communications acoustiques, CNPS, Université Paris XI, CNRS UMR 8195, Orsay, France S. Martin La Ferme aux Crocodiles, Pierrelatte, France abilities. This result reinforced the idea that crocodilians have developed adaptations to use sounds efficiently for communication needs. Keywords Crocodiles Caimans Acoustic communication Species recognition Information coding Conditioning experiments Introduction In crocodiles, acoustic signalling is used for social interactions, particularly between adults during courtship and territorial defence (Garrick et al. 1982), and within family groups where survival of the young depends on maternal care (Campbell 1973; Herzog and Burghardt 1977; Vergne and Mathevon 2008; Senter 2008; Vergne et al. 2011; for a review of acoustic communication in crocodilians, see Vergne et al. 2009). Juvenile crocodilians produce various sounds that have been classified into three main functional categories. First, hatching calls (with the sub-categories, pre-hatching, hatching and post-hatching) that are uttered by embryos and newborns to solicit the mother to open the nest and have been shown to fine-tune hatching synchrony among siblings (Vergne and Mathevon 2008); second, contact calls produced by juveniles from hatching to several days or weeks old. These mainly support cohesion between juveniles (Vergne et al. 2011). And the third is distress calls that induce parental (mother) protection (for an experimental approach of the biological roles of contact and distress calls, see Vergne et al. 2011). Finally, juveniles may emit threat and disturbance calls when threatened (Britton 2001; Vergne et al. 2009). Although the acoustic structure and the biological roles of these juvenile signals have recently started to be studied (Britton 2001;

2 1096 Anim Cogn (2012) 15: Vergne and Mathevon 2008; Vergne et al. 2007, 2009, 2011), we still do not know the specific acoustic features that make these sounds relevant to crocodilians. Species specificity is one of the most consistent characteristic of animal vocalizations, especially in the context of reproduction (Dooling et al. 1992). This is related to the fact that most sounds are directed towards conspecifics whose reactions should be appropriate to the content of the message. The species specificity of vocalizations is of primary importance when two close-related species live in sympatry and when risks of hybridization or mis-directed parental care can occur. Among the 23 species of crocodilians spread over the subtropical and tropical regions of the New and Old World (Trutnau and Sommerlad 2006), sympatry is rare and most interactions are likely to be intraspecific (though there are some exceptions, for example, Spectacled caiman Caiman crocodilus and Black caiman Melanosuchus niger in South America). Due to genetic drift, differences in the acoustic structure of vocalizations may also arise in allopatry. Except Britton (2001) who conducted a preliminary structural analysis of juvenile calls from different species, there is no published investigation on the species specificity of juvenile calls. In the present study, we focused on juvenile contact calls. These vocalizations, used by the group to gather, seem to be mostly directed towards other juveniles (Vergne et al. 2011). By investigating the acoustic structure of contact calls from different species, our goal was to discover whether calls were species-specific or whether they share a common rule for encoding information. First, we compared the acoustic structure of contact calls from three different species (one species belonging to the family Crocodylidae: Nile crocodile Crocodylus niloticus; two species from the family Alligatoridae: Black caiman, Melanosuchus niger and Spectacled caiman, Caiman crocodilus), and we looked for the presence of speciesspecific information in these recorded calls using playback experiments. Then, modified signals were broadcast to find the acoustic features that trigger the receiver s behavioural response. Finally, conditioning experiments helped to determine the details of some aspects of crocodile auditory discriminatory abilities. Experiment 1: species specificity of juvenile calls Methods Sound analysis We analysed calls from juveniles of 3 different species (Nile Crocodile, Spectacled Caiman and Black Caiman; 25 calls per species, from 5 individuals/species). Animals were recorded either at the laboratory (Nile crocodiles and Spectacled caimans, provided by the zoo La Ferme aux Crocodiles, Pierrelatte, France), or in the field (Black caimans, studied along the Rupununi river, Guyana, South America). All recorded calls were spontaneously emitted in contexts where juveniles were gathered as a group, without any visible disturbance. They thus can be considered as typical juvenile contact calls (Vergne et al. 2009, 2011). These acoustic signals are complex sounds, that is, composed of a fundamental frequency and a harmonic series, modulated in frequency and amplitude (Fig. 1). The Nile crocodiles recorded were 4 6 days old. They hatched in our laboratory, and vocalizations were recorded at the time the crocodiles were released into a home tank where they could meet other siblings for the first time. Spectacled caimans were older individuals (2 3 weeks old). They hatched at the zoo La Ferme aux Crocodiles (Pierrelatte, France) and were brought to the laboratory 2 weeks later. Their vocalizations were recorded when we released them together in their own home tank. The Black caimans contact calls were recorded during a field trip to Guyana (South America). Individuals were about 1 week old. Contact calls were emitted spontaneously by juveniles gathered on the riverbank. All recordings were performed at a distance of cm from the animals with an omnidirectional microphone SENNHEISER MD42 connected to a Marantz PMD670 tape recorder. We conducted a sound analysis to describe the calls acoustic structure. For the Nile crocodile contact calls, the analysis had already been performed and published elsewhere (Vergne et al. 2009). For comparison purposes, we used the same method to analyse the contact calls of the two other species. Briefly, 7 acoustic variables were measured in both temporal and frequency domains using Avisoft-SAS Lab-Pro ( and Praat ( praat.org) (Fig. 2). The total duration of the call (DT) was measured from the oscillogram (Fig. 2b). We performed a spectrographic analysis (window size: 1,024, sampling frequency: 48 khz, overlap 90 %) and measured two variables related to the fundamental frequency: the maximal frequency value (F0max, Hz) and the ending frequency value (Fend, Hz; Fig. 2a). To describe the frequency modulation, we calculated the slope of the first temporal quartile of the call (Slope 1) and the slope of the last three temporal quartiles of the call (Slope 2; Fig. 2a). To describe the spectral energy distribution, we measured two parameters on a spectrum of the entire call: the bandpass (Bandpass, Hz) and the frequency at the maximum amplitude (PicF, Hz) (Fig. 2c). Statistical tests were conducted using the Statistica package (version 6, Statsoft France). First, a one-way MANOVA based on the 7 measured acoustic variables was performed to compare the calls between the 3 species. This

