Kith or Kin? Familiarity as a Cue to Kinship in Social Birds

Interacting with relatives provides opportunities for fitness benefits via kin-selected cooperation, but also creates potential costs through kin competition and inbreeding. Therefore, a mechanism for the discrimination of kin from non-kin is likely to be critical for individuals of many social species to maximize their inclusive fitness. Evidence suggests that genetic cues to kinship are rare and that learned or environmental cues offer a more parsimonious explanation for kin recognition in most contexts. This is particularly true among cooperatively breeding birds, where recognition of familiar individuals is usually regarded as the most plausible mechanism for kin discrimination. In this article, we first review the evidence that familiarity provides an effective decision rule for discrimination of kin from non-kin in social birds. We then consider some of the complexities of familiarity as a cue to kinship, especially the problems of how individuals become familiar, and how familiar individuals are recognized. We conclude that while familiarity as a mechanism for kin recognition may be more parsimonious and widespread than genetic mechanisms, its apparent simplicity as a decision rule governing social interactions may be deceptive. Finally, we identify directions for future research on familiarity as a kin recognition mechanism in social birds and other taxa.


INTRODUCTION
Kin selection is often invoked to explain the evolution of cooperation among relatives in social animals (Rubenstein and Abbott, 2017). Here, we use 'social' to describe species that exhibit cooperative breeding, following the widely used definition of cooperative breeding as a reproductive system in which more than a pair of individuals collectively raise young in a single brood or litter (Emlen and Vehrencamp, 1985;Koenig and Dickinson, 2016). Hamilton's rule predicts that cooperation confers indirect fitness benefits and will be selected for providing that the coefficient of relatedness between actor and recipient, multiplied by the benefits of cooperation to the recipient exceed the costs to the actor (Hamilton, 1964). Therefore, differential treatment of conspecifics that vary in genetic relatedness, i.e., kin discrimination (Sherman et al., 1997), is an important consideration in studies of social evolution. In addition to kin-selected fitness benefits, kin discrimination may also play an important role in inbreeding avoidance when passive processes, such as sex-biased dispersal, are insufficient to reduce inbreeding risk (Pusey and Wolf, 1996). These functional benefits of discriminating kin from non-kin are well-established, but the mechanisms through which this is realized are keenly debated.

Term Definition
Kin discrimination The differential treatment of conspecifics within a population that differ in their genetic relatedness (Sherman et al., 1997) Kin recognition The mechanism by which kin discrimination is achieved. A discriminating individual, or actor, acquires cues to kinship from a referent (itself, a subset of kin, or the local environment) and uses these cues to form a template (Reeve, 1989). This template is compared with the phenotype of an encountered conspecific, or recipient, and an assessment about kinship is made based on the perceived similarity between the template and the recipient's phenotype (Lacy and Sherman, 1983). A specific action is then taken, based on this assessment Recognition cue A phenotypic trait expressed by a individual that acts as a reliable signal of kinship, whereby within populations, similarity at the phenotypic trait is correlated with genetic similarity across the genome

Recognition template
An internal representation of kin traits with which the phenotypes of encountered conspecifics can be compared. Templates are usually formed by learning the recognition cues of putative kin, or ones own cues. Templates may also conceivably be genetically determined

