The comparative study of empathy: sympathetic concern and empathic perspective‐taking in non‐human animals

While empathy is a century‐old psychological concept, its study in non‐human animals has become the focus of much recent scientific interest, as it promises to provide the clues to understand the evolutionary origins of our social and moral nature. A review of the comparative study of empathy is thus timely to complement and constrain anthropocentric views, and to integrate current findings. However, this is not an easy task. The study of animal empathy has developed using different paradigms, different concepts of the phenomena involved, and the absence of a systematic program. Herein, we carry out a comprehensive review of the literature on complex forms of empathy in non‐human animals: sympathetic concern and empathic perspective‐taking. In particular, we focus on consolation and targeted helping, as the best examples of each category. In so doing, we try to shed light on the current debate concerning whether these phenomena are exclusively human traits. First, we try to clarify the terminology and taxonomy of forms of empathy, providing operative criteria for these phenomena that are applicable to both human and non‐human animals. Second, we discuss whether the available evidence qualifies such behaviour as empathic. Third, we aim to provide an integrative view of the field, clarifying the challenges and conditions to satisfy. We also hope to highlight the importance of the study of these processes for elucidating the evolutionary history of this capacity across the animal kingdom.


I. INTRODUCTION
Rats preferring to free trapped companions than to eat chocolate chips, voles 'consoling' distressed partners, dolphins helping injured companions to breathe . . . reports like these have fascinated us since ancient times. In his Historia Animalium (Aristotle, 1897), the Greek philosopher Aristotle considered the temper of animals, highlighting the gentleness of the lion, the sensitivity of the elephant or the kindness of the dolphin in saving companions from fishermen or compassionately supporting dead calves to prevent them from being devoured. However, fear of anthropomorphism restricted research on non-human animal empathy for decades (Panksepp, 1998;de Waal, 1999). Recently the field has overcome this concern, stimulating public interest and promoting observational and empirical studies.
Although there is still a lack of consensus on the precise nature and definition of empathy, broad views of this term are increasingly being accepted and used, especially in the study of animal empathy. For example, Frans de Waal (2008, p. 281) defines it as: 'the capacity to be affected by and share the emotional state of another, assess the reasons for the other's state, and identify with the other adopting his or her perspective'. de Waal contends that even if a behaviour only satisfies the first element of this definition, it can still be considered empathic. This amounts to an incremental view of empathy, dividing this capacity into several interrelated processes of varying complexity, both cognitive and affective. This is a useful strategy for the study of homologous empathic behaviours in other species (Panksepp & Lahvis, 2011), as well as for assessing the phylogenetic continuity of this capability (Preston & de Waal, 2002). According to this conception, the simplest forms of empathy, such as emotional contagion (emotional state-matching of a subject with another; de Waal, 2008), should be found in a wide variety of species, while the most complex empathic behaviours, i.e. those involving not only emotional processes but also higher cognitive abilities, are not expected to be widespread across the animal kingdom. Sympathetic concern and empathic perspective-taking are usually considered to be the most complex forms of empathy (de Waal, 2008), comprising behaviours such as consolation or empathic targeted helping, which are traditionally treated as uniquely human. Recent results suggesting the occurrence of these high-level behaviours in species such as rodents call for a deeper review of the nature and mechanisms of these behaviours.
Even though research on empathic processes in animals, mostly focused on mammals and birds, has experienced a surge of interest in recent years, it has developed using multiple paradigms and concepts of the phenomena involved, and the lack of a systematic program. In addition, an overwhelming number of terms are used to refer to empathic-related behaviours, which lack common currency or understanding. Sometimes these terms are used interchangeably, assuming that they refer to a single process, while at others the same terms are used to differentiate between processes. Certain terms appear only in the human empathy literature while others are used to study similar behaviours in non-human animals. This lack of consensus complicates the comparative study of empathic behaviours. Another issue in empathy research is assessing the motivations underlying empathic behaviours. While it is not easy to distinguish between the motivations and consequences of a behaviour in humans, this task may be even harder in other animals. Although the motivational aspects of empathy are crucial, most studies on empathy in non-human animals do not provide data on the motivations underlying the observed or measured responses. We focus herein on the available data (reactions, responses and outcomes/effects), leaving aside the underlying behavioural motives in defining operative criteria to distinguish empathic from non-empathic behaviours in non-human animals. It is not our aim to propose new definitions of the terms 'sympathetic concern', 'empathic perspective-taking' and associated processes. Instead, we provide operative criteria that represent the best available approach for reviewing the existing evidence. By using these criteria, we can separate empathic behaviours from similar behaviours that are not empathy-based. We also carry out a comprehensive review of the literature on two complex forms of empathy in non-human animals: sympathetic concern and empathic perspective-taking. In particular, we focus on consolation and empathic targeted helping, as examples of these two major categories. First, we clarify the terminology used in this field and provide operative criteria of application. Second, we review the most relevant evidence and discuss whether these studies really offer examples of complex empathic behaviours comparable to those found in humans. We thus hope to provide an integrative view of the field, and discuss the challenges and conditions to be addressed.

II. SYMPATHETIC CONCERN
(1) Definition of sympathetic concern, personal distress and consolation Sympathetic concern has been defined in opposition to personal distress. Definitions by Batson, Fultz & Schoenrade (1987), de Waal (2008 or Eisenberg et al. (1989) underline that sympathetic concern is an emotional reaction elicited in response to the negative state of another, that motivates to improve its situation (a prosocial response), while Table 1. Definition and main components of consolation (sympathetic concern), personal distress and empathic targeted helping (empathic perspective-taking) Components 1. Reaction 2. Response 3. Outcome

