the irrelevance of individual discrimination in meerkat alarm calls

10
The irrelevance of individual discrimination in meerkat alarm calls FABIAN SCHIBLER & MARTA B. MANSER Verhaltensbiologie, Zoologisches Institut, Universita ¨t Zu ¨rich (Received 22 July 2006; initial acceptance 19 September 2006; final acceptance 23 February 2007; published online 1 November 2007; MS. number: 9054R) Individual discrimination is an important element in the evolution of social behaviour and is particularly important in social living species which show intense intragroup interactions. Numerous previous studies, particularly with nonhuman primates, ground squirrels and marmots, demonstrate the widespread ability of various species to signal and perceive individual identity from vocalizations. The function of individu- ally different alarm calls is thought to assist in the detection of unreliable individuals. This would allow individuals to optimize the benefits of antipredator behaviour by self-assessing the relative predation risk, and responding selectively to reliable callers. In this study we investigated whether meerkats, Suricata suricatta, a social mongoose, discriminated among alarm callers individually, and adjusted their response accordingly. Several parameters of the acoustic structure of meerkat alarm calls were measured and ana- lysed with multivariate statistics. Within groups, the 10-fold cross-validation of a multinomial regression yielded, on average, 90% correct assignment. This strongly suggests that meerkats have individually dis- tinct alarm calls. With a habituationedishabituation playback experiment, we then tested whether meer- kats use this information to discriminate between individual callers. Receivers did not distinguish between different individuals in the playback experiments. In meerkats, unreliable callers appear to be uncommon, and the cost of being predated might exceed the costs of responding to an unreliable caller, thus rendering a reliability-based discrimination mechanism unnecessary. Although meerkat alarm calls contain informa- tion on individual identity, this information does not appear to be important to the receivers in this context. Ó 2007 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved. Keywords: alarm call; acoustic structure; habituationedishabituation playback; individual discrimination; meerkat; reliability; Suricata suricatta The signalling and perception of identity are of major importance in the evolution of social behaviour (Hamil- ton 1963; Trivers 1971; Wilson 1979) and may be the basic requirement for complex mechanisms of social communi- cation (Beecher 1982; Cheney & Seyfarth 1990). Hence, we might expect to find such recognition abilities espe- cially in systems where coalitions or subgroups are present and where the interactions within and between such social categories are essential. Above all, nonhuman pri- mates, with their complex social behaviour, have received most attention in regard to individual recognition mecha- nisms. Many of the calls emitted in various social contexts show acoustic properties that differ between individuals (Hauser 1991), and based on these individual differences, many nonhuman primates are able to differentiate be- tween callers (Snowdon & Cleveland 1980; Rendall et al. 1996; Semple 2001; Weiss et al. 2001; Ceugniet & Izumi 2004). The function of individual differences in alarm calls is less clear. Cheney & Seyfarth (1988) proposed the concept of signal reliability to explain such differences. They rea- soned that in systems where individuals give deceptive alarm calls, ‘selection should favour the ability of recipi- ents to compare signals on the basis of their meaning and to transfer information about the reliability of a sig- naller’s calls from one context to another’. The same selec- tive pressure should apply in systems where variable thresholds for emitting such calls occur, that is high and low false alarm rates (Blumstein et al. 2004). In both cases, Correspondence: M. B. Manser, Verhaltensbiologie, Zoologisches Institut, Universita ¨t Zu ¨rich, Winterthurerstrasse 190, 8057 Zu ¨rich, Switzerland (email: [email protected]). 1259 0003e 3472/07/$30.00/0 Ó 2007 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved. ANIMAL BEHAVIOUR, 2007, 74, 1259e1268 doi:10.1016/j.anbehav.2007.02.026

Upload: fabian-schibler

Post on 04-Sep-2016

215 views

Category:

Documents


3 download

TRANSCRIPT

The irrelevance of individual discrimination in meerkat alarm calls

FABIAN SCHIBLER & MARTA B. MANSER

Verhaltensbiologie, Zoologisches Institut, Universitat Zurich

(Received 22 July 2006; initial acceptance 19 September 2006;

final acceptance 23 February 2007; published online 1 November 2007; MS. number: 9054R)

Individual discrimination is an important element in the evolution of social behaviour and is particularlyimportant in social living species which show intense intragroup interactions. Numerous previous studies,particularly with nonhuman primates, ground squirrels and marmots, demonstrate the widespread abilityof various species to signal and perceive individual identity from vocalizations. The function of individu-ally different alarm calls is thought to assist in the detection of unreliable individuals. This would allowindividuals to optimize the benefits of antipredator behaviour by self-assessing the relative predationrisk, and responding selectively to reliable callers. In this study we investigated whether meerkats, Suricatasuricatta, a social mongoose, discriminated among alarm callers individually, and adjusted their responseaccordingly. Several parameters of the acoustic structure of meerkat alarm calls were measured and ana-lysed with multivariate statistics. Within groups, the 10-fold cross-validation of a multinomial regressionyielded, on average, 90% correct assignment. This strongly suggests that meerkats have individually dis-tinct alarm calls. With a habituationedishabituation playback experiment, we then tested whether meer-kats use this information to discriminate between individual callers. Receivers did not distinguish betweendifferent individuals in the playback experiments. In meerkats, unreliable callers appear to be uncommon,and the cost of being predated might exceed the costs of responding to an unreliable caller, thus renderinga reliability-based discrimination mechanism unnecessary. Although meerkat alarm calls contain informa-tion on individual identity, this information does not appear to be important to the receivers in thiscontext.

� 2007 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

Keywords: alarm call; acoustic structure; habituationedishabituation playback; individual discrimination; meerkat;reliability; Suricata suricatta

ANIMAL BEHAVIOUR, 2007, 74, 1259e1268doi:10.1016/j.anbehav.2007.02.026

The signalling and perception of identity are of majorimportance in the evolution of social behaviour (Hamil-ton 1963; Trivers 1971; Wilson 1979) and may be the basicrequirement for complex mechanisms of social communi-cation (Beecher 1982; Cheney & Seyfarth 1990). Hence,we might expect to find such recognition abilities espe-cially in systems where coalitions or subgroups are presentand where the interactions within and between suchsocial categories are essential. Above all, nonhuman pri-mates, with their complex social behaviour, have receivedmost attention in regard to individual recognition mecha-nisms. Many of the calls emitted in various social contexts

Correspondence: M. B. Manser, Verhaltensbiologie, ZoologischesInstitut, Universitat Zurich, Winterthurerstrasse 190, 8057 Zurich,Switzerland (email: [email protected]).