Recognition errors
Desirable recipients are those which, following acceptance, provide greater fitness pay-offs to the actor than undesirable recipients (Reeve, 1989). Within populations, recognition templates are matched against a finite set of cues which overlap in desirable and undesirable recipients due to individual variation (Lacy and Sherman, 1983). Therefore, any recognition system will involve a certain amount of acceptance errors, where undesirable recipients are accepted, and rejection errors, where desirable recipients are rejected (Reeve, 1989) Our current framework for understanding kin recognition systems involves three components: the production of external cues; the perception of these cues and formation of recognition templates; and the action taken based on the perceived similarity between a template and an encountered phenotype (Beecher, 1982;Reeve, 1989;Gamboa et al., 1991; Table 1). Both the cue and the template may be either genetically determined or acquired from the biotic or abiotic environment (Sherman et al., 1997). Recognition systems will also be prone to errors; in the case of positive discrimination in favor of kin for helping behavior, these will be either rejection errors, in which kin are not recognized as such and rejected as social partners, or acceptance errors in which non-kin are erroneously recognized as kin and accepted as social partners (Reeve, 1989; Table 1). The extent to which cues and templates are determined genetically and/or environmentally, and the risk of making rejection/acceptance errors will vary greatly between and within species (Sherman et al., 1997;Komdeur et al., 2008).
This framework leads to three broad categories of kin recognition mechanism. Recognition may be based on familiarity, in which discriminating individuals learn the recognition cues of relatives (e.g., parents and/or siblings) at a sensitive phase during development (Komdeur and Hatchwell, 1999) and discriminate these familiar individuals from unfamiliar ones later in life. Second, recognition may be based on phenotype matching, whereby individuals use their own phenotype and/or those of their familiar kin to form a generalized template with which to compare the phenotypes of other individuals (Lacy and Sherman, 1983). Familiarity and phenotype-matching are considered alternative processes (Holmes and Sherman, 1983), but both involve matching phenotypes to learned templates; the two mechanisms differ only in the specificity of the template employed (Reeve, 1989). Thirdly, it is also possible that both cues and templates are genetically-determined rather than environmentally-acquired or learned, thereby satisfying Grafen's (1990) definition of kin recognition as requiring discrimination of true genetic relatives, although note that here we use the less restrictive definition of Sherman et al. (1997), as stated above.
The ecological and social circumstances in which a recognition system evolves is likely to have a profound effect on the probable mechanism of recognition (Komdeur et al., 2008). Likewise, a species' kin recognition mechanism will have consequences for the accuracy of discrimination and the degree of resolution between different categories of kin. For example, kin recognition that requires prior association for the learning of cues or templates allows individuals to recognize familiar kin only, whereas recognition that is based on phenotype matching may permit recognition of unfamiliar kin (Mateo, 2004). Among cooperatively breeding birds, recognition of familiar individuals is usually regarded as the most plausible mechanism for kin recognition (Komdeur and Hatchwell, 1999). However, the term familiarity is often ill-defined, the recognition cues are poorly understood, and very little is known about the conditions under which a previous association constitutes familiarity in the context of kin recognition. In this article, we first review the evidence for alternative kin recognition mechanisms in social birds, concluding that recognition based on familiarity is the best-supported decision rule for discrimination of kin from non-kin in most studies. We then consider some of the complexities of familiarity as a cue to kinship, suggesting that while such a mechanism for kin recognition may appear more parsimonious and widespread than phenotype matching, its apparent simplicity is deceptive. Finally, we discuss possible directions for future research on familiarity as a kin recognition mechanism in social birds and other taxa.

MECHANISMS OF KIN RECOGNITION
Kin recognition may be achieved via a variety of mechanisms that range from simple to complex. In the simplest form of recognition, individuals encountered in a particular area are recognized as kin. As long as relatives are predictably distributed in space, location can correlate reliably with genetic relatedness (Komdeur and Hatchwell, 1999). Some researchers suggest this is not a true form of kin recognition, as individuals are responding to location, rather than phenotypic cues (Halpin, 1991;Tang-Martinez, 2001). However, in many natural populations, it is rare for unrelated individuals to be encountered in the nest for example, and a simple decision rule such as "treat anything in my nest as kin, " is an effective and widely used mechanism for offspring recognition in birds (Beecher, 1991), despite its

Genetic kin recognition
All three components of the recognition system are determined by a specific allele or gene complex (Mateo, 2004). Genetic kin recognition does not require a period of learning in order for templates to form Phenotype matching Individuals use their own phenotype and/or those of their familiar kin to form a generalized template with which to compare the phenotypes of other individuals (Greenberg, 1979;Holmes and Sherman, 1982). Because learned templates are generalized, a period of previous association is not required for kin to recognize one another. Instead, a positive correlation between cue similarity and level of genetic relatedness is required, so the recipients with phenotypes that most closely match the actor's general template are its closest kin (Tang-Martinez, 2001) Familiarity The recognition cues of putative relatives are learned and used to form templates during a sensitive phase during development (Komdeur and Hatchwell, 1999), within which associating individuals are likely to be kin. Individuals are subsequently able to discriminate these familiar individuals from unfamiliar ones outside of the association period. Recognition based on familiarity therefore requires a period of prior association for individuals to be categorized as kin potential for exploitation by intra-and inter-specific brood parasites (Davies, 2000). Other contextual cues may modify this simple rule; for example, polyandrous male dunnocks Prunella modularis are more likely to feed the young of females with which they mated during their fertile period (Burke et al., 1989;Davies et al., 1992), thereby maximizing their chance of directing their care toward offspring. Spatial cues to offspring recognition may be superseded by individual recognition when fledglings leave the nest (Beecher, 1988), but, in most cases, parent-offspring recognition does not persist beyond the period of offspring dependence. Such simple rules work well in non-social species, in which there is little or weak selective pressure to recognize kin beyond offspring independence. However, in social species there are often indirect fitness benefits to be gained from cooperating with close kin during adulthood or fitness costs of inbreeding, and, consequently, selection for mechanisms of kin recognition that persist beyond the period of parental care (Komdeur and Hatchwell, 1999;Cornwallis et al., 2009). In this review, we focus on mechanisms in social birds that might permit kin recognition over an individual's lifetime, or at least the period over which cooperative behavior or the risk of inbreeding exists. Such mechanisms may be based on genetic kin recognition, phenotype matching or familiarity ( Table 2).

