R2.1. Psychological and neural mechanistic explanations

I first address the commentaries that propose extensions to the target article's framework in ways that would explain how teaching works in terms of proximate, causal mechanisms. These include mental, cognitive, neural, hormonal, and behavioral mechanisms.

Commentaries by Ronfard & Harris, Koenig, Kruger, and Corriveau, each, respectively, argue for greater coverage of a topic not covered in detail in the target article: the behavior of pupils as learners. I agree that this is an important topic underserved in the target article, and I thank these commentators for highlighting it. Ronfard & Harris as well as Corriveau emphasize that among humans, pupils may solicit teaching, may monitor or direct ongoing instruction, and may seek out reliable informants (see sect. 6.2 in the target article for brief coverage of these ideas). Ronfard & Harris even nominate pupil's behavior as another important piece of the puzzle in understanding human cultural capacities, and cite a broad body of work showing that children seek out adult attention. This raises the question of whether such a practice occurs in nonhuman pupils, as well. In some cases, I argue that we know it does: teaching via social tolerance depends by definition upon pupil initiation, and seems to be relatively common among nonhuman animals. The question of pupil behavior extends even beyond the suggestions made by Ronfard & Harris, and could inspire comparative work in the future.

In addition, Corriveau and Kruger argue that teacher and pupil should not be envisioned as independent actors. Both these commentators take issue with the target article's asymmetrical treatment of teacher and pupil behavior. They make an important point in recognizing how humans (and other animals) coordinate, and that this may be observable in teacher–pupil interactions. It may be especially interesting to describe how the behavioral and psychological mechanisms for coordination do or do not differ across human populations, where social hierarchies may restrict the entreaties of young learners (see, e.g., Lancy Reference Lancy2008), or where norms about verbal communication and eye contact mean coordination is more subtly communicated, as is common in Pacific societies (Watson-Gegeo & White Reference Watson-Gegeo and White1990). Explaining the communicative mechanisms by which teachers and pupils jointly interact is not mutually exclusive with tracking the fitness costs and benefits separately. Even for a public good created through cooperation, and even if all parties benefit equally, the reproductive or fitness consequences of behavior must be tallied individually, since it is their fitness relative to each other, rather than absolute reproductive fitness, that predicts the outcomes of natural selection. It is this practical concern that drives the target article's focus on the evolution of teacher behavior over questions regarding pupil behavior. In the context of the target article, this is as a parallel line of inquiry into a mechanistic explanation of teacher and pupil behavior that benefits from the proposed functional framework, even as it requires shifting focus toward pupil behavior.

Pasquinelli, Zalla, Gvodzic, Potier-Watkins, & Piazza (Pasquinelli et al.) discuss how little is known about the neurocognitive mechanisms of teaching, and make a worthwhile appeal for translational, applied research using established methods in cognitive psychology and neuroscience. Similarly, Kim, Torres-Garcia, & Swain (Kim et al.) focus on the neurocognitive and hormonal bases of teaching. These are growing fields, and better integration of their techniques may advance the study of teaching. Both commentaries, however, treat teaching behavior as a unitary phenomenon rather than a set of behaviors. Kim et al. even subsume “parenting” under teaching, which is either a narrow definition of parenting, or a loose definition of teaching that does not coincide with the one proposed in the target article. While Pasquinelli et al.'s focus is a valuable and promising research direction, I urge researchers to step away from the entrenched questions their commentary highlights: whether humans have a “teaching instinct,” and its dependence on theory of mind. A central tenet of the target article is that teaching behaviors are not monolithic in their form and mechanism – and as a result we should not assume that proximate teaching psychologies are uniform, either. This demands that we take a more textured look at the variety of teaching behaviors used by a given study species, by linking the proximate psychological mechanisms to their ultimate functional explanations.