120003e3472/07/$30.00/0 � 2007 The Association for the

show acoustic properties that differ between individuals(Hauser 1991), and based on these individual differences,many nonhuman primates are able to differentiate be-tween callers (Snowdon & Cleveland 1980; Rendall et al.1996; Semple 2001; Weiss et al. 2001; Ceugniet & Izumi2004).

The function of individual differences in alarm calls isless clear. Cheney & Seyfarth (1988) proposed the conceptof signal reliability to explain such differences. They rea-soned that in systems where individuals give deceptivealarm calls, ‘selection should favour the ability of recipi-ents to compare signals on the basis of their meaningand to transfer information about the reliability of a sig-naller’s calls from one context to another’. The same selec-tive pressure should apply in systems where variablethresholds for emitting such calls occur, that is high andlow false alarm rates (Blumstein et al. 2004). In both cases,

59Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

ANIMAL BEHAVIOUR, 74, 51260

individuals experience a severe fitness cost when respond-ing to unreliable callers, compared with a situation wherethey only respond to reliable callers. Therefore, animalsthat are able to recognize unreliable individuals on the ba-sis of their alarm calls may benefit by optimizing their an-tipredator behaviour and so allocate more time to otherimportant activities. In yellow-bellied marmots, Marmotaflaviventris, this leads to an increase in vigilance behaviourafter being exposed to artificially created unreliable callers(Blumstein et al. 2004); that is, individuals respond to re-liable callers, but make their own independent assessmentof relative risk after hearing alarm calls by unreliable cal-lers, thereby optimizing their foraging efficiency. Addi-tionally, both Richardson’s ground squirrels, Spermophilusrichardsonii, which are thought to benefit socially throughreciprocal altruism (Hare & Atkins 2001), and vervet mon-keys, Cercopithecus aethiops (Cheney & Seyfarth 1988)show an ability to discriminate between unreliable and re-liable callers based on an association with their previousantipredator behaviour.

In this study, we investigated whether consistent differ-ences exist between the alarm calls of individual meerkats,Suricata suricatta, and if so whether they were important indetermining an individual’s response to them. Meerkatsare highly cooperative small mammals, which inhabitthe drier, open regions of southern Africa (Estes 1991),and live in groups up to 50 individuals (Clutton-Brocket al. 2005). They typically forage for 5e8 h a day, diggingfor invertebrates and small vertebrates in the sand (Doo-lan & Macdonald 1996), and while doing so they areunable to detect predators, and annual mortality rate ishigh (Clutton-Brock et al. 1999a). Their main aerial pred-ators are martial eagles, Polemaetus bellicosus, and palechanting goshawks, Melierax canorus. Group members al-ternate in guarding from a raised position (Manser 1999;Clutton-Brock et al. 1999b), alerting other group membersby emitting a variety of alarm calls (Manser 2001), andforaging animals frequently scan their surroundings forpredators, and warn other group members, in the samemanner. Meerkat alarm calls not only encode informa-tion about the predator type, but also about the relativerisk that the calling individual is exposed to (Manser2001). Therefore, it is likely that variable thresholds foremitting alarm calls exist, and hence the potential for dis-crimination between individuals arises (Blumstein et al.2004).

To investigate whether individual discrimination is ofimportance in meerkat alarm calling and response behav-iour, we first determined whether they emit individuallydistinctive alarm calls using a multinomial discriminantfunction. Secondly, we conducted a modified habitua-tionedishabituation playback experiment (Cheney & Sey-farth 1988; Johnston & Jernigan 1994; Gheusi et al. 1997;Hauser 1998; Weiss et al. 2001; Blumstein & Daniel 2004),to investigate whether receivers use the information onindividuality to differentiate amongst each other. We fo-cused on ‘medium-urgent aerial alarm calls’ (see Fig. 1a),which are usually emitted when aerial predators approach(Manser 2001). On hearing this call, meerkats immedi-ately stop foraging, and either stand up on their rearlegs, or run to the nearest shelter (Manser et al. 2001).

METHODS

Study Site and Animals

Recordings and playbacks were conducted on ninegroups of free-living meerkats at the Kuruman RiverReserve, which lies about 30 km east of Van Zylsrus alongthe dry riverbed of the Kuruman River (26�580S, 21�490E)(Clutton-Brock et al. 1998). Groups consisted of six to23 individuals, resulting in a total of approximately 144individuals, all of whose genealogy is known as theyhave been marked and observed since emergence. We fo-cused on adult subordinate individuals, preferably fromthe same sex within the same group. As part of the Kala-hari Meerkat Project’s long-term study, all animals in everygroup were marked with a subcutaneous transponder(Clutton-Brock et al. 2001) and with superficial dye spotsfor individual identification (details see Jordan et al.2007), and were habituated to a level that allowed record-ings and observations within 0.5 m.

Acoustic Analysis

Audio recordingsRecordings of medium-urgent aerial alarm calls were

performed between February and August 2005 through-out the day, with a Sennheiser microphone ME66/K6connected to a portable Marantz PMD670 digitalprofessional solid state recorder (D&M Professional, IL,U.S.A.). The majority of all recorded calls were obtainedduring naturally occurring predator encounters. For threetest groups call sequences were insufficient in numbers,and we therefore presented a model kite painted in black ina distance of 200 m and a height of up to 50 m, to elicitalarm calls. The kite resembled a hovering aerial predator

100 200 300

kHz

4

2

0

100 200 300ms

4

2

0

(a)

(b)

Figure 1. Spectrogram of one call out of an alarm call sequence: (a)medium-urgent aerial alarm; (b) medium-urgent terrestrial alarm.

