R2.1. What are the evolutionary origins of human speech?

It seems clear that, in the target article, the authors have stumbled into a robust discourse about the evolutionary origins of speech. Mayer et al. provide comprehensive analyses with evidence that “there are at least some core speech movements which are direct ontogenetic adaptations of preexisting digestive movements.” A different, though structurally similar proposal for the ontogenetic adaptation of speech from pre-existing movements can be found in the work of Ghazanfar et al. (Reference Ghazanfar, Takahashi, Mathur and Fitch2012), derived from MacNeilage's view (Borjon et al. Reference Borjon, Takahashi, Cervantes and Ghazanfar2016; Chandrasekaran et al. Reference Chandrasekaran, Trubanova, Stillittano, Caplier and Ghazanfar2009; Ghazanfar et al. Reference Ghazanfar, Takahashi, Mathur and Fitch2012; Shepherd et al. Reference Shepherd, Lanzilotto and Ghazanfar2012), that speech has evolved from rhythmic facial expressions to which vocalizations have been added. Murray et al. emphasize that these mouth movements were long ago co-opted into early infant-parent dynamic interaction.

Although we are not experts on the evolution of speech, a few things seem clear. Our aerodigestive theory is not meant to be an evolutionary theory of speech but, rather, is meant to explain why neonatal imitation is unlikely to occur given the facts of aerodigestive development. However, our theory is compatible with both sides of this debate, with the evolution of speech from either fetal aerodigestive behaviours or orofacial behaviours such as yawning, blinking, scowling, smiling, and so on. (Dai & Hata Reference Dai and Hata2006; Kanenishi et al. Reference Kanenishi, Hanaoka, Noguchi, Marumo and Hata2013; Kurjak et al. Reference Kurjak, Azumendi, Andonotopo and Salihagic-Kadic2007; Sato et al. Reference Sato, Kanenishi, Hanaoka, Noguchi, Marumo and Hata2014; Yigiter & Kavak Reference Yigiter and Kavak2006).

In our target article, we concentrated on aerodigestive behaviours in the belief that TP/R fits neatly into the class of aerodigestive behaviours. Apart from the neonatal imitation literature, and unlike lip-smacking among New World Primates, there is little to suggest that TP/R is a universal, affiliative human behaviour (cf. Murray et al.). That said, facial expressions of the fetus and neonate have a developmental trajectory that is parallel to aerodigestive development, that is, from individual stereotypies to sensorimotor sequences. In a set of experiments, Reissland et al. (Reissland et al. Reference Reissland, Francis, Mason and Lincoln2011; Reissland et al. Reference Reissland, Francis and Mason2012; Reference Reissland, Francis and Mason2013) showed that from 24 to 36 weeks gestational age (GA) these individual action units begin to coalesce into “Gestalts” of emotional expressions, such as a “happy face” or a “cry face.” Prior to 24 weeks, single-action units predominate facial expressions, whereas by 36 weeks, 85% of action units co-occur with two to four other units. Just as aerodigestive stereotypies coalesce into motor runs by 36 weeks GA, coherent facial expressions emerge over the same time period. Moreover, just as suckling and respiration are “practiced” in utero without, for example, air to breathe, the “pain face” of a fetus at 36 weeks GA occurs spontaneously and independently of any (visible) harmful event. So, aerodigestive behaviours and facial expressions share the same developmental trajectory. Thus, our theory is consistent with either or both evolutionary theories.

R2.2. Is the aerodigestive development consistent with early, pre-linguistic behaviour of the infant?

We know that prior to speech acquisition, infants younger than 4 months of age begin to vocalize. Therefore, according to our view, infants must be able to learn such behaviours prior to gaining cortical control of articulatory structures. Meltzoff argues that infants younger than 4 months “produce diverse cooing sounds, which require tongue movements markedly different from suckling and tongue stereotypies.” Consequently, if “infants cannot control their tongues prior to 4 months of age” and “all tongue movements are purported to be the stereotypic thrust/retraction involved in suckling,” then the aerodigestive theory must be false.