3 Anim Cogn (2012) 15: Fig. 1 Spectrograms (top) and oscillograms (bottom) of juvenile contact calls (a: Nile crocodile; b: spectacled caiman; c: black caiman) Fig. 2 Acoustic parameters used for the analysis of juvenile calls. a Spectrogram. b Oscillogram. c Average frequency spectrum. DT (s): total duration, F0max (Hz): maximal frequency of the fundamental, Fend (Hz): final frequency of the fundamental, slopes 1 and 2 (Hz/s): frequency modulation slopes, PicF (Hz): frequency at the maximum power amplitude of the spectrum analysis was followed by a one-way ANOVA on each acoustic variable and a post hoc Fisher s LSD procedure at a 95 % confidence level to identify statistically homogeneous groups. A cross-validated discriminant function analysis using the 7 measured acoustic variables, preceded by a principal component analysis, provided a classification procedure that assigned each call to its appropriate species (correct assignment) or to one of the two others (incorrect assignment). Playback experiments Five young Nile crocodiles (age: 1 month old, sex unknown) were tested. These individuals were kept together in a home tank (dimensions: cm; water temperature: 30 C; air temperature: C; appropriate luminance to this tropical species was programmed 12 h day/12 h night). During experimental periods, the tested individual was isolated from the others in a test tank whose characteristics particularly temperature, luminance and feeding conditions were the same as for the home tank. Both tanks had a shelter at one end and the crocodiles hid in it most of the time. The test tank was located in an acoustically isolated room. To limit the possible stress on the crocodiles due to carrying them from on tank to another, we always moved the tested individual in the test tank on the morning before the playback so that there were several hours of non-disturbance before the

4 1098 Anim Cogn (2012) 15: tests. All playback tests were conducted during the night (from 8 p.m. to 2 a.m.) a preliminary study of the crocodile circadian behavioural activity having shown a preferential nocturnal activity. A low intensity red neon light was placed above the aquarium to allow the monitoring of behaviours at night. To test the species specificity of calls, each of the 5 individuals was successively challenged with Nile crocodile calls, black caiman calls, spectacled caiman calls and a control noise (white noise band-pass filtered between 100 and 5,000 Hz; duration = 0.15 s; emitted at 50 db SPL ; built with Syntana software, Aubin 1994). The spectrum bandwidth, duration and intensity of the control noise were chosen to mimic those of a juvenile natural contact call. A loudspeaker was suspended at one end of the test tank (at the opposite side from the shelter). Bubble wrap covered the tank s inner walls in order to minimize reverberation of acoustic waves. The loudspeaker was connected to a computer located outside the experimental room to minimize disturbance for the isolated crocodile. This computer was programmed to control the emission of acoustic stimuli. Emission level conformed to the natural intensity of calls (51 ± 4dB SPL at 1 metre from the loudspeaker, measured with a sound level meter SL-4001, Digital Instruments). Every 2 h starting at 8 p.m., the tested young Nile crocodile was exposed to a 90-second playback of an experimental acoustic sequence. Each sequence was composed of 5 identical series of experimental signals; interval between series = 12 s; 1 series = 4 repetitions of a given experimental signal with a natural emission rhythm, approximately one call every 2 s; total duration of one series = 8s. During each experimental night, each tested individual was challenged with 1 sequence of each species calls and 1 sequence of control noise. The order of sequences was changed for each individual. Pseudo-replication within sequences of a given species calls was avoided by building sequences using different calls from different individuals. Analysis of behavioural responses All behavioural and vocal responses were recorded. We were able to follow precisely the crocodile s movements with two webcams (IP DCS-900) suspended above the two opposite sides of the tank to get a complete view. Go1984 software ( was programmed to activate the webcams and record the crocodile s activity during sequences of 15 min (5 min before the start of the playback sequence, during the 90 s of playback and 10 min after). Any vocal activity was recorded during the same duration via a microphone (Labtec desk microphone). Two parameters were chosen to measure the responses to playbacks: (1) Latency (t), the time between the first call played back and the first observable reaction (head or body movement) of the tested individual. A 4-level scale was defined: 0 if t [ 60 s or no reaction, 1 if 41 B t B 60 s, 2 if 21 B t B 40 s, and 3 if t B 20 s; (2) animal displacements were also quantified according to a 4-levels scale: 0 was scored if the animal did not move at all during the 90 s of playback, 1 if we observed a head and/or body orientation towards the loudspeaker but no displacement, 2 if the individual moved towards the loudspeaker but stayed at more than 20 cm from it, and 3 if the individual came at less than 20 cm from the loudspeaker. Note that we only took into consideration experiments when the crocodile was at the opposite side of the speaker before the start of a test (that was the case 99 % of the time). Also, we did not need negative scores to assess displacements because once a tested animal moved towards the speaker, it never moved away from it during our recordings. These two scores (head or body movement and displacements) were then summed to obtain a general score representing the behavioural response to playback, varying from 0 to 6, with 0 corresponding to the weakest behavioural response and 6 to the strongest. Using the behavioural response scores, we first performed a Friedman two-way repeated measures analysis of variance by ranks to detect differences in Nile crocodiles responses throughout the different playback