Consolation (sympathetic concern)
-Other-oriented reaction in response to other's distress/situation/needs: moderate level of arousal; non-emotional match -Some level of emotional control regulation needed -Other-oriented response: attempts to ameliorate the other's state (approach responses) -Alleviation of the distressed party Personal distress -Self-focused reaction in response to other's distress/situation/needs: over-arousal; emotional state-matching -Emotional control regulation not needed -Self-focused response: attempts to ameliorate one's own distress (escape responses) -Alleviation of the individual's own distress Empathic targeted helping (empathic perspective-taking) -Other-oriented reaction in response to other's distress and after a cognitive appreciation of the situation: moderate level of arousal -Emotional control regulation needed -Flexible other-oriented response: fine-tuned help or care appropriate to the situation of the distressed party -Improvement of the situation of the distressed party personal distress involves a motivation to overcome such an emotional reaction. Sympathetic concern is seen as an other-oriented emotional reaction, which -in contrast to emotional contagion -does not necessarily mimic that of the distressed partner, and leads to a prosocial response in order to alleviate the other's negative state. Personal distress is a self-focused reaction, which mimics the aversive state of the distressed partner. This emotionally aversive over-arousal focusses the observer on its own negative state, generally triggering a non-prosocial response which aims to alleviate its own distress. Apparently, then, these processes respond to opposite motivations. While sympathetic concern is viewed as an other-oriented emotion fostering altruistic behaviour, personal distress is thought to be an ego-oriented reaction that motivates care for oneself, aiming to alleviate one's own distress, while disregarding the other's (Eisenberg, 2010). However, this distinction may be oversimplistic. Even if sympathetic concern involves an other orientation, it may also produce self benefits, such as punishment avoidance, aversive-arousal reduction or social reward seeking (Batson, 2010). Therefore, some authors claim that sympathetic concern is also self-interested. Conversely, personal distress behaviours, even if self-oriented, may benefit others as well. Herein, we do not consider underlying motives of behaviour in characterizing these phenomena. Instead, we attempt to operationalize sympathetic concern and personal distress in terms of the type of reaction and response triggered by the perception of others' distress and its main effects (Table 1). In this way, sympathetic concern and personal distress can be more easily distinguished: we can view sympathetic concern as a moderate emotional response elicited by the perception of another individual in distress, which gives rise to an other-oriented reaction; and personal distress as an over-arousal and intense negative emotional reaction that mimics that of the distressed party, and gives rise to a self-focused response. It is useful to make this distinction because the emotional reactions and behavioural dispositions involved in each case are different: an other-oriented response that alleviates the other's distress and perhaps the mild distress vicariously felt by the observer (sympathetic concern), or a self-focused reaction that generally elicits escape responses in order to move away from the source of distress (personal distress). Motivations play no role in drawing this distinction. Of course, there will always be the possibility of ambiguous cases. In some situations (for example, when the observer cannot escape from the distressed partner), personal distress can lead to prosocial effects if this is the easiest way to reduce the observer's negative emotional state (Eisenberg & Eggum, 2009). In such cases, it will be more difficult to differentiate personal distress from sympathetic concern; doing so would require experimental manipulation to change the context and to determine whether the reaction and response change accordingly.
It has also been claimed that an important element that differentiates sympathetic concern from personal distress is the respective level of effective emotional regulation. Some authors point out that one of the fundamental characteristics of sympathetic concern is the ability to distinguish, to some extent, between internally and externally generated emotions, that is, between personal and vicarious emotions (de Waal, 2008). This does not mean that an individual must perceive another's internal state (Batson, 2010), but be able to tell whether its affective state was triggered by its own, or somebody else's, situation. This ability allows individuals to respond differently depending on the source of the emotion and explains why only observers perform an other-oriented response, not the ones being observed. Development, extended life history and prefrontal functions, such as working memory, increase the flexibility and control of the emotional response, thus facilitating the emergence of sympathetic concern (de Waal, 2008). In general, human data point to a positive relationship between sympathetic concern and regulatory capacities and an inverse relationship between personal distress and emotional regulation (Eisenberg, 2002). Higher forms of inhibitory control are thought to depend upon the frontal lobe of the cortex. As prefrontal size correlates with emotional regulation skills in development and phylogeny (Preston & de Waal, 2002), sympathetic concern processes are expected to be displayed only in species with a well-developed prefrontal cortex (Preston & de Waal, 2002). However, it remains to be shown that the basic form of emotional regulation presupposed for sympathetic concern can only be achieved by means of advanced prefrontal abilities. It is in principle possible that the distinction between vicariously generated and personal emotions could be done in a more automatic way. Indeed, there is evidence that different neuronal populations are activated in the human anterior cingulate cortex (ACC) when experiencing or when observing pain (Morrison & Downing, 2007). This allows the possibility that different neuronal populations could trigger different automatic responses depending on the source of the emotion (vicarious or personal). If so, then sympathetic concern could be found in a wider range of animal species than possible if prefrontal control structures are necessary. Whether sympathetic concern can be mediated by different neural circuits should be subject to further investigation.
Consolation or comfort behaviour (consolation-like behaviour outside the post-conflict context) is considered to be a major example of sympathetic concern. de Waal & van Roosmalen (1979) first applied this term to non-human primates, defining it as contact of the aggressed party with a third animal. They distinguished consolation from reconciliation, which is contact between former opponents, although both interactions seemed to have a calming effect. Their definition excluded affiliative contacts between winners and a third party and did not consider which individual (aggressed or third party) initiated the affiliative contacts.
However, three types of post-conflict bystander affiliation can be distinguished: (i) appeasement -post-conflict contact offered by bystanders to aggressors (Romero, Castellanos & de Waal, 2011); (ii) solicited bystander/third-party affiliation -solicited affiliation by the distressed party to a bystander (Fraser & Bugnyar, 2010); and (iii) unsolicited bystander/third-party affiliation -post-conflict affinitive contact spontaneously directed to victims by bystanders (Palagi & Cordoni, 2009). Each of these post-conflict contacts might have different functions: (i) although appeasement could also alleviate the aggressors' post-conflict stress, it is not usually considered as an example of consolation since it may function to reduce aggressive tendencies in the aggressor. (ii) This behaviour would provide direct benefits to bystanders by preventing them from becoming the target of redirected aggression. If so, appeasement could be qualitatively different from affiliation between bystanders and victims (Romero et al., 2011). (iii) To alleviate post-conflict distress: victims of aggression are expected to experience higher levels of anxiety and more distress after the conflict than aggressors, and hence, they are likely to have a greater need for consolation (Romero et al., 2011). Solicited bystander contacts may be functionally similar to unsolicited ones in that they could also reduce post-conflict distress (Palagi & Cordoni, 2009). But the underlying mechanism is likely to differ from consolation in that it may not require empathy since the initiative is not taken by the bystander but by the distressed party (Fraser & Bugnyar, 2010). Solicited affiliative contacts could range from the mere approach of the victim to the bystander to the emission of specific signals by the distressed party (Cordoni, Palagi & Tarli, 2006). These solicitation signals carry a specific meaning: approach the sender. Therefore, solicited third-party affiliation simply requires that the bystander tolerates the approach of the other individual or recognizes the specific approach signal emitted by the other animal, irrespective of whether it is distressed or not. On the other hand, to be considered true consolation, the uninvolved party must spontaneously realize that the other needs reassurance and provide it. That is, the prosocial response should be driven by the perception of the other's distress and not by an explicit request from the distressed party for attention and support from a bystander.
Unsolicited bystander affiliation is expected to be performed by individuals who share a close relationship with the distressed party, as they are more likely to be responsive to the other's distress (Fraser & Bugnyar, 2010). Much attention has been paid to the role of familiarity in consolatory responses because sympathetic concern is related to empathy and, according to the literature, familiarity has robust effects in processes facilitating empathy in several contexts (Preston & de Waal, 2002).
In the field of animal empathy research, the term consolation thus has come to define a specific type of unsolicited third-party affiliation characterized by the alleviation of the victim's distress and the directionality of the bystanders' affiliations, directed mostly towards familiar individuals (kin or friends) (Palagi et al., 2014). Furthermore, this view assumes that consolation is based on empathy. Therefore, some authors still prefer the ethology descriptive tradition of using the more neutral concept 'unsolicited third-party affiliation' (Palagi et al., 2014). The above definition of consolation excludes reconciliation, appeasement and solicited affiliation, considering only the contacts of bystanders towards victims of aggression.
(2) Sympathetic concern and consolation studies in non-human animals We consider a behaviour to be an example of sympathetic concern if it involves the following three components: (i) a moderate other-oriented emotional reaction in response to another's distress, situation or needs that does not mimic that of the distressed party; (ii) a prosocial response including attempts to ameliorate the other's state (approach responses); (iii) alleviation of the distressed party (Table 1). We focus first on features of the emotional reaction triggered by the perception of another's distress. Both behavioural and neurophysiological data give us information about the type of response. For example, in sympathetic concern, the behavioural displays of the bystander are expected to be less overt that those of the distressed party (moderate other-oriented emotional reaction). The bystander's anxiety indicators (e.g. level of stress hormones) also should be lower than those of the stressed individual. Secondly, we can assess the responses of the observers: sympathetic concern responses must be directed towards the distressed party. And third, we consider whether this response alleviates the other's distress. We can test for stress alleviation by observing the termination of agonistic or stress behaviours in the distressed party after the occurrence of a prosocial response or/and by physiological measures that test for decreases in physiological indicators of anxiety.
From Zahn-Waxler's seminal studies on the development of concern in children (Zahn-Waxler, Radke-Yarrow & King, 1979;Zahn-Waxler, Hollenbeck & Radke-Yarrow, 1984;Zahn-Waxler & Radke-Yarrow, 1990;Zahn-Waxler et al., 1992), the study of sympathetic concern processes in humans and non-human animals, has been almost exclusively focused on consolation behaviour, as it is the main example of this category (de Waal, 2008). Therefore, here we review only consolation studies. Apart from the involvement of a third party, the crucial feature of consolation or comfort behaviour is its distress-alleviation effect. When this effect cannot be established, then we resort to the common practice of referring to unsolicited third-party affiliation (within the post-conflict context). As the study of consolation in non-human animals is mostly focused on unsolicited bystander affiliative contacts towards victims of aggression, there are few data available on consolatory behaviours outside the post-conflict context.

(a) Non-human primates
Consolation has been studied in detail in chimpanzees, starting from de Waal & van Roosmalen (1979). They analysed hundreds of post-conflict contacts in a semifree-ranging colony of chimpanzees (Pan troglodytes) and pointed out that, after agonistic interactions, chimpanzees usually engaged in non-violent body contacts. de Waal & van Roosmalen (1979) distinguished between reconciliation among former opponents, and consolation, which consisted of contacts of an uninvolved individual towards the aggressed party. They stated that such interactions seemed to serve an important distress alleviation function in the short term, and a longer-term social homeostasis function. Because of the short duration of the post-conflict observation period and the lack of control data, de Waal & Aureli (1996) replicated this study. They proposed that, if post-conflict affiliative contact with bystanders alleviates the distressed party, three predictions follow: (i) more affiliative contacts are expected within a few minutes of the conflict than after longer time periods or during control periods; (ii) more contacts are expected following intense aggressive encounters (given that the distress of the victim is likely to be proportional to conflict intensity); and (iii) more contacts are expected between bystanders and victims of aggression than between bystanders and aggressors. Their data on chimpanzees supported the results of de Waal & van Roosmalen (1979) and met their three predictions. Furthermore, they observed that bystanders often did not show overt signs of distress before approaching the victim, suggesting that the main role of this behaviour is not the bystander's comfort but alleviation of the distressed party. These results led the authors to conclude that chimpanzees console victims of aggression. Recent work has provided further evidence for the existence of this type of behaviour in chimpanzees showing that, in captivity, third-party affiliation occurred more often between individuals that were socially close (kin and affiliation partners) (Fraser, Stahl & Aureli, 2008;Romero, Castellanos & de Waal, 2010;, was more typical of females  and reduced behavioural measures of stress in recipients of aggression (Fraser et al., 2008). However, studies with wild chimpanzees do not provide corroborative evidence: in one wild group third-party affiliation was found to be very rare (Arnold & Whiten, 2001), and in another, third-party affiliation was observed but was delayed compared to captivity data (Kutsukake & Castles, 2004).
Several studies have described the occurrence of unsolicited third-party affiliations after conflicts in other ape species: bonobos (Pan paniscus) (Palagi, Paoli & Tarli, 2004;Clay & de Waal, 2013;Palagi & Norscia, 2013) and gorillas (Gorilla gorilla and Gorilla beringei) (Watts, 1995a,b;Cordoni et al., 2006;Mallavarapu et al., 2006). In bonobos, close relationships positively affected only unsolicited affiliation rates along an empathic gradient, being mostly offered to kin, then friends and, finally to acquaintances (Palagi et al., 2004;Clay & de Waal, 2013;Palagi & Norscia, 2013). Unsolicited bystander affiliation effectively reduced the distressed party's anxiety in bonobos (measured through self-scratching behaviour, an indicator of anxiety) (Palagi & Norscia, 2013). Data on unsolicited bystander affiliation in gorillas are not conclusive due to fewer studies and inconsistent results between wild mountain gorillas (Watts, 1995a,b) and captive groups of western gorillas (Cordoni et al., 2006;Mallavarapu et al., 2006). Overall, unsolicited third-party affiliation has been reported for both wild and captive gorillas but the extent to which this behaviour leads to stress alleviation in the distressed party is not yet clear.
de Waal & Aureli (1996) compared their chimpanzees' results with data on post-conflict behaviour of several species of macaques in the wild and in captivity, and found important differences in unsolicited bystander affiliation. Unlike chimpanzees, affiliative contacts between macaque bystanders and victims of aggression did not occur more often following a conflict that during control periods. Instead, macaques avoided the scene of a conflict in which they were not involved. Other studies have also failed to find positive evidence of unsolicited third-party affiliation as a post-conflict consolatory mechanism (reviewed by Watts, Colmenares & Arnold, 2000). These negative results led to the conclusion that consolation behaviour is virtually absent in monkeys. Recent studies have challenged this, reporting the existence of unsolicited bystander affiliation in stump-tailed macaques (Macaca arctoides) (Call, Aureli & de Waal, 2002), mandrills (Papio anubis) (Schino & Marini, 2012) and Barbary macaques (M. sylvanus) (McFarland & Majolo, 2012). In the latter study bystanders initiated post-conflict affiliations more frequently with victims with whom they shared high-quality relationships than with other individuals, just as in great apes and humans. None of these three studies found evidence of stress-alleviation effects of bystander affiliation. Palagi et al. (2014) reported such an effect in Tonkean macaques (M. tonkeana), one of the most tolerant macaque species, but not in the despotic Japanese macaque (M. fuscata). In Tonkean macaques, unsolicited third-party affiliation reduced indicators of anxiety (scratching) in the conflict victims and was mainly directed towards friends (measured by the amount of grooming exchanged between individuals), especially towards females who experienced higher levels of anxiety compared to males. These results support the 'Social Constraints Hypothesis' (de Waal & Aureli, 1996), which proposes that unsolicited bystander affiliation depends on the social structure of the species, being more likely to be found in tolerant than in despotic species since the risks of further aggression for potential bystanders are lower (Fraser et al., 2009).
Thus, unsolicited third-party affiliation in chimpanzees (Fraser et al., 2008), bonobos (Clay & de Waal, 2013;Palagi & Norscia, 2013) and Tonkean macaques (Palagi et al., 2014) reduced behavioural indicators of stress in the victims, satisfying a crucial feature of consolation: distress alleviation. It also met the familiarity bias criterion, being offered mostly to familiar conspecifics. However, in the case of bonobos and Tonkean macaques, no data were collected assessing features of the observers' reaction towards the distressed victim. As noted above, this is crucial to distinguishing personal distress from sympathetic concern. For chimpanzees however, observers' reactions are available. de Waal & Aureli (1996) reported that bystanders often did not show overt signs of distress before approaching the distressed individual and waited until the most intense displays of distress stopped. While de Waal & Aureli (1996) did not detail subtle indicators of emotion (behavioural or physiological) in the bystander, their observations of its reaction towards the distressed party suggest that chimpanzees have some inhibitory control over their emotional reactions, so that their approaching responses were not due to personal distress, but to sympathetic concern. Thus, to date, chimpanzees are the only species that show consoling behaviour according to the three proposed criteria. There are only two studies (Koski & Sterck, 2007;Fraser et al., 2008) that assess stress-alleviation effects of third-party affiliation in chimpanzees. These studies used behavioural indicators of stress, not psychophysiological ones. Furthermore, Fraser et al. (2008) found positive evidence for a stress-alleviating effect of third-party affiliation, whereas Koski & Sterck (2007) did not. Therefore, it would be of great interest to measure not only behavioural but also physiological parameters in order to characterize the emotional reaction of the bystander better and the effectivity of the observer's responses in alleviating the victim's distress. Such research, especially in wild animals, could clarify the occurrence and effects of this type of behaviour in chimpanzees and other primate species and outside post-conflict contexts.