Pasquinelli et al. also suggest that teaching behavior might directly benefit teachers in two ways. First, that “learning in order to teach” creates direct benefits to the teacher, through enhancing the teacher's own learning. The evidence cited in favor of this idea is based on grade-school children and on undergraduates asked to learn in order to teach in a classroom context. This cannot distinguish between the impact of teaching versus the impact of preparing-to-teach, on the teacher's comprehension. Thus, while this research may impact in-classroom teaching techniques, it seems unlikely to affect functional explanations of the cost/benefit structure of teaching behavior. After all, learning for the anticipated purpose of teaching is a special case in human teaching, and almost certainly non-existent other animals. Second, Pasquinelli et al. suggest that teachers might manipulate others for their own direct gain, and that manipulation in a broad sense may be considered teaching. Communicative behaviors used in persuasion and manipulation are superficially similar to those displayed in teaching. Further, it may be that the two share a proximate psychology: The same psychological features that make learners receptive to teaching also make them vulnerable to manipulation, and manipulators may use the same communication techniques that can be useful for teaching. This is a broader problem in human communication, and demands epistemic vigilance among communicators (for deeper discussion, see Sperber et al. Reference Sperber, Clément, Heintz, Mascaro, Mercier, Origgi and Wilson2010). However, integrating mechanistic psychological descriptions with functional explanations makes it clear that manipulative influence should not be categorized as teaching. Teaching, as functionally defined, is a cooperative behavior in which the teacher benefits only indirectly, through the learning benefits the pupil derives. This illustrates why a multi-level approach to the study of teaching is important. For further discussion, see deceptive teachers and skeptical pupils (sect. 6.2) in the target article.

Ravignani & Sonnweber make an ambitious proposal to dissect the proximate mechanisms of the full range of teaching behaviors through hormonal and neuromodulator evidence, as well as time-series analyses. They argue that studying the physiological mechanisms will help scientists to identify the motivations and emotions involved in teaching behavior. While this is one promising tool scientists may use to study teaching, I caution against relying too heavily on proximate mechanisms to diagnose teaching. It poses the same sort of problem discussed above – and repeats the same stumbling blocks for using theory of mind capacities as a measuring stick. In the target article (sect. 5), I argue that there is more than one possible psychological mechanism that could reliably produce a given type of teaching behavior. This suggests that using a single proximate mechanism as a heuristic for identifying teaching runs the risk of excluding behavior that fits the functional definition of teaching.

This same critique applies to the use of teacher–pupil synchrony to diagnose teaching behavior. Marin states that motor movement synchrony (and indeed a wide range of mutual influence between pupil and teacher) can happen during teaching. However, this does not mean it is particularly diagnostic of all types of teaching behavior in all species. For example, for teaching types that promote pupils' independent exploration of novel stimuli (teaching via stimulus enhancement and via opportunity provisioning), behavioral synchrony would not be expected. It also does not establish that synchrony is unique to teaching, as opposed to being common in communication more generally. Similarly, Badets & Osiurak state that perceptual resonance is common across animals, but it is not clear that it is especially relevant to the evolution of teaching, as opposed to perception or social interaction more broadly.

Finally, Palagi, Stanyon, & Demuru (Palagi et al.) highlight play as one of many “forms of teaching and learning.” I think this is a mistake. Although play certainly facilitates learning, and although teaching behavior may happen in the context of play, behaviors commonly accepted as “play” can fall outside of any of the definitions of teaching discussed in the target article. For example, solitary play cannot possibly be categorized as teaching, since it takes (at least) two to teach. While some kinds of teaching may resemble some kinds of play, behaviors sharing proximate mechanisms do not necessarily share an ultimate function. For this reason, a functional explanation of teaching behavior makes an ideal structure upon which to build a framework for comparative research. For example, not all displays of ostensive cues constitute teaching, and not all teaching relies upon ostensive cues. Some behavioral cues, like “motherese,” may be teaching-specific, but this is an empirical question requiring further study. Palagi et al. describe gestural motherese in gorillas as being “direct” and “active,” but this does not categorize it as direct active teaching as defined in the target article. Instead, teaching behaviors should be categorized based on the learning problem they are thought to solve, with consideration of the learner's access and attention. Within the target article, I explicitly classify motherese as a form of teaching via stimulus enhancement (see sect. 3.1).

R2.2. Social, cultural, and other contextual mechanistic explanations

Another group of researchers focuses on a proximate description of the social, cultural, or other contexts of teaching behavior. Rogoff begins by pointing out that cultural anthropologists and others have been richly describing a broad array of teaching and learning behaviors for decades, and that this includes behaviors that fit into my five teaching type categories. This is right – and in the target article I credit such researchers for their achievement (see sect. 2.2), while drawing heavily on their descriptions in the empirical review of teaching in humans (sect. 4.2). To clarify, I do not argue that culture-based approaches are without descriptions of the range of behaviors I discuss in the target article. In fact, there is a richness in that literature that I could not begin to address in the target article. This includes one-to-many and many-to-one teaching, which Rogoff discusses, and questions that draw on social theories of power, knowledge-making, modes of communication, and meaning, to name a few. It also includes aspects of teaching behavior featured in Rogoff's graphic, ranging from community organization, to learning goals, learner and others' behavior, and the mental state or motivation of the learner and others (see Rogoff's Figure 1). These are all valid domains of study, at the level of describing sociocultural contexts. However, descriptions and analyses made without a shared framework do very little to integrate the study of teaching behavior across levels of explanation. Further, much of this qualitative data is exclusive to human teaching and is analyzed without respect to the features we do share with other animals.