tests. Second, we did Wilcoxon tests to compare responses to the Nile crocodile calls with responses to the two other signals. Statistical tests were conducted using Statistica package. Results Acoustic differences between the contact calls of the three species Although the acoustic structure of calls of the three species show strong similarities, as shown by spectrographic representations (Fig. 1), the analysis sheds light on significant differences between the measured parameters from one species to another (Table 1, MANOVA, F(14,132) = 21.2, P \ 0.001). The one-way ANOVAs and the Fisher s LSD tests reveal that all measured acoustic parameters except the slopes of the frequency modulation allow differentiating a Nile crocodile contact call from a Black or a spectacled caiman s call (ANOVAs: df = 2,72; P \ for each of the seven acoustic variables, DT : F = 178, F0max : F = 21.2, Fend : F = 9.29, Bandpass : F = 31.2, Slope 1 : F = 7.65, Slope 2 : F = 8.30, PicF : F = 22.9; homogeneous groups: see Table 2). Nile crocodile calls are higher pitched (F0max, Fend, Bandpass), possess more energy towards high frequencies (PicF) and have a longer duration (DT) than calls of the other tested species. Differences were also measured between contact calls of Spectacled and Black caimans. For instance, the maximum of the fundamental frequency (F0max) and the

5 Anim Cogn (2012) 15: Table 1 Mean ± SD of acoustic variables from contact calls of three species of crocodilians Nile crocodiles, Spectacled caimans and Black caimans (5 individuals, 5 calls/individual) DT(s) F0max (Hz) Fend (Hz) Bandpass (Hz) Slope 1 (Hz/s) Slope 2 (Hz/s) PicF (Hz) Nile crocodiles (N = 25) ± ± ± 61 3,925 ± 760-2,225 ± 1,536-1,291 ± ± 222 Black caimans (N = 25) ± ± ± 46 2,573 ± 193-2,784 ± 1,607-1,437 ± ± 242 Spectacled caimans (N = 25) 0.09 ± ± ± 49 3,289 ± 694-4,005 ± 1,784-2,091 ± 1, ± DT (s): total duration, F0max (Hz): maximal frequency of the fundamental, Fend (Hz): final frequency of the fundamental, slopes 1 and 2 (Hz/s): frequency modulation slopes, PicF (Hz): frequency at the maximum power amplitude of the spectrum frequency bandpass (Bandpass) are higher pitched in Spectacled caiman calls than in Black caimans ones. The frequency modulation slopes (Slope 1 and 2) also differentiate Spectacled caiman contact calls from the two others, the frequency modulation of the calls from this species being more pronounced (Table 2). In summary, though the basic acoustic structure of crocodilian calls (a complex sound with multiple harmonic, modulated in frequency and amplitude) appears similar between different species, differences exist and are substantial enough to assign each call in its proper category with confidence. A cross-validated discriminant function analysis correctly classified 96 % of the Nile crocodile contact calls, 88 % of the Black Caiman contact calls and 80 % of the Spectacled caiman calls (Fig. 3). Behavioural response to contact calls The Friedman analysis found significant differences between Nile crocodiles responses to the broadcast signals (P \ 0.02, Fig. 4): four individuals out of the five tested did not react with the noise stimulus; conversely, the three types of crocodilians calls elicited strong behavioural responses from all juvenile Nile crocodiles. There was no significant difference between responses to Nile crocodile, Spectacled caiman and Black caiman calls. Because acoustic analysis had shown significant structural differences between the calls of the three species, one hypothesis that could explain why Nile crocodiles individuals respond similarly to calls of their own species and to heterospecific calls is that all these signals encode the same information. Investigation of the acoustic parameters that are responsible for the crocodiles behavioural responses may shed some light on this matter. Experiment 2: identification of calls salient acoustic parameters Methods From playback experiments 1, it appeared that Nile crocodile juveniles react similarly to calls of their own species and heterospecific calls. However, they differentiate crocodilian calls from noise, meaning that calls convey some crocodilian information. In order to identify the relevant parameters of calls (in the frequency and/or temporal domains), we ran playback experiments using acoustic lures (modified signals). Each acoustic parameter was tested independently. The original sounds used to prepare experimental signals were selected from the bank of recorded contact calls previously described. Adult crocodiles (especially females) are also known to react to juvenile calls (Vergne and Mathevon 2008). Thus, we decided to also test adult crocodilians in order to see whether the coding rules might change with age. Experiments on juveniles Twenty-two naïve individuals (6 Spectacled caimans 2 months old, 16 Nile crocodiles 1 month old, sex unknown) were available for our experiments. To avoid habituation due to multiple testing, we performed playback experiments on two different sets of individuals: The first experimental set was composed of 6 Spectacled caimans and 6 Nile crocodiles. We tested them individually with a series of 7 experimental signals (Fig. 5). Each individual was tested with natural and modified signals from its own species. To avoid pseudo-replication, we used a different series of experimental signals for each individual each series being built from a different natural contact call (NAT). Spectacled caiman natural calls came from 2- to 3-week-old individuals while Nile crocodile natural calls came from 4- to 6-days old individuals (see details in Experiment 1 above). We worked with synthetic copies of calls so that we could modify all parameters of the signal. Modified versions of natural contact calls were created with Avisoft-SAS Lab-Pro software. The six following signals were built from a NAT signal (Fig. 5): (1) a synthetic copy of the natural call (Scontrol; control signal), (2) a signal without amplitude modulation (noam; the frequency modulation of the NAT was retained), (3) a signal without frequency modulation (FM1; the amplitude modulation as well as the energy distribution between the different harmonics was identical to the NAT), (4) a signal without any harmonic structure (1H, only the first