(b) Canids
Wolves (Canis lupus) and dogs (Canis familiaris) engage in post-conflict third-party affiliation and the dynamics of this behaviour show a strong similarity with that observed in great apes (Palagi & Cordoni, 2009). Wolves show solicited and unsolicited bystander affiliation (Palagi & Cordoni, 2009;Baan et al., 2014). Unsolicited third-party contacts occurred between friends (measured by body contact rates) (Palagi & Cordoni, 2009). In addition, third-party affiliation provided immediate benefits to the victims reducing the probability of renewed aggression and restoring social cohesiveness, suggesting that unsolicited affiliation with a third party buffers social tension in the group. In dogs, third-party affiliations were more often directed towards the victim of a conflict than towards the aggressor (Cools, Van Hout & Nelissen, 2008) but the influence of familiarity in these post-conflict contacts could not be assessed.
Outside the post-conflict context, Custance & Mayer (2012) evaluated the reactions of dogs towards apparently distressed humans. Dogs were exposed to four conditions: two in which their owner or a stranger pretended to cry, and two in which their owner or a stranger hummed a song. All conditions were preceded by a 2-min period during which the owner and stranger chatted. Dogs oriented significantly more often towards the person (looking at, approaching and touching them) during the crying condition than during the humming or talking periods regardless of the person's identity. Dogs approached the crying person in a submissive way, producing in some cases mild distress vocalizations. When the stranger was pretending to cry, dogs sniffed, nuzzled and licked them (instead of seeking out their owner for comfort), even if the stranger ignored them. This evidence is inconclusive since, as suggested by Custance & Mayer (2012), the dogs may have previously been positively reinforced for approaching crying individuals and thus, their response could be the result of previous learning rather than spontaneous comfort behaviour.
Quervel-Chaumette et al. (2016) assessed whether dogs displayed comfort behaviours towards familiar conspecifics (dogs that lived together) after being exposed to conspecifics' distress vocalizations or control sounds. The distress vocalizations consisted of whines produced by their familiar partners or by stranger dogs. Dogs presented higher behavioural alertness, more stress-related behaviours, and an increase in comfort behaviours towards the familiar dog (longer time spent in proximity and expressing affiliative behaviour) after being exposed to conspecifics' whines than when exposed to control sounds. Dogs displayed more comfort behaviours towards their partner after hearing familiar whines compared to stranger whines. Furthermore, familiar whines tended to maintain a dog's cortisol (a stress-related hormone) concentration at a higher level than stranger's whines.
Overall, none of the above studies provided clear positive evidence of a distress-alleviation effect of third-party affiliation in dogs or wolves. Thus, for the moment, we cannot talk of consolation in these species. However, canids seem to be affected by and respond to other's distressing situations, providing unsolicited third-party contacts to conspecifics or even to members of other species like humans, instead of trying to alleviate their own stress. Furthermore, the results of Quervel-Chaumette et al. (2016) suggest that dogs' comfort behaviours were not just an automatic response driven by emotional contagion. After being exposed to whines, dogs opted to comfort their familiar partner without any solicitation or distress signal emitted by the other dog, instead of going to their owner for comfort. Future studies should examine the effect of unsolicited affiliation in canids and features such as the existence of a familiarity bias in those contacts.

(c) Elephants
Despite anecdotal reports describing possible empathic behaviours in elephants, empirical studies assessing the occurrence of emotional behaviours such as consolation are scarce. Bates et al. (2008) used observational reports collected over a 35-year period in Kenya with African elephants (Loxodonta africana). These reports suggest that adult females tend to comfort immature individuals, more precisely calves under two years old. Comfort behaviour was preceded, in most cases, by the calf giving a distress vocalization or displaying alarm postures. In some situations, calves received comfort without displaying any obvious distress cues. In all but one case, the calf and the adult shared a valuable relationship. Plotnik & de Waal (2014) studied affiliative tendencies in Asian elephants (Elephas maximus) after a distressing event. They assessed the occurrence of other-regarding responses to an individual's distress irrespective of its cause. Results showed that elephants affiliated significantly more with a conspecific through directed physical contact and vocal communication following a distress event than in control periods. But, in many cases, the authors could not identify a clear stimulus for a distressing event. Therefore, it was difficult to distinguish whether the individuals were reacting directly to the stimulus or to the conspecific's distress. However, the researchers contend that the directionality of the contacts and their nature strongly suggest attention towards the emotional states of other elephants and that this behaviour could be akin to consolation responses observed in apes.
Research on sympathetic concern behaviour in elephants is still in its infancy. For the moment, the few existing studies suggest that elephants react to perceived distress cues in their conspecifics providing in some cases comfort-like behaviours. It remains to be further clarified whether this response is empathically mediated and has the effect of reducing the stress of the receptor.

(d) Cetaceans
Third-party affiliation with victims of conflicts has been reported in bottlenose dolphins (Tursiops truncatus) (Tamaki, Morisaka & Taki, 2006;Yamamoto et al., 2015) but its effects and main features, such as the direction of affiliative contacts, require further research. Outside post-conflict contexts, there are anecdotal instances of cetaceans appearing to comfort distressed individuals. Pace, Miragliuolo & Mussi (2005) provided a systematic report on this type of interaction. They observed a group of five sperm whales (Physeter macrocephalus), entrapped in a drift net during a two-day period. All the animals showed clear signs of stress and had numerous lesions on their body, in particular on their tails. Divers managed to free two individuals on the first day. These individuals remained near the trapped whales watching the divers' work, touching the entangled tails and rubbing their sides with the melon. The function of tactile contacts in cetacean species has not been studied in detail but it may have an affiliative role analogous to that of allogrooming in primates (Dudzinski et al., 2009). For example, Herzing (1996) reported that a spotted dolphin calf (Stenella frontalis) repeatedly emitted distress calls until another animal calmed him by pectoral rubbing. During post-conflict periods, flipper-rubbing among bottlenose dolphins seemed to ease tension and restore friendly relationships between former opponents (Tamaki et al., 2006). Thus, the rubbing behaviour provided by the free individuals towards the distressed ones may have a comforting effect and be displayed by cetaceans towards distressed individuals. It would be of great interest if more systematic studies assessed comfort behaviour and the possible consolatory effect of rubbing in cetaceans.