This is a problem for an integrative framework for use across human and animal populations. Drawing categories of social learning and teaching in terms of social meaning and cultural context can work as a coherent mechanistic-level analysis. These categories have affordances for discussing the questions that cultural psychologists, for example, are interested in. However, it is incompatible with functional and phylogenetic analyses, and with any comparison with nonhuman animals. This incompatibility limits the usability of data from the culture-based approach by other researchers. In turn, this confines the questions one can ask from a culture-based approach to the mechanistic level. For instance, only a shared framework allows the combination of evidence from cognitive developmental and ethnographic descriptions for a better understanding of how humans, as a single biological species, can demonstrate so much behavioral variation in teaching, across the lifespan. This variation can be described in detail from a culture-based approach, and its subjective meaning can be examined. However, if we want to know how humans came to be this way, or to shift the goals of educational systems towards creativity and away from uniformity, then ontogenetic and functional perspectives are key. Since developing these explanations depends in part upon comparison with other species, a multi-level approach grounded in phylogeny is the best option. This is not a call for cultural psychologists and anthropologists to abandon their questions – it is merely a call to integrate their questions, and their very valuable data, with complementary work on teaching from other approaches.

In many ways, the culture-based approach can be compatible with the functional framework proposed here, so long as we recognize it is a different level of explanation, and seek to reconcile those levels of explanations. Lancy argues that the definition of teaching used in the target article is too broad, and that researchers should persist in the culture-based approach to the study of teaching. However, the breadth of a functional definition is only problematic if one interprets the framework as obligating researchers to use functional definitions exclusively, and to use these definitions without operationalizing criteria for identifying teaching behavior. I am not persuaded, and refer readers to the target article's discussion of the benefits of integration (sect. 2.4). A key consideration is that the functional framework would not prevent researchers from focusing on specific contexts or genres of teaching should they have the intellectual motivation to do so. The teaching typology means research focused on a particular kind of teaching – for example, direct active teaching – can be fitted into the framework with minor adjustments. In fact, the functional framework is perfectly open to the study of when and how learners may solicit teaching – a concern Lancy highlights. Further, distinctions between what Koenig calls “mode” of learning – hearing, watching, seeing – are compatible with a functionalist approach as a mechanistic explanation. Finally, the framework also does not require granting privileged status to teaching as a means of learning, so it can be integrated into the ethnographic and theoretical study of social learning more broadly. Research on social learning has itself benefited from the kind of integrative framework I propose here, and has allowed for cross-species studies. As is clear from my own empirical work (e.g., Kline et al. Reference Kline, Boyd and Henrich2013), teaching is but one of many kinds of social learning.

Lancy also suggests that the illustrative examples I use from my field work in Fiji are irrelevant, because (as I explain in the target article and elsewhere in my published work) the indigenous Fijian people I work with do attend British-style formal schooling. However, I include these details precisely because I think they are important for contextualizing my qualitative observations and quantitative data (see, e.g., Kline et al. Reference Kline, Boyd and Henrich2013). Moreover, the presence of formal schooling does not render these examples useless, nor does it undermine the overarching goal of the target article, which is to propose a synthetic framework. These ethnographic examples are single events, chosen specifically to illustrate teaching types. Examples like this cannot be used to generalize about a population, and disregarding these examples would not topple the proposed framework. The target article includes a range of examples of teaching behavior from the broader ethnographic literature (see sect. 4.2). Further, argument by example cannot tell us whether Western education correlates with (let alone causes) teaching behavior outside of the classroom. To test this kind of relationship would require a large-scale cross-cultural comparative study with quantitative measures of Western schooling and the prevalence of teaching types per population. Alternatively, we might test whether within populations, exposure to Western schooling for the learner or the teacher predicts likelihood of teaching. For the Fijian population in which I work, the answer is that it does not (see Kline Reference Kline2013). Still, Lancy's question highlights the need for a broad, quantitative comparative study of teaching behavior across human populations. The target article has outlined the framework that makes this kind of study possible.