Genetic Recognition
Genetic kin recognition requires discrimination of kin from non-kin based entirely on genetically acquired cues without a period of associative learning. Here, recognition alleles, dubbed "greenbeard genes" by Dawkins (1976) or gene complexes encode the production of phenotypic cues, the templates and the perception of the cue and performance of a discriminatory action. Such a system relies on polymorphic recognition genes for reliable discrimination, yet paradoxically, kin-selected fitness benefits are predicted to reduce allelic diversity at these loci. This is because in cooperative contexts, individuals bearing common cues are more likely to encounter equivalent individuals and receive altruistic benefits than those with rare cues. These individuals will gain higher fitness, and eventually the common alleles become fixed and the recognition system breaks down (Crozier, 1986). Alternatively, mutation will interfere with genetic kin recognition, and mutant cheats who carry the phenotypic cues but not the associated relatedness, may evolve and spread through the population (Hamilton, 1964). Finally, in the case of a gene complex orchestrating recognition, recombination could disrupt kin recognition. In each of these theoretical scenarios, the required correlation between similarity in the inherited phenotypic cue and kinship among pairs of individuals would decrease over time, rendering such a cue useless for kin recognition (Gardner and West, 2007). There are no convincing cases of genetic kin recognition in cooperatively breeding birds (Table 3). Indeed, empirical evidence of genetic kin recognition across taxa is scarce, the clear exceptions being the slime mold, Dictyostelium discoideum (Queller et al., 2003) and fire ant Solenopsis invicta (Keller and Ross, 1998;Wang et al., 2013).

Phenotype Matching
The second candidate mechanism for kin recognition is phenotype matching. The distinction between phenotype matching and genetic kin recognition is that template formation requires the learning of phenotypic cues that reliably reflect genetic similarity. However, because individuals can use their own phenotype or the phenotypes of a subset of known kin to learn a generalized "kin" template, this does not require a period of prior association, or familiarity between matching individuals. Phenotype matching is an attractive potential mechanism for kin recognition, particularly in the context of inbreeding avoidance, because it allows individuals to recognize unfamiliar kin. Phenotype matching has been demonstrated in the decorated cricket Gryllodes sigillatus (Capodeanu-Nägler et al., 2014) and in several social mammals (e.g., Boyse et al., 1991;Pfefferle et al., 2013). Although in some species, such as the Belding's ground squirrel Spermophilus beldingi both phenotype-matching and familiarity seem to play a role (Holmes and Sherman, 1982). However, empirical support for phenotype matching in cooperatively breeding birds remains rare and inconclusive ( Table 3).
One of the first studies to suggest phenotype matching as a plausible kin recognition mechanism in a cooperative bird was conducted by Price (1998Price ( , 1999 on stripe-backed wrens Campylorhynchus nuchalis. A series of playback experiments demonstrated that wrens were able to discriminate between vocalizations made by their own group, familiar neighboring groups and unfamiliar groups, consistent with a recognition system based on familiarity (Price, 1998). Subsequent experiments showed that the behavioral responses of wrens to calls from patrilineal relatives in the unfamiliar groups did not differ from their responses to calls from patrilineal relatives in their own group, which could indicate phenotype matching  Familiar kin are actively avoided as mates when pairing occurs in winter groups. Males respond more aggressively toward songs of non-kin than songs of kin, but call similarity does not indicate kinship Açkay et al., 2013Açkay et al., , 2014Dickinson et al., 2016 White-fronted bee eater Merops bullockoides Familiarity Field observations Helpers preferentially help kin; the probability of helping decreases with relatedness between helper and potential recipient. Recognition is based on early associations formed during nestling or fledgling development Emlen and Wrege, 1988 Frontiers in Ecology and Evolution | www.frontiersin.org (Price, 1999). However, patrilineal relatives in unfamiliar groups are dominant males that have dispersed from their natal group, so a period of association between the dominant male in each group cannot be ruled out. As male helpers may follow the dominant male in their behavioral responses to intruders, this result could be achieved through recognition based on familiarity.
A recent study on Siberian jays Perisoreus infaustus, a species that exhibits kin-based sociality although not cooperative breeding, has suggested that phenotype matching is used to recognize kin in some contexts. Within family groups, breeders are more aggressive toward immigrants than to their own offspring, but aggression of breeders toward immigrants was negatively associated with the immigrant's genetic relatedness to the breeding male (Griesser et al., 2015). In this study, individuals were considered unfamiliar if they had not interacted between fledging and dispersal, although the possibility that individuals had prior association could not be ruled out unequivocally.
Studies of bell miners Manorina melanophrys provide the best evidence for kin recognition via phenotype matching in cooperatively breeding birds (McDonald and Wright, 2011). Certain features of the bell miner's social system have important consequences for their recognition systems. They form large colonies, often comprising hundreds of individuals, within which individuals are organized into coteries of numerous breeding pairs assisted by non-breeding helpers of varying relatedness that provision multiple nests within their coterie. Like many cooperative breeders, kinship appears to be the most important factor in explaining the patterns of cooperation between breeders and helpers (Wright et al., 2010) and the shared provisioning efforts of helpers within social networks (McDonald et al., 2016). From an early age, however, young interact with both related and unrelated group members, making spatial or associationbased recognition unreliable. Instead, the provisioning effort of helpers correlates with their vocal similarity to the breeding male, an apparently innate signal that also correlates with genetic relatedness (McDonald and Wright, 2011). However, whether vocal similarity permits kin recognition on a continuous scale or on a binary scale, whereby conspecifics are categorized as either kin or non-kin based on a threshold of templatephenotype similarity, remains unclear. Furthermore, although no evidence of call learning has been found, a putative association period during which kin may be learned has not been excluded empirically.
In the closely related noisy miner Manorina melanocephala, which has a similar social system, helpers direct their help toward genetic relatives (Barati et al., 2018), and discriminate between individuals based on acoustic cues (McDonald, 2012). Still, individuals may also rely on prior association to identify relatives, and whether kin recognition is based on phenotype matching or familiarity remains untested in this species.
The problem with recognition via phenotype matching of inherited cues is that, like genetic kin recognition, it is vulnerable to mutation and recombination, and requires sufficient polymorphism to permit precise discrimination. Another important consideration is that there may be selection for individuals to conceal kinship at certain life stages or in certain situations. For example, when paternity is uncertain, effective kin recognition by parents would be adaptive in order to direct care toward genetic offspring. However, from the offsprings' perspective, it would not be beneficial to display an obvious cue to genetic relatedness, as this could exclude cuckolded care-givers (Beecher, 1988;Davies et al., 1992). This conflict of interest between parent and offspring may make it difficult for phenotype matching of genetic cues to evolve as a recognition mechanism. Even if recognition cues are learned, the formation of a generalized template may still select for convergence, as individuals with a more common phenotype are more likely to be accepted as social partners than those with rarer cues. On the other hand, theory suggests that genetic diversity at recognition loci may be maintained if rare alleles confer an extrinsic selective advantage, such as resistance to certain parasites (Rousset and Roze, 2007). Indeed, the highly polymorphic major histocompatibility complex (MHC), has been implicated as a kinship marker during mate choice in vertebrates, detected through odor cues. MHC diversity affects parasite resistance (Kurtz et al., 2004), perhaps explaining how MHC polymorphism is maintained despite its putative role in kin recognition. However, the role of MHC in kin recognition is contested, as disassortative mate preference based on MHC haplotype may arise from the improved immunity associated with heterozygosity at MHC loci itself, rather than MHC haplotype acting as a reliable signal of genetic similarity across the genome (Green et al., 2015).