(e) Rodents
In rodents, behaviours likely to relieve pain in conspecifics such as nursing, grooming, licking or physical proximity, have been observed in studies assessing the emotional responses of individuals to another's distress (Walker et al., 2003;Knapska et al., 2010;Atsak et al., 2011). For example, Knapska et al. (2010) observed that rats (Rattus norvegicus) spent significantly more time directing allogrooming (social behaviour considered to have an investigative and affiliative role; towards partners that had undergone an electric shock than towards non-shocked individuals. Atsak et al. (2011) noticed that rat witnesses that had previous experience with foot-shocks spent significantly more time in the area close to a distressed rat than any other witness groups (non-shock groups or non-experienced groups) during and after the shock period. In a first attempt to test consolation-like behaviours empirically in rodents Burkett et al. (2016) compared the spontaneous responses of two species of voles, the social and monogamous prairie vole (Microtus ochrogaster) and the solitary and promiscuous meadow vole (M. pennsylvanicus), towards stressed conspecifics (which received foot-shocks). Prairie voles spent significantly more time directing allogrooming towards distressed familiar partners (mates or siblings) than to control individuals (non-shocked), but not towards stressed strangers. This affiliative pattern was also observed in a previous study on social buffering of the stress response in prairie voles (Smith, Lieberwirth & Wang, 2013). Conversely, meadow voles did not show differences in the duration of allogrooming based on the state of their partner. Anxiety behaviour was normalized in distressed prairie voles that interacted with observers, pointing to an alleviation effect of the observers' affiliative behaviours. According to Burkett et al. (2016), the combination of an increase in prairie voles' directed affiliation with a social-buffering effect supports the designation of this behaviour as a consolation behaviour.
However, this study has important limitations (Pérez-Manrique & Gomila, 2016): only the responses of male meadow voles were tested despite potential gender differences in the allogrooming response. In fact, female and male meadow voles' responses towards familiar individuals differed in a previous study (Ferkin, 1988). Ultrasonic vocalizations (USVs) were not controlled, so that it cannot be excluded that the affiliation was solicited. In addition, the affiliative responses of observer voles might be a by-product of the observer's mechanism of coping with its own distress. As discussed in Section II.1, in situations in which the observer cannot escape from the distressed party, like the voles in these two experiments (tested in small cages), personal distress could lead to apparently prosocial behaviours, if this is the easiest way to reduce one's own stress. In fact, in the study of Burkett et al. (2016), bystanders separated from the stressed party by a clear barrier had significantly higher levels of plasma corticosterone which correlated strongly with that of the distressed vole. This is indicative of physiological state-matching. This stress reaction was absent in observers in full contact with distressed voles, suggesting that the contact with the distressed party served to alleviate the observer's own stress. Finally, meadow voles undergo seasonal changes in their social structure: in winter (short photoperiods) they become socially tolerant, often participating in communal nursing and rearing of pups (especially females) (Anacker & Beery, 2013). These seasonal behavioural changes are associated with changes in oxytocin receptor levels and influenced by oxytocin administration (Parker et al., 2001;Beery & Zucker, 2010;Anacker & Beery, 2013). However, Burkett et al. (2016) only tested male meadow voles under long photoperiods, i.e. when they are expected to show lower levels of social tolerance. Therefore, the possibility that meadow voles display affiliative responses towards distressed individuals cannot yet be ruled out. Assessing the existence of these responses in meadow voles (males and females) under short photoperiods would help to clarify the nature of voles' affiliative responses towards distressed individuals. It would also be of great interest for investigating the function of consolation-like responses as an adaptive mechanism facilitating group living.
In summary, evidence of sympathetic concern in rodents is still highly limited but research on emotional processes in these animals emerges as a promising field of study allowing the use of several experimental paradigms and controls that cannot be applied to other species.

(f ) Corvids
Two studies have found strong evidence for unsolicited third-party affiliation in social species of birds: rooks (Corvus frugilegus) (Seed, Clayton & Emery, 2007) and ravens (Corvus corax) (Fraser & Bugnyar, 2010). During the post-conflict period, rooks engaged in third-party affiliation employing a specific behaviour named 'bill twining', in which two birds interlock their beaks. Both former opponents and uninvolved bystanders initiated the affiliative contacts. These behaviours appeared significantly more during the post-conflict period, suggesting that the use of this behaviour could serve to advertise third-party affiliation to other members of the group. In ravens, Fraser & Bugnyar (2010) found that post-conflict affiliation offered to the victim was more likely to occur if the bystander shared a valuable relationship with the victim and after intense conflict. Subsequent studies (Logan, Emery & Clayton, 2013a;Logan, Ostojić & Clayton, 2013b) using a slightly different methodology, showed that jackdaws (Corvus monedula) and Eurasian jays (Garrulus glandarius) engage in third-party affiliation initiated by both combatants and bystanders. Colonial jackdaws showed third-party affiliation with mates; despotic and territorial jays with any individual, not just mates. This result partially fitted the authors' hypothesis that jackdaws will show third-party affiliation but not jays since these birds lack high-quality relationships outside of the breeding season, when they were tested. Young jays were involved in most of the conflicts suggesting that aggression and post-conflict affiliation resulted from the formation of pairs and the establishment of a dominance hierarchy, so that once it has been established, conflicts and post-conflict affiliation disappear.
However, the observed pattern of third-party affiliation in these birds is very similar to that of primate species, suggesting that they could be sensitive to the emotions of others. At present we have no evidence for a consolatory effect of this post-conflict behaviour in corvids. Furthermore, other main features defining consolation behaviour are missing, such as specific data on the directionality of third-party affiliation in jackdaws and jays or the nature of the reaction of the bystander in these affiliative interactions.

(g) Parrots
Budgerigars (Melopsittacus undulatus), a social species of Australian parrot, also show post-conflict third-party affiliation initiated by both bystanders and combatants (Ikkatai, Watanabe & Izawa, 2016). The occurrence patterns of initiated bystander affiliation were specific to the post-conflict context. Third-party affiliations were initiated significantly more often by males towards loser partner females but less towards winner females than would be expected outside the post-conflict context. Conversely, observer females did not behave in this manner, initiating affiliations more often with winner partner males. Moreover, combatants tended to initiate affiliations with partners but loser males suppressed this behaviour. According to the authors, these gender differences in post-conflict behaviour could be rooted in sex differences regarding socio-behavioural features, like female dominance over males. Therefore, third-party affiliation in budgerigars seems to be a more fixed behaviour dependent on specific signals and gender than true consolation behaviour.

III. EMPATHIC PERSPECTIVE-TAKING (1) Definition of empathic perspective-taking and empathic targeted helping
Empathic perspective-taking is a term traditionally used to refer to an ensemble of complex cognitive and emotional abilities including perspective-taking, self-awareness, mental-state attribution, emotional-responding and well-developed emotional control regulation. For example, de Waal (1996) defined it as the capacity to take another's perspective (understanding another's specific situation and needs as different from one's own) combined with vicarious emotional arousal. It has also been said that a heightened self-awareness allows individuals to relate to the other's emotional state without losing sight of the actual source of this state (de Waal, 2008). Animals capable of self-recognition could have an introspective awareness of their own mental states and be able to adopt other's perspective (Decety & Moriguchi, 2007). So, in theory, animals endowed with these capacities possess the requirements to not only respond emotionally towards an individual in distress, but also to carry out a cognitive appreciation of the situation and the specific needs of the distressed individual, acting accordingly within the relevant context.
Empathic targeted helping is considered to be the primary example of empathic perspective-taking (de Waal, 2008). It has been defined as 'help and care based on a cognitive appreciation of the other's specific need or situation' (de Waal, 2008, p. 285). By definition, targeted helping requires the animal to understand, to some extent, the others' needs or goals given the situation, in order to help effectively (Yamamoto, Humle & Tanaka, 2012). The thorny issue is how to distinguish real empathic targeted helping from simpler forms of empathy. A way to address this issue is to focus on the flexibility of the prosocial response: to qualify as targeted helping, a behaviour must be fine-tuned to the situation of the distressed individual, and change as the situation changes. We can assume then, that empathic targeted helping involves the combination of sympathetic concern with a cognitive appreciation of the specific situation (de Waal, 2008).
Targeted helping can appear in two forms: rescue behaviour and epimeletic behaviour. The former is defined as 'the rescue of one animal by another' (Nowbahari et al., 2009, p. 1) and it involves four components: (i) the victim is in distress (is in a situation that implies immediate physical risk to itself); (ii) the rescuer's behaviour is suited to the circumstances of the victim's distress; (iii) the rescuer places itself at risk by engaging in this behaviour; and (iv) the rescue is not inherently rewarding to the helper [see Nowbahari & Hollis (2010) for a deeper look at rescue behaviour]. Epimeletic behaviour is also known as care behaviour and is characterized as 'intra-, interspecific, or intergeneric behaviour that involves attention or assistance' (Lodi, 1992, p. 284) or 'helping and caregiving directed to distressed individuals' (de Waal & Aureli, 1996, p. 81). It has been called 'nurturant' if attention or care is directed towards young individuals, and 'succorant' if directed towards adult individuals in distress (Caldwell & Caldwell, 1966). Examples of epimeletic behaviour are assisting mobility, care giving or assisting ill, disabled or stunned individuals, and pre-emptive protection. This last behaviour is defined as protection provided by an individual before any harm has come to a conspecific (Bates et al., 2008). In these cases, the protector may be empathic in the sense of predicting the distress that the other will suffer if the situation is not stopped (Bates et al., 2008). It might be argued that, in some cases, consolation could be considered as an example of epimeletic behaviour as well (de Waal & Aureli, 1996). Such is the case if reassurance contacts are the most adequate response that an observer could provide to a distressed party in a specific situation.
The study of well-developed cognitive capacities in non-human animals is fraught with difficulties and controversies. Therefore, we will assume as a working hypothesis that perspective-taking can take place in different ways, so that a wider set of social species can be considered.
(2) Empathic targeted helping studies in non-human animals According to Table 1, a behaviour should meet the following three requirements in order to be considered an example of targeted helping: (i) a moderate emotional reaction towards another in distress; (ii) a fine-tuned help response appropriate to the situation of the distressed animal, which should lead to (iii) the improvement of the situation of the distressed animal.

