Experimental Designs and Research Questions
Experimental designs can vary widely in terms of their complexity, ranging from simple presentations of a single call from one speaker, to repeated presentations at flexible intervals, to complicated variation of contextual and acoustic information. The single-speaker design is technically the simplest case, where only one speaker is set up, from which either a single call, a call bout, or even a combination of calls recorded from different animals can be played. The use of two different speakers delivering different calls from different locations is also possible, and not surprisingly, much more difficult to execute, especially when the exact timing of events is crucial. Below, we broadly group the different experimental designs by research question, namely the investigation of (1) context- or caller-related acoustic differences, (2) the importance of the social relationship between the putative caller and the subject, including the social function of calls, and (3) social knowledge. Two special cases, the habituation-recovery design and the two-speaker choice test are discussed separately, as they cut across these broad categories.
Acoustic Variation in Relation to Context or Quality of the Sender
The typical research question would be whether observed acoustic differences are salient to the listeners; the experiments thus consist of the presentation of acoustically distinct calls that were originally given in response to different stimuli, such as different predator categories, in different social contexts, or that reflect variation in caller condition.
Following the playback study on vervet monkey responses to conspecific alarm calls [Seyfarth et al., 1980a], a number of studies investigated whether other species show similarly distinct responses to variation in different alarm calls. For instance, Fichtel and Kappeler played back different calls of aerial and terrestrial predators to redfronted lemurs, Eulemur fulvus rufus, and to white sifakas, Propithecus verreauxi verreauxi. Both species responded with specific alarm calls only to the calls of aerial predators, whereas the calls given in response to calls from terrestrial predators also occurred in other contexts [Fichtel & Kappeler, 2002]. Similar studies were conducted in Barbary macaques [Fischer et al., 1995], tamarins [Kirchhof & Hammerschmidt, 2006] and forest guenons [Zuberbühler, 2000d, 2001].
In numerous species, the relevant acoustic variation does not appear to be the structure of single calls, but the composition of the entire call bout [Arnold & Zuberbühler, 2006, 2008; Schel et al., 2009]. To test whether listeners can extract this potential information, the playback consists of broadcasting natural or edited sequences of sounds. For instance, Arnold and Zuberbühler found that putty-nosed monkeys, Cercopithecus nictitans, give sequences of “pyows” and “hacks” in response to different aerial and terrestrial predators, or when they initiate a group movement. The composition of the sequences varies systematically in relation to the original context, and listeners respond differently to different types of sequences, thus supporting the view that listeners attach different meaning to different call combinations [Arnold & Zuberbühler, 2008].
Primate vocalizations may not only vary in relation to predators but also the quality of an interaction. For instance, the copulation calls of female Barbary macaques, Macaca sylvanus, differ acoustically in relation to cycle stage: calls at the beginning of the cycle are significantly shorter and have a lower mean dominant frequency than during peak swelling, when the ovulation is most likely to occur, and this variation is salient to male Barbary macaques, who respond more strongly to playbacks of calls recorded during peak swelling [Semple & McComb, 2000]. Note, however, that the calls do not clearly indicate the timing of ovulation [Pfefferle et al., 2008a]. Intriguingly, female Barbary macaque copulation calls given during copulation that resulted in ejaculations and those given during non-ejaculatory copulations differ acoustically. Playback experiments revealed that males looked significantly longer towards the loudspeaker after hearing ejaculatory calls than after hearing non-ejaculatory calls. Also, they spent more time walking and more time in close proximity to females after having heard ejaculatory calls [Pfefferle et al., 2008b].
Acoustic variation in relation to physiological state can also be found in the loud calls (“wahoos”) of male chacma baboons, Papio ursinus. These vary with the rank of the caller [Fischer et al., 2004]. When playback experiments systematically varied different features of the calls, listeners attended to variation in the two most prominent features that correlated with rank, namely the fundamental frequency and the length of the “hoo” syllable [Kitchen, D., Cheney, D.L., Engh, A., Fischer, J., Moscovice, L., Seyfarth, R.M., unpublished data]. Calls not only correlate with physiological but also with emotional state. For instance, chimpanzee, Pan troglodytes schweinfurthii, screams differ in relation to the role in a social conflict [Slocombe & Zuberbühler, 2005], and this variation appears to be perceptually salient [Slocombe et al., 2010].
The results of these experiments indicate that there are no fundamental differences in the experimental design and the assessment of the responses in relation to whether the calls “refer to” some external event such as the presence of a predator, or variation in arousal or internal state [Wheeler, 2010]. The implications for the mechanisms underlying the so-called “functionally referential signaling” are discussed elsewhere in more detail [Wheeler & Fischer, 2012], but it should be noted that the cognitive mechanisms supporting the responses to different alarm calls or different mating calls are probably the same.