6 1100 Anim Cogn (2012) 15: Table 2 Results of Fisher LSD tests to identify statistically homogeneous groups of calls on the basis of 7 measured acoustic parameters Acoustic parameter F0max (Hz) Bandpass (Hz) Slope 1 (Hz/s) Slope 2 (Hz/s) DT (s) Fend (Hz) PicF (Hz) Homogeneous groups (Nile) (Black) (Spec) (Nile) (Black) (Spec) (Nile, Black) (Spec) (Nile, Black) (Spec) (Nile) (Black, Spec) (Nile) (Black, Spec) (Nile) (Black, Spec) Fig. 4 Behavioural reaction of Nile crocodile juveniles to contact calls from different species and to white noise (each of the 5 tested individuals is represented by a different colour) (colour figure online) Nile : Nile crocodile calls; Black : black caimans calls; Spec : spectacled caiman calls. Calls within the same brackets cannot be distinguished on the basis of the parameter considered Fig. 3 Classification of crocodilian calls by discriminant function analysis. The DFA correctly classified 96 % of the Nile crocodile contact calls, 88 % of the Black Caiman contact calls and 80 % of the spectacled caiman calls harmonic, which is also the one of highest energy, was retained), (5) a temporally reversed natural call (Srev; a temporally mirror of the frequency modulation slope of the original call), and (6) a temporally reversed version of 1H (1Hrev). The second experimental set was composed of 10 naïve Nile crocodiles (2 weeks old). Besides the control NAT signal, they were challenged with 6 synthetic signals (Fig. 5): (1) a synthetic copy of the natural call (synthetic control signal, Scontrol); (2) a signal with a frequency modulation slope reduced by one-third compared to the natural slope (SLOPE1): in the Nile crocodile SLO- PE1 = 1,017 ± 384 Hz s -1, in the Black caiman: SLO- PE1 = 1,182 ± 417 Hz s -1 and in the Spectacled caiman: SLOPE1 = 1,713 ± 557 Hz s -1 ; (3) a signal with a frequency modulation slope reduced by a half compared to the natural slope (SLOPE2): in the Nile crocodile SLO- PE2 = 763 ± 288 Hz s -1, in the Black caiman: SLOPE2 = 887 ± 313 Hz s -1 and in the Spectacled caiman: SLOPE2 = 1,284 ± 418 Hz s -1 ; (4) a signal without any frequency modulation or harmonics series (FM2); (5) a signal with a modified energy distribution across the spectrum (ENER, with 80 % of energy between 2,500 and 5,000 Hz instead of between 200 and 2,500 Hz as in the control signal); (6) a noise having the same duration, frequency bandpass and amplitude modulation as the control signal (NOISE). The first four and the NOISE experimental signals were built using the Avisoft-SAS Lab-Pro software. The ENER signal was built using PRAAT software. The experimental procedures and the assessment of behavioural responses were identical to those described in the experiments testing the species specificity of calls (Experiment 1; air and water temperatures were and 30 C, respectively). All statistical tests were conducted using Statistica package. We performed a Friedman twoway repeated measures analysis of variance by ranks to detect differences in crocodilian responses between the different playback tests, and Wilcoxon two-tailed tests for comparisons of behavioural responses to the Natural call with responses to other signals tested. Experiments on adults Adult experiments were performed at the zoo La Ferme aux Crocodiles (hosting 350 adult Nile crocodiles in a 8,000 m 2 tropical greenhouse; 80 % of the crocodiles are mature females of 3 8 years old; the air and water temperatures are and 30 C, respectively). Although it was possible to approach animals conveniently, it was impossible to test them individually. We thus played back experimental signals to clusters of 7 24 adults and monitored the response of all individuals together. Five clusters were carefully chosen within the greenhouse to avoid possible interferences between them: they were located as far away as possible from each other (more than 30 m apart). To limit the beam size of propagated sounds, we used a directional loudspeaker (Audax) and only the tested cluster was situated in front of the loudspeaker (the closest

7 Anim Cogn (2012) 15: Fig. 5 Experimental signals used for playback experiments (NAT: natural contact call; Scontrol: synthetic copy of the natural call = control signal; noam: signal without amplitude modulation; FM1: signal without frequency modulation; 1H: signal without any harmonic structure; Srev: temporally reversed natural call; 1Hrev: temporally reversed 1H signal; SLOPE1: signal with a frequency modulation slope one-third reduced compared to the natural slope; SLOPE2: signal with a frequency modulation slope half reduced compared to the natural slope; FM2: signal without any frequency modulation nor harmonics series; ENER: signal with a modified energy distribution among the spectrum; NOISE: noise having the same duration, frequency bandpass and amplitude modulation as the control signal) individual of the cluster was at least 10 m from the loudspeaker). The amplitude of emitted sounds was of 61 ± 4dB SPL at 1 m from the loudspeaker. The maximum range of propagated signals from the loudspeaker was approximately 20 m (at this distance, the intensity level was below the background noise). Given the minimum distance between clusters (30 m), we assumed that individuals within a cluster were not able to hear the signals emitted during another cluster s test. The experiments took place in April, a month when most adults are sexually active. Females had already started to lay their eggs; males were defending their harems and thus tended to stay around the same area in the greenhouse (±10 m). In addition, all the experiments only lasted 2 days, so it is likely that the composition of clusters remained stable during this time and that no animal was tested twice. To avoid habituation, each cluster was challenged once, with only one experimental signal. Due to the limited number of clusters that could be tested independently (N = 5), we thus only used 5 experimental signals chosen from those used for Nile crocodile juvenile tests: (1) a natural contact call (control signal, NAT, number of adult crocodiles within the tested cluster n = 24), (2) a signal without amplitude modulation (noam, n = 7), (3) a signal without any harmonic structure (1H, n = 20), (4) a signal without frequency modulation (FM1, n = 7), (5) a signal without any frequency modulation nor harmonic series (FM2, n = 20). Behavioural responses were assessed according to a 4-level scale: 0 was scored if the playback did not provoke any observable response during the 90 s of playback, 1 if there was a head and/or body orientation towards the loudspeaker but no displacement, 2 if the individual moved towards the loudspeaker but stayed at more than 5 m from it, and 3 if it came at less than 5 m from the loudspeaker. The behaviour of each animal within a cluster was assessed independently, allowing calculation of the proportion of females with scores of 0, 1, 2 and 3 (expressed in % to the total number of observed females) for each tested cluster and thus each experimental signal. Due to the limited number of tests, no statistical tests were performed on these adult data and only raw results are presented. Results The frequency modulation is a key parameter for both juveniles and adults Playback experiments run on juvenile Spectacled caimans and Nile crocodiles showed that the shape of the frequency modulation seems to be a biologically relevant call parameter: the responses to FM1, 1Hrev and Srev were significantly weaker than the responses to the synthetic copy of the natural call (Scontrol) for both species (Fig. 6a; Wilcoxon tests, in Spectacled caimans, N = 6, T = 0.0, P = 0.028; N = 6, T = 1, P = 0.046; N = 6, T = 0.0,