Familiarity
Familiarity is the most widely supported mechanism of kin recognition in cooperatively breeding birds (Komdeur and Hatchwell, 1999;Komdeur et al., 2008;Riehl and Stern, 2015; Table 3). Kin association during extended brood care provides a sensitive period during which reliable recognition templates can form. This period of association also offers an opportunity for learning of cues that are more similar within a family than in the general population, termed a family or kin "signature" (Beecher, 1982). Once recognition cues are fixed, individuals are potentially able to recognize familiar kin outside of the association context. When extra-pair paternity (EPP) and brood parasitism is rare, association during this period accurately reflects kinship, and a simple rule such as "assist anyone who was present in my natal nest" can be selected for (Komdeur and Hatchwell, 1999). For example, in cooperative contexts, Galápagos mockingbirds Nesomimus parvulus and white-fronted bee-eaters Merops bullockoides discriminate based on previous association, rather than kinship (Curry, 1988;Emlen and Wrege, 1988). In complex societies, a more precise rule, such as "assist anyone that fed me as a nestling" may be more reliable (Komdeur, 1994). In most cooperatively breeding birds, males are the predominant helping sex, but in the Seychelles warbler Acrocephalus sechellensis females are more likely to help than males, and choose to help at nests belonging to female breeders who fed them as nestlings, even if they are not the closest genetic relatives (Komdeur, 1994;Richardson et al., 2003). This makes evolutionary sense in species with high levels of extra-pair paternity, such as Seychelles warblers, because helpers are often unrelated to the male that fed them (Richardson et al., 2003).
Cross-fostering experiments confirm that female subordinates base their helping decisions on associative learning and it is unlikely that young can discriminate between their mother and any other female helper (Komdeur et al., 2004).
Playback experiments show that cues enabling recognition of familiar individuals beyond the association period are encoded vocally (Table 3). An early study on the splendid fairy-wren Malurus splendens showed that while fairy-wrens responded aggressively to the songs of fairy-wrens from other social groups, they exhibited a similar response to the songs of both non-kin and unfamiliar close kin (Payne et al., 1988). More recent experiments have demonstrated that vocalizations signal group membership in Mexican jays Aphelocoma wollweberi (Hopp et al., 2001), green woodhoopoes Phoeniculus purpureus (Radford, 2005) and superb starlings Lamprotornis superbus (Keen et al., 2013). These studies suggest that vocalizations reflect social association rather than kinship per se, as would be expected if cues and templates are learned within groups.
In the context of inbreeding avoidance, good evidence for avoidance of kin as reproductive partners based on familiarity comes from studies of two species of social woodpecker: acorn woodpeckers Melanerpes formicivorus and red-cockaded woodpeckers Picoides borealis. Acorn woodpeckers exhibit high within-group relatedness, with most individuals being parents, siblings or offspring of everyone else within the group (Koenig and Haydock, 2004). Acorn woodpecker females do not breed in their natal group when the reproductive male in their natal group at the time of their birth (their assumed father) is still present (Koenig and Pitelka, 1979). Furthermore, when a dominant male or female dies, reproductive vacancies remain unfilled when nonbreeding helpers of the missing sex are present, and breeding does not usually occur until the vacancy is filled by immigrants from outside the group (Koenig et al., 1999). Similarly, red-cockaded woodpecker females will abdicate a breeding position following the death of their mate when the remaining males are their sons, but will remain when they are unrelated to the male helpers (Walters et al., 1988). The mechanism behind these decisions has not been examined experimentally in either species.
The most compelling cases of kin recognition based on familiarity come from cooperative breeders in which helping occurs within kin neighborhoods (Dickinson and Hatchwell, 2004), where individuals routinely interact socially with both kin and non-kin so that selection for effective kin discrimination is likely to be strong (Cornwallis et al., 2009). In western bluebirds Sialia mexicana there is a strong kin preference in helping behavior (Dickinson et al., 1996) and active kin avoidance during mate choice . However, males do not reduce their provisioning effort in response to behavioral cues to paternity loss, such as extra-pair male intrusion or witnessing female acceptance of extra-pair copulations (Dickinson, 2003). This suggests, along with earlier studies (Leonard et al., 1995), that males do not recognize their own offspring, and that discrimination by both parents and offspring is based on social experience in the nest, rather than genetic relatedness (Dickinson, 2003). Playback experiments have shown that individuals discriminate kin based on vocal cues (Açkay et al., 2013) even though these vocalizations are poor indicators of genetic relatedness, because they are most similar among neighbors, regardless of kinship (Açkay et al., 2014). These findings collectively suggest that western bluebirds recognize familiar individuals, rather than kin, using vocal cues.