In sum, there are a number of factors that can give rise to variation in calls, and depending on the hypothesis to be tested, the simple playback of a call or a call sequence can shed light on listeners' assessment and responses to these calls. The majority of the published data support the view that nonhuman primates are adept at distinguishing minor variation in calls, if this is related to ecologically or socially relevant variation in the environment or caller state.
Integration of contextual information
Subjects' responses may not only depend on the variation in the acoustic structure of the calls but also on other factors, such as variation in context [Fischer, 2013]. This has important repercussions for the identification of “call meaning,” which is typically inferred from the responses. An attempt to disentangle the relative contributions of variation in the signal and variation in context can be made by systematically pairing calls with different context. One early such attempt was made by Rendall et al. , who played so-called “move” and “infant” grunts of baboons in situations, in which either actual move or infant grunts are typically heard. The responses clearly depended on both aspects. For instance, subjects were most likely to respond to the playback of move grunts by uttering move grunts themselves when the calls were presented at the edge of an “island,” where animals typically embark on crossings of the surrounding flood plains [Rendall et al., 1999]. Interestingly, Pfefferle et al. [2008b] found that male Barbary macaques only paid attention to variation in copulation call characteristics, while they ignored other potential sources of information about the receptivity of the female. Specifically, they presented copulation calls recorded during a time of high likelihood to conceive during the time of maximum swelling, and outside the females' sexual cycle. Male responses did not differ in relation to cycle stage.
Prime probe experiments
A more elaborate way to assess the integration of contextual and acoustic variation is to experimentally manipulate the context, for instance by presenting stuffed predator models before the playback, or, alternatively by first presenting some calls, and then later some others. This approach has become known as the “prime-probe” design. In a series of elegant experiments, Klaus Zuberbühler applied this method to explore the knowledge and communication of forest guenons [Zuberbühler, 2000b]. Among other aspects, he investigated whether male Diana monkeys, Cercopithecus diana, understand the semantic content of the alarm calls that chimpanzees give in the presence of leopards, Panthera pardus. These monkeys have two important predators, chimpanzees and leopards, and use two strategies to defend themselves against them: a conspicuous strategy in the presence of leopards (i.e. alarm calling), and a cryptic strategy in the presence of chimpanzees (i.e. silently vanishing). The conspicuous strategy tends to result in the leopards leaving the area upon being detected, whereas the cryptic strategy seems to result in the chimpanzees not being able to detect their prey. Chimpanzees themselves sometimes fall prey to leopards, and use a loud alarm call upon detection of a leopard in the area. This led to the question of whether Diana monkeys only took the leopard alarm calls of chimpanzees as signs of a chimpanzee being present, or whether they also understand that the chimpanzee signal contains the information that a leopard is present in the area. Zuberbühler [2000b] presented two playback stimuli, a prime and a probe, separated by 5 min of silence, to different Diana monkey groups. In the decisive test condition, males heard chimpanzee alarm calls followed by leopard growls. In this case, prime and probe stimuli were acoustically different but correlated with the same event, namely the presence of a leopard. About half of the males responded conspicuously towards the chimpanzee alarm calls, and then weakly towards the leopard growls, whereas the other half responded weakly towards the chimpanzee alarm calls, but then strongly towards the leopard growl [Zuberbühler, 2000b].
Variation in social relationships
Playback experiments that tap into the modulation of behavior in relation to the social relationship either address the effects of long-term relationships such as kinship or friendship, or may explore more short-term contingencies such as the effect of the last interaction before the experiment. In the first category, the context in which playbacks are conducted is typically held constant, while the nature of the relationship varies, whereas the reverse is true for the latter category. Regarding the importance of long-term relationships, rhesus monkeys, Macaca mulatta, for instance looked longer towards playbacks of their kin compared to other young group members [Rendall et al., 1996], while young Barbary macaques responded more strongly to playbacks of their mothers' calls [Fischer, 2004]. Palombit et al.  investigated the function of baboon friendships. In one of their playback experiments, they examined whether the responses of males towards the distress calls of females varied systematically with their relationships to those females. For this purpose, female distress calls were played back to male friends and to non-friends (control) of similar social rank who had a friendship with another female. As expected, females' screams elicited significantly stronger responses from friends than from the control males [see also Lemasson et al., 2008].