8 1102 Anim Cogn (2012) 15: P = 0.028, respectively; in Nile crocodiles, N = 6, T = 0.00, P = for the three signals). This observation was verified by the second run of experiments on juvenile Nile crocodiles alone: individuals responded significantly less to FM2 compared to the Scontrol signal (Wilcoxon test, N = 10, T = 0.0, P = 0.005; Fig. 6b). In addition, an interesting point is that, while the juveniles continued to respond to a signal whose frequency modulation slope was slightly modified (SLOPE1, Wilcoxon test: N = 10, T = 4, P = 0.091), a stronger modification significantly reduced their responses (SLOPE2, Wilcoxon test: N = 10, T = 0.0, P = 0.008; Fig. 6b). Figure 7 illustrates the proportion of juveniles and adults who expressed a strong behavioural response to signals (score C4 for juveniles, C3 for adults, see Methods ). Only signals with no or weak modifications of the frequency modulation elicited such strong responses. Conversely, FM1 and FM2 (no frequency modulation), Srev and 1Hrev (temporally reversed frequency modulation), as well as SLOPE2 (half reduced frequency modulation slope) rarely elicited strong responses. These results emphasize the importance of frequency modulation as an acoustic parameter for the crocodilian information in both juveniles and adults. Conversely, our experiments demonstrated that amplitude modulation is not a determinant parameter: more than 80 % of both juveniles and adults expressed strong responses to the noam signal (no significant difference between juveniles responses to noam compared to Scontrol: in juvenile Spectacled caimans, N = 6, T = 5, P = 1; in juvenile Nile crocodiles, N = 6, T = 0.0, P = 0.109; Figs. 6, 7). Moreover, the presence of harmonics did not seem necessary: all tested adults strongly responded to 1H and more than 70 % of the juveniles also did (Fig. 7; no significant difference from Scontrol, Wilcoxon tests, in Spectacled caimans, N = 6, T = 8, P = 0.60; in Nile crocodiles, N = 6, T = 2, P = 0.075, Fig. 6a). We also did not detect any differences in juveniles responses when the energy of the calls was moved towards high frequencies (responses to ENER were not significantly different to Scontrol responses, Wilcoxon test: N = 10, T = 1, P = 0.285; Figs. 6b, 7). Thus, our experiments showed that the presence of the frequency modulation of the call is necessary to elicit a behavioural response in crocodilians from the youngest to adult age. However, amplitude modulation, the presence of harmonics and the energy distribution across the frequency spectrum do not seem to be key parameters. Experiment 3: test of juveniles discrimination abilities using conditioning experiments Methods From the playback experiments, it appears that signals that differ in their acoustic structure elicit a similar behavioural response. It is worth asking whether that could arise from a lack of discriminatory abilities (i.e. the tested animals are not able to differentiate between the signals) or whether it Fig. 6 Juveniles responses to playback of natural and modified calls. a Tests performed on 6 Nile crocodile juveniles and 6 Spectacled caiman juveniles. b Tests performed on 10 naive Nile crocodile juveniles. The stimuli eliciting the weakest behavioural responses are those with no frequency modulation (NOISE) or a modified one (1Hrev, FM1, Srev, SLOPE2, FM2). Black stars highlight behavioural responses that are significantly different from the control ones (i.e. Scontrol signal)

9 Anim Cogn (2012) 15: Fig. 7 Proportion of responding individuals to experimental signals (score C 4 for juveniles, C3 for adults). The behavioural scores were established according to a scale (see text). Only few individuals showed a strong behavioural response to the signals with no or disrupted frequency modulation (Srev, 1Hrev, Noise, SLOPE2, FM1 and FM2). n = total number of tested individuals (Nil = juvenile Nile crocodiles; Spec = juvenile Spectacled caimans; Ad = adult Nile crocodile) could be the result of a true behavioural choice (i.e. the tested animals perceive the difference between calls of different acoustic structures but do not show any differential responses to them). This situation (similar responses to structurally different calls) occurred during our playback experiments on juvenile Nile Crocodiles when, for instance, we compared the behavioural responses to the control synthetic Scontrol signals and to the ENER signals (Scontrol signals = synthetic copies of natural calls; ENER = signals with a modified energy distribution among the spectrum, that is, 80 % of energy between 2,500 and 5,000 Hz instead of between 200 and 2,500 Hz in the Scontrol signals). With the setup of experiment 2, behavioural responses elicited by Scontrol and ENER signals were not different (see Results of experiment 2 above); however, this result does not prove that the tested animals are not able to distinguish between both the two. It is possible that both signals have similar biological relevance. Conditioning experiments are a way to test whether this undifferentiated response is linked to a something other than discriminatory ability. Although conditioning experiments have already been used in several situations with crocodiles (Davidson 1966; Williams 1968; Burghardt 1977, to our best knowledge, this method had never been used on crocodiles to test their discrimination abilities with acoustic signals. To test whether young crocodiles can learn to respond to a sound stimulus, we first started with a simple GO/NOGO task. Second, we used a GO/NOGO procedure to test the crocodile s ability to discriminate between Scontrol and ENER signals. Animals and experimental conditions Three naïve young Nile crocodiles (1 month old) participated in the experiments. The home tank in which the individuals were kept had the same characteristic as in the previous experiments. Before the beginning of conditioning experiments, we stopped feeding the young crocodiles in their home tank. Feeding was used as a reinforcer during the conditioning tasks and thus