SCHIBLER & MANSER: INDIVIDUAL DISCRIMINATION IN ALARM CALLS 1261

and always evoked strong acoustical responses (aerial alarmcalls) by the meerkats. To avoid any habituation to suchkite experiments, we performed two experiments at mostwithin each group, with a 14-day break in between them.

Caller identity, group response, the context (predatoridentity and distance) and time of day that each call wasgiven were recorded. All recordings were performed atclose distance (within 2 m) by focusing on one specific in-dividual until an alarm call had been successfully recordedfrom it. All digitized recordings were then transferred toa DELL Inspiron 9100 portable PC.

Call selectionFor acoustic analyses, we used 37 different call se-

quences, each consisting of four to six calls. All callsequences were recorded in separate predator encounterevents from 24 individuals (19 males, five females) inseven groups. In the two other groups used for theplayback experiments, we were only able to record alarmcall sequences repeatedly from one individual, and onesequence from other group members, and therefore didnot use these calls for the acoustic analysis on individualdifferences. From 10 of the individuals we had been ableto obtain call sequences repeatedly from several differentcontexts. To obtain a nonbiased selection of appropriatecalls, we took consecutive four-call-sequences, whichstarted right after the fourth call of the whole recordedalarm call sequence. By applying this method, we ensuredthat all analysed calls represented the natural occurringvariation within this particular call type.

Using Avisoft-SASLab Pro 4.38 software (Avisoft Bio-acoustics, Berlin, Germany), calls were sampled at11.025 kHz with a fast Fourier transform length of 1024points a frequency resolution of 10 Hz and a temporalresolution of 2.9 ms. If background noise did not allowan accurate measurement, all four consecutive calls werehigh-pass filtered at 400 Hz. All sequences were controlledand analysed call after call, to ensure the most accurateresults. Automatic measurements finally yielded 55 sepa-rate temporal, spectral and energy parameters, most ofwhich were measured at four different locations, namelyat the start of each element, at its centre, at the locationof the maximum amplitude, and at its end.

Statistical analysisTo reduce the measured acoustic parameters from all

individuals and to avoid correlations, we performed a prin-cipal component analysis (PCA) using SPSS software(version 13.0 SPSS Inc., Chicago, IC, U.S.A.). Principalcomponent analyses serve to reduce many parameters toa subset of derived new parameters (principal components),allowing a large part of the information from a given dataset to be summarized in fewer variables (Quinn & Keough2002).

A multinomial regression analysis based on the princi-pal components was then conducted to test whether the24 individuals can be separated, by generating logisticcombinations of the derived components, which maxi-mize the probability of correctly assigning observations totheir actual groups. To avoid an overfitting of the data,

a 10-fold cross-validation was performed. In this process,the data are randomly split into 10 parts, nine of whichare used to estimate the model parameters. The 10th partis then being predicted by the model of the other nine.The whole process is repeated 10 times, so that the entiredata set has been predicted once. Only through thismethod, the function’s generality can be assured. Becauseof the low levels of correct assignment in the 10-fold cross-validation across all individuals, the same tests wereconducted within groups (n ¼ 7). This approach is coher-ent with the biological idea that individuals must be rec-ognized within, rather than between, groups.

Playback Experiments

Experimental designTo test whether meerkats discriminate between individ-

uals based on their alarm call vocalizations, we performeda habituationedishabituation playback (Cheney & Sey-farth 1988; Johnston & Jernigan 1994; Gheusi et al.1997; Hauser 1998; Weiss et al. 2001; Blumstein & Daniel2004). This particular type of playback is a popular methodused to test the discrimination abilities of animals onspecific categories, such as individual differences. Themethod involves the successive presentation of severalsingular stimuli in a specific time interval (habituationphase), followed by the presentation of a different, ornovel, stimulus (dishabituation phase). The difference inthe reaction intensity between the habituation and disha-bituation stimulus is a predictor for the ability to discrim-inate between the two stimuli. In other words, a reboundin the reaction to the dishabituation phase, comparedwith the reaction in the previous habituation phase im-plies that the recipient is able to discriminate betweenthe two stimuli. In our case we tested a single subject(focal individual) to the alarm calls of two of its own groupmembers.

To prevent habituation to the playback experiments perse, we included the control within one complete playbackexperiment instead of performing an additional series.This control, also called rehabituation test, ensures that arebound is not due to a spontaneous recovery of prehabi-tuation levels (Rendall et al. 1996). It consists of an addi-tional stimulus from the same type and individual as inthe habituation series, and is presented directly after thetest call. In our set-up shown in Table 1, this is representedby the second last call (RHB1 ¼ rehabituation call 1) ofthe whole experiment. In addition, we also presented adifferent type of call to control for the possibility thatmeerkats might just habituate to the playback procedurerather than the actual individual. We therefore playeda bout of medium-urgent terrestrial alarm calls (Table 1,TERRESTRIAL) after the dishabituation (Table 1, TEST)stimulus. Meerkats respond to terrestrial alarm calls bygathering together, scanning the area and often movingaway as a group (for more details see Manser et al.2001). This control stimulus is especially important insystems where no obvious discrimination between habi-tuation and dishabituation phase is found, and wherea lack of response may be due to the experimental design.

ANIMAL BEHAVIOUR, 74, 51262

Table 1. Standardized set-up of playback experiment in each group

Order of call sequence Period Played call sequence Abbreviations in text and results

1 Habituation Natural call (a) of individual 1 HB12 Natural or synthesized call of individual 1 HB23 HB34 HB45 Foreign aerial call FOREIGN6e14 Natural or synthesized call of individual 1 HB6 to HB14Second last (7e15) 2.last_HBLast (8e16) Natural call (b) of individual 1 last_HB1 (9e17) Dishabituation Natural call of individual 2 TEST1 (10e18) Control Foreign terrestrial call TERRESTRIAL2 (11e19) Natural call (c) of individual 1 RHB13 (12e20) Synthesized call of individual 1 RHB2

Playbacks of specific call sequences in the habituation, dishabituation (or test) and control period.