(a) Non-human primates
There are many anecdotal reports describing situations where non-human primates, particularly chimpanzees, seemed to help others (de Waal, 1996). However, due to ethical and methodological issues most of the available experimental studies on helping behaviour in primates are focused on instrumental helping and other-regarding preferences. Instrumental helping, a behaviour performed by an individual that enables another to reach a goal that cannot be achieved otherwise (Greenberg et al., 2010), differs from empathic targeted helping in that the latter always requires some emotional involvement of the observer, which reacts towards the perceived distress of the other individual. Both instrumental helping and empathic targeted helping require a cognitive evaluation of the other's situation and may require other-regarding preferences. Concern for the welfare of others (other-regarding preferences), refers to whether an individual is sensitive to another's pay-offs (Yamamoto & Tanaka, 2010). Helping others at a very low or no cost to oneself is an example of this capacity. Experimental studies on other-regarding preferences in non-human primates have provided mixed results depending on the context and paradigm used. In chimpanzees, studies testing other-regarding preferences in a food context have produced mostly negative evidence (Silk et al., 2005;Jensen et al., 2006;Vonk et al., 2008;Amici, Visalberghi & Call, 2014). However, other studies have shown that chimpanzees do seem to help others even in contexts where there is food involved (Warneken & Tomasello, 2006;de Waal, Leimgruber & Greenberg, 2008;Yamamoto, Humle & Tanaka, 2009;Greenberg et al., 2010;Horner et al., 2011;Melis et al., 2011;Yamamoto et al., 2012;Claidière et al., 2015). In some studies, orangutans (Pongo abelii, Pongo pygmaeus) (Amici et al., 2014;Kim et al., 2015), bonobos, western gorillas, spider monkeys (Ateles geoffroyi) (Amici et al., 2014), cotton-top tamarins (Saguinus oedipus) (Cronin et al., 2009;Stevens, 2010), Japanese macaques (Burkart & van Schaik, 2013), and capuchin monkeys (Cebus apella) (Burkart & van Schaik, 2013;Drayton & Santos, 2014) did not show other-regarding preferences. However, marmosets (Callithrix jacchus) (Burkart et al., 2007;Burkart & van Schaik, 2013), capuchin monkeys Lakshminarayanan & Santos, 2008;Brosnan et al., 2010;Takimoto, Kuroshima & Fujita, 2010;Suchak & de Waal, 2012;Claidière et al., 2015) and bonobos (Tan & Hare, 2013) were found to show other-regarding preferences under different experimental conditions. In the context of instrumental helping, chimpanzees spontaneously helped an unrelated conspecific, or a human, irrespective of being rewarded or whether helping was costly (Warneken & Tomasello, 2006;Warneken et al., 2007). These apes also showed flexible helping depending on conspecifics' needs and this helping behaviour varied depending on the situation (Yamamoto et al., 2009(Yamamoto et al., , 2012Melis et al., 2011). Some studies emphasize the importance of a request behaviour for prompting this type of behaviour (Yamamoto et al., 2009(Yamamoto et al., , 2012 and others showed that chimpanzees seldom help others without a direct request (Greenberg et al., 2010). A study with capuchin monkeys (Cebus apella) (Barnes et al., 2008) demonstrated that they are able to instrumentally help human experimenters in the absence of a reward but failed to take into account the perspective of others when they expect to get food for themselves. Furthermore, they prioritized the potential cost to themselves over the benefit to the conspecific that needs help. Capuchin monkeys also failed to help a conspecific in a minimal-cost instrumental helping task, not sharing tokens that allow the other monkey to obtain food from an apparatus (Skerry, Sheskin & Santos, 2011).
Apart from the above data on instrumental helping and other-regarding preferences, there is a lack of systematic studies on empathic targeted helping per se in non-human primates. Pruetz (2011) provides an anecdotal report which could be one of the few detailed cases of empathic targeted helping in chimpanzees in the wild. Over two days, an adolescent male was seen helping an injured mother carry her infant. The mother was unable to keep up with the rest of the group and stopped frequently, leaving her infant on the ground each time. The young male picked up the infant and carried her for approximately one-third of the distance the group moved. He was not observed to carry the baby again during a subsequent 8-week period. Furthermore, this adolescent had never been observed to carry this or any other baby before this incident. In this community of chimpanzees, other males had been observed to carry infants but never for the duration shown by this young male. The male was not matrilineally related to the mother but the cohesion of this community is three times greater than that reported for chimpanzees at other study sites. Pruetz (2011) highlights that the issue of concern here is the recognition of the mother's state and her needs by the male. Neither the mother nor the infant emitted emotional signals to indicate distress, so the young male apparently perceived the difficulty of the mother and anticipated her goal. Vogel & Fuentes-Jiménez (2006) described a case of rescue behaviour in white-faced capuchin monkeys (Cebus capucinus). A mother and her baby were rescued by the beta male of their social group after being isolated by male attackers of another group that were attempting to grab the infant. The beta male appeared soon after the female emitted alarm calls, and the attackers left the female and infant to chase him. Marzec et al. (2016) reported a striking case of possible rescue behaviour in wild orangutans (Pongo pygmaeus wurmbii). An old female with her infant received protection from an unrelated male while she was being attacked by a young female which, in turn, received active support from another male. The defender male could have been attracted by the screams of the victim. The intervention of this male reduced the physical aggression directed at the old female and he remained and protected the distressed orangutan and her infant during a 3-day period.
Other reports provide possible instances of 'pre-emptive' protection during mother-infant interactions. Hirata (2009) describes cases where mothers pull the infants away from potentially dangerous plants, and a situation in which a mother witnessed her infant playing with a chain and removed it from the infant. The infant was still clumsy in her motor actions so her hand or fingers could possibly have been damaged. It is usual that chimpanzee, orangutan and spider monkey mothers help young individuals to move from one tree to another (Carpenter, 1935;Bard, 1995;de Waal, 2008). Although in many cases the infant or the juvenile vocalizes, soliciting help, the mother seems to observe and analyse the situation, anticipating possible danger, the reason for the infant or juvenile's distress, and intervenes appropriately to address it. Other examples of possible nurturant or succorant behaviour are provided by caring for sick, injured or disabled individuals. In the Taï forest chimpanzees have been seen licking, cleaning and preventing flies from accessing the wounds of other individuals, sometimes for extended periods (Boesch, 1992). These chimpanzees seem to adapt their behaviour to the specific needs of the injured individual: dominant males prevent other group members from disturbing the wounded individual, and the whole group wait for the injured individual to begin to walk again (Boesch, 1992). There are also reports in which chimpanzees and Japanese macaques care for disabled infants (Turner, Gould & Duffus, 2005;Matsumoto et al., 2016). In such cases, mothers seemed to respond to the abnormal behaviours, limited physical abilities, and particular needs of the infant.
The lack of empirical and systematic observational data on empathic targeted helping in non-human primates prevents us from reaching robust conclusions. The available information suggests that at least some of the components listed in Table 1 for empathic targeted helping are present in apes' helping behaviour. Experiments on instrumental helping in chimpanzees and anecdotal descriptions suggest that these apes are able to make a cognitive evaluation of the situation and needs of other individuals, adapting their helping behaviour to the specific context. Helping behaviour in monkeys does not seem to be as flexible as that of chimpanzees. The observations of Pruetz (2011), Boesch (1992 and Marzec et al. (2016) also imply that great apes, at least chimpanzees and orangutans, understand the situation of others in distress, providing help in a flexible way. In these reports, the helping behaviour provided by chimpanzees improved the wounded individual's situation, allowing it to keep up with the group and reach the feeding site, and in the orangutan's case, the intervention of the male stopped the physical attacks towards the old female. These reports do not provide information about the helping individuals' emotional reaction towards the distressed animal. In particular, the young chimpanzee seemed to recognize the needs of the injured female without any emitted distress signal and the male orangutan may have been attracted by the screams of the old orangutan. It is thus difficult to establish the nature of these males' reactions towards the distressed females.