Our claim is that orofacial stereotypies develop via “practice” prior to birth and are then incorporated into complex, sensorimotor sequences through networks of interacting oscillators. Both learning and tongue control occur before and after birth, the result of the developing oscillatory networks in the brainstem. Importantly, tongue thrust is a primitive reflex, not a stereotypy, a difference that makes a difference. Unlike primitive reflexes, infant stereotypies are highly variable – for example, small or large (just over the lip line or far beyond it) and in all directions (to the right or the left, or straight down the midline). In our view, stereotypies that continue to occur alone, independently of sensorimotor runs, organize cortical motor space through the somatosensory and proprioceptive feedback from the full range of possible movement. Accordingly, TP/R explores the deformation space of tongue movement, across the full range of protrusive tongue movements. This variation provides a bridge between aerodigestive and articulatory behaviours. For example, Mayer et al. suggest that tongue-bracing contacts during swallowing are a subset of the tongue-bracing contacts in speech. Stated differently, the motor activation space of tongue-brace during swallowing falls primarily within the activation space of tongue-brace for speech. If tongue-brace is a stereotypy – and it is – this transition is not mysterious. The stereotypy has already explored the sensorimotor space of tongue-brace prior to speech learning. Note that we are not claiming that Mayer et al.'s view is necessarily true. Our claim is that sensorimotor learning occurs subcortically in early human development; therefore, some articulatory movements need not require cortical input to be learned. Which tongue movements require cortical input depends upon the type of tongue movement. Ballistic movements such as catching a drip of ice cream as it escapes from the cone probably require cortical input to learn or initiate such directed tongue movements. This is our claim in the target article.

R2.3. How might human speech develop out of neonatal stereotypies?

Neither of us is a specialist in language development, but the commentators' questions (cf. Choi et al., Murray et al.) sparked our interest in the recent literature on speech development and oscillatory entrainment. If perinatal behaviour is a function of subcortical networks of central pattern generators (CPGs), and if mature speech also involves repetitive, rhythmic movements, then perhaps interaction between parents and infants form coupled systems of oscillators; that is, parental speech entrains the pre-linguistic behaviour of infants. Perhaps the best example of this research project is found in the work of Ghazanfar et al., who characterized human speech as an inherently multimodal capacity that has evolved from facial expressions/stereotypies (i.e., the MacNeilage view; see Borjon et al. Reference Borjon, Takahashi, Cervantes and Ghazanfar2016; Chandrasekaran et al. Reference Chandrasekaran, Trubanova, Stillittano, Caplier and Ghazanfar2009; Ghazanfar et al. Reference Ghazanfar, Takahashi, Mathur and Fitch2012; Shepherd et al. Reference Shepherd, Lanzilotto and Ghazanfar2012). Their experiments suggest that at least three types of pre-linguistic learning can be explained in terms of coupled oscillation modulated by speaker and listener arousal, namely cooperative vocal adjustment at a distance, the maturation of turn-taking, and the development of the “phee” call in infant marmosets. Several of these experimental results are directly applicable to the human case.

First, human infants might learn conversational turn-taking in much the same way as infant marmosets. Ghazanfar et al. (Reference Ghazanfar, Chandrasekaran and Logothetis2008) reported that infant marmosets, during the first postnatal days, respond to recordings of their own voices and parental cries with the same frequency of response and response time. This suggests the inability of the infant to distinguish between its own voice and the voices of others. By 2 months of age, infant marmosets attained the adult temporal pattern of vocal turn-taking, with ever decreasing response time to the parental voice. The authors explain this reduction in terms of the vocal entrainment of the infant's call by the parent. In the human case, the dorsal cochlear nucleus receives somatosensory feedback from the face and parts of the vocal tract, thus a rudimentary mechanism to identify one's own vocalizations. Auditory stimulation with vocal tract and facial proprioceptive feedback is “my voice,” auditory stimulation without feedback equals “someone else's voice.” Once a human infant differentiates her own voice from another speaker's, auditory entrainment can begin.

Second, self-entrainment might explain the importance of infant cooing. Cooing, the production of vowel-like sounds, involves large facial movements – the pursing of lips (“oo”), a wide-open jaw (“ah”), lip retraction (“ee”), and so on. Such expressions are not unknown to the infant. Many of these expressions can be mapped directly onto the stereotypies that comprise the newborn's behavioural repertoire or fall within the variations seen in the first stages of stereotypy acquisition. Fagan (Reference Fagan2014) reported that infants deaf from birth make less frequent vocalizations of all kinds. After early cochlear implantation, they resumed a normal pattern of pre-linguistic vocalizing. However, deaf newborns did not resume normal rates of crying and other vocal signals of distress. Fagan concludes that infant cooing/babbling “are primarily motivated by auditory feedback.” To put this slightly differently, the infant's own vocalizations might create a self-sustaining training-cycle – of vocal production followed by both auditory and somatosensory encodings. As long as arousal by self-vocalization is speech sound specific, that is, causes the vocalization of multiple “vowels,” learning paired associations between somatosensory (S1) and auditory (A1) encodings will occur. This suggestion may explain why 4-month-old infants cannot judge auditory-visual mismatches when prevented from making articulatory movements. As long as the categorization of speech sounds is primarily somatosensory – or if somatosensory categorization is required to entrain speech production – an infant will not be able to classify phonemes or differentiate between them without self-production. Nor would an infant be able to learn a new phoneme without first trying to reproduce it and thus see what it feels like.