Because of a lack of sufficient terrestrial alarm calls fromour test groups, we had to present calls of foreign groups,and some calls were presented to more than one group.Consequently, to ensure that a response to this call typewas not caused by the fact that it was from a foreigngroup, we included one call sequence of medium-urgentaerial alarm calls from a foreign group, in the middle ofthe habituation sequence (Table1, FOREIGN). This notonly controls for a recovery in the response to the medium-urgent terrestrial call due to an identification of anothergroup, but also provides information about individual dis-crimination, since this stimulus also represents the call ofa different individual.

Every playback stimulus consisted of the actual alarmcall sequence of four to six calls (representing the range ofnatural occurring call sequences of medium-urgent aerialcalls), inserted between two 30-s periods of backgroundnoise. Because of a lack of sufficiently variable callsequences from all individuals, we avoided pseudorepli-cation by repeated playbacks of a limited number of callsequences, by synthesizing additional sequences out ofthe existing recordings using CoolEdit 2000 v.1.1 (Syntril-lium Software Corporation, Phoenix, U.S.A.) on a DELLInspiron 9100 laptop. Thereby we duplicated and rear-ranged the single calls within a recorded sequence. Toreduce the risk of habituation we duplicated two calls atthe most within one sequence. Per group, we createda maximum of seven synthesized call sequences, whichtypically together with the natural call sequences wasenough for the habituation. Only in three groups we hadto play more than 10 call sequences for complete habit-uation, and therefore had to play a maximum of foursequences a second time. To exclude the possibility thatthese synthesized sequences were treated unnaturally, weincluded three natural, unaltered playback stimuli in thewhole experiment (Table 1, Fig. 2): one at the beginning(HB1), one just before the dishabituation stimulus(last_HB), and one at the very end (RHB2). By doing so,any negative effect on the focal individual’s behaviourdue to the perception of an artificial call, can be detected.

Each playback experiment included recordings froma maximum of four different individuals (habituationcalls, dishabituation calls, foreign alarm call and otheralarm call type), which were either ‘medium-urgent aerial

alarm calls’ (Fig. 1a) or ‘medium-urgent terrestrial alarmcalls’ (Fig. 1b). Both call types were generally taken fromthe initial recording phase of this study, except in thecase of the terrestrial types, from older recordings (audiodatabase Manser). Moreover, all stimuli were recordedwithin the same group the playback experiments wereconducted (except for foreign alarm call and terrestrialcall). All call sequences were edited with CoolEdit 2000.To achieve highest playback quality, background noisewas effectively reduced by applying a 400-Hz high-passfilter. All calls were matched for amplitude within andbetween consecutive call sequences.

All call sequences in the habituation series were placedin random order, apart from the first (natural recording),the fifth stimulus (from a different group) and the lastone (natural recording). However, through the above-mentioned modifications and controls, we were able toprovide enough certainty to detect a variation in themeerkats responses to the two different call origins.

Experimental procedureWe performed playback experiments in nine different

groups, using medium-urgent aerial alarm calls from 18different individuals (16 males, two females). In additionwe presented terrestrial alarm calls from five differentindividuals (four males, one female). On average we playednine habituation call sequences (ranging from six to 16) tothe point where the focal individual’s response on twofollowing sequences resulted in little response comparedwith that to the preceding call sequence.

Median interval duration between playbacks of callsequences was 1001800 (range from 504200 up to 3604800)varying according to the momentary arousal state of thewhole group; only when 90% of the individuals in thegroup were foraging and the focal individual showed novigilance behaviour, were consecutive stimuli presented.The same procedure was followed after natural occurringpredator encounters or other disturbance events (approx-imately eight events per playback session), for examplepassing cars, group encounters, and mobbing behaviour(Graw & Manser 2007), when we waited at least an addi-tional 5 min until the group and the focal individualresumed foraging.

SCHIBLER & MANSER: INDIVIDUAL DISCRIMINATION IN ALARM CALLS 1263

RHB2RHB1

TERRESTRIA

LTEST

last_H

B

2.last_

HBHB7

HB6

FOREIG

NHB4

HB3HB2

HB1

140

120

100

80

60

40

20 *

* *

*0

Rel

ativ

e ti

me

to r

elax

(%

)

Figure 2. Median and quartiles for the relative time to relax (%), that is time to resume foraging, after the presentation of playback stimuli. The

strongest response to the medium-aerial alarm calls within one playback series was set as 100% (terrestrial alarm call was excluded; 75th per-

centile at 172.1%; outlier at 396.5%), with the following responses being plotted relatively to it. Playback stimuli: HB1e7, habituation se-quences one to seven; FOREIGN, habituation sequence from a foreign group; 2.last_HB, second last habituation sequence before test

stimulus; last_HB, last habituation sequence before test stimulus; TEST, test sequence from a second individual; TERRESTRIAL, terrestrial alarm

sequence; RHB1, rehabituation sequence one with a natural sequence; RHB2, rehabituation sequence two with a synthesized sequence.

In each group the response of the same focal individualwas tested to the playbacks of all the sequences over thewhole experiment. The focal subject was randomly cho-sen (excluding those two whose calls were played back),but it was always a subordinate, adult group member.These individuals were followed during the whole exper-iment and their reaction recorded on a digital videocamera (Sony DC PC109). Filming began 30 s before play-backs started, and stopped when the individual resumedforaging or appeared relaxed again. An assistant playedback all stimuli at a distance of 8e10 m to the focal indi-vidual, using a portable Marantz PMD670 and a portableSony walkman SR A60 speaker. Playback volume was ad-justed prior to trials to match the same amplitude for callsobserved during naturally occurring predator encounters(58e62 dB measured 1 m in front of the speaker), usinga sound level meter (Voltcraft 329 Sound Level Meter,Conrad Electronic, Hirschau, Germany; accuracy � 2 dBat 94 dB), and according to varying wind conditions.Experiments were conducted in the mornings between0900 and 1300 hours, after the group had left the burrowand been foraging for 30 min. Playbacks were only con-ducted when the target animals and most of the othergroup members were further than 10 m from any shelterlocation. Through the presentation of alarm calls duringthis study, meerkats could have been subject to an in-creased level of emotional stress as well as to a decreasedlevel of vigilance behaviour. To minimize these two fac-tors, playbacks were cancelled or delayed if the group

markedly increased their level of vigilance, if the wind be-came too intense or potential predators approached. Forthese reasons, in eight groups we had to repeat the exper-iment a second time, which resulted in a total of 17 play-back attempts over nine groups. Consecutive experimentsat each group were conducted at least 14 days apart toavoid any habituation to the playback procedure.