Kin recognition has also been extensively studied in another species that helps within kin neighborhoods, the longtailed tit Aegithalos caudatus. Long-tailed tits have a kinselected cooperative breeding system in which failed breeders preferentially redirect their care to help relatives (Russell and Hatchwell, 2001;Hatchwell et al., 2014). Playback experiments show that long-tailed tits are able to discriminate between the calls of close kin and non-kin , and the calls thought to be used as recognition cues are individually distinctive, repeatable and more similar among close kin than among non-kin Leedale et al., 2020). Cross-fostering experiments showed that nestlings and/or fledglings acquire their recognition templates from familiar kin during an associative learning period, when the cues themselves develop , and that cross-fostered offspring subsequently help at the nest of foster siblings . Moreover, there is strong evidence for effective discrimination of first-order kin, but not second-order kin, both in the context of helping behavior and mate choice (Leedale, 2018;Leedale et al., 2020). These results are all consistent with the idea that long-tailed tits categorize conspecifics as either kin or non-kin based on early association in the context of brood care . On the other hand, Nam et al. (2010) and Leedale et al. (2020) both found that long-tailed tit helpers modified their effort according to their relatedness to the helped brood, suggesting that assessment of kinship is not based on a simple dichotomous rule of familiar (kin) vs. unfamiliar (non-kin) birds. Indeed, this suggests a mechanism of phenotype matching, with a gradation of similarity in vocalizations providing a fine-grained, continuous estimation of kinship. However, bioacoustic analysis did not support this suggestion (Leedale et al., 2020), so even in this relatively well-studied system, the mechanism underlying graded discrimination remains unknown.
This review focuses on kin recognition, but familiarity also provides a potential mechanism by which individual recognition may be achieved; for example, some cooperative bird species, such as the chestnut-crowned babbler Pomatostomus ruficeps have individually distinct vocalizations (Crane et al., 2015). However, although individual recognition has been identified in several social mammals, including chacma baboons Papio hamadryas (Bergman, 2003) and golden hamsters Mesocricetus auratus (Johnston and Bullock, 2001), there are no conclusive examples of individual recognition in cooperatively breeding birds ( Table 3). The difference between individual and group recognition depends on the specificity of the templates acquired during the association period, which in turn depends on the nature of the interactions that occur between individuals during that time. In practice, this makes distinguishing individual from kin or group recognition difficult (Tibbetts and Dale, 2007). We discuss this in more detail in the following section.
Overall, there is substantial evidence that familiarity is a widespread kin recognition mechanism in cooperatively breeding birds. The limitation of familiarity is that non-kin will be considered kin if they are encountered during the putative associative learning stage, and kin not encountered during this period will not be recognized as such. However, in most cases, proximity at certain life stages is a reliable indicator of kinship. This is particularly true of birds, which have a prolonged period of parental care at the nest where encountered individuals are likely to be close kin. A second assumed limitation of recognition based on familiarity is that it may result in a binary recognition rule, in which individuals are categorized as either kin or nonkin. A more sophisticated mechanism that permits relatedness to be assessed on a continuous scale would be adaptive, in accordance with Hamilton's rule (Hamilton, 1964), although, as already discussed, such mechanisms may be evolutionarily unstable. Kin recognition through familiarity or prior association is also considered the most likely mechanism of kin recognition in social birds because it is simpler to evolve and arguably less cognitively demanding than an assessment of genetic relatedness based on phenotypic similarity. Yet, while a recognition system based on familiarity may be more parsimonious and widespread than phenotype matching and genetic mechanisms, we argue below that its apparent simplicity is deceptive.