Third, entrainment may explain the development of audiovisual matching between the voice and lip movement of a speaker with the lips, the capacity that underlies the McGurk effect (Chandrasekaran et al. Reference Chandrasekaran, Lemus, Trubanova, Gondan and Ghazanfar2011). The 2-month-old infants look longer at matches between facial movements and speech sounds (Patterson & Werker Reference Patterson and Werker2003); at 4 months, infants reverse this correlation, paying more attention to mismatches (Kuhl & Meltzoff Reference Kuhl and Meltzoff1984). Intuitively, this looks like an initial period of multimodal learning followed by the ability to perceive exceptions based upon learned audiovisual associations. In speech, the motion of the lips and the resultant auditory signal are both rhythmic events with the same periods of oscillation, synchronized in space and time. Thus, temporally synchronized auditory and visual signals, matched in periodicity, indicate a common source/speaker of these signals. But from where does this visual information about dynamic faces come? We know that A1 and its adjacent areas contain a high percentage of multimodal (audiovisual) cells in adult primates. But at birth, the visual system has yet to begin processing dynamic patterns of visual stimuli, a necessary requirement for training up A1 cells that associate facial movements and their respective sounds. In the superior colliculus, the alignment of audiovisual maps is also experience dependent and will not occur until about 4 months of age.

One plausible route for this association is by way of the pulvinar, a division of the thalamus. In adult primates, human and otherwise, dynamic facial expressions are matched to speech through input from the middle temporal (MT) visual area in the posterior bank of the superior temporal sulcus (STS) (Ghazanfar et al. Reference Ghazanfar, Chandrasekaran and Logothetis2008). If cortical vision developed hierarchically, one would expect area MT to lag behind visual area 1 (V1) in maturity. But directional sensitivity arises in V1 and MT at the same time, around postnatal 4–7 weeks. The key here may lie in the pulvinar (Kaas Reference Kaas2015). Newborn primates have a transient pathway from retina to pulvinar and then from pulvinar to visual associative area STS including area MT (Warner et al. Reference Warner, Kwan and Bourne2012). So visual motion information via the pulvinar could be critical to the formation of multimodal, audiovisual cells in auditory cortex. The timing here is highly suggestive. If direction selectivity begins by 4–7 weeks, infants in the Patterson and Werker study (2003) would have had, on average, a few weeks of associative training and thus the basis for preferential looking.

R3.1. The cross-target methodology: Should we trust it?

Several commentators argue that neonatal imitation can still be a viable theory when a wider array of data is taken into account (Meltzoff; Simpson, Maylott, Heimann, Subiaul, Paukner, Suomi, & Ferrari [Simpson et al.]; Vincini, Jhang, Buder, & Gallagher [Vincini et al.]). This evidence comes largely from experiments designed to test neonatal imitation using a methodology that has changed very little after the first-wave improvements. When an experimental methodology continues to produce a tangle of positive and negative results over a 35-year period, this is deeply worrisome. It suggests that the methodology is somehow flawed and that its continued use will not produce definitive results in the future. Witness here the competing results of two recent, careful studies by Oostenbroek et al. (Reference Oostenbroek, Suddendorf, Nielsen, Redshaw, Kennedy-Costantini, Davis, Clark and Slaughter2016) and Simpson et al. (Reference Simpson, Murray, Paukner and Ferrari2014a).

In the target article, we attempted to set this issue aside in order to develop a positive theory. But ours is not a theory of how or why neonatal imitation occurs: It is meant to explain away the positive results of neonatal imitation experiments, not to vindicate them. As the commentaries demonstrate, we must address some issues of procedure and statistical analysis in the neonatal imitation work. Given the constraints of space we will mention one example that has relation to the aerodigestive mechanisms.