R2.3. Applied uses for mechanistic explanations

Several commentaries have discussed human socio-cultural institutions, such as classrooms, apprenticeships, or styles of parenting, that can affect the performance of teaching. While the target article focuses on kinds of teaching behavior rather than on the human socio-cultural institutions that structure and influence teaching behavior, the study of those institutions is not in conflict with the approach I propose. In fact, this focus on context is one of the most valuable contributions of the culture-based approach can contribute to the study of teaching more broadly. While socio-cultural institutions can be analyzed at other levels, most of the commentaries on these topics focused on how it is that teaching happens, and in what contexts. In this way, they make use of mechanistic accounts of teaching behavior and its contexts in humans.

Shafto & Seifert take an applied approach, asking how we might optimize learning through a combination of teaching and self- and peer-directed learning. In an example regarding peer-directed learning, they contrast peers with teachers as important sources of knowledge. However, in the analysis in the target article, teachers and pupils are not separate, mutually exclusive populations: they are fluid roles that are shorthand ways of describing behavior. A classroom in NASA or at Google may contain only one person employed as a teacher, but within the framework I propose, students can just as easily be performing teaching behavior, and a single individual can be both a learner and teacher. Like a classroom, an apprenticeship is a social arrangement and a context, and not a type of behavior.

Similarly, a syllabus or overarching goals of an educational system are social institutions designed by humans in order to shape teaching behavior to reach a given end – for example, conformity or creativity. Poddiakov likewise describes a number of contexts in which teaching may happen, for example, when a teacher's interests are in conflict with a pupil's. Because teaching evolved as a cooperative behavior, this context poses a cheater-detection problem for pupils, discussed at length in the target article (sect. 6.2). Despite this, what Poddiakov terms “negative work” in teaching, and the kind of false-information that Koenig discusses, are not separate kind of teaching behavior – both are contexts in which teaching behavior happens, and may be examples of teacher-pupil conflict where the pupil's role as skeptic could be studied. For example, Müller suggests a specific domain in which teaching and social learning may proliferate unhealthy behavior: psychiatric disorders.

Researchers whose goal is to define and create strategies for teaching that maximize a given outcome – be it creativity or conformity – would do well to consider teaching from more than one level of analysis. Shafto & Seifert point out that humans are good teachers but not optimal ones. This makes sense from a functional perspective because natural selection produces traits that are best-suited to maximize actors' reproductive success. This is not the same as optimizing teaching for the best possible learning outcome, because there are cost/benefit tradeoffs that affect teachers and pupils. We should not assume that evolved teaching behaviors are optimal teaching behaviors; instead, we can think of them as tendencies or biases that should be taken into consideration in attempts to optimize the teaching process.

R5.1. Extensions within a functionalist approach

Commentators have also made suggestions for extensions of the functional analysis offered in the target article. For example, Johnston, McAuliffe, & Santos (Johnston et al.) make a compelling argument for how dogs may serve as a useful study species, because they have evolved to learn from humans, and to attend to ostensive cues. Their proposal is intriguing and novel – though, as with any cross-species comparison, caveats apply. For example, basic differences in dog and human cognitive and communicative capacities mean that the study of human direct active teaching with dogs as pupils is probably not possible.

Scott-Phillips & Sperber contribute richness to the theoretical foundations of the target article, highlighting the importance of psychological, biological, and environmental factors of cultural attraction. Abrams details transmission probability interactions as a system that makes it possible to build abstract mathematical models of teaching and of cultural attractors. No matter their origins, attractors can preserve and construct what is to be learned – and may work in concert with or against what is being taught. A future analysis might delve deeper into the importance of attractors for the range of different teaching types. For example, for teaching that enables individual learning, attractors may play a major role in determining learning outcomes, while teaching affects whether learning is possible. These authors also point out that the study of teaching behavior is especially compatible with cultural attraction theory because teaching (a) functions to influence the frequency of practices or variants other than itself, and (b) the impact of teaching is in part dependent on other sources of cultural attraction. In the context of cross-species comparisons, a major caveat is that not all teaching promotes cultural learning in particular, and only some forms of teaching may be culturally learned. Cultural attractors may be especially relevant for the study of direct active teaching, and for applications of this theory to improving the effectiveness of teaching methods.