(b) Elephants
Elephants live in very closely bonded societies and their interest in ailing companions or in the carcasses and bones of dead conspecifics is well known (Bates et al., 2008). There is a remarkable number of anecdotal reports describing elephants assisting suffering conspecifics (Douglas-Hamilton et al., 2006). Elephants possess complex cognitive capacities (Plotnik, de Waal & Reiss, 2006), and it has been demonstrated that they are able to solve tasks acting cooperatively (Plotnik et al., 2011). Thus, it seems reasonable to hypothesize that their behaviour towards distressed conspecifics could be an example of empathic targeted helping. Observational reports of wild elephants include instances in which familiar or unrelated elephants assisted or tried to lift suffering companions (Schulte, 2000). For example, Douglas-Hamilton et al. (2006) reported the behavioural responses of different elephants and their families towards a dying matriarch. After the distressed animal collapsed, an unrelated matriarch rapidly approached her in some degree of excitement [tail raised, vocalizing and showing temporal gland secretion (related to stress, excitement or fear in elephants; Gorman, 1986)]. She tried to lift the matriarch and stayed with her even though her family was gone. These behaviours have also been observed in captivity, even between an Asian and an African elephant (Schulte, 2000), so this helping behaviour can occur irrespective of genetic similarity (Douglas-Hamilton et al., 2006). Mothers or other adult elephants have also been reported to help to lift calves when they could not stand by themselves, had fallen into ditches or could not climb in or out of rivers or swamps (Bates et al., 2008). In most cases the elephants' behaviour seemed to adapt specifically to the problem faced by the calves (Bates et al., 2008). Elephants protect young or injured individuals in potentially dangerous situations, when the victims are unable to defend themselves effectively (Bates et al., 2008). Most reports involve mothers protecting their calves or retrieving them, but there are also instances in which darted adults were guarded and helped by familiar or unrelated individuals (Douglas-Hamilton et al., 2006;Bates et al., 2008). There are also reports of 'pre-emptive' protection in which the individuals acted before any harm had come to the conspecific, such as stopping play fights between calves, pushing individuals or predators away from newborns, keeping offspring away from young individuals and preventing calves from moving into dangerous areas (Bates et al., 2008). The protector could be using its past experience in order to act effectively in each situation, rather than simply be responding to the direct perception of distress signals emitted by the calf, because in most of these instances distress signals were absent. Another behaviour that could be an example of empathic targeted helping is the removal of foreign objects from the body of other elephants. Bates et al. (2008) described three examples of this behaviour including the removal of a dart, a spear and a plastic bag. The authors claim that it could be possible that in situations where there is a visible wound or damage, or the object presents a potential danger, elephants are compelled to touch the object. They may be able to detect dangerous foreign objects using emotional and behavioural signals from the injured animal and this recognition triggers the action to remove them. However, interaction with foreign objects could be dangerous to the individual that removes or touches the object (Bates et al., 2008).
Some of the anecdotal data described above could be considered as positive evidence for the three components listed in Table 1 for targeted helping. In some situations, elephants react emotionally towards distressed conspecifics, as in the case of the dying matriarch described by Douglas-Hamilton et al. (2006). The helping animal displayed moderate signs of stress and temporal gland secretion, suggesting that the behavioural changes were accompanied by autonomic modifications driven by the perception of distress in the other individual and thus, fitting the Biological Reviews 93 (2018)  first requirement. Regarding the second component, these reports show that elephants responded appropriately in many situations, adjusting their behaviour to the specific case and, more surprisingly, seeming to anticipate the future consequences of potentially dangerous situations (Bates et al., 2008). Elephants not only helped familiar individuals but also unrelated conspecifics as humans do, suggesting a generalized response to distress (Douglas-Hamilton et al., 2006). Finally, some of the helping behaviours described above could have improved the distressed animal's situation, meeting the requirements of the third component. However, almost all of the examples discussed above are from only two studies. It is clearly desirable to carry out systematic studies focusing on the emotional and cognitive components of these helping responses in elephants to provide convincing evidence and expand our knowledge of empathic targeted helping in these social animals.
Nurturant behaviour has been observed towards both live and dead calves (reviewed by Reggente et al., 2016). The most commonly observed nurturant behaviour is the mother's habit of holding her calf at the surface, supposedly to aid it in breathing (Siebenaler & Caldwell, 1956). But cases in which individuals support sick or injured calves at the surface have been also reported. For instance, a female bottlenose dolphin (presumed mother) assisted an injured juvenile, while three other dolphins placed themselves between the couple and a research vessel in a protective manner. When the juvenile was euthanized the group appeared agitated, showing short dives of irregular direction (Warren-Smith & Dunn, 2006). Joan Gonzalvo and his team observed a group of bottlenose dolphins supporting a sick calf. The group appeared stressed (swimming erratically). After the calf's death, they let the corpse sink and left the area (Hooper, 2011). Interspecific cases of nurturant behaviour have also been observed: two Pacific white-sided dolphins (Lagenorhynchus obliquidens) took turns to support a neonatal harbour porpoise (Phocoena phocoena) at the surface (Baird, 1998) and a group of Indo-Pacific humpback dolphins (Sousa chinensis) were reported to assist a finless porpoise (Neophocaena phocaenoides sunameri) calf for more than 3 h . The group seemed to adjust to the calf's situation by swimming slowly and showing no signs of aggressive behaviour.
There are many anecdotal reports of succorant and rescue behaviour in cetaceans, in particular of bottlenose dolphins (Siebenaler & Caldwell, 1956;Caldwell & Caldwell, 1966). One group stayed with an individual injured in an explosion buoying it to the surface to breathe, instead of leaving the area as they did on another occasion in which dynamite exploded but none of them was injured. When the animal was recovered they rapidly left the area. A school of dolphins charged and milled around a vessel where fishermen were trying to release an individual hooked on a fisherman's line until the dolphin was freed. A sick captive dolphin ceased to echolocate and consistently rammed the side of the tank, sometimes causing bleeding. During these occasions, another dolphin put itself between the sick dolphin and the tank wall. It circled the tank with the sick animal until it started swimming normally. When the dolphin's health worsened and it started sinking and listing, the other dolphin intermittently buoyed up it until it died. Park et al. (2012) filmed a group of long-beaked common dolphins (Delphinus capensis) apparently trying to help a paralyzed female. The group supported the female by forming a sort of raft, swimming while carrying her on their backs. Kuczaj et al. (2015) described a similar incident in bottlenose dolphins where helping behaviours towards a distressed female included raft formation, lifts, and stimulating pushes. These behaviours seemed to be directed towards bringing the dolphin to the surface to breathe and appeared to stimulate it to move. Sometimes, four or five individuals pushed the distressed dolphin towards the surface together, suggesting possible cooperative behaviour among the helpers. While some of the dolphins supported the distressed individual others stayed nearby swimming slowly or, on several occasions, circling or milling, behaviours that have been interpreted as signals of distress or excitement in cetaceans (Siebenaler & Caldwell, 1956;Cockcroft & Sauer, 1990;Fertl & Schiro, 1994;de Moura, da Silva Rodrigues & Siciliano, 2009). The distressed dolphin frequently produced what were almost certainly distress calls. The frequency of these whistles was positively correlated with the frequency of the supporting behaviours of the helper dolphins, suggesting an important role for these calls in eliciting helping behaviours. Lilly (1963) recorded dolphin distress calls containing two whistles repeated until a distressed animal received sufficient aid. He reported that the helper usually assisted the distressed dolphin in an appropriate way and adapted to its specific needs. By contrast reports also exist in which individuals did not help a distressed animal, such as a male bottlenose dolphin that, instead of assisting a female emitting distress signals, abandoned her (Connor & Norris, 1982). Finally, there are also instances of some dolphin species apparently helping others to give birth or removing foreign objects from the body of others as in the case of a captive rough-teethed dolphin (Steno bredanensis) that pulled a hypodermic syringe from a pen-mate during medical treatment and then behaved aggressively towards the veterinarian (Connor & Norris, 1982).
It has been reported that humpback whales (Megaptera novaeangliae) may engage in a type of interspecific rescue behavior (Pitman et al., 2016). These whales have been observed to harass attacking killer whales even if their prey was from another species (other whales, pinnipeds or teleost fish). Their mobbing response sometimes differed depending on the prey species. If the prey was a humpback whale, the approaching humpbacks always acted aggressively towards the killer whales, driving them off on some occasions. But if the prey was not a humpback, the whales showed a range of responses: moving away, staying nearby, or aggressively confronting the orcas. Pitman et al. (2016) suggested that humpbacks may respond to acoustic cues of attacking mammal-eating orcas. Sometimes, the prey could escape thanks to this harassing response. However, there was no apparent benefit for the humpbacks, rather the contrary: they were often attacked by orcas during these interactions, sometimes they travelled long distances to interfere with the attack and, on some occasions, they spent a huge amount of time harassing the orcas. There is also a report of a humpback whale harassing short-finned pilot whales (Globicephala macrorhynchus), when a pod were attacking a pair of gray whales (Eschrichtius robustus) (Pitman et al., 2016), suggesting that such harassing is not only restricted to killer whales.
All three requirements defined for targeted helping in Table 1 thus seem to be present in some of the reports described above. Regarding the first component, it is important to note that there are virtually no data on emotional responses or emotional contagion in cetaceans. Their facial musculature prevents facial expressions and thus it is difficult to detect emotional behaviours within the members of this order (Simmonds, 2006). However, some surfacing behaviours, and bubble and swimming patterns like circling or milling seem to signal stress or excitement in these animals and, in some of the reports cited above, these behaviours were observed. Kuczaj et al. (2015) reported that several dolphins stayed near a distressed female, taking turns to assist it, swimming slowly and, on several occasions, circling or milling. These responses suggest that the observers were not over-aroused due to emotional contagion. Nonetheless, solid evidence is needed to conclude that such behaviours are an indication of a moderate emotional reaction rather than signs of personal distress.
Several hypotheses have been proposed to explain the cetacean behaviour of pushing or holding dead or alive conspecifics: displaced nurturant behaviour (Deakos et al., 2010), curiosity towards novel objects (Siebenaler & Caldwell, 1956), surrogate behaviour in response to perinatal mortality (Smith & Sleno, 1986). All of these assume that this behaviour is an automatic response triggered by a specific stimulus such as a calf's death or the perception of an immobile or floating object. However, an automatic response is not always elicited by the presence of a corpse or an object (Caldwell & Caldwell, 1964). A more complex interpretation of this pushing behaviour is that it depends on an appreciation or understanding of the circumstances of the distressed individual, differing depending on context. This is the second requirement for a behaviour to be considered as true targeted helping. In most reports of potential cases of rescue and epimeletic behaviour, animals seem to behave flexibly and according to the specific situation (lifting an injured companion in the correct manner to allow it to breathe, preventing a sick individual from injury from ramming against the side of the tank, interposing themselves between an individual and a threat, acting together to form a raft to carry a paralyzed individual, etc.). Moreover, such supporting behaviour was directed towards individuals requiring different types of assistance or even towards distressed individuals from other species. Sometimes the helper took risks in doing so and in several cases the entire group became involved, apparently adjusting their behaviour to the circumstances of the distressed individual. These features could indicate that cetaceans are aware of the needs of the distressed individual, and act to alleviate its suffering (Ritter, 2007). The assisting behaviour usually ceased when the individual recovered or died, indicating that the response was restricted to the circumstances in which the individual needed help, meeting the third criterion for targeted helping. Recent studies have highlighted the potential of signature whistle rate and loop number as indicators of stress in bottlenose dolphins (Esch et al., 2009) and, as suggested by Lilly (1963) and Kuczaj et al. (2015), such calls may give rise to helping behaviour. Although distress calls are likely to play an important role in rescue and epimeletic behaviour in cetaceans, they do not alone explain the observed behavioural flexibility of the responses and why these behaviours are sometimes directed towards other species or towards corpses that do not emit distress calls. Further studies should include auditory recordings to clarify the role and spectrum of distress calls better in cetacean species. Finally, some of these reports detail cases in which individuals appeared to risk their life to help another animal. Apparently, these behaviours did not provide immediate benefits to the rescuers, on the contrary the helping responses implied high costs to the helpers. The problem here is to determine whether these rescue responses were triggered by the perception of distress in others, in turn eliciting a moderate emotional response in the bystanders that elicited their helping. In the case of rescuer humpback whales, for instance, we do not have data on any emotional nature of their reaction and thus, we cannot yet define these as empathic targeted-helping.