A standard part of the cross-target methodology is the “burst-pause procedure” that in Meltzoff and Moore (Reference Meltzoff and Moore1989, p. 955) was justified as follows: “In previous work with newborns it was reported that attention and responsivity were maximized if adult gesturing was alternated with an interval in which the adult remained passive …” (Meltzoff & Moore Reference Meltzoff and Moore1983). If we look back at Meltzoff and Moore (Reference Meltzoff and Moore1983, p. 707), the authors reported that: “We found in preliminary work that a constant demonstration of the target gesture was not maximally effective in eliciting imitation. Therefore, in our design the experimenter alternated between the presentation of the gesture and a passive face. We are not certain why our burst-pause procedure is the more powerful, but we can suggest three possibilities.” The three possibilities then given were that (a) it gave infants several periods of time over which to organize their motor response; (b) the burst-pause paradigm nicely “bookended” the demonstrated gesture, thereby accentuating what was to be imitated; or (c) the paradigm mimicked the give and take of conversational turn-taking, thus encouraging infant response. All of these options assume that imitation occurs and that the question about this methodology is why it increases imitation.

The right question to ask is why the new methodology changes infant behaviour and whether those factors are related to imitation or to extraneous factors. For example, it could be that in addition to increasing attention, the methodology also increases arousal. Or perhaps the methodology increases negative affect and thus responsiveness. In the still-face-effect experiment, an adult who stares without facial expression or movement causes anxiety in infants as measured by fussing or crying, increased heart rate, skin conductance, and vagal tone (Bertin & Striano Reference Bertin and Striano2006; Ham & Tronick Reference Ham and Tronick2006; Moore & Calkins Reference Moore and Calkins2004; Striano Reference Striano2004). This effect begins in infants between 1 and 1.5 months of age (Bertin & Striano Reference Bertin and Striano2006; Bigelow & Power Reference Bigelow and Power2012), just around the age of testing (Meltzoff & Moore Reference Meltzoff and Moore1992; Reference Meltzoff and Moore1994). Yet imitation experiments use what amounts to a still face stimulus as a control condition. We now also know that neonates prefer their mother's face to that of a stranger (Bartrip et al. Reference Bartrip, Morton and Schonen2001) and to prefer a new face over a previously viewed face or a noncommunicative face (Cecchini et al. Reference Cecchini, Baroni, Di Vito, Piccolo and Lai2011). So multiple aspects of the burst-pause paradigm are linked to negative affect.

Finally, as we saw from the research on marmoset turn-taking, the third option may well have hit the nail on the head. Six-week-old infants may already have learned turn-taking behaviour through interaction with their parents, what amounts to a “call-and-response” sequence in which behaviour is inhibited during the “call” and the inhibition is released when the static gesture ends, thus producing more gestures thereafter. But this need not involve imitation per se, merely a move towards conversational turn-taking or a reflexive “wait and listen/see” disposition that reduces noise while in the listening phase. Because the authors assumed that the burst-pause paradigm increased imitation, they did not investigate any hypotheses about why “attention and responsivity were maximized” by the new paradigm. In our opinion, it is time to go back and look carefully at the assumptions incorporated into the cross-target paradigm.

R3.2. If your explanation is in terms of arousal, what explains the differential response?

As a number of our commentators have pointed out, the primary empirical question is not whether arousal increases the production of infant stereotypies, that is, the base rate of production, but whether there is evidence for a differential response – more mouth opening (MO) in response to modeled MO, not TP; more TP in response to TP, not MO (Meltzoff, Simpson et al., Vincini et al.). As Simpson et al. assert, “differential imitation in neonates is incompatible with aerodigestive or arousal-driven explanations.” Moreover, the search for differential responses should not focus on TP alone, given that one needs to show a pattern of differential responses across neonatal behaviours.

Let's take the easiest question first, why we have focused primarily on TP. We do think the case of TP/R is special. It is the only gesture that has consistently garnered more positive than negative results in imitation experiments for which there is no ready, alternative explanation. In fact, TP/R is the most commonly modelled gesture across studies; approximately 85% of studies investigating imitation in neonates up to 6 weeks of age have included this gesture (Oostenbroek et al. Reference Oostenbroek, Slaughter, Nielsen and Suddendorf2013). Many reports of a differential response to other gestures have failed on replication, even in Meltzoff and Moore's own experiments (i.e., MO in the 1994 experiments). Moreover, TP/R is the standard comparison condition for the imitation of mouth opening, another gesture often cited as known to elicit imitation. So, if TP/R is not imitated, then a large and convincing body of evidence for the existence of neonatal imitation vanishes. All things being equal, such a result would probably spell the end of the neonatal imitation (NI) research project. A vindication of the TP/R results, in contrast, would require the further investigation of differential responses to other gestures.