Tatone & Csibra argue that despite the synthetic value of the framework, ostension must still be key to understanding human teaching. While I agree that in some contexts, ostension is a key part of human teaching behavior, I think it does not cover the range of teaching behaviors that exists. The goal of the framework above is not to rule out ostension as a feature of a subset of teaching behavior, but it does include a range of teaching behaviors that do not necessarily rely upon ostension. I also do not see ostension as a reliable cue to teaching, as distinct from other kinds of communication.

Hernik & Gergely argue that teaching fills the function of enabling a naïve individual to learn content that is opaque and generic. Their commentary questions the validity of basing the target article's framework in a learner's access and attention, and redirects the focus to four problems of opacity. These indeed do pose problems for learners, but the identification of these problems is relative to the learner and the context (i.e., to the learner's access and attention). The framework in the target article explicitly captures this, by taking the learner's access and attention into account. In this sense, opacity is not treated as an objective feature of the world, because it depends upon the learner's existing store of relevant knowledge. Considering the attention and access of the learner can characterize the learning situation as more or less opaque, rendering particular types of teaching useful or irrelevant. Conceptualizing the pupil's access and attention are not in conflict with identifying kinds of opacity, and in some ways these concepts are addressing the same problem: Why is it the pupil cannot learn, or will learn more slowly, without aid from a teacher?

Finally, while Caro endorses the need for a broadly comparative database, he is not interested in friendly campfire tunes, instead arguing that the mentalistic and culture-based approaches require constraints that make the kind of synthesis I outline impossible. I maintain that a functionalist approach can benefit from integration with the research questions and the empirical observations made across specialized fields (sect. 2.4). Further, I think the application of multiple levels of explanation in this response makes that compatibility explicit rather than hypothetical. In the end, I propose we employ the framework from the target article and evaluate its usefulness empirically.

R5.2. Operational definitions

Another group of commentaries generally approves of the functionalist approach, but expresses doubts about the practical application of the proposed framework, given that operational definitions are only addressed briefly in the target article.

Dean & Kendal argue that the framework focuses too heavily on teacher behaviors as opposed to learner behavior. However, a key component to the framework is to match teacher behaviors to learning problems. These learning problems are defined from the perspective of the learner – whether the learner lacks access or attention to information. From a practical standpoint, this may mean that observational and experimental studies will need to focus on the learner. From a theoretical perspective, it means that the adaptive problem is always with respect to the learner's status. For this reason, learner behavior is indeed an important source of information in arguing for or against alternate explanations of candidate teaching behaviors. In fact, learner behavior can serve both as evidence of an existing learning problem that might be solved by teaching, and as a measure of the fitness-relevant effects of teaching should it be observed. Both kinds of evidence are important in identifying cases of teaching behavior.

Alternatively, Dean & Kendal return to the cost/benefit criteria first proposed by Caro and Hauser (Reference Caro and Hauser1992). Caro and Hauser's criteria are admirably simple and should serve as one of many kinds of evidence for the function of behavior as teaching (see target article, sect. 2.3, para. 1). However, it does not follow that these criteria are an objective rule by which to identify teaching. Take, for example, the criteria that there is a cost, or at least no immediate benefit to the teacher: applied without a researcher's inference or subjective interpretation, this would neatly rule out all instances of (adaptive) teaching as cases of teaching. This is because, for teaching to evolve, there must be some fitness benefit to teachers as opposed to non-teachers. The fact that Caro and Hauser's criteria have been used to identify and confirm cases of teaching is evidence that researchers (quite rightly, I think) apply it in selective ways that make sense within the context of their studies. This is not to say we should do away with Caro and Hauser's criteria. Instead, we should regard it as only one of many lines of evidence for specialized design: that the form of the behavior fits the proposed function of teaching. Further, we should aim to show a behavior fits this proposed function better than it does an alternate function.

In their argument against the adoption of the target article's framework, Dean & Kendal do not consider the discussion of learning problems upon which the framework rests. Specifically, they do not consider that teaching types correspond to a learner's access and/or attention. As a result, they suppose that when otters and sea lions drag their offspring into the water, this fits the framework's definition of either teaching by stimulus/local enhancement or teaching by opportunity provisioning equally well. They take this as an indication of broader problems with the framework. Ironically, given their suggestion for greater focus on learners, the distinction between these two forms of teaching is in the learning problem, from the perspective of the learner. In these examples, the pups are not prevented by circumstances or by conspecifics from entering the water. This means there is no access problem for the pups. However, they only seem to learn to swim as a result of being forced into the water by their mothers. This suggests that they lack attention to the water, and only engage in swimming when their attention is forced. We can imagine that pups who are never pulled into the water by their mothers might never learn to swim, or might learn much later without interference. In either outcome, opportunity provisioning does not fit the scenario, since it corresponds to cases where learners lack access to a stimulus, and do not lack attention. Opportunity provisioning would apply if, for example, otter pups consistently attempted to enter the water but were not able to do so because they were physically prevented from accessing it by their small size, and could only reach the water through assistance in gaining access by their larger-bodied mothers.