THE COMPLEXITIES OF FAMILIARITY AS A CUE TO KINSHIP
Despite the general acceptance of familiarity as an important means of kin recognition and discrimination, much remains unknown about how associating individuals are categorized as kin and how familiar individuals are recognized after the associative learning period. Here, we suggest that progress will be made in understanding familiarity as a mechanism of kin recognition only when certain gaps in knowledge can be addressed: (i) the meaning of "familiarity, " (ii) the sensitive period for association; (iii) the cues used for recognition; and (iv) the distinction between familiarity and phenotype matching.

What Is "Familiarity"?
Familiarity in the context of kin recognition is difficult to define and to quantify. What is the specific series of events during which an individual learns who is familiar? In the kin recognition literature, familiarity generally refers to some previous social association among individuals, usually during early life stages (Hepper, 1986;Komdeur and Hatchwell, 1999), but the nature of this association is often vague. For instance, is spatial proximity sufficient, or do individuals need to interact in specific ways in order to become familiar? In studies of social birds, such as long-tailed tits, spatial proximity of nestlings may provide the basis for future helping among siblings, but helping also occurs across generations indicating that association when provisioning a brood or when being provisioned also provides the basis for future helping Nam et al., 2010). Precisely when the interactions took place, how many interactions there were, their duration, and the specific behavior and information transfer that took place during these interactions may influence how individuals are recognized and treated later in life. A critical issue here is that individuals often become familiar with and recognize many conspecifics through their lifetime, including mates (Blumenrath et al., 2007), territorial neighbors (Stoddard, 1996) or flock mates (Nowicki, 1983), so is it the timing, frequency or nature of the social interaction that results in some individuals being treated as kin and others not? A particularly nice example of such context-specificity in kin recognition is suggested by Komdeur et al.'s (2004) finding that Seychelles warbler helpers assist in the rearing of half-siblings that are the offspring of their mother but not those of their father, even though both parents would have provisioned the helper when it was young.
It may also be possible for individuals to acquire cues to kinship based on observations of the behavior of their familiar relatives toward other individuals. For example, unfamiliar individuals observed engaging in positive interactions with one's parents could be treated as kin. Indeed, such "indirect familiarity" could provide a kin recognition mechanism through which individuals recognize their younger siblings, despite not being reared together. Although we are not aware of any evidence for indirect familiarity among cooperative breeders, this idea parallels the social interaction expected under indirect reciprocity, in which help is directed toward an individual who has been observed providing help to others (Nowak and Sigmund, 2005). However, indirect cues to kinship are likely to be more error-prone than those learned through direct association because the link between kinship and familiarity will tend to be diluted. For example, in the case of direct association among parents, offspring and siblings during rearing, kinship of familiar individuals will usually be consistently high. But, if an offspring observes their parent interacting positively with an uncle, say, its relatedness to the "indirectly familiar" individual is lower than that between directly familiar individuals. If the offspring subsequently helps its uncle, and this is observed by their offspring, the relatedness between such "indirectly familiar" individuals is further reduced. As with direct familiarity, the frequency and nature of the interactions observed must also be considered, which, overall, may make the behavior of others a noisy and unstable cue to kinship.
Social network analysis is being used increasingly to quantify the strength of association between individuals and can be applied at different life history stages (Kurvers et al., 2013;McDonald et al., 2016). A social network inevitably reflects the nature of the behavior used to construct it (Madden et al., 2012), and they do not necessarily reflect genetic relatedness alone (Godfrey et al., 2014). For example, Napper and Hatchwell (2016) found that helping decisions in long-tailed tits reflected not only kinship, but also individuals' spatial distribution and their social associations during the previous winter. More work is needed to evaluate how prior association affects kin-directed behaviors using precisely quantified social networks in different contexts and life history stages.