Alternatively, the differences in the previous interaction between the caller and the recipient, or a relative of the caller and the recipient can be systematically exploited, as in the experiment by Engh et al. , mentioned above. In another experiment, Wittig et al. examined whether vocal support in agonistic interactions function as vocal alliances. Female baboons heard the same female's threat-grunts after being threatened by a signaler's relative or an unrelated female. As a control, the threat grunts were played after an affiliative interaction with the signaler's kin. Subjects avoided both the signaler and her relatives in the condition mimicking vocal support by kin than in the control conditions [Wittig et al., 2007a]. Cheney et al. used a similar approach to assess the “contingent reciprocity” in chacma baboons. They played back threat grunts of a female baboon after she had previously engaged in grooming with the subject, or after she had threatened the subject. They found that subjects were more likely to approach the speaker after a previous positive compared to an agonistic interaction—presumably in support of the partner [Cheney et al., 2010].
An experimental design using combinations of stimuli has been successfully used to investigate baboon knowledge of third-party relationships. The general (and ingenious) idea is to present not just one call or call series given by one individual, but instead to mock an interaction between two (or more) different animals, such as a threat vocalization followed by a fear bark (signaling submission). Since both realistic and unrealistic combinations of calls are used as playback stimulus, the paradigm opens the way to presenting stimuli that are violating the expectations of subjects [Hauser, 1998b].
One of the first such experiments was devised by Cheney et al. , who played sequences of calls mimicking a social interaction. Specifically, they exploited the fact that dominant female baboons, when approaching subordinate females with infants, typically emit a series of grunts, to which mothers sometimes respond with submissive fear barks. In the experiments, the roles were now reversed: now a lower ranking female appeared to be grunting to a higher ranking female, who responded with fear barks. In the control condition, the test subjects heard the same sequence, but at the end, grunts from a third female that was higher ranking than the two other subjects were added, making the sequence of interactions plausible again. The analysis of looking time revealed that subjects responded more strongly to the implausible interaction than to the plausible one, indicating that they are aware of the hierarchical relationships in their social group.
Bergman et al.  used a variant of this design to show that these baboons simultaneously classify group mates by social rank and by affiliation to a matriline, a form of higher order classification that up to that point had been viewed as uniquely human. The sequence of calls that they used mimicked a fight between two females, and consisted of a series of threat-grunts of one female, followed by the screams of the other. These baboons live in a society with a stringent and nested hierarchy where all members of a given matriline occupy adjacent ranks. Because threat/submission interactions are clearly unidirectional, some call combinations are very likely to occur in daily life while others are virtually non-existent. The study rested on the idea to broadcast call sequences mimicking highly unlikely female rank reversals, namely a lower-ranking female uttering threat calls to which a higher-ranking female responded with screaming. Subjects were tested in two test conditions, the first suggesting a rank reversal within a matriline, and the second a rank reversal between matrilines. In control trials, the sequence was consistent with the actual dominance hierarchy. Subjects responded most strongly to rank reversals that cut across matrilines, while there was no significant difference in the responses to rank reversals within matrilines or control trials. The experiments demonstrate that subjects are highly sensitive to putative upheavals that could potentially affect the entire female dominance structure in that baboon group, while they appear to care less about rank changes within matrilines [Bergman et al., 2003].
Crockford et al.  investigated whether male chacma baboons pay attention to sexual consortships, that is, to highly transient relationships between a male and a receptive female that last up to several days. They placed two loudspeakers 40 m apart, with the test subject sitting calmly in the middle while feeding or resting. Each of the speakers broadcast a different call. In the test condition, subordinate males heard the consorting male's contact grunts from one speaker, and after a few seconds, the consorting female's copulation call from the other. This playback sequence simulated the fact that the consort pair had unexpectedly spatially separated, and that another male was now copulating with that female. Since a subordinate male possessing such information could gain a rare opportunity to mate opportunistically with the receptive female, strong responses to this scenario were expected. In the first control condition, the contact grunts of a non-consorting male were played from one speaker, and the receptive female's copulation calls from the other. This playback suggested that the consorting pair were busy copulating, while another male grunted 40 m apart. Therefore, only a weak response to this scenario was expected. The second control trial was conducted 24 hr after the actual consort had ended. Again, subjects heard the former consort male's grunts from one speaker, and the consort female's copulation call from the second. Thus, this control suggested that the consort pair had separated, and the female was now copulating with another male. Again, weak responses were expected to this scenario, since this scenario suggested what informed subjects already knew: that the consort had ended.
Indeed, males responded significantly more strongly to the test condition, compared to the control conditions. The number of looks towards the “female” loudspeaker was significantly higher during the test condition, compared to the control conditions, whereas the number of looks to the “male” loudspeaker did not vary across conditions. In the test condition, more approaches were scored to the speaker that had broadcasted the female copulation call, compared to the control conditions. These results suggested that male baboons routinely monitor other males' consortships, and Crockford et al.  proposed that eavesdropping upon the temporal and spatial properties of male and female vocalizations uttered during consortship might be a strategy by which male baboons achieve sneaky matings.