Responses to playbacksThe video recordings of the focal individual were

digitized and analysed with frame by frame measurements(frame ¼ 0.03 s) using Windows Movie Maker (version1.1). The ‘time to relax’ served as response variable, andrepresented the time that the focal individual took toresume the pretrial behaviour after its initial response.This initial response involved either, just raising its headand looking around, standing up on its rear legs and scan-ning the surroundings, or running towards a shelteredposition. We only used one response variable, as this rep-resents a quantitative measurement of otherwise correlat-ing response categories, and as just raising the head andscanning is typically much shorter than running towardsa sheltered position.

Statistical analysisFor each group only the one successful playback

experiment (n ¼ 9) where all the different playback condi-tions could be performed was included in the analysis.

ANIMAL BEHAVIOUR, 74, 51264

A Friedman test was used to test against the null hypoth-esis that the medians of all compared response categorieswere equal. HochbergeBonferronieadjusted Wilcoxonsigned-ranks tests were performed post hoc to comparethe response of individuals between the four main test cat-egories. The first test compared differences between habit-uation-, dishabituation- and rehabituation trials. Thesecond test compared the natural versus the synthesizedcalls, whereby because of failure of equipment the dataduring the habituation period were only available for sixgroups and not nine as in the rehabituation period, andfor all other comparisons. The third series of tests com-pared terrestrial and aerial calls, and the last test comparedthe responses to own group calls versus calls from foreigngroups.

RESULTS

Acoustic Analysis

The PCA across all individuals reduced the measured 55parameters into seven main components that explained81.7% of the variability of the whole data set (Table 2). The25th percentile and several frequency parameters (mini-mum frequency, amplitude at the maximum frequency,frequency at the maximum amplitude and fundamentalfrequency) strongly loaded on the first component, whereasentropy and bandwidth correlated with the second com-ponent. For the third and fourth component, correlationswith energy parameters, such as the 50th and 75th per-centile, respectively, peak-to-peak amplitude, root meansquare and the energy of the element, were found.

The multinomial regression analysis with those com-ponents was highly significant (Chi-squared test: c2 ¼784.76, P < 0.001) and indicated that individuals differin the acoustic characteristics of their alarm calls. Multino-mial regression correctly classified the seven classificationfunctions of 24 individuals with an accuracy of 94.1%(Table 3). The according 10-fold cross-validation provedthe validation of the model with an average of 45.6%correct assignment, which is higher than by chance alone(i.e. 4.2%). Tests were repeated within each of sevengroups, including multiple call sequences from severalindividuals (ranging from two to five). On average, multi-nomial regression analysis correctly assigned 100% of theclassification functions to the individuals within theseven groups. This result was verified through the accord-ing 10-fold cross-validation, which correctly classified10% of the data with an accuracy of 90.2 � 8.55% SD,n ¼ 7 (Table 3).

Playback Experiments

During playbacks, the focal individuals habituated tothe first series of alarm calls from the same individual(Wilcoxon signed-ranks test: T ¼ 0, N ¼ 9, P < 0.01), andshowed no rebound in the reaction to the test stimulusfrom the different individual (Wilcoxon signed-rankstest: T ¼ 15, N ¼ 9, P ¼ 0.67; Fig. 2). No difference be-tween test and rehabituation trials was found (Wilcoxon

signed-ranks test: T ¼ 14, N ¼ 9, P ¼ 0.57), nor betweenthe last two calls in the habituation and the two callsfrom the rehabituation trials (Wilcoxon signed-rankstest: T ¼ 9, N ¼ 9, P ¼ 0.11). Furthermore, there was nodifference between synthesized calls and unaltered naturalcalls, neither in the habituation phase (Wilcoxon signed-ranks test: T ¼ 2, N ¼ 6, P ¼ 0.13), nor in the rehabitua-tion phase (Wilcoxon signed-ranks test: T ¼ 15, N ¼ 9,P ¼ 0.67).

Meerkats did not habituate to the actual experimentalset-up itself, as they still responded strongly when con-fronted with terrestrial alarm calls (Wilcoxon signed-rankstest: T ¼ 0, N ¼ 9, P < 0.01). The control stimulus consist-ing of a foreign medium-aerial alarm sequence confirmedthat the strong response to the terrestrial alarm call wasnot due to an unknown individual, but due to a differentcall type (Wilcoxon signed-ranks test: T ¼ 0, N ¼ 9,P < 0.01). The foreign control stimulus in the habituationphase tended to elicit a less strong response than the pre-ceding two habituation calls (Wilcoxon signed-ranks test:T ¼ 7, N ¼ 9, P ¼ 0.08), and did not differ to the followinghabituation stimuli (Wilcoxon signed-ranks test: T ¼ 15,N ¼ 9, P ¼ 0.57).

DISCUSSION

Meerkat medium urgency aerial alarm calls encode ampleinformation about the identity of the calling individual,yet receivers appear not to discriminate between callerson the basis of these differences. Although our analysesacross all individuals from several groups gave a ratherlow correct assignment percentage to the specific individ-uals, the high correct assignment within groups showed aclear individual distinction between group members. Thisconfirms previous work on meerkats where three otheralarm call types have been described to show individualvoice characteristics (Manser 1998). Hence, from the sig-naller’s point of view, the potential for successful discrim-ination of individuals by their alarm calls exists (Beecher1982).