When Is the Sensitive Period?
There is good evidence that kin recognition requires a period of learning, but when is this critical period? Many vocal learners have a sensory learning phase or window when they learn songs that they sing during adulthood (Kroodsma, 1978). Once this window closes, most songbirds are unable to learn new songs, although their repertoire may later be modified in some species (Mooney et al., 2008). Studies of songbirds show that that the window can be very short with a long delay between the sensory learning phase and the sensorimotor phase, during which the song is rehearsed and perfected, e.g., swamp sparrows Melospiza georgiana (Marler and Peters, 1982). Likewise, offspring that imprint on parents have a sensitive imprinting period (Bateson, 1964), and it has been suggested that learning of parental calls may even precede hatching, resulting in a parent-specific password, in superb fairy-wrens Malurus cyaneus (Colombelli-Négrel et al., 2012). This is interpreted as defense against inter-specific brood parasitism, but selection for early parentoffspring recognition would also be expected whenever there is a substantial risk of mis-directed parental care. For example, parents in colonial bank swallows Riparia riparia accept any offspring in their nest before 15 days, then recognize their own offspring at 15-17 days, i.e., just before fledging (Beecher, 1982(Beecher, , 1988(Beecher, , 1991. Based on these parallels between bird song learning and associative learning of kin, we postulate that the critical period for learning the template for recognition of kin through familiarity is similar to the sensory learning phase in many vocal learners. Thus, individuals could discriminate kin from non-kin even though they also associate with non-kin before they start vocalizing (Radford, 2005) or cooperating, and any associations that occur after the sensory learning phase (but before the sensorimotor phase) might result in non-kin being disregarded as social partners (i.e., associated but not "familiar"). Crossfostering experiments provide strong empirical support for this putative learning period Komdeur et al., 2004;. However, although the time of call development is known in some species (e.g., , the precise timing of kin recognition template formation has not been identified in any cooperatively breeding species. Furthermore, while this mechanism may be effective as a rule for reliably directing care toward kin when mature offspring help their parents or siblings to raise subsequent broods, as is typical of many cooperatively breeding birds (Cockburn et al., 2017), there are species in which helpers care for the offspring of a younger generation of breeders (e.g., Richardson et al., 2007;Nam et al., 2010), suggesting that older birds can learn the identity of younger relatives, a process that must occur outside the putative critical learning period. A similar conclusion must be drawn when parents avoid breeding with younger relatives, as in acorn woodpeckers (Koenig et al., 1999).
Therefore, while the parallels with song-learning are intuitive and appealing, there are clearly situations in which a single sensitive period for learning kin identity do not apply. Crossfostering experiments targeted at different life history stages and social network analysis across lifetimes provide invaluable tools with which to address this problem, but there remain formidable challenges to achieving a better understanding of the putative learning phase in natural populations.

What Are the Recognition Cues?
Another challenge when determining the role of familiarity is determining the cues used in kin recognition. Vocal cues are the most likely mechanism in birds ( Table 3), but this has been the default sensory modality in all of the cited studies, so visual and olfactory cues cannot be ruled out. Kin recognition mechanisms in several non-cooperatively breeding birds, particularly in the context of inbreeding avoidance, have focused on odor cues (Coffin et al., 2011;Krause et al., 2012). Storm petrels Hydrobates pelagicus prefer non-kin odors when choosing mates (Bonadonna and Sanz-Aguilar, 2012) and odor has also been suggested as a recognition cue in zebra finches Taenopygia guttata (Caspers et al., 2013, but see Ihle and Forstmeier, 2013). These studies should encourage future work on olfactory kin recognition in cooperatively breeding birds for two reasons. First, most recent evidence of odor-based kin recognition comes from species with enclosed nests, which may retain odor more readily than open nests, thereby promoting the learning and familiarization of nest odors. Many cooperative breeders nest in domed nests or cavities (Price and Griffith, 2017), suggesting that olfactory cues to kinship are plausible. Second, most species in which odorbased kin recognition has been identified live in flocks or breed in colonies, even though they do not breed cooperatively, suggesting that there might be common selection pressures for odor-based kin recognition to evolve. Interestingly, the finding that preen gland secretion chemicals are positively correlated with MHC relatedness in black-legged kittiwakes Rissa tridactyla (Leclaire et al., 2014) suggests that phenotype-matching of odor cues is a feasible recognition mechanism, just as in mammals (Green et al., 2015). However, it should also be noted that even less is known about the timing of development, individuality and repeatability of odor profiles than is known about vocal cues.
For any kin recognition cue, whether vocal or odor, to be effective it must carry either an individual or family signature and be individually repeatable from its initial development to the time of discrimination; the same logic applies to a recognition template. Signal convergence therefore presents a significant problem for the stability of any recognition system. Frequent interactions may lead to an increase in phenotypic similarity among individuals. Vocal convergence can be adaptive for coordinated foraging (Bradbury and Balsby, 2016), particularly when birds forage in annual winter flocks that disband each spring. For example, black-capped chickadees Parus atricapillus, exhibit vocal plasticity throughout adulthood and vocal convergence can occur within a week of winter flock formation (Nowicki, 1989). However, such species do not breed cooperatively and individuals do not gain indirect fitness benefits from associating with kin. In kin-selected systems, kin recognition cues must be fixed during early development and cannot be updated during adulthood, even when interactions with non-kin are frequent (Radford, 2005;Keen et al., 2013). In long-tailed tits, vocalizations do not change significantly over an individual's lifetime once learned , but more studies that investigate the plasticity of putative recognition cues are needed. In addition, while the idea of a signature system, a specific profile of phenotypic components that vary in their combination from individual to individual, is well-established (Beecher, 1982), most studies continue to focus on a single recognition modality, rather than recognizing that familiarity is likely to be based on a combination of cues, which may minimize convergence and maintain recognition cue diversity and integrity.