(d) Rodents
Targeted helping requires sophisticated cognitive capacities that are not thought to be present in rodents. However, their behavioural ecology and new empirical evidence suggest that some rodents, particularly rats, could be interesting subjects to investigate the roots of helping behaviour (Panksepp & Lahvis, 2011;Panksepp & Panksepp, 2013;Kirk, McMillan & Roberts, 2014;Hernandez-Lallement et al., 2015;Sivaselvachandran et al., 2016). Rice & Gainer (1962) assessed the ability of rats to act effectively in reducing the suffering of a conspecific. In this pioneering work, a rat was suspended in the air (stressful situation) and another individual could help by pressing a lever that lowered the distressed rat. The helper rat lowered the distressed rat more often than a plastic block (control stimulus) and the helper stayed close to and oriented towards the suffering conspecific. Rats that had previous experience with a conspecific's distress were 10 times more likely to lower the distressed partner compared to controls. Lavery & Foley (1963) suggested that such helping responses were due to auditory stimulation in the experimental condition (distress calls of the suspended rat), as opposed to the control situation (silent object). More recently, Schneeberger, Dietz & Taborsky (2012) tested rats using an iterated Prisoner's Dilemma. In the first experiment, actor rats were paired with cooperative or defective partners. The actor rat could help its partner by pulling a tray with food towards the cage, the cost of which was manipulated by varying the resistance of pulling. In the second experiment, the researchers manipulated the levels of food deprivation of partners and assessed the potential effects of conspecifics' condition (body mass) in the actor's willingness to help. Rats provided more help to cooperative than to defective conspecifics, and the amount of help decreased with increasing costs when the partner was a defector. The need (hunger) and body condition of the partner influenced the amount of help provided by the actor rat: hungry rats received more help if they were underweight, whereas if the receiver was satiated, actor rats provided more help for heavy partners. The authors hypothesized that, as body mass correlates with dominance in rats, helping a dominant partner could reduce the probability of punishment for defection or might improve the social reputation of the helper.
In order to test for targeted helping behaviour in rodents, Ben-Ami Bartal, Decety & Mason (2011) placed a free rat in a chamber with a distressed cagemate trapped in a restrainer. The free rat learned, after several sessions, to open the restrainer and free the cagemate. Rats opened the restrainers containing cagemates even when social contact was prevented, but, in general, did not open empty or object-containing restrainers. When free rats were presented with two containers, one containing chocolate and the other containing their cagemate, they opened both and typically shared the chocolate. Consistent with the hypothesis that females are more responsive to emotional stimuli than males (Eisenberg, 2000), the proportion of females that opened the restrainers was significantly higher. Observer rats apparently acted in a more agitated way when their cagemates were restrained, which could be considered evidence for emotional contagion. Further work complemented and supported these results (Ben-Ami Bartal et al., 2014) showing that rats also helped trapped strangers, but did not help strangers of a different strain. Nevertheless, after being housed with rats of a different strain, these rodents helped strangers of that strain, demonstrating that rats can expand helping behaviour from one familiar individual to phenotypically similar others. Genetic relatedness alone was not required for helping, proving, according to the authors, that strain familiarity is a requisite for the expression of pro-social behaviour in rats.
The study of Ben-Ami Bartal et al. (2011) has been criticized. Silberberg et al. (2014) in a follow-up study concluded that their results and those of Ben-Ami Bartal et al. (2011) could be better explained in terms of neophobia and the pursuit of social contact, instead of in terms of empathic helping. Silberberg et al. (2014) showed that: (i) free rats did not learn to release the trapped rat when social contact was prevented during the learning phase; (ii) trapped rats often returned to the restraint tube following release, suggesting that the tube was not aversive later in the experiment; and (iii) observer rats continued to make the freeing response even when this action was no longer effective. These findings do not support a conclusion of empathy. It should be noted, however, that in the experiment of Silberberg et al. (2014), the door could be opened by making contact with a metal strip attached to the door, so it is possible that some of the responses were accidental.
If rats' helping behaviour is mainly driven by social contact as proposed by Silberberg et al. (2014), rats should open restrainers similarly for distressed and non-distressed cagemates (Sivaselvachandran et al., 2016). Sato et al. (2015) tested this prediction. In a first experiment, the cagemate was soaked in a pool of water (distressing condition) and the observer could open a door to allow their cagemate into a safe area. Rats opened the door significantly faster in the soaking condition than in control conditions (no cagemate in the pool area, neither cagemate nor water in the pool area, stuffed toy rat in the empty pool area). The experiment also showed that the rats disliked being soaked and that those with previous experience of soaking learned how to open the door faster. In a second experiment, the cagemate was just placed in an adjacent cage. Only one out of eight observer rats showed door-opening behaviour suggesting that rats did not learn door-opening behaviour when the cagemate was not in distress. In a third experiment, a group of rats were conditioned to open the door for a soaked cagemate and another group was conditioned to open the door for a food reward. Both groups were then tested in an arrangement in which they had to choose between opening a door that freed the cagemate from the pool or a door that gave them access to a food reward. Only the first group showed a significantly higher proportion of door opening to the cagemate, suggesting that the learning phase influences later behaviour. Sato et al. (2015) concluded that their findings support the view that rats show empathic helping. Recently, Schwartz et al. (2016) replicated the study of Sato et al. (2015) and disputed their conclusions. They highlighted that the use of a different apparatus in the second experiment of Sato et al. (2015) may have influenced the results obtained.
Using an E maze, Schwartz et al. (2016) also showed that rats preferred to release another rat standing in a pool of water than a rat standing on dry ground but this preference was maintained by two separate outcomes: the social contact offered by the freed rat and the reinforcing value of a pool of water. Although in Sato et al.'s first experiment observer rats sometimes opened the pool chamber even when there was no distressed cagemate present, they did not observe a rats' preference for the pool area, rather noting that the rats showed a tendency to avoid water.
In sum, the results from these rat studies remain controversial and inconclusive. The fact that rats only help distressed individuals (Sato et al., 2015) or provide help even when social contact was prevented (Ben-Ami Bartal et al., 2011) is compatible with an empathy hypothesis. By contrast, the possibility that the outcome experienced during the learning process may influence future helping responses, as illustrated by the rats that failed to learn how to help distressed partners when the outcome was not social contact (Silberberg et al., 2014), supports the social-contact hypothesis. Another limitation of some of these studies is that the same rats were tested repeatedly across sessions using the same restraining tube (Ben-Ami Bartal et al., 2011Bartal et al., , 2014Silberberg et al., 2014). The trapped rats showed habituation, even returning to the tube when the helper rat had opened the door. This casts doubt on whether the observer rat was really helping the trapped rat in response to its distress, if the latter was not in an aversive situation.
The presence of distress calls during the learning process may have influenced these results. Lavery & Foley (1963) showed that rats spontaneously learned to press a bar to stop the auditory playback of distress squeals, suggesting that squeals are aversive to some extent and rats will act to stop them. These calls can evoke emotional contagion processes, inducing negative emotional states in the receivers (Kim et al., 2010;Atsak et al., 2011;Saito et al., 2016). Therefore, it is possible that, after hearing the calls of their cagemates, observer rats experienced personal distress during the early learning phase and acted to stop emission of the squeals. If their actions during this phase not only stop the disturbing calls but also allow them to interact with their cagemate, rats might perceive these helping responses as a highly rewarding experience (Vasconcelos et al., 2012). Physical contact and social interaction are known to have a calming effect in these rodents (Hofer & Shair, 1980;Carden & Hofer, 1992;Beery & Kaufer, 2015). Thus, in situations in which they experience personal distress, rats could act not only to stop the distressing stimuli but also to look for social interaction with their partners to reduce their own negative arousal. Therefore, knowing the behavioural and physiological features of the emotional reaction of the observer rat towards a distressed partner is necessary to distinguish between personal distress and a moderate emotional reaction associated with empathetic targeted helping.
Rats respond differently depending on the needs of their partner (Schneeberger et al., 2012) and their previous experience with the distressing procedure (Rice & Gainer, 1962;Sato et al., 2015) suggesting some degree of flexibility in their response and some cognitive evaluation of the situation. Sato et al. (2015) found that rats that had previously been soaked were quicker to learn how to open the door for a soaked cagemate than were naïve rats. However, this paradigm consisted of role-reversal sessions, and soaked rats could have learned about door opening through observation of their own helper rat, thus learning more quickly than naïve rats (Sato et al., 2015). When rats' responses lead to the termination of a distressing situation such as being soaked or suspended, there is obviously an improvement of the distressed party's state, fulfilling the third requirement of empathic targeted helping (Table 1). However, this outcome could be a by-product of attempts by the rat to reduce its own stress and thus is not yet a true example of empathic targeted helping.
In summary, rodents represent promising models to study the roots of empathic target helping, allowing researchers to measure behavioural and physiological data, use different controls and test the role of distress calls in prompting helping responses in social animals. Auditory recordings and physiological measurements are crucial for assessing the nature of the reaction and responses of these animals towards distressed others to allow us to determine whether empathic targeted helping occurs in rodents.

IV. DISCUSSION
Our objectives herein were to (i) clarify the terminology used in this field by providing operative criteria of application; (ii) review relevant studies on sympathetic concern and empathic perspective-taking in non-human animals using these criteria; and (iii) discuss whether the available evidence offers examples of complex empathic behaviours.
We reviewed relevant studies in light of whether they meet the three requirements established in Table 1 to qualify as examples of consolation and empathic targeted helping. A summary of our main findings is provided in Table 2. Note that in most of the reviewed studies, the first requirement is lacking: most studies assessing consolation or empathic targeted helping in non-human animals did not record the existence of an emotional reaction of the observer towards the distressed party. We are not suggesting that there is a lack of studies assessing emotional reactions in non-human animals per se, nor that animals do not display emotional reactions to a distressed individual. Instead, we note that studies assessing consolation and empathic targeted helping usually overlook the emotional component of the behaviour by not reporting data on the emotional response of the involved individuals, undermining our ability to reach firm conclusions. Furthermore, among studies that did report the observers' emotional reactions, only a minority included physiological data, making it difficult to establish solid conclusions regarding the nature of those reactions. As discussed in Section II.1, the distinction between a moderate emotional reaction and an intense negative emotional reaction that mimics that of the distressed party could help us to distinguish between empathic behaviours and personal distress processes. The lack of assessment of this emotional component in most studies thus is remarkable and should be addressed in future work, especially considering that, for humans, there is a broad agreement on the presence of an emotional response to another person as a crucial component of empathy (Lamm, Batson & Decety, 2007).
It is important to establish the compliance of all three criteria in order to interpret a certain behaviour as empathic. We use as an example the helping behaviour of some ?