Our playback experiments did not confirm the findingsof several other studies that describe the use of individualdiscrimination based on individually distinctive calls(Snowdon & Cleveland 1980; Rendall et al. 1996; Semple2001; Weiss et al. 2001; Ceugniet & Izumi 2004). Meerkatsshowed no rebound in responsiveness when confrontedwith the test stimulus of a different individual in thehabituationedishabituation playback. Additional evidenceis provided by the low response to the call of a differentgroup, and supports these results. This stimulus not onlyacted as a control for foreign alarm calls, but was alsoused as a call from a different individual, emphasizingthat meerkats did not dishabituate and appeared not todistinguish between callers. However, the habituation inresponse and the according transfer of such can be inter-preted in three ways: (1) as a habituation to the calltype, because they did not recognize the two differentindividuals; (2) as a habituation to the call type, eventhough they recognized the two different individuals,but caller identity had no influence on the response;

SCHIBLER & MANSER: INDIVIDUAL DISCRIMINATION IN ALARM CALLS 1265

Table 2. Factor loadings for the acoustic variables

Acoustic parameter

Components

1 2 3 4 5 6 7 8 9

Quart 25 (mean) 0.945Quart 25 (centre) 0.933Min freq (max) 0.895Quart 25 (max) 0.894Quart 25 (start) 0.857Quart 25 (end) 0.839Min freq (centre) 0.835Peak freq (end) 0.828Min freq (mean) 0.825Peak freq (centre) 0.815Max freq (end) 0.807Min freq (end) 0.807Quart 50 (end) 0.784Peak freq (mean) 0.783Peak freq (max) 0.779Max freq (centre) 0.760 0.492Max freq (start) 0.743Fundamental (start) 0.742 �0.491Min freq (start) 0.739Max freq (max) 0.732 0.553Peak freq (start) 0.729Peak ampl (start) 0.669 0.55Peak ampl (end) 0.665 0.552Peak ampl (mean) 0.626 �0.439 0.561Peak ampl (max) 0.609 �0.406 0.591Fundamental (max) 0.606 0.454Peak ampl (centre) 0.594 �0.439 0.592Fundamental (mean) 0.576Fundamental (end) 0.56 0.448Fundamental (centre) 0.479Entropy (centre) 0.839Bandw (max) 0.879Entropy (mean) 0.868Bandw (centre) 0.864Entropy (max) 0.844Bandw (mean) 0.813Max freq (mean) 0.544 0.636Quart 75 (mean) 0.809Quart 75 (max) 0.807Quart 75 (centre) 0.796Quart 75 (start) 0.436 0.625Quart 50 (centre) 0.624 �0.461Quart 50 (start) 0.551 0.576Quart 50 (max) 0.574 �0.554Quart 75 (end) 0.447 0.552Quart 50 (mean) 0.523 0.55 �0.441Peak to peak 0.521 0.67Rms 0.549 0.626Energy 0.457 0.568Entropy (start) 0.425 0.444 0.468Disttomax 0.468 0.615 0.403Bandw (end) 0.446 �0.493Entropy (end) 0.401 0.436 0.406 �0.447Bandw (start) 0.697Duration 0.495

Measured at five locations: start, start point of the element; end, end point of the element; centre, centre of the element; max, location of themaximum amplitude of the element; mean, mean spectrum of the entire element. Measured parameters: bandw, bandwidth; disttomax, dis-tance from start point to the maximum amplitude; duration, duration from start to end; energy, energy of the element; entropy, entropy of theentire element; fundamental, fundamental frequency; maxfreq, highest frequency; minfreq, lowest frequency; peak ampl, amplitude at thepeak frequency; peakfreq, frequency of the maximum amplitude; peak to peak, the peak-to-peak amplitude; quart25, quartile at 25% ofthe energy; quart50, quartile at 50% of the energy; quart75, quartile at 75% of the energy; rms, the root mean square.

ANIMAL BEHAVIOUR, 74, 51266

and (3) as a habituation to the experimental set-up. Thelatter explanation can clearly be rejected because ofthe strong response towards the played back terrestrialalarm call. However, with our current experimental set-up, we were unable to distinguish between the first twopossibilities. As Gheusi et al. (1997) stated, a lack of re-bound does not exclude the possibility that the abilityto recognize individuals exists. This would have to betested in further experiments, for example conditioningtests out of the typical context. Nevertheless, whethermeerkats recognized the different individuals, individualdiscrimination of alarm callers appears to be of little orno importance to the receivers, contrary to previouswork described for other species (Cheney & Seyfarth1988; Blumstein et al. 2004).

However, there are reasons to believe that these hy-potheses of advantages to discriminate unreliable callersmay be of a certain importance in this study system.Meerkat pups (up to 3 months old) and juveniles (up to 6months old) emit alarm calls in situations where the riskof predation is low or nonexistent (Hollen & Manser2006). Adults appear to be less likely to respond to callsof these younger group members. However, future re-search will show whether this might be because of a differ-ent acoustic structure of their calls, or because of a lesspronounced signal to background noise ratio. Younger in-dividuals might simply be unable to produce as loud callsthat are as loud as those of adults thereby falsely commu-nicating less urgent situations, and therefore other groupmembers are less likely to respond. It may not have todo anything with recognition of unreliable callers at all,but simply be an effect of amplitude of the emitted call.Furthermore, this discrimination would not necessarilycorrespond to individual identity, but to age categories.

To really understand why individual discriminationdoes not appear to be used in the context of alarm callsone has to address the costs and benefits of antipredatorbehaviour. Predation pressure in meerkats is high (Clut-ton-Brock et al. 1999a), and the cost of not responding

Table 3. Classification table of the Multinomial Regression Analysis(MNR) and the 10-fold cross-validation

Group N MNR

Ten-fold

cross-validation By chance

Across allindividuals ofseven groups

24 94.1 45.6 4.2

Balrog 4 100 94.7 25Young ones 4 100 100 25Zappa 5 100 91.6 20Elveera 4 100 80 25Gattaca 2 100 80 50Commandos 3 100 85.7 33.3Frisky 2 100 100 50

Mean 100 90.2 32.6

First row, across N individuals from all seven groups. Afterwards, onlywithin-group members in each group, with the mean of within-group classifications (n ¼ 7). A correct classification due to chance(1/N ) is given in the last column.