Familiarity vs. Phenotype Matching
Although in principle the mechanisms of familiarity and phenotype matching are readily distinguished, in practice this may often not be the case. The two mechanisms differ in the predictions they make about whether the ability to discriminate requires prior association and about the resolution of discrimination. First, familiarity is explicitly dependent on social partners having prior knowledge of each other, whereas phenotype-matching allows recognition of unfamiliar kin. In practice, it is extremely difficult to rule out prior association in most field studies, even in cross-fostering experiments where there is often a period of association between parents and offspring prior to separation (e.g., Hatchwell et al., 2001). Kin recognition cues may even develop during gestation (e.g., Hepper, 1987) or incubation (e.g., Colombelli-Négrel et al., 2012;Dowling et al., 2016). Secondly, familiarity is generally assumed to result in dichotomous classification of conspecifics into familiar (kin) and unfamiliar (non-kin) individuals, while cue-template similarity under phenotype-matching is assumed to be continuous. However, if the recognition system involves a threshold for acceptance/rejection of social partners (Reeve, 1989), then discrimination based on phenotype-matching and familiarity may appear very similar in practice. Equally, it is possible that familiarity could be assessed as a continuous trait, with conspecifics discriminated according to their degree of familiarity.
Thus, the extent to which recognition cues permit kinship to be perceived on a continuous or binary scale is an important aspect of the kin recognition mechanism. When group membership is used to categorize relatives, as in Arabian babblers Turdoides squamiceps (Wright et al., 1999), kin discrimination is binary. When recognition is based on phenotype, e.g., whitefronted bee eaters Merops bullockoides (Emlen and Wrege, 1988), it may be binary or continuous, depending on the algorithm used to assess kinship. Binary or threshold kin discrimination will be effective in most cooperative breeders living on stable territories that, at least with regard to the helping sex, are mostly made up of first-order relatives, facilitating a decision rule based on prior association (Curry, 1988;Payne et al., 1988;Komdeur et al., 2004). In contrast, a recognition cue that permits individuals to discriminate kin varying in relatedness has been identified only in the bell miner (Wright et al., 2010), even though such fine-scale discrimination has been reported in at least one other species (Nam et al., 2010;Leedale et al., 2020).
The ability of helpers to assess the relatedness of conspecifics continuously may have been overlooked in some cases because of the way in which cooperative behavior is measured. For example, some studies focus on the probability of helping (Curry, 1988;Creel et al., 1991;Dickinson et al., 1996), whereas others measure the amount of help given (Dunn et al., 1995;Wright et al., 1999;Clutton-Brock et al., 2001), and both have been measured in just a few (Emlen and Wrege, 1988;Komdeur, 1994;Russell and Hatchwell, 2001;Nam et al., 2010). Moreover, consideration must also be given to how relatedness is assessed by helpers, especially the possibility of error and degree of resolution achievable (Leedale et al., 2020). These problems pose formidable challenges to empiricists, with more sophisticated observations and experiments required to determine how relatedness is perceived.

CONCLUSIONS
Familiarity is an intuitively plausible mechanism of kin recognition in social birds that, at first sight, appears more parsimonious than alternatives. However, we think that this apparent parsimony is deceptive, so that although most empirical studies support familiarity as the most likely mechanism, we argue that there is a great deal we do not understand about this process. In particular, we have identified four specific issues that would benefit from further investigation, although in making these recommendations, we acknowledge the difficulty of addressing them in natural populations.

AUTHOR CONTRIBUTIONS
AL collated research and drafted the manuscript with BH. AL, BH, and JL contributed to the final version of the manuscript.