That said, our central worry about the criterion of differential imitation concerns the comparison class and the statistical analysis of the cross-target experiments. As Kennedy-Costantini, Oostenbroek, Suddendorf, Nielsen, Redshaw, Davis, Clark, & Slaughter (Kennedy-Costantini et al.) point out in their commentary, a recent longitudinal study of neonatal imitation at ages 1, 3, 6, and 9 weeks (Oostenbroek et al. Reference Oostenbroek, Suddendorf, Nielsen, Redshaw, Kennedy-Costantini, Davis, Clark and Slaughter2016) reported that “(infants) were just as likely to produce the gestures in response to control models as they were to matching models” (p. 1334). By looking at a limited, cross-section of the data, the Oostenbroek researchers were able to reproduce the same positive effects as reported in earlier studies. They concluded that failure to include adequate control conditions or test infants across multiple time points in previous studies has resulted in the false impression that infants selectively copy tongue protrusions, thereby perpetuating the idea that newborn imitation exists. For our part, given that we see the “gestures” of infants as neonatal stereotypies, this makes a good deal of sense. Because we can actually categorize and then count up the kinds of neonatal stereotypies, just as Thelen did (Thelen Reference Thelen1981b), we have the right comparison class in hand. So, at least for neonates, the question cannot be “tongue protrusion or mouth-opening.” This doesn't represent the statistical landscape. Of course, no one knows whether these results will themselves stand the test of time. But one possibility – the one on which we are betting – is that there will be no positive results at all. Hence, no explanation will be needed.

If we are wrong, and the positive results prove robust, then we must give an alternative explanation. At present, we do not have a single reason, but we do have several we think are worth pursuing. For example, there is the intriguing result in Meltzoff and Moore (Reference Meltzoff and Moore1992) that MO duration is longer than TP duration. When the authors tested MO using a dynamic stimulus, MO duration increased; it was twice as long on average than when the static stimulus was used. However, a dynamic presentation of TP produced only slightly more frequent TPs than in response to the static display (which were not timed only for duration). This difference in response suggests that the causal mechanisms of MO and TP differ. Therein could lie a tale of how arousal could differentially affect the rate and duration of TP and MO: Perhaps arousal/apprehension inhibits MO and increases TP. At this point in time, we have no clear explanation of why this should occur. But note that arousal plays a crucial role in Ghazanfar's examples of speech development as well, both in how infant marmosets learn to adjust the intensity of their voices as a function of listener distance (Choi et al. Reference Choi, Takahashi and Ghazanfar2015) and in how parental calls, timed to infant vocalization, produce mature “phee” calls in the infant over time (Ghazanfar & Zhang Reference Ghazanfar and Zhang2016). These events occur beyond the neonatal period, of course, so they are only illustrative of how the context of arousal differentiates its causal effects.

We have also wondered whether Anisfeld (Reference Anisfeld1996) and others might be right in their explanation that the higher rate of MO in the mouth opening condition, compared to its rate in the tongue protrusion condition, may be explained as a by-product of infants' TP/R responses. TP/R and MO/C co-vary with each other: Increase in one results in a decrease in the other. Given that overall level of oral activity stays roughly constant in the two conditions, the rise in the rate of TP/R in the tongue protrusion condition seems to reduce the rate of MO/C in this condition. The higher rate of MO in the mouth opening condition compared to its rate in the tongue protrusion condition may then be on account of the lowering of the MO/C rate in the tongue protrusion condition, not to its raising in the mouth opening condition. For instance, in Meltzoff and Moore (Reference Meltzoff and Moore1983) the rates of MO/C and TP/R are about the same in the mouth opening condition, whereas the rate of TP/R is substantially higher than the rate of MO/C in the tongue protrusion condition. So what may rise and fall under different conditions is actually the rate of TP/R. More generally, we worry that even neonatal arousal is not a homogeneous state. Here the presupposition is that increasing arousal is like turning up the speed dial on an oscillating fan: It makes everything go faster. But there is no reason why this should be so, that the effects of arousal should be undifferentiated. Why, exactly, given the complexity of the neonatal brain, must this be so?

R3.3. Is our argument for the aerodigestive theory a form of Occam's razor?

Several commentators argue that we should apply Occam's razor to decide between the imitation and aerodigestive theories (Beisert, Zmyj, & Daum [Beisert et al.]; Jones). Originally, Occam's razor was the dictum that we ought not to multiply explanatory entities beyond those needed. As the old medical dictum goes: If you see hoofprints, look for horses not unicorns. Certainly part of our argument has followed Occam's razor understood in this way. We know that subcortical oscillators control the central aerodigestive behaviours of early infancy; neurodevelopmental research also suggests that the maps of primary visual cortex, S1 and M1, are still highly immature at birth. So, taken together, if infants respond differentially to modelled tongue protrusion with more frequent tongue protrusions (than to other gestures), we ought not to look for the complex representational structures required by the hypothesis of opaque imitation: that is, visual representations of facial gestures, somatosensory patterns of self-produced facial gestures, cortical motor commands that produce TP, plus the various mechanisms of mapping and association that yield genuine imitation. There is no independent evidence for the existence of such structures or abilities. Consequently, we should train our attention on the kinds of processes that we know to exist – or think are likely to exist – at birth in human infants. This would be an argument of the classic form.