to a given alarm can be substantial. If these costs exceedthe cost of responding to an unreliable caller, individualsshould react independent of the caller’s reliability. How-ever, meerkats experience great costs in responding toalarm calls, as they forage for mobile prey, which are easilylost within a short time period. If meerkats regularly aban-doned foraging spots where they had been digging for sev-eral minutes, a substantial amount of time and energycould be lost. In support of this explanation, digging indi-viduals often do not respond to low urgency alarm callsand hesitate longer to medium urgency calls, while indi-viduals foraging on the surface (with less to lose) run forshelter immediately (M. B. Manser, personal observation).Furthermore, meerkats use individually distinctive alarmcalls which grade along the level of urgency from onecall type into another (Manser 2001). Therefore, individu-ally distinctive alarm calls might be emitted according toindividually distinctive threshold levels on the perceptionof danger. According to Blumstein et al. (2004), both ofthese factors would suggest a great advantage in the abilityto recognize individuals based on their level of reliability.However, this study is the first to observe that the discrim-ination of individuals in species with high predation riskis of no major importance. Meerkats show a transfer of ha-bituation when confronted with playbacks of a differentalarm calling individual. Whether the habituation isbecause of the individual or the presented call type, theconcept of reliability (Cheney & Seyfarth 1988) and itsimportance in explaining individual recognition, as it isdiscussed in marmots (Blumstein et al. 2004), do not cor-respond with our results. Despite intense observation ofthis population of meerkats, cheating behaviour has notbeen described, and only few anecdotal observations indi-cate that cheating may occur at all (M. B. Manser, personalobservation). Therefore, common cheaters are unlikely toexist in this cooperative living mongoose species, andmeerkats probably have no need to assess caller reliabilityin any form. This allows us to conclude, that individualdiscrimination in the context of alarm calling is irrelevantin meerkats. However, the rapid decline of meerkats’ re-sponse to repeated playbacks of the same call type doesnot support such a conclusion and needs alternativeexplanations.

Following repeated playbacks of the same call type,meerkats might quickly realize that there is no potentialpredator present, by associating a particular call type witha potentially nondangerous context, which leads to a re-duction in their vigilance behaviour. It is possible thatonly by experiencing a real predator threat themselves, forexample through visual contact, do they maintain theirvigilance behaviour. Personal observations (Schibler)indeed suggest that after experiencing a real predatorencounter, a meerkat’s level of habituation initially de-creased dramatically, before it continued to increase again.Furthermore, support for this ‘context-reliability assess-ment’ is evidenced by the strong response on presentationof the terrestrial alarm call. On hearing this different calltype meerkats might link it to a novel predation threat,which requires an individual to act appropriately (Cheney& Seyfarth 1988). The fast habituation expressed by thedifference between the response duration to the first call

SCHIBLER & MANSER: INDIVIDUAL DISCRIMINATION IN ALARM CALLS 1267

and the consecutive calls can be explained by the type ofantipredator behaviour. In response to these calls, the cor-rect response would be to run to a bolthole and go belowground. However, when habituation occurs the reactiontime is greatly reduced as the response becomes a quickscan of the environment. By the quick assessment of theabsence of predators, meerkats avoid such time consum-ing escapes and instead show a more stationary response,by shortly looking up or standing up on two legs. Previousexperiments on responses to alarm calls have shown thatmeerkats appear to be able to detect a raptor by scanningthe sky within short time periods (Manser & Fletcher2004).

The question of the purpose of individually distinctalarm calls without the according discrimination stillremains. It is possible that these individual differencesmay simply result from a correlation with morphologicaltraits, such as body size, body weight, vocal tract length,etc., and that the differences themselves have no adaptivefunction. Individual differences have been shown for othermeerkat vocalizations, including tonal call types, such asthe sentinel calls (Manser 1999) and also the barks, whichhave a more noisy structure (Manser 1998). This indicatesthat similar to, for example humans, the different types ofmeerkat vocalizations may be in general individually dis-tinct. Nevertheless, detectable individual differences inthe acoustic structure do not necessarily imply a specificbiological function (Beecher 1991). Before we make anyconclusions on vocal signatures, call production, percep-tion and discrimination by receivers must be tested andtheir relevance demonstrated. On the other hand, simplybecause receivers do not show an obvious difference intheir response to different individuals does not meanthat they are not able to distinguish between them (Gheusiet al. 1997; Tang-Martinez 2001), it merely shows that thisinformation is not used in this specific context.

Acknowledgments

We would like to thank Tim Clutton-Brock for allowing usto work on the meerkat population of the KalahariMeerkat Project. Thanks to Family Hennie Kotze forallowing the work to be carried out on his land, JohanDu Toit and Martin Haupt at the Mammal ResearchInstitute, University of Pretoria, for logistical support.We are grateful to Tom Flower as field manager, and toall of the students and volunteers who contributed to datacollection. Thanks to Lorenz Gygax and Hansjoerg Kuncfor statistical advice, and Neil Jordan and two anonymousreferees for comments on the manuscript. This project wasfunded by a grant given to MBM from the Swiss NationalScience Foundations, SNF-Forderprofessur Nr 631-066129.The study was carried out under licences issued by theNorthern Cape Conservation Service and ethical commit-tee of Pretoria University, South Africa.

References

Beecher, M. D. 1982. Signature systems and kin recognition.

American Zoologist, 22, 477e490.

Beecher, M. D. 1991. Successes and failures of parent-offspring rec-

ognition in animals. In: Kin Recognition in Animals (Ed. by P. G.

Hepper), pp. 94e124. Cambridge: Cambridge University Press.

Blumstein, D. T. & Daniel, J. C. 2004. Yellow-bellied marmots

discriminate between the alarm calls of individuals and aremore responsive to calls from juveniles. Animal Behaviour, 68,

1257e1265.

Blumstein, D. T., Verneyre, L. & Daniel, J. C. 2004. Reliability and

the adaptive utility of discrimination among callers. Proceedings of

the Royal Society of London, Series B, 241, 1851e1857.

Ceugniet, M. & Izumi, A. 2004. Vocal individual discrimination in

Japanese monkeys. Primates, 45, 119e128.