occurred only during experimental periods when the individual was isolated from the two others in the test tank. As the crocodiles were used to being fed once every 3 days before the experiments, we maintained this feeding rhythm and tested each of the three crocodiles every 3 days. All experiments occurred during the night between 8 p.m. and 5 a.m. (water temperature: 30 C; air temperature: C). All trials were video-recorded. The tested individual was isolated in the test tank the morning before the test. We used 4 speaking feeders (CAT Gato) located in the four corners of the test tank (Fig. 8a). The feeders were put on bricks at about 5 cm above the water level, not only to protect them from getting wet but also so that it would require a physical effort for the crocodile to access them and check for the presence of food. Each speaking feeder was composed of four compartments of exactly the same size (Fig. 8b). The opening of each compartment was electronically programmable. It took 15 s for each compartment to open completely, and during that time, a soft and regular mechanical noise could be heard. Once the compartment was open, we could choose to programme the feeder to play 20 s of a sound sequence or to remain silent. Each feeder possessed an integrated microphone and a loudspeaker (Fig. 8b). Experimental signals and conditioning protocol The first conditioning protocol (pre-training) aimed at testing whether it was possible to condition a crocodile to respond to a feeder opening associated with a sound

10 1104 Anim Cogn (2012) 15: stimulus. On the morning before any nocturnal test and before we brought the crocodile in the experimental room, the test tank was organized with two food feeders filled with food (the four compartments of each feeder were each filled with a piece of food, for example, fish, frog s legs, shrimps, chicken, meat ) and programmed to play 20 s of a pre-recorded sound sequence (3 repetitions of the same series; one series = 7 successive Scontrol signals from different individuals; interval between Scontrol signals = 1 s; interval between series = 1.5 s). The prerecorded sound sequences differed between the 2 food feeders (different Scontrol signals), and we also made sure not to use the same sequences from one night to another for the same individual. The two other no-food feeders were soaked with the smell of food (the same as the one placed in the other feeders) but contained no food, and their compartment opening was not programmed to be followed by sounds. The position of the 4 feeders in each corner of the tank was chosen at random and changed for every test. We programmed the opening times in alternation between the feeders so that every 30 min one opening occurred from a given feeder. The order of openings between feeders was randomly determined and changed for every test. The experiment was run during 45 consecutive nights. Each of the three individuals was tested every three nights (16 openings per night 9 15 nights/individual = 240 learning trials/individual). The second conditioning protocol (training) involved the same three pre-trained crocodiles. There was no rest night between the two protocols. The presence of food was still associated with the Scontrol signal. However, we replaced the silent signal associated with no food by a sound sequence built with the ENER signal previously used during the coding-decoding tests. The pre-recorded sound sequences were different for each feeder (new Scontrol and ENER signals), and we also made sure not to use the same sequences from one night to another one on the same individual. As in the first experiment, the two no-food feeders were soaked with the smell of the same kinds of food as those placed in the other feeders, and the position of the 4 feeders in each corner of the tank was chosen at random and changed every day. Measurements and analysis of behavioural responses As for experiments 1 and 2, all conditioning tests were controlled by a computer located outside the experimental room. The audio and video recordings were set up to start 5 min before each feeder s opening and to stay on for a total of 15 min. We quantified the responses of the crocodiles using the video recordings by blindly assessing the following 3 parameters (blind evaluation was achieved by analyzing the behavioural responses without knowing what was the sound stimuli that triggered them): (1) first, reaction (head or body movement) of the tested individual following the beginning of the feeder s opening. We expressed the number of positive responses (i.e. the event crocodile reaction occurred) as a percentage of the total number of tests per category (i.e. no-food feeder or food feeder). Every night, a crocodile was exposed to 16 feeder s openings per category. In addition, we measured the reaction latency T0 (i.e. the time in seconds between the beginning of a feeder opening and the animal s first reaction). A 3-level scale was set up with a score of 3 assigned when T0 B 10 s, 2 when 30 s B T0 B 11 s, 1 when 60 s B T0 B 31 s or 0 when T0 [ 60 s (or no response). (2) approach: the crocodile displacements towards the feeder. A displacement was considered as an approach if the crocodile came within 5 cm of the active feeder. As for the reaction parameter, we expressed the number of approaches as a percentage of the total number of tests per feeder s category. We also measured the time to approach the feeder T1 (i.e. the time in seconds from the first reaction to come within the 5 cm around a feeder). We used the same scale as for the reaction parameter with 3 if T0 B 10 s, 2 if 30 s B T0 B 11 s, 1 if 60 s B T0 B 31 s, 0 if T0[ 60 s. (3) climbing behaviour: once the crocodile had reached the feeder, it had to decide whether or not to Fig. 8 Test tank and speaking feeders used during conditioning experiments. a Position of the 4 speaking feeders at each corner of the test tank. b Detail of a speaking feeder. It is composed of 4 compartments whose openings are programmable. The integrated microphone was used to record a 20 s sound sequence which could be played via the integrated loudspeaker