A paradigm that cuts across several of the research questions outlined above is the habituation-recovery paradigm, which is also known as “habituation-dishabituation” paradigm. The latter term, however, originally refers to a method to distinguish habituation from exhaustion at the neuronal level, and differs with regard to methodological details. We therefore prefer the term “habituation-recovery.” Because of its potential power but also the many problems associated with it, we discuss it separately. It has been used to study call categorization and the detection of unreliable signalers [Cheney & Seyfarth, 1988], the discrimination of different individuals within a group of kin [Rendall et al., 1996], the importance of acoustic similarity and “external reference” [Hauser, 1998a], and the discrimination of acoustically similar call types [Fischer, 1998; Fischer et al., 2001b]. The general idea is to establish one category by repeatedly presenting different call exemplars until the subjects begins to habituate (Fig. 5). The presentation ceases after either a predetermined number of calls or after the subject shows no discernible response in a number of presentations. Moreover, the spacing of the calls can be predetermined (fixed or jittered, with some variation in inter-call-interval) or interactive, where the next call is played when the subject turns its attention away from the calls. Finally, when the criterion has been reached, the test call, putatively belonging to a different category, is presented. Fischer  tested how Barbary macaques responded to acoustic variation between two call types that form a graded continuum. In the tests, she presented calls that (i) stemmed from the same category, and revealed only minor acoustic variation, (ii) revealed the same amount of variation, but cut across the acoustic boundary, or (iii) a test call that was clearly acoustically distinct from the calls used for habituation. She found that animals did not show renewed responses if the novel call was from the same category, while they did respond both to small and large acoustic difference when they straddled the category boundary [Fischer, 1998]. Overall, the findings supported the view that the monkeys categorized these calls in a fashion similar to humans, when they categorize continuous acoustic variation in speech.
Figure 5. Time-course of looking time duration in a habituation-recovery experiment. A: Subjects were habituated with different Barbary macaque alarm call exemplars given in response to a human observer, and then tested with a call given in response to a dog after a variable number of habituation trials, depending on the behavior of the subject. After the subject had failed to respond in three trials, the test call was broadcast. The looking time shows a clear rebound of the animals' interest. B: As before, subjects were habituated with calls given in response to the human observer, and then tested with a novel call from the same category. The subjects showed no renewed interest. Data from Fischer .
Download figure to PowerPoint
Although this approach is extremely powerful, a high number of trials needed to be aborted because the subjects moved away in the middle of the habituation sequence. Moreover, the repeated presentation of long and perhaps unnatural sequences of calls may lead to a general decline in response strength, as the subjects recognize the experimental situation as such. We made one further attempt to use this technique to explore the social knowledge of Barbary macaque females (C. Teufel and J. Fischer, unpublished data), only to find that most subjects hardly responded in the second or third trial, and that the overall response strength declined with an increasing number of trials. We therefore caution against the usage of this technique in the field. Clearly though, it can be highly useful in more controlled settings where the animals can be separated from their group and tested under confined conditions [Fitch & Hauser, 2004].
Two-speaker choice test
In the two-speaker choice paradigm, two stimuli are broadcasted from two spatially separated speakers simultaneously. This paradigm may work well to find out which signal out of two is more relevant than the other. It has successfully been used in birds [reviewed in Douglas & Mennill, 2010] and frogs [Gerhardt, 1994], and more recently also in deer [Wyman et al., 2011]. Hammerschmidt and Fischer  used this approach to assess infant recognition by female Barbary macaques. Two speakers were set up in different directions, with the mother in the center, and screams of the mother's and another infant of the same social group were played simultaneously. Females looked longer towards the loudspeaker playing the screams of their own infants compared to the other. The advantage of such a design is that the subject is in the same motivational state—if the subject responds, the preference is usually unequivocal. Setting up the experiment is, however, difficult, and the question came up whether one could reliably determine what the subjects actually perceived.
In summary, acoustic playback experiments represent one of the most powerful types of field experiments. They can provide invaluable insights into the cognitive mechanisms that underlie primate communication and social behavior, and shed light onto the signals that are meaningful to the animals in an ecologically rich setting. However, experiments should not be done lightheartedly, but instead sparingly and only after collecting solid background data on the natural behavior of the species, which is a prerequisite for the solid interpretation of the results. Patience is crucial, as is the choice of the experimental design and the appropriate controls. If all these conditions are met, playback experiments are tremendously rewarding because they provide us with a glimpse into the animals' minds, and help us to understand the evolution of communication, intelligence, and social behavior.