Cheney, D. L. & Seyfarth, R. M. 1988. Assessment of meaning and

the detection of unreliable signals by vervet monkeys. Animal Be-

haviour, 36, 477e486.

Cheney, D. L. & Seyfarth, R. M. 1990. How Monkeys See the World.

Chicago: University of Chicago Press.

Clutton-Brock, T. H., Gaynor, D., Kansky, R., MacColl, A. D. C.,McIlrath, G., Chadwick, P., Brotherton, P. N. M., O’Riain,J. M., Manser, M. & Skinner, J. D. 1998. Costs of cooperative be-

haviour in suricates (Suricata suricatta). Proceedings of the Royal So-

ciety of London, Series B, 265, 185e190.

Clutton-Brock, T. H., Gaynor, D., McIlrath, G. M., Maccoll,A. D. C., Kansky, R., Chadwick, P., Manser, M., Skinner,J. D. & Brotherton, P. N. M. 1999a. Predation, group size andmortality in a cooperative mongoose, Suricata suricatta. Journal

of Animal Ecology, 68, 672e683.

Clutton-Brock, T. H., O’Riain, J. M., Brotherton, P. N. M., Kansky,R., Griffin, A. S. & Manser, M. 1999b. Selfish sentinels in cooper-

ative mammals. Science, 284, 1640e1644.

Clutton-Brock, T. H., Brotherton, P. N. M., O’Riain, J. M., Griffin,A. S., Gaynor, D., Kansky, R., Sharpe, L. & McIlrath, G. M. 2001.

Contributions to cooperative rearing in meerkats. Animal Behav-iour, 61, 705e710.

Clutton-Brock, T. H., Russell, A. F., Sharpe, L. L. & Jordan, N. R.2005. ‘False feeding’ and aggression in meerkat societies. Animal

Behaviour, 69, 1273e1284.

Doolan, S. P. & Macdonald, D. W. 1996. Diet and foraging

behaviour in group-living meerkats, Suricata suricatta, in the

south-western Kalahari. Journal of Zoology, London, 239,

697e716.

Estes, R. D. 1991. The Behaviour Guide to African Mammals. Berkeley

and Los Angeles: University of California Press.

Gheusi, G., Goodall, G. & Dantzer, R. 1997. Individually distinctive

odours represent individual conspecifics in rats. Animal Behaviour,53, 935e944.

Graw, B. & Manser, M. B. 2007. The function of mobbing in meer-kats. Animal Behaviour, 74, 507e517.

Hamilton, W. D. 1963. The evolution of altruistic behavior. AmericanNaturalist, 97, 354e356.

Hare, J. F. & Atkins, B. A. 2001. The squirrel that cried wolf:

reliability detection by juvenile Richardson’s ground squirrels(Spermophilus richardsonii). Behavioral Ecology and Sociobiology,

51, 108e112.

Hauser, M. D. 1991. Sources of acoustic variation in rhesus macaque

(Macaca mulata) vocalizations. Ethology, 89, 29e46.

Hauser, M. D. 1998. Functional referents and acoustic similarity:

field playback experiments with rhesus monkeys. Animal Behav-

iour, 55, 1647e1658.

Hollen, L. & Manser, M. B. 2006. Ontogeny of alarm call responses

in meerkats (Suricata suricatta): the role of age, sex and nearby

conspecifics. Animal Behaviour, 72, 1345e1353.

Johnston, R. E. & Jernigan, P. 1994. Golden hamsters recognize

individuals, not just individual scents. Animal Behaviour, 48,129e136.

ANIMAL BEHAVIOUR, 74, 51268

Jordan, N. R., Cherry, M. I. & Manser, M. B. 2007. Latrine distribu-

tion and patterns of use by wild meerkats: implications territory

and mate defence. Animal Behaviour, 73, 613e622.

Manser, M. B. 1998. The evolution of auditory communication in

suricates, Suricata suricatta. Ph.D. thesis, University of Cambridge.

Manser, M. B. 1999. Response of foraging group members to sen-

tinel calls in suricates, Suricata suricatta. Proceedings of the RoyalSociety of London, Series B, 266, 1013e1019.

Manser, M. B. 2001. The acoustic structure of suricates’ alarm callsvaries with predator type and the level of response urgency.

Proceedings of the Royal Society of London, Series B, 268, 2315e

2324.

Manser, M. B. & Fletcher, L. B. 2004. Vocalise to localise e a test on

functionally referential alarm calls. Interactions Studies, 5, 325e342.

Manser, M. B., Bell, M. B. & Fletcher, L. B. 2001. The information

that receivers extract from alarm calls in suricates. Proceedings of

the Royal Society of London, Series B, 268, 2485e2491.

Quinn, G. P. & Keough, M. J. 2002. Experimental Design and

Data Analysis for Biologists. Cambridge: Cambridge UniversityPress.

Rendall, D., Rodman, P. S. & Emond, R. E. 1996. Vocal recognition

of individuals and kin in free-ranging rhesus monkeys. Animal Be-

haviour, 51, 1007e1015.

Semple, S. 2001. Individuality and male discrimination of female

copulation calls in the yellow baboon. Animal Behaviour, 61,1023e1028.

Snowdon, C. T. & Cleveland, J. 1980. Individual recognition ofcontact calls by pygmy marmosets. Animal Behaviour, 28,

717e727.

Tang-Martinez, Z. 2001. The mechanisms of kin discrimination and

the evolution of kin recognition in vertebrates: a critical re-evalua-

tion. Behavioural Processes, 53, 21e40.

Trivers, R. L. 1971. Evolution of reciprocal altruism. Quarterly Review

of Biology, 46, 35e57.

Weiss, D. J., Garibaldi, B. T. & Hauser, M. D. 2001. The production

and perception of long calls by cotton-top tamarins (Saguinus oe-

dipus): acoustic analyses and playback experiments. Journal ofComparative Psychology, 115, 258e271.

Wilson, E. O. 1979. Sociobiology. Cambridge, Massachusetts: TheBelknap Press of Harvard University Press.