Still, when dealing with the massive complexity of the human brain – and here we include the neonatal brain as well – we rarely have conclusive facts in hand of the sort that would support a simple version of Occam's razor. Instead, what we suggest is more akin to inference to the best explanation than Occam's razor. The question is this: What if we were to stand back from the specific and highly contested results of neonatal imitation experiments and look instead at the distance traveled over 35 years of neurodevelopmental research, in psychophysics, neurophysiology, and neuroanatomy. What picture emerges? We know that neural mitosis and migration begins via chemical cues and/or via the use of nearby cells for pathfinding; we also know that prior to environmental input, Ca2+ activity influences every aspect of neural development from cell mitosis, migration, arborization, transmitter expression, axonal growth – and that once transducers are in place and functioning, development will continue based upon patterned environmental stimulation. We also know that this process of neural scaffolding can involve “two steps forwards, one step back” – the expression and re-expression of transmitters, the growth of temporary structures such as the subcortical plate, and of transient neural pathways, connection areas that will disappear when no longer needed. Although this developmental process proceeds simultaneously in multiple systems, by birth the human cortex is still immature. The alternating columns in V1 that segregated visual input by eye have yet to form, an essential organizational structure for stereoscopic vision. Although the subcortical motor system is well developed at birth, the cerebrospinal system has yet to establish functional connections between M1 and the spinal gray matter, a process that will require protracted development after birth.

It is this cumulative picture that seems at odds with the existence of the infrastructure, at birth, required for cross-modal or amodal information transfer. Whether one supposes that such structures are innate or learned, the results of genetic transcription or neural activity, there is no reason to think that the neonatal brains “comes with” such resources. It is also at odds with the kind of intentional explanations of neonatal behaviour often given by researchers. There is no recognition that your gesture is like mine, or probing a model to reveal model identity (Meltzoff & Moore Reference Meltzoff and Moore2002) or understanding that you are a thing “like me” (Meltzoff Reference Meltzoff2007). So, either we “deflate” what is normally meant by, for example, a motor command (for there are no motor commands in the cerebrospinal system of the neonate), or we look towards the kinds of mechanisms that we know to exist. We then ask how the neonatal brain bridges the seemingly vast gap between non-intentional and intentional cognitive processes. Zappettini asks whether we, the authors, are fond of deflationary or “killjoy” accounts, of accounts that take prima facie cognitive tasks and explain them in non-cognitive terms. We deny that we have a predilection for killjoy accounts. We are not killjoys by nature. Rather, we agree with Booth, Beisert et al., and Provine that we should avoid cognitively rich interpretations when there are more plausible cognitively lean interpretations available.

R3.4. Is neonatal imitation connected to later social skills?

Simpson et al. argue that neonatal imitation is connected to later social skills. If there is a connection between neonatal imitation and later social skills, first and foremost there should be a connection with the later imitation skills. However, many studies failed to find a connection between neonatal imitation and later imitation (Abravanel & Sigafoos Reference Abravanel and Sigafoos1984; Fontaine Reference Fontaine1984; Heimann et al. Reference Heimann, Nelson and Schaller1989; Jacobson Reference Jacobson1979; Kugiumutzakis Reference Kugiumutzakis, Nadel and Butterworth1999). On the contrary, neonatal imitation drops out after 3 months only to re-appear after 6 months. Among researchers who accept neonatal imitation as a fact, it is controversial whether this “drop-out” is significant. Does the imitation of tongue-protrusion simply end abruptly at 2–3 months? Or does the infant merely move on to other forms of interaction with adults and resume a different repertoire of imitative behaviors a few months later? This is Meltzoff and Moore's (Reference Meltzoff and Moore1992) explanation of the phenomenon of “drop-out.”