Similarly, commentators Moore & Tennie and Eschar & Fragaszy favor an evolutionary approach focused on adaptive problems, but take issue with the framework's categories. First, Moore & Tennie suggest my definition of teaching by stimulus enhancement is meant to be a new definition of stimulus enhancement more generally. As discussed above, I am not arguing for a new definition of stimulus enhancement and would not categorize all learning by stimulus enhancement as teaching. They go on to argue that lumping teaching behavior by function rather than by cognitive difficulty is “not intuitively a useful tool for understanding evolution.” This is demonstrably untrue, since evolutionary reasoning is regularly and fruitfully applied according to functional definitions. For example, the reproductive habits of soapbox bugs and of humans rely upon wildly different cognitive mechanisms, yet we can understand frequency-dependent mating strategies in both species with the same kind of cost/benefit rationale. On the other hand, a functional framework is just one level of analysis. This shared functional analysis does not prevent us from studying the cognitive complexity of each teaching type, or of teaching behaviors that cut across functional categories. On the contrary, it facilitates phylogenetic comparisons, which help researchers to understand the details of how and when individuals in each species do or do not adjust their behavioral strategies. This gives us insight into the behavior's proper domain and the constraints on its flexibility, including cognitive constraints.

Moore & Tennie go on to argue that the flossing behavior of Japanese macaques, as discussed in the target article, could fit the definition of “direct active teaching.” In contrast, they make the suggestion that Warao fathers' repositioning of their sons' hands during canoe carving should fall under evaluative feedback rather than direct active teaching. This is only possible because they substitute their own intuitive definition of what constitutes “direct teaching,” for the one I propose, stating “we often engage in direct teaching by drawing others' attention to important features of objects.” In asserting this, Moore & Tennie step away from the definition of direct active teaching proposed in the framework above, such that these specific critiques cannot apply in a meaningful way. The source of confusion here seems to be similar to Dean & Kendal's, in that the focus is on apparent behavioral similarity (i.e., something can be described as “direct” and “active” in lay terms) rather than how researchers might tease apart behaviors based on their function in facilitating learning. This problem is not unique to the evolution of teaching behavior. For example, researchers have had to create ways to discriminate mounting behaviors that function to establish dominance from those that function as reproductive sex. As in the framework I propose above, distinctions are made by contextual indicators, including others' behaviors, rather than by a distinguishing feature of the actor's behavior alone.

Eshchar & Fragaszy state that “[f]raming teaching in terms of its evolutionary sources does not help the empirical researcher to recognize it.” I strongly disagree. Observed teaching behaviors should, all else equal, represent a specialized functional match to the context of the adaptive problem, and, if functioning in its proper domain, should have fitness-enhancing consequences for the teacher and pupil. This form–function fit can generate insights into the ways cognition can operate to reliably produce teaching behavior in teachers and pupils.

Mapping conditions under which the behavior's functionality breaks down can serve as evidence of the adaptation's proper domain, and experimental methods may help us to probe that landscape to better understand the flexibility of an adaptation (Barrett Reference Barrett2014). Similarly, ontogenetic evidence can be used to infer a trait's function and further detail its flexibility or lack thereof. Phylogenetic analyses can temper adaptationist reasoning with an understanding of the evolutionary origins of underlying traits, as suggested by Smaldino & Newton. Further, comparative studies based on phylogenetic analyses may help us to identify recurring social learning problems that can be solved by teaching, for a set of socio-ecological niches. All of this comes with a better understanding of the evolutionary sources of teaching behavior, and these multiple lines of evidence can be used to design better operational criteria for distinguishing among alternative explanations for potential teaching behavior.