11 Anim Cogn (2012) 15: climb on the feeder s platform to look for the presence of food. Note that every time a crocodile climbed onto a feeder, it ate the food if it was available. We expressed the number of climbing behaviours as a percentage of the total number of tests per feeder category. We also measured the climbing time T3 (i.e. the time in seconds from the end of the approach to the feeder to come up onto the feeder to check for the presence of food). The same scale as for the two previous parameters was used. For data analysis and statistical tests, we grouped each three consecutive days of test (i.e. from day 1 to day 3 together, day 4 to day 6 until day 13 to day 15). For each of these groups, we used Wilcoxon tests to compare crocodiles responses to food feeders (associated to the Scontrol signal) with their responses to no-food feeders (associated with Silence in the pre-training protocol or ENER signals in the training protocol). Results Discrimination between calls with different energy distribution over the frequency spectrum The pre-training tests showed that it was possible to teach the crocodiles to respond to a sound stimulation versus silence. Results of the pre-training conditioning protocol are shown Fig. 9. For the reaction parameter (Fig. 9a), statistical tests show significant differences between crocodiles reaction to Scontrol feeders compared to Silent feeders from the first day of test until the last: crocodiles reactions were stronger in response to Scontrol feeders. However, during the whole experiment, the tested individuals were highly motivated and we have observed a response in more than 80 % of all tests for both silent and Scontrol feeders triggers. Times to react did not improve with time or depend on the type of feeder (graphs not shown). Conversely, the approach parameter (Fig. 9b) shows that the crocodiles quickly learnt to associate the presence of food with the Scontrol signal. This learning was apparent in a significant decrease in approach behaviours towards the silent feeders. Until day 9, we still observed a high percentage of errors (crocodiles approached both types of feeders in more than 80 % of the time: no significant differences between responses to Scontrol compared to Silence, D7 D9: P = 0.14, z = 2.38). However, from day 10, the number of approaches in response to silent feeders decreased significantly (from 68 % of approaches from day 10 to 12 to 55 % during the last 3 days of tests) while responses to Scontrol feeders remain constant and close to 100 %. The speed of approaching the active feeder did not change with time or depend on the type of feeder. The climbing parameter was even more discriminatory. Crocodiles climbed significantly less on silent feeders compared to Scontrol feeders (Fig. 9c). Indeed, while crocodiles climbed on Scontrol feeders more than 90 % of the time, we observed a decrease from above Fig. 9 Results of the pre-training tests: crocodiles learnt to associate the presence of food with the Scontrol signal compared to silence. a Reaction parameter: crocodiles answered during all the tests period ([80 % of reaction for both conditions) but analyses showed a significantly stronger response to the Scontrol feeders compared to the Silence feeders from the first days of experiment. b Approach parameter: from day 10 the number of approaches in response to Silence feeders decreased significantly (from 83 % of approaches at day 7 9 to 55 % at day of tests) while the number of approaches to Scontrol feeders remained constant and close to 100 %. c Climbing parameter: crocodiles climbed significantly less on Silence feeders compared to Scontrol feeders and got better with time (while crocodiles climbed on Scontrol feeders in more than 90 % of the time, we observed a decrease from more than 60 % of climbing behaviours at days 1 3 to less than 20 % at days of tests in response to Silence feeders). Results for Wilcoxon tests are presented on each chart, *P \ 0.05, **P \ 0.02, ***P \ 0.01

12 1106 Anim Cogn (2012) 15: % of climbing behaviours the first 3 days to below 20 % the last 3 days of tests in response to silent feeders. Also, times to climb did not change with time or depend on the type of feeder. It is thus likely that crocodiles learnt quickly to associate the presence of food with the Scontrol sound sequence. Second and interestingly, the training protocol showed that crocodiles are able to discriminate between Scontrol and ENER signals (Fig. 10). For the reaction parameter, there is no significant difference between crocodiles responses to Scontrol feeders compared to ENER feeders from the beginning to the end of our conditioning tests (Fig. 10a). Furthermore, reaction times comparisons still did not bring any information about the learning process. The reaction parameter was thus not the most sensitive for detecting learning. The crocodiles also seemed to approach Scontrol feeders as much as ENER feeders (no significant differences, from the beginning to the end; Fig. 10b). Thus, the approach parameter did not help to demonstrate any associative learning nor did the time to make these displacements towards the two types of feeders. However, the climbing parameter did show that crocodiles learnt to differentiate between Scontrol and ENER signals (Fig. 10c). From day 7, the number of climbing behaviours in response to ENER feeders decreased significantly compared to those in response to Scontrol feeders (between 70 and 92 % of climbing behaviours in response to Scontrol feeders during the 15 days versus a decrease from 85 % the first 3 days to about 50 % on the last 3 days of tests in response to ENER feeders). Hence, crocodiles learnt that the presence of food was associated with Scontrol and not ENER and are thus able to differentiate between these acoustic signals. From a methodological point of view, note that playing the sound after the feeder was opened gave us another clue as to whether the crocodiles were able to learn that the discrimination was based on sounds. The opening of the feeder was relatively silent but crocodiles did react to it. However, after a while we could see that the crocodiles merely reacted to the sound and stayed motionless during the feeder operation. This process confirmed that crocodiles were attending to the broadcast sounds and not to the opening of the feeders. General discussion The aim of this study was to investigate the acoustic structure and information coding in the contact calls of juvenile crocodilians. On the basis of acoustic analyses and playback experiments, the following points have been demonstrated: First, in spite of common acoustic structure between the contact calls of juvenile Nile crocodiles, Spectacled caimans and Black caimans, most of the measured acoustic parameters showed significant inter-specific differences and it is straightforward to classify calls according to their species origin using a multivariate analysis. Although calls of all crocodilian species share the same overall acoustic features, previous studies had already shown inter-specific Fig. 10 Results of the training test: crocodiles were able to learn discriminating between Scontrol and ENER signals. a Reaction parameter: crocodiles did not express any significant differences in their responses to Scontrol versus ENER feeders. They reacted to both feeders in more than 80 % of the tests. b Approach parameter: they approached both feeders and did not seem to learn with time not to approach ENER feeders. c Climbing parameter: conversely, from day 7, crocodiles started climbing significantly less on ENER feeders compared to Scontrol ones (between 70 and 92 % of climbing behaviours in response to Scontrol feeders during the 15 days versus a decrease from 85 % at days 1 3 to about 50 % at days in response to ENER feeders). Results for Wilcoxon tests are presented on each chart, *P \ 0.05, **P \ 0.02, ***P \ 0.01

