The cumulative nature of human culture appears to be unique within the animal kingdom 
. This quality requires a high level of copying fidelity to every stage involved; it has been suggested that cumulative culture requires individuals to rely on imitation learning as this leads to learning not only the products, but also the detailed actions necessary to acquire a certain behaviour (i.e., the process leading to the product 
). Academics remain undecided as to whether non-enculturated (i.e., untrained in any way) chimpanzees learn socially in a comparable way to humans, with some arguing that these chimpanzees engage in imitative learning (e.g. 
), and others remaining more sceptical (e.g. 
). And it is these un-enculturated chimpanzees which represent more closely the state of wild living chimpanzees, since wild-living apes do not have the option of human raising or training (thus, if ecological validity is sought, non-enculturated chimpanzees should be studied). The question of whether or not such chimpanzees can socially learn from others using imitation remains an important debate as it may shed light on what sets human culture apart from other types of cultures in non-human species 
Imitation is considered a complex form of social learning that involves copying the demonstrator's bodily actions 
. An alternative form of social learning hypothesized to underlie ape traditions is emulation learning (
, see also 
). When emulation takes place, the observer “picks up” on changes in the environment that result from the demonstrator's actions, hence the term “results copying” may also be used to describe emulation learning 
. These results might be perceived and computed in different ways, ranging from so-called “object-movement re-enactment” 
to insight learning (for a general overview see 
). An emulator ignores the actions of the demonstrator, and focuses primarily (if not solely) on the changes in the environment. As a consequence, if anything is copied following a demonstration, it will be the results, but not the actions involved.
It is, however, possible that observers are not completely blind to actions insofar as these actions can transmit information about the demonstrator's goals. Observers may therefore learn something about the demonstrator's goals based on the observed actions and in combination with what they understand about the observed results, determine how to achieve the same results (or, if the demonstration failed, they may achieve the opposite result instead; see 
). The specific details of the actions would however still be lost since focus would be placed on the goals of the demonstrator and not on the actions themselves. Because these goals typically (though not always) revolve around changes in the environment, emulation learning will be required as well. In these cases, the resulting learning type may represent a mixture of goal copying and results copying (named differently by different researchers; e.g., “goal emulation”–with emphasis on the goal copying part 
, but see also 
; “teleological emulation”–which weighs both learning mechanisms equally 
Empirically, it has proven difficult to separate the effects of actions and results because frequently the two are presented simultaneously (and this is especially problematic if both are somewhat redundant; e.g., a finger pressing a recessed button; see also 
). For example, Whiten et al. claimed chimpanzees imitated the actions performed by demonstrators during a so-called two-action foraging task 
. The apparatus in their study had a block obstructing a food chute, which could be moved to release a piece of food by either lifting or pushing the block. In one experimental group, a chimpanzee demonstrator lifted the block by levering an attached bar with a stick, and in another experimental group, another demonstrator used the same stick to push through a hole at the block itself (two different demonstrations–hence two
-action task). The concern with defining this as imitation, is that if a demonstrator pokes a stick into a hole and the accompanying results (possibly together with the goal of inserting a stick) are copied by an observer, then the observer's behaviour will appear as if they had also copied the underlying actions (the “human eye” seems prone to this type of error). Because of the redundant actions, this experimental design does not rule out emulation as the underlying learning force (which is why it would be more precise to call this method the “two-actions/two-results task”). This kind of methodological issue is not unavoidable, as evidenced by studies on birds (e.g. 
), dogs 
and marmosets 
which overcame these problems by introducing action style components into demonstrations. For great apes, however, most observational learning studies which found copying are unable to distinguish between imitation and emulation learning, because it is unclear exactly what element has been copied (e.g. 
). To complicate matters, some such studies often add further potential information types, such as local or stimulus enhancement (e.g. 
), which again increase the number of possible underlying learning mechanisms (e.g., the two locations in 
). In addition to the typical confound of mixing action with results information (see above), two-action tasks often involve the introduction of relatively trivial differences between groups (e.g., move a lever to the left or right), which can hardly be regarded as a full blown culture even if the respective methods spread (e.g. 