As Rapaport discusses, there are a range of criteria that can and should be used to identify and characterize teaching behaviors. This is directly addressed in the target article (sect. 3). This section of the paper makes clear that, contrary to Eshchar & Fragaszy's assertions, the framework in the target article is not designed to include any and all learning that happens in any social context as teaching. Eschar & Fragaszy's critique is that the framework does not come with a generalizable operational rule to distinguish teaching behavior from all alternative explanations of behavior, so it must be impossible to differentiate between teaching and other kinds of learning under this framework. This argument is flawed. Our operational definitions are necessarily going to be variable and piecemeal, because we are studying a selection of behaviors spread across a range of species and contexts. Creating operational criteria will not be simple or easy, and I do not provide a universal operational criterion because I believe there is none that will apply across all study contexts. Again, this does not mean that researchers should subjectively judge what seems to be teaching and what is not. Instead, it means they should design tools of measurement and experimental and field methodologies to distinguish teaching from other forms of social learning and from other forms of communication. These methodologies should enable researchers to provide evidence that a behavior likely functions as teaching, and to collect evidence of functional specialization for teaching, as opposed to some alternate function.

As Fogarty highlights, this is a challenging problem for researchers, especially those conducting observational work without the option for experimental manipulation of their subjects. In mathematical modeling, Caro and Hauser's (Reference Caro and Hauser1992) criteria may be the easiest and therefore most fruitful to apply. However, this represents a tradeoff, since those models will only be able to tell us about a subset of socio-ecological contexts that the underlying assumptions accurately describe. To explore the evolutionary dynamics that could have given rise to the full range of teaching behaviors, a more differentiated set of models is needed. Nonetheless, this kind of model can be a powerful tool for generating theoretical insight beyond what verbal reasoning can tell us.

Rapaport meets the difficult question of what operational criteria we might use head on. She proposes we use Hoppitt et al.'s (Reference Hoppitt, Brown, Kendal, Rendell, Thornton, Webster and Laland2008) standard of only including “active” demonstration as teaching. However, this approach struggles with precisely the same questions as the framework proposed in the target article. Hoppitt et al.'s “active” or “advertent/inadvertent” distinction is simply another way of weighing alternative explanations of behavior against one another. Rapaport rightly notes that distinguishing social tolerance that evolved as teaching from social tolerance more generally is a difficult challenge in primate species that have reached ceiling-high levels of social tolerance. I agree with Rapaport's assertion that it is not scientifically sound to categorize all types of social tolerance as instances of teaching by social tolerance; I also agree with similar assertions made by Moore & Tennie with respect to teaching by stimulus enhancement, and by Eshchar & Fragaszy about the teaching types in general. However, these objections are somewhat surprising since they echo the argument made in the target article, in the very introduction to the framework itself (see sect. 3). In each of these cases, the commentators misinterpret some of the examples of teaching and as a result infer that the framework equates social learning with teaching. Here I discuss these misinterpretations in turn.

First, Rapaport notes that, in the example of the Balinese music teacher who plays for his pupils without adjusting his style of play, there is no behavioral adjustment whatsoever by the teacher. In fact, in this example, the music teacher puts on a performance exclusively for the pupils to observe – were the pupils not present, he would not play. The modification is that he is playing, not how he plays. Aside from the specifics of interpreting ethnographic examples, there is a broader point. The practical difficulty of identifying teaching by social tolerance does not necessarily lessen its theoretical validity. Because of the minimalist nature of teaching by social tolerance, it may predate the evolution of other, more derived forms of teaching. The conceptual role of teaching by social tolerance in this framework may be important in considering the socio-ecologies in which we would expect teaching to evolve, even if we are unable to reach a consensus on how to quantify and observe cases of teaching by social tolerance.

Eshchar & Fragaszy provide their own examples, presenting them as problematic for the framework. These examples are worth discussing in that they highlight some important misunderstandings. First, they propose that in a case of chimpanzee termite fishing, when a juvenile physically intrudes upon a foraging adult, it will be impossible to tell the “purpose” of the adult's social tolerance. They argue that teaching by social tolerance will be indistinguishable from the social tolerance that evolved because intolerant adults would incur direct costs. This is a testable hypothesis, provided the researchers can come up with a proxy measure of the cost or threat of costs (e.g., presence and rank of the juvenile's kin). This question can be settled by testing the competing hypotheses against each other, using measures developed for the study species and context. In fact, it is difficult to see why Eshchar & Fragaszy think this problem is specific to the framework described in the target article, as the same kind of challenge exists for all studies of teaching behavior, and more broadly for the measurement of behavior in general.

