We praise the Keven & Akins (K&A) target article for emphasizing that neonatal imitation findings must be read in the broader context of sensorimotor development, especially as portrayed by Thelen (Reference Thelen1979; Reference Thelen1981b). By describing tongue protrusion as one of many rhythmic stereotypies whose rate can increase in relation to arousal, and by indicating a precise timeline for the onset, development, and dropout of spontaneous tongue protrusion, K&A strengthen the arousal explanation of the neonatal imitation findings (Jones Reference Jones2009). Nonetheless, their support for the arousal theory presents some shortcomings. Brief examination of some empirical studies and theoretical alternatives suggests that the debates about neonatal imitation, and its relevance to social cognition, remain more open than K&A imply.

K&A's characterization of the operational definition of neonate imitation (as “producing the modeled gesture more often than an unrelated one,” (sect. 2, para. 2) is imprecise. Most empirical studies of neonatal imitation operationalize imitation as greater frequency of a gesture in response to the same gesture than in response to other gestures. The operational definition entails reference to a plurality of gestures exhibiting the comparative increase just described. This point is of critical importance because Meltzoff and Moore (Reference Meltzoff and Moore1977) were well aware that, if only one gesture was matched, arousal would be the most plausible explanation, and they envisaged the operational definition of differential imitation precisely to exclude this explanation. Thus, the primary question is an empirical one: Is there evidence for differential imitation? If the answer is no, if only tongue protrusion matching is evidenced, then arousal is the most plausible hypothesis; if the answer is yes, the arousal explanation is no longer viable (Anisfeld Reference Anisfeld, Hurley and Chater2005; Jones Reference Jones2009; Meltzoff Reference Meltzoff and Goswami2002; Ray & Heyes Reference Ray and Heyes2011).

In this regard, because current empirical literature remains ambiguous, K&A may be too hasty in siding with the negative answer. K&A do not mention a recent study providing evidence for differential imitation (Coulon et al. Reference Coulon, Hemimou and Streri2013); instead, they cite Oostenbroek et al. (Reference Oostenbroek, Suddendorf, Nielsen, Redshaw, Kennedy-Costantini, Davis, Clark and Slaughter2016), which did produce negative results but whose theoretical assumptions and experimental design are questionable. Relying on the supposition that early imitation has a major foundational role for social cognition and implies a strong motivation to imitate, Oostenbroek et al. (Reference Oostenbroek, Suddendorf, Nielsen, Redshaw, Kennedy-Costantini, Davis, Clark and Slaughter2016) averaged data across a large number of infants in their domestic environments, presenting 11 models one after the other. This design did not control variables potentially affecting infant behavior, introducing interference in the measurement of imitative responses (“delayed” imitative responses count as non-imitative because they occur when a different model is presented). A more realistic line of research takes each infant as its own control (e.g., Meltzoff & Moore Reference Meltzoff and Moore1992) and minimizes interference.

K&A's examination of the theoretical alternatives is wanting in at least three aspects. First, explanations of differential imitation are not as narrow or impoverished as K&A suppose; some already rely on conceptions of the body schema formed through spontaneous prenatal behavior (Gallagher & Meltzoff Reference Gallagher and Meltzoff1996; Meltzoff Reference Meltzoff, Cicchetti and Gunnar2009; Meltzoff & Moore Reference Meltzoff and Moore1997), including involvement of gustatory processes (Gallagher et al. Reference Gallagher, Butterworth, Lew and Cole1998).

Second, a plausible account for differential imitation remains possible. An explanation of differential imitation does not have to postulate computational processes for the recognition of self-other similarities – which we agree is a major fault in Meltzoff and Moore's (Reference Meltzoff and Moore1997) account – nor does it have to postulate a module specifically evolved for imitation or related functions. In contrast, the commonalities between visual experiences and corresponding motor experiences may operate tacitly as the means by which spontaneous behaviors can be differentially induced (Vincini et al. Reference Vincini, Jhang, Buder and Gallagher2017; Vincini & Jhang, revised and resubmitted). From this perspective, infants do not actively intend to match the behavior of others but, rather. tend to respond in a way that is more passively elicited. An implication of this perspective is that the more frequent an action is in spontaneous behavior, the easier it will be to induce it. Moreover, if early imitation is differential induction of spontaneous behavior, then it is inappropriate to use 11 control models (which entails provoking uneven levels of arousal). Rather, a small number of control models is sufficient to establish that specific models can induce their corresponding actions more than other models.

A third way in which K&A's examination is wanting is that K&A reject a resonance or “entrainment” approach to differential imitation for debatable reasons. They delineate a model of “imitation without representation” (sect. 7.2 heading) that requires visual encodings of models to be matched with corresponding central pattern generators, and then they note that this systematic matching is a “tall order” (sect. 7.2, para. 7.2.1). In synchronization theory, entrainment requires participation of endogenously activated (“autonomous”) oscillations, whereas resonance can occur in a system that would not spontaneously oscillate by itself (Pikovsky et al. Reference Pikovsky, Rosenblum and Kurths2001). We therefore consider rhythmic stereotypies as a suitable basis for inducing differential imitation precisely because behaviors that are already active but not cognitively modulated are most susceptible to entrainment, and because distinct dynamic trajectories are the medium of oscillator coupling, not requiring mediation by recognition processes.

Finally, even if the arousal theory were correct, the link between infant arousal responses and social cognition should not be dismissed altogether. A caregiver may respond to the infant's spontaneous tongue protrusion with a facial expression, which in turn provokes arousal in the infant. In this way, the infant is drawn into an interaction – most readily, moreover, when the behaviors presented match and entrain spontaneous rhythmic stereotypies. Indeed, a caregiver may over-interpret the infant's response as a social response, which promotes further interaction (Gallagher Reference Gallagher2008). Therefore, a spontaneous behavior serving neurodevelopment for aerodigestive function may assume a new social-cognitive function – analogous to processes of recycling or reuse (Anderson Reference Anderson2016; Dehaene Reference Dehaene2009).

In conclusion, we bring the question of differential imitation back to the foreground, and we discuss how it should be addressed empirically with the aim of developing a more plausible account. One should also consider previous theorizing on spontaneous behavior and explore the possible explanatory power of entrainment and subtler connections with social cognition. All this suggests that K&A's discussion, although praiseworthy for its emphasis on rhythmic stereotypies, does not do justice to the complexity of the empirical and theoretical issues surrounding neonatal imitation.