As we have shown in the target article, that mammalian aerodigestion develops in two phases: (1) from the onset of isolated orofacial movements in utero to the postnatal mastery of suckling at 3 months after birth; and (2) thereafter, from preparation to the mastery of mastication and deglutition of solid foods. This division in the maturation of the mammalian aerodigestive system has important consequences for the question of neonate imitation drop-out. Suppose, now, that tongue protrusion qua spontaneous neonatal behaviour itself ends between 2 and 3 months after birth. And suppose it does so because the developmental phase of which it is but one part comes to an end as a whole. This fact would suggest that the “imitation” of tongue protrusion does not end because the infant loses interest in copying orofacial gestures, but because spontaneous tongue protrusion itself declines as this first phase of aerodigestive maturation draws to a close. This fact – that an aerodigestive developmental stage, involving a period of spontaneous tongue protrusion, coincides with the period during which neonates “imitate” tongue protrusion – is highly significant. This coincident phase makes it more plausible that the increase in neonate tongue protrusion in the experimental setting is the result of some extraneous cause – for example, general arousal in the face of an interesting stimulus. Proof of an independent but coincident developmental phase, then, again raises the spectre that we have mistaken a spontaneous behaviour for an imitative one.

R3.5. Given that your theory posits a dynamical system of aerodigestive CPGs, why did you dismiss imitation via entrainment in the target article? What about mirror neurons?

One of the central features of the cross-target methodology is the alternation of each static display of gesture modeling with a period of static neutral face during which the response of the infant is recorded. During both periods, the model works to minimize any differences in presentation, paying particular attention to inadvertent social signals that might cue the production of stereotypies and/or create experimental artefacts. The experiment seems almost perfectly designed to preclude the dynamic entrainment of the infant's behaviour by the model's behaviour. There is no entrainment without oscillatory activity and oscillatory activity is conspicuously absent from the standard cross-target methodology. This is why, in the target article, we dismissed entrainment as an explanation of the reported NI results.

In retrospect, a more nuanced answer is possible. There is one stage in the static cross-target paradigm that is dynamic, namely at the temporal boundary between the presentation of the static gesture and the model's transition to the static face: for example, when retraction of the tongue marks the completion of static TP modeling or after one full period of TP/R oscillation. Whether this has any effect on the neonate's behaviour is an open question. But entrainment theory could explain why the burst-pause methodology seems to promote imitation as well as why the imitative effects of gesture modelling are so weak (because there is only one very slow period of oscillation). That said, confirming a dynamical theory of behavioural entrainment would require a quite different experimental setup, one that uses dynamic stimuli and records neonatal behaviour concomitantly.

As to mirror neurons, we agree entirely with Fitch's commentary. Were it not for the discovery of neonatal imitation in macaques, it would have been hard to explain why adult macaques, a species notoriously lacking in robust imitative behaviour, had mirror neurons at all. So neonatal imitation in macaques has served to bolster claims that mirror neurons underlie imitative behaviours in primates more generally. We do not wish to deny the existence of mirror neurons. But we are sceptical of any claim that mirror neurons explain the NI experimental results or that mirror neurons must be present in neonatal macaques/humans because of the NI experimental results. Independent evidence of either claim is needed before proceeding down that explanatory path. Still, contra Leisman, it is hard to imagine that mirror neurons exist in the human neonate. Mirror neurons require a functional, mirror neuron network. If active intermodal matching (AIM) is unlikely to be true given the immaturity of the neonatal human cortex, then the same arguments apply to the cortical network/s required to drive mirror neurons – and thus to the existence of mirror neurons in the perinatal period.

R4.1. Social engagement

A number of our commentators have stressed the importance of social interaction for infant development (Aitken; Buck; Desseilles; Libertus, Libertus, Einspieler, & Marschik [Libertus et al.]; Murray et al.; Simpson et al.). We are entirely in agreement with this view and, more generally, with the interactive nature of human infants. At birth, infants come into the world entirely dependent upon the caretaking of adults. Infant survival requires the constant attention of their caregivers. This makes care as important to human infants as normal physiological development, for example, of a functioning aerodigestive system at birth. But looking beyond bare survival, the extended period of postnatal motor development has the consequence that what is learned during this period rests heavily on parental interaction.

Parents (or caregivers) facilitate the lion's share of the infant's interaction with the world prior to the attainment of goal-oriented action. For the neonate, being carried, coddled, cuddled, changed, fed, bathed, bounced, and generally responded to “in words or deed” are the rich events that foster immediate infant learning. Moreover, this learning involves multiple dimensions, the social and emotional no less than learning within the standard sensory and motor domains. Infants learn to be soothed by touch and voice, “read” the prosody of human speech and the emotional “temperature” of their social environments, to make eye contact and visually explore a human face, distinguish between the friend and “foe” (the still-face effect), and to relish in human interaction. Last, but not least, they learn how to engage their caregivers – or, as my (Akins) mother-in-law used to say, “how to run a household from the cradle.” The inert, non-interactive infant is an infant at risk. But such infants will also learn far less about the world in general and about human social relations (cf. Casartelli & Parma).