). If such traditions are induced they can at best be described as mere “founder effects” of binary types of information/traditions (compare 
), which (even though important in their own rights) may not get at the heart of the question of whether ape traditions have much in common with human culture. Therefore, in an attempt to answer these questions, less trivial tasks should be used (compare also 
In order to truly investigate whether great apes copy actions spontaneously, a study needs to do one of three things: 1) demonstrate pure actions without any results information at all (“esture copying studies” see 
); 2) demonstrate pure results without any action information at all (“host control studies” see 
) or 3) decouple actions and results (this study). In the latter case, (non-redundant) demonstrated actions Y would lead to the result X (e.g., an approach to a hole walking on one's hands and the insertion of the stick using the foot). Due to them being decoupled from the resulting effect, the peculiar actions Y would later only materialize in the observers if they were indeed copying actions. This would then be a direct test of action copying. The logical counter-variant of such studies–used here–may directly test for emulation learning instead. Here, the demonstrator demonstrates the action Y, but, crucially, the setup is such that observers can only perform an unobserved action Z (because action Y is blocked/unavailable to observers). Again, both actions do lead to the same result (X). Here, if observers produce action Z, then action copying cannot have been responsible–because the observers never have seen action Z being demonstrated. Thus, the observers must have used different types of learning (i.e., emulation learning: reproducing the same result–by necessarily re-inventing an unseen action).
There is only one published study to date that uses the “esture copying”method in non-enculturated chimpanzees (i.e., to demonstrate pure actions without results). In this study, chimpanzees failed to copy a novel action (a “begging gesture” from a conspecific model despite potentially high levels of rewards 
. Although these findings provided no evidence that chimpanzees copy actions spontaneously in problem solving tasks, subsequent ghost control studies have produced a more ambiguous picture. Three ghost control studies with chimpanzees have now been published, all using conspecific-demonstrator conditions as comparisons (i.e. 
). In one study, observers showed no evidence of observational learning regardless of the condition 
. In a different study, chimpanzees learned in the full-demonstration condition, but did not learn in the ghost condition 
. However, in the third study, there was evidence for observational learning in both conditions (i.e., evidence also for emulation in the ghost condition), but with stronger observational learning in the full demonstration condition 
. In sum, non-enculturated chimpanzees (henceforth simply chimpanzees) do not seem to copy pure actions without results information 
–but they also seem to be reluctant to copy pure results (see above). The underlying reason might be that a third factor may be responsible for these discrepant findings, and we believe this factor could be social.
In reviewing the ghost condition literature in chimpanzees, we noticed that these studies systematically differed with respect to social factors, which might explain their conflicting findings. Besides actions and results, there is a third–social-type of information that observers may learn about and copy during demonstrations: goal information 
. Goal information describes the state of the world that the demonstrator tries to achieve. Ghost control studies typically lack such goal information (one cannot gather goals from ‘ghosts’). Recent studies suggest that chimpanzees may be able to perceive more about goals than was previously thought 
. In the light of such recent findings, it is conceivable that the absence of this type of information may prove detrimental to the observational learning process for chimpanzees, and if that is the case, it is to be expected that chimpanzees' ability to copy will be negatively affected in ghost conditions. In addition to the potential lack of goal information, previous ghost condition studies have also lacked social presence during the demonstration phase. Yet, having a conspecific present during ghost demonstrations may enhance learning by way of social support 
. Social ghost conditions may act as general motivation enhancers; the only ape study to provide evidence for emulation in a ghost condition found copying only in its social
ghost condition (i.e., “enhanced ghost condition” 
). Finally, in those cases where a conspecific is present during demonstrations, it may matter whether there is a separation between observer and demonstrator. For example, in Tennie et al., the demonstrator and observer were separated by a glass/mesh during full model demonstrations, which may have led to this study's negative finding 
. In both Hopper et al. studies, there was no conspecific present in the ghost condition, and no copying was found there 
. However, in the Hopper et al. “enhanced” ghost condition (i.e., “ocial”ghost control 
) a conspecific was present in the same room as the observer (Lydia Hopper, pers. comm.), and it was here that clear evidence for emulation was found in chimpanzees. The reason that little evidence of emulation (i.e., only in the 1st trial) was found even in this “nhanced”condition might also be simply because in this condition the demonstrators did not directly interact with the apparatus, and thus probably did not transmit any goal information. To summarize, it is conceivable that these three social factors play a role in emulation learning in chimpanzees: 1) goal information, 2) the presence of a conspecific during demonstrations, 3) if a conspecific is present, physical proximity between conspecific and observer.