13 Anim Cogn (2012) 15: variations (reviewed in Vergne et al. 2009). However, these inter-specific differences may be biased in the present study due to the relative age heterogeneity of our animals: we showed in a previous study with newborn Nile crocodiles distress calls that acoustic variables related to call duration and to the fundamental frequency vary with the individual s age (Vergne et al. 2007). Specifically, the youngest/smallest individuals produce the highest-pitched calls. This may explain why the contact calls of the Nile crocodile we studied possessed more energy towards high frequencies and a broader frequency bandwidth than those of the spectacled or black caimans. Nevertheless, the slope of the frequency modulation has been found to remain stable with the age in Nile crocodile s distress calls (Vergne et al. 2007), and in the present study, this parameter was one of those that distinguished most strongly between the contact calls of the different species. Hence, the variability in age between the animals recorded is unlikely to explain all the observed differences and it is not unreasonable to assume that inter-specific differences in the fine structure of juvenile calls do exist. Second, playback experiments showed that these interspecific structural differences between calls seemed not to be relevant to the animals. We observed no significant behavioural differences in juvenile Nile crocodile s responses to calls of their own species compared to calls of other species. We have to be cautious with this result because of the small sample size and unknown power of the experimental system: in a more subtle paradigm and a fortiori in the wild, discrimination between species calls remains conceivable. Nevertheless, this result supports a previously established hypothesis about a possible interspecific repertoire of crocodilian calls with behavioural responses not restricted to the conspecific calls (Britton 2001; Campbell 1973) and is in accordance with several preliminary experiments we have made in the wild. That gives us several reasons to think that species-specific recognition based on juvenile calls is extremely weak or nonexistent in crocodiles. The species specificity of juvenile calls may be irrelevant information for these animals. One main point is that effective sympatry is rare among crocodiles. The Nile crocodile lives in Africa while Black and Spectacled caimans are American species. This geographical distribution might explain why the Nile crocodiles have not developed the ability to discriminate their calls from those of other species. The Black and the Spectacled caimans are sympatric but do not usually meet in the field, their preferential habitat being slightly different (and observations have been made of Black caimans chasing the other species, P. Taylor pers.com.). Based on our own observations in the field, it is likely that acoustic exchanges between juveniles and between adults and juveniles occur only within family groups (i.e. a female and her young; for example, Black caiman family clusters stay away from other individuals, Vergne et al. 2011). Third, our study shed light on the key acoustic parameters responsible for the biological relevance of the juveniles contact calls and yielded information on the process of coding and decoding crocodilian information by juveniles and adults. The importance of the slope of the frequency modulation and the tolerance of tested individuals to slight modifications of this parameter is in accordance with the inter-specific responses we observed during Experiments 1, as FM slopes differed slightly between species. An information encoding process using frequency modulation is widespread in animals using acoustic to communicate (Becker 1982). A coding based on a slow frequency modulation has the advantage of being robust, particularly in the face of propagation, because modulation characteristics are only slightly damaged during transmission over long ranges (Wiley and Richards 1982). Also, as the crocodilian contact call s frequency modulation extends over a large frequency band, its characteristics are useful for improving the localization of the sound source (Aubin and Jouventin 2002b). In addition, playback experiments using modified signals showed that amplitude modulation does not play a major role in inducing a behavioural response. It is known that this parameter is quickly modified during signal transmission throughout the environment (Aubin et al. 2000; Mathevon and Dabelsteen 2002; Wiley and Richards 1982). Previous propagation experiments using crocodilian calls have confirmed this result (Vergne et al. pers. obs.). In birds (Jouventin et al. 1999) and mammals (Charrier et al. 2002), amplitude modulation is never a parameter encoding specific or individual identity. Nevertheless, this parameter can play a crucial role during sound localization as has been demonstrated for instance in the barn owl Tyto alba (Konishi 1973; Shalter and Schleidt 1977) and in the King Penguin Aptenodytes patagonicus (Aubin and Jouventin 2002b). Further experiments would be necessary to determine whether this is also the case in crocodilians. Playback experiments also showed that the entire frequency spectrum is not necessary to induce a behavioural response. One unique harmonic signal seems to be enough to maintain the biological effectiveness of the signal. Experiments with birds and mammals have shown that tolerance towards such a modification is extremely variable from one species to another. For instance in Adelie Penguin Pygoscelis adeliae, parent-offspring recognition needs the entire frequency spectrum while the same type of recognition in Subantarctic Fur Seal Arctocephalus tropicalis or in King Penguins is effective with a reduced number of harmonics (Aubin and Jouventin 2002a; Charrier et al. 2002; Searby et al. 2004). Young crocodilians seem to be quite tolerant towards modifications of this parameter. Just