Infant-maternalFootnote ¹ bonding is thus essential for normal development. In the target article, we mentioned the multitude of ways in which this occurs (see sect. 7.3). These processes arise in the neonate alone (e.g., the olfactory recognition of the mother's colostrum at birth), the mother alone (e.g., oxytocin release, increased sensitivity to the infant's cry), and through interaction between the two (e.g., kangaroo care, the coordination of mother and infant heartbeat). Social interaction begins immediately at birth – turning towards the mother's voice, visually exploring her face, making facial expressions such as smiles or grimaces – and gradually grows more sophisticated. We suggest that this interaction “works” – creates and maintains a bond – at least in part because of the automatic human propensity for intentional interpretation. We see in that now-famous, first Pixar video, two lamps (one a large anglepoise lamp and the other a small gooseneck) interacting as a mother and child. They play, are watchful, talk, remonstrate, and even sulk, cycling through the gamut of parent-child interactions. Indeed, we cannot help but see them as persons despite knowing that lamps cannot have intentional states (or have children for that matter).

Our interactions with the newborn are no less intentionally infused; we see, and cannot help but see, the crying, grimacing, and smiling of a human infant as actions. This is not an argument that we think infants are vegetative automatons, as Aitken would have us say. Rather, in practical terms, our own propensity to see intentional states is so deeply engrained that we see intentional behaviour even when our perceptions conflict with what we know. For example, we see the neonate smiling even though we know that the transition to the social smile takes ~8 weeks after birth to develop. Add infant-adult turn-taking to the mix once the infant can distinguish self- from other- produced voices – that is, remaining silent/inactive when the mother acts/speaks and then continuing activity when the mother stops – and any infant becomes entrancing to his or her caregivers (or at least to those disposed to be entranced at the outset). The net result of this and other such processes is maternal bonding, a state essential to the infant in all respects. In other words, the interactive baby has little need for a complex capacity for imitation. Nor is this possible, as Campos and Neito say, “their perceptive and attentional capacities (Volpe Reference Volpe2008), and face processing and intersensory processing abilities (Bahrick et al. Reference Bahrick, Lickliter and Flom2004; Johnson et al. Reference Johnson, Senju and Tomalski2015; Lewkowicz Reference Lewkowicz2014; Morton & Johnson Reference Morton and Johnson1991), among others, are too weak yet.” The infant who reacts, in whatever way, to our attention reaps the benefits that advocates of neonatal imitation so often invoke.

R4.2. Infants as automatons

A related misconception of our view is that we are throwbacks to the bad old days of psychological behaviorism. Or worse, that we see infants as automatons that exhibit reflexive behaviours without variation or learning. To quote Aitken (who quotes Polani & MacKeith [1960]), we are tainted with the view that “the newborn infant may be described as a tonic animal with oropharyngeal automatisms and neurovegetative mechanisms.” This is not our view although we would be hard pressed to say exactly what it is like to be a neonate. One misconception concerns our talk of stereotypies, a term we have adopted from the literature and which distinguishes between neonatal reflexes and stereotypies. Few sensorimotor neuroscientists talk in terms of reflexes anymore, at least not in the classical Sherrington sense, except for the patellar reflex arc in children and adults. Indeed, neonatal “reflexes” are notoriously difficult to evoke, requiring an experienced clinician.

In any event, neonatal stereotypies are not invariant or “released” by specific stimuli. On the contrary, our suggestion is that so-called neonatal stereotypes are useful precisely because they are not invariant. Precipitated by general arousal, their constant variation serves to explore the full -range of sensorimotor space, and, as O'Sullivan and Caldwell point out, the stereotypic actions are likely to play a significant role in the development of associations between sensory and motor representations of the same behaviour. Infant stereotypies occur precisely when the somatosensory and motor cortices are developing functional connections within and between the M1 and S1 regions, as well as developing the functionality of the corticospinal tract. Of course, it is an essential part of our view that proprioceptive and motor learning begins in utero. But in just the way that “suckling” in utero is a faint facsimile of actual suckling after birth, sensorimotor activity in utero fails to replicate the physics (and freedom) of terrestrial locomotion after birth. It's a whole new ballgame, as they say, once on land. So, for this reason, it is not a bad idea to “wire” the sensorimotor system “in place,” at least for any species with a large and complex sensorimotor repertoire. At least, a priori, that would seem to be the case.