Rat (Rattus
and Walker et al. ?
Burkett et al.
Burkett et al.
Asian elephant (E.
? Ben-Ami Bartal et al. (2011Bartal et al. ( , 2014 and Silberberg et al. Biological Reviews 93 (2018) (Nowbahari et al., 2009;Nowbahari, Hollis & Durand, 2012;Miler, 2016) and in the field Taylor et al., 2013) have shown that several species of ants engage in a type of helping behaviour, termed by the authors 'precision rescue', to free entrapped nestmates. Ants directed rescue behaviours such as sand-digging and limb-pulling towards individuals restrained by a nylon snare in a sand field. Interestingly, when these behaviours are ineffective, rescuer ants begin to bite precisely at the nylon snare. Furthermore, rescuers discriminate between victims based on their state and identity, providing help only to distressed nestmates and not to motionless nestmates, members of different colonies, common prey items, ants from a different species or the empty apparatus (Nowbahari et al., 2009). Thus, ants displayed what seem to be sophisticated helping responses towards familiar distressed individuals. This rescue behaviour is fine-tuned to the specific situation of the victim and, if the biting attempts directed toward the snare are ultimately successful, the rescue is likely to alleviate the victim's distress. Therefore, it appears that the second requirement (fine-tuned help appropriate to the situation of the distressed party) and the third requirement (improvement of the distressed party's situation) established in the definition of targeted helping (Table 1) are met. These responses are very similar to those displayed by rats in the studies of Ben-Ami Bartal et al. (2011Bartal et al. ( , 2014 and Sato et al. (2015). However, as Vasconcelos et al. (2012) pointed out, these two sets of results led to quite different interpretations. Whereas Ben-Ami Bartal et al. (2011Bartal et al. ( , 2014 discuss their findings in terms of empathically motivated helping, Nowbahari et al. (2009) refrained from hypothesizing about the motivation underlying the ants' behaviour and focused on the functional aspects of the rescue behaviour. It is true that ants' responses seemed to be other-directed and fine-tuned to the specific needs of the victim, but we should also consider the flexibility of those responses. These help responses do not appear to change radically depending on the context. Subsequent studies revealed that this rescue behaviour is elicited only under specific circumstances related to the ant species' behavioural ecology: loose sandy-soil habitats and close proximity to pit-digging antlions, a common predator of many ant species, are critical factors in whether an ant species displays rescue behaviour . Thus, rescue behaviour has evolved in those ant species that face high risks of possible entrapment by sand collapse and as an antipredator strategy (Taylor et al., 2013). Furthermore, rescue behaviour is constrained by division of labour: help was received and dispensed only by certain colony members . Ants belonging to a different colony were vigorously attacked and, individuals belonging to the youngest inactive caste were ignored. The mechanism underlying this rescue behaviour remains unknown but it has been hypothesized that it is mediated by the release of facilitating pheromones by the entrapped ant (Miler, 2016). When they are in distress, ants secret alarm pheromones through the Dufour and poison glands and these two glands seem to be involved in the observed rescue behaviour . The alarm signals are colony-specific and their secretion increases with worker age Miler, 2016). In addition, the fact that anaesthetized ants elicited no help supports the hypothesis that active pheromone release by the victim triggers the rescue behaviour (Miler, 2016). Therefore, ants' helping responses are likely elicited by the detection of alarm pheromones and not by a moderate emotional reaction in response to other's distress or requiring the well-developed cognitive abilities expected for empathic targeted helping. Thus, ants' rescue behaviour does not meet the first requirement. Ants do display a sophisticated form of rescue but it is not based on complex empathic processes, their responses lack the emotional component. The ants' helping behaviour is an adaptive specialization finely tuned to their precise ecological circumstances: the caste member's probability of becoming, or finding, a nestmate in need of help . Overall, this example shows us that some behaviours that appear cognitively complex can be explained by much simpler mechanisms (Vasconcelos et al., 2012). Simple mechanisms can lead to responses and outcomes very similar to those expected for empathic behaviours but this does not mean that they are the same. Thus, we should be careful when comparing similar behaviours displayed by different species since, although the observable features may be identical, their nature may differ and be derived from very different mechanisms. In attributing empathic motivations to a specific behaviour, we tend to be more permissive for mammals than for other classes such as insects. But, we must attempt to avoid such species biases and consistently apply the same criteria. The use of different criteria, definitions and paradigms in the field of animal empathy may mean that behaviours that rely on different mechanisms but that have similar features are considered to be identical. Returning to the example of ants, it should not be surprising that many social animals display seemingly complex helping behaviours that are not empathy-based. For example, in social species distress calls could play the same role as alarm pheromones in ants and automatically elicit sophisticated approaching or helping responses in the receivers. These responses will be related to the behavioural ecology of the species and fined-tuned to its specific socio-behavioural traits. For example, the affiliative behaviours observed in voles, budgerigars and Eurasian jays towards distressed group members is consistent with the socio-behavioural features and group structure of these species. Chimpanzees' caring behaviour for wounded individuals also differs depending on the ecological features of their habitat. While Taï chimpanzees care for wounded group members independent of kin relationship, a non-kin member with a wound elicits a reaction of fear or disgust in Gombe chimpanzees. This differential pattern of behaviour may be explained by the high predation pressure that Taï, but not Gombe, chimpanzees experience from leopards (Boesch, 1992). Animals could obtain survival and reproductive benefits by caring or helping group members with whom they are likely to be genetically related (Vasconcelos et al., 2012). Thus, for group-living species, the adaptive value of this affiliative or helping behaviour is evident. If certain selective pressures are present, it is likely that social animals may develop strategies that favour survival not only for the individual but also for the rest of the group. These behaviours could range from simple, automatic and fixed responses not based on empathy, such as perception-action mechanisms involving alarm signals that trigger approach and rescue responses, to flexible behaviours requiring more complex emotional and cognitive processes. With increased emotional and cognitive abilities and a greater level of effective emotional regulation, animals can respond flexibly to distressed individuals in different contexts, even if alarm signals are identical or absent.
It seems, then, that the ability to attend and respond to the emotional state of others was a major evolutionary step for group-living animals. This capacity is an important adaptive strategy in the maintenance of complex social systems allowing animals to predict others' future behaviour (Waller, Whitehouse & Micheletta, 2016) and to attend to others' needs in changing circumstances. The affective experiences may help to build, strengthen or repair social bonds, also providing a variety of reproductive benefits (Lahvis, 2016). The combination of a capacity to respond to others' emotional states with the ability to carry out a cognitive evaluation of the situation will lead to a more robust and flexible behavioural response to an urgent situation (Lahvis, 2016).

V. CONCLUSIONS
(1) Sympathetic concern, empathic perspective-taking and the two major behaviours in these categories, consolation and empathic targeted helping, have been traditionally defined in terms of their underlying motivations. This approach cannot be easily applied to the available data on empathic behaviours in non-human animals. Thus, we operationalize these behaviours in terms of the type of reaction and response triggered by the perception of distress in others and its main effect or outcome. Separating the underlying motivations of the behaviours and focusing instead on the set of behavioural and physiological features of animals' reactions and responses towards distressed individuals allows us to distinguish empathic from non-empathic behaviours better in non-human animals. We note that this method may still not be sufficient in ambiguous cases.
(2) A wide range of social species can perform sophisticated behaviours that calm, help or rescue other group members from distressing situations. These behaviours are related to the species' behavioural ecology and may be simple, not involving empathy-based mechanisms, such as rescue behaviour in ants, or more complex behaviours requiring an emotional component and which, in turn, can be graded into different levels of complexity (de Waal, 2008). The combination of emotional and cognitive elements allows animals to respond effectively even in new situations, which is presumably an important adaptation for animals living in unpredictable environments.
(3) In general, empathic behaviours seem to follow a familiarity gradient, offered mostly to familiar individuals but, for animals living in fission-fusion groups like some primates or cetaceans, responses to others' needs irrespective of their identity may also be an important adaptation. For these species, the involvement of emotion may provide a more generalized response to others' emotional states. Humans regularly help strangers (Raihani & Bshary, 2015) and some studies reviewed herein record cases in which primates, dogs and cetaceans helped non-familiar individuals or even members of other species. Thus, whereas familiarity could facilitate empathy in several contexts, expanding these emotional responses towards a wider range of individuals including strangers might indicate an increased empathic capacity as in humans.
(4) Compliance with all three requirements defined in Table 1 is essential to distinguish empathic from non-empathic processes. Only some reports involving primates, cetaceans and elephants seem to meet all three criteria. More research is needed to characterize these behaviours better. For primates or cetaceans, obtaining physiological measures could be extremely difficult due to methodological and ethical issues and we must therefore rely on behavioural parameters. Interpreting behaviour is contentious but by combining such information with data on the level of effective emotional regulation for each species and with features of the two other requirements (response and outcome) we could develop a more complete picture of the observed phenomena. Although we have not discussed evidence for neural mechanisms of empathy in non-human animals herein (most of the available data comes from rodent studies, and there are several reviews available: Panksepp & Lahvis, 2011;Panksepp & Panksepp, 2013;Keum & Shin, 2016;Meyza et al., 2016;Sivaselvachandran et al., 2016), the study of the neural basis of empathy is very promising for assessing continuity between empathy in humans and other species. Finally, future studies assessing the motives underlying possible empathic behaviours (e.g. using tests of conditioned place preference/aversion) will provide useful information to clarify the nature and features of empathic behaviours in non-human animals and their phylogenetic continuity across the animal kingdom.

VI. ACKNOWLEDGEMENTS
This research received support from the Spanish Ministry of Economy, Industry and Competitiveness through project FFI 2013-44007-P. A. P. M. was supported by a FPU PhD scholarship (AP2012-3501) from the Spanish Ministry of Education, Culture and Sport. We greatly acknowledge the constructive comments of the two anonymous reviewers and the Assistant Editor that have improved this manuscript.