In the present study, we tried to include all of these social factors in our demonstration phases. We also aimed to avoid the potential pitfalls of studies that employ the “two-action task” procedure, especially the problem of triviality of task (e.g. 
). When using less trivial tasks, one can use two different approaches, the first of which would be to use tasks incapable of being solved by individual chimpanzees (i.e., where low-fidelity learning mechanisms alone do not help). These studies are interesting because they can uncover cases of cumulative culture. Currently, only one such study has been conducted, and the four species of great apes tested failed to show evidence of such copying 
. A second approach would be to use tasks in which most, but not all, chimpanzees fail during baseline trials–an approach chosen for this study. In these experiments one can use the few successful subjects as “natural demonstrators” since they learned the technique during baseline trials and do not need to be trained further (it should be noted that such a practice may introduce a bias against good inventors. However, this is less problematic as long as several experimental conditions are compared with each other. Here each condition then tests apes with comparable performances). The added benefit of this situation is that the demonstrated technique is potentially more representative of a real ape tradition, and hence more ecologically valid (see 
; see also discussion
). Here, we tested chimpanzee subjects in such a difficult (but not impossible) problem solving task (“floating peanut task”; see 
). This task consists of a Plexiglas tube mounted vertically to the mesh of a cage, with only the top end open. Shelled peanuts are placed inside the tube resting at the closed bottom. The peanut could not be extracted from the top unless subjects added water to the tube thus causing the peanut to float high enough so that it could then be extracted. Prior to our study, Hanus et al. tested 25 chimpanzees using this task on Ngamba Island, Uganda; as well as 19 chimpanzees at the WKPRC, Leipzig, Germany (total n
). Overall, in Hanus et al. 's study, only 7 subjects were successful, and all of these invented the solution either in their first or their second trial. Thus, in Hanus et al. 's study, subjects either learned early on in the trials or never at all, despite the fact that all subjects received four to eight trials.
We adapted the Hanus et al. study into a social learning experiment in order to ascertain whether chimpanzees are best described as emulators or imitators. All subjects were first tested in a baseline period, in which no previous information was provided to subjects (partly data from Hanus et al. and partly novel baseline trials established by us). Subjects then entered one of two experimental conditions: the full demonstration condition (providing information about actions, goals and results), or the emulation condition (“ater bottle”, providing only information about results and goals). In the full demonstration condition subjects witnessed a model pouring water from the mouth to the tube in order to get access to the peanut. In the emulation condition, subjects were shown how to solve the task by pouring water from a bottle into the tube. Thus, observers were required to produce the alternative, unobserved action (spitting water into the tube) in order to achieve the demonstrated result (i.e., making the peanut float up to the top with water).
By using these three conditions, we set out to disentangle the contributions of different learning processes potentially involved in the floating peanut task–as a model for behavioural traditions in chimpanzees in the field (e.g. 
). Comparison of the subjects' performances allowed us to do the following: a) measure the probability of innovation in these subjects over trials as a potential general means of solving the problem–and the rate of innovation was determined by baseline performance, b) measure the effects of different demonstration types compared to baseline performance (i.e., whether one or both demonstration types led to more solutions than had occurred during baseline; in other words, whether observational learning could help elicit the behaviour), c) determine the most plausible underlying learning mechanism (imitation or emulation) by comparing the effects of the two demonstration conditions. The underlying logic was that one type of demonstration (the full model; actions, goals and results) would only constitute an advantage if subjects were engaging in action copying (imitation) in order to learn the solutions. However, if no difference between the demonstration conditions could be found the most parsimonious explanation would be that subjects had made use of the same type of information in both conditions (i.e., results information (possibly spurred by goal information)–since this was the only type of information that was present in both experimental conditions).