Eshchar & Fragaszy also suggest that under the proposed framework, teaching would include any behavior done in proximity to juveniles, even if there is no behavioral modification contingent on the learner's presence. This directly contradicts the target article's discussion of distinguishing teaching behavior from baseline behavior (sect. 3). These commentators also highlight the need to know whether a behavior is “for the purpose” or with the “intention” of teaching. This is not a reliable standard, for reasons discussed throughout the target article, and has been discarded by functionalists since Caro and Hauser's (Reference Caro and Hauser1992) critique. On the other hand, some of the ethnographic examples I draw upon in the empirical review section (sect. 4.2) do give accounts of the mental states of human teachers. As discussed in that same section, I include these examples because ethnographic data are necessarily impoverished of quantitative data on behavioral details, and the ethnographer's qualitative interpretation of the actor's intention can lend valuable insight when applying quantitative criteria is not possible. Using these examples as an act of necessity is not the same as endorsing intentionality as a criterion for identifying teaching.

Jacquet raises punishment as an example, asking how punishment can be distinguished from teaching by evaluative feedback and suggesting that this may necessarily be a grey area. Jacquet notes that others have distinguished punishment from teaching according to the timing of the fitness payoffs, reasoning that behavior that results in an immediate benefit to the actor should be considered to function as punishment, rather than as teaching (see, e.g., Thornton & Raihani Reference Thornton and Raihani2008). Evident in Jacquet's commentary are a range of definitions of what actually constitutes punishment, and I suspect this topic could inspire a review piece of its own. For the sake of clarity, I conceptualize punishment here as “negative feedback” in which the actor creates and imposes a cost on its recipient. The specific definition of teaching by evaluative feedback (sect. 3.1) is not as relevant here as the definition of teaching more generally as a cooperative behavior between teacher and learner (sect. 6). With these definitions, teaching via evaluative feedback is neither equivalent with punishment nor mutually exclusive with it. That is, some punishment may function as teaching by evaluative feedback, and some may not. Differentiating between these cases will require considering the costs and benefits to the candidate teacher and pupil in a particular context.

A defining feature of teaching as discussed in the target article is that the benefit to the teacher is indirect, derived from benefits to the pupil. In this sense, teaching is a cooperative behavior. This means punitive behavior that benefits the actor/punisher at the expense of the recipient/punished – for example, parking tickets or reputational damage – does not function as teaching. This is more precise than timing of costs and benefits since, in reality, teachers must benefit at some point regardless of timing, and punishers can accrue delayed benefits. Likewise, punishment that is not teaching can still create direct benefits for the punisher, as a result of the punished individual modifying her behavior. To continue the example, if those who receive parking tickets learn not to park in permit-only zones, the punisher gains parking availability. In this case, punishment is not equivalent to teaching because it is not cooperative: the punisher (the permitted parker) benefits at a cost to the punished individual (the ticketed, unpermitted parker). Where punishers derive indirect benefits from the modified behavior of the punished individual, and the punished individual also benefits, punishment is a form of teaching by evaluative feedback, albeit a relatively costly form. Of course, these questions become more complex when we consider that teaching, like other cooperative behaviors, is open to manipulation for selfish means by teachers. The waters are further muddied by considerations of how to measure the scale and flow of benefits and costs – but these are matters of methodological rather than theoretical distinctions and are beyond the reach of this response.

Sharkey provides an outside-the-box commentary, asking whether robots are teachers. Since robots are not living, reproducing creatures, they do not evolve or behave in the way that organisms do. Thus, in an evolutionary approach, robots are best characterized not as being teachers themselves, but as culturally evolved tools that humans use for teaching. This is because robot actions are not shaped by natural selection on robot behavior and reproduction, but instead are the result of natural and cultural selection on human behavior and culture. So, for example, where robots are deployed to provide an English-speaking companion for students studying English as a second language, this is an example of humans teaching via opportunity provisioning. Humans may even undertake direct active teaching via a robot, by equipping a robot with a pre-planned lecture or demonstration. Likewise, some book-based teaching may fall into this category since, like robots, books are cultural artifacts created by humans. Similarly, the online quizzes with feedback that are common in large university courses at the present can be considered teaching via evaluative feedback. From an applied perspective, the important difference between teaching that uses robots as tools and teaching that happens between two humans face-to-face, is the relatively poor ability of a robot to tailor its actions to the context and the learner's status. As noted in other commentaries, this reactivity between teacher and pupil (or more broadly, in acts of human communication) is important for the effectiveness of communication. This has implications for the use of robots as teachers-by-proxy: Teaching types that are more dependent on adjustment and tailoring may be less effective through the use of robots. This poses a problem for the use of robots in the classroom, because direct active teaching can be tailored, intensive, and quite common in a typical Western-model classroom.