Twelve-Month ‘‘Social Revolution’’ Emerges from Mother-Infant Sensorimotor Coordination: A Longitudinal Investigation

Previous accounts of the development of triadic attention identify a ‘‘curious’’ shift around nine to twelve months. We introduce a novel approach inspired by distributed and embodied cognition frameworks. In a longitudinal study of five mother-infant dyads, videos of home play interactions were recorded over the infants’ first year. We scrutinized the real-time organization of mother-infant sensorimotor activity, including the targets of hands, gaze, and mouth, as the dyad members attended to one another and to toys. We identified a pervasive developmental pattern: At four months, infants converged all sensory modalities on objects introduced by the mother. From six to twelve months, infants showed increasing decoupling of hands and eyes and increasingly elaborate sequences in multi -object play. Concurrently, dyads engaged in increasingly elaborate social exchanges (e.g., turn-taking) as mothers adapted to infants’ sensorimotor skills. We therefore theorize that triadic attention emerges not as a novel form of social cognition but as a continuous product of sensorimotor development, scaffolded by parents’ expanding social actions. © 2013 S. Karger AG, Basel The period around an infant’s first birthday marks a qualitative change in how the infant responds to and participates in the activity of adults. While, from much earlier months, infants will share gaze with a partner or gaze at objects manipulated Kaya de Barbaro Department of Cognitive Science University of California, San Diego 9500 Gilman Dr. # 0515, La Jolla, CA 92093-0515 (USA) E-Mail kaya @ cogsci.ucsd.edu © 2013 S. Karger AG, Basel 0018–716X/13/0564–0223$38.00/0 www.karger.com/hde E-Mail [email protected] This work was part of a doctoral thesis submitted by the first author to the University of California, San Diego, La Jolla, Calif., USA. D ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Human Development 2013;56:223–248 DOI: 10.1159/000351313 224 de Barbaro/Johnson/Deák by a partner, at around twelve months they begin more actively to engage in shared actions on objects such as imitation and games [Bakeman & Adamson, 1984; Hay, 1979; Piaget, 1962; Ratner & Bruner, 1978; Stern, 1985; Trevarthen & Hubley, 1978]. These activities are ‘‘triadic’’ in the sense that they all involve shared attention and activity between the infant, an adult, and an external locus of attention such as an object or an event. Previous research has shown that triadic attention is a foundation for later learning, including language development [Bruner, 1983], social skills [Bornstein & Tamis LeMonda, 1989], and cultural learning [Tomasello, Carpenter, Call, Behne, & Moll, 2005]. However, we know less about how triadic attention develops. Specifically, previous methods have puzzled researchers eager for a coherent developmental account of the processes that bridge to the dramatic ‘‘triadic shift’’ at twelve months [Adamson & Bakeman, 1991, p. 34; Fogel & DeKoeyer-Laros, 2007]. We aim, in this longitudinal study of naturalistic mother-infant interaction, to demonstrate how a novel approach to studying interaction can make headway on this putatively intractable developmental question. Our approach comes out of recent theoretical and empirical work in cognitive science that takes as its premise that cognition is fundamentally embodied and distributed [Hutchins, 1995], and can be observed through the systematic microanalysis of multimodal, multiparty interaction. We will first give an overview of methods traditionally used to study triadic attention and consider how they have shaped researchers’ interpretation of its development. Next, we review microstudies of infant attention, followed by an overview and motivation of an embodied and distributed cognition approach. Finally, we will describe the results of our application of a microanalysis of multiple attentional modalities in mother-infant interactions, and detail the implications of this approach for the distributed and embodied cognition that develops. The Triadic ‘‘Shift’’: Previous Accounts There is a broad consensus that mother-infant face-to-face interaction follows a three-part trajectory [Adamson & Bakeman, 1991; Tomasello et al., 2005]. First, from about two months, infants begin to be able to engage in dyadic states of ‘‘shared attention’’ with their caregivers. Here, shared attention is construed as jointly attending to one another’s faces via gaze. Infants both initiate and respond to various facial expressions their caregivers make with increasing amounts of positive engagement. This has been observed both in observational and experimental studies. For example, when mothers are asked to stop responding to their infants by ‘‘freezing’’ their facial expressions, even very young infants will quickly become less positive and animated [Tronick, Als, Adamson, Wise, & Brazelton, 1978] and will make active bids to reengage the mother [Tronick, Ricks, & Cohn, 1982]. Next, starting at around six months, infants show a marked decrease in gaze and positive affect to their mother’s face [Kaye & Fogel, 1980; Trevarthen & Hubley, 1978] and become much more attentive to the objects in front of them [Bakeman & Adamson, 1984; Bruner, 1983]. Thus again, it is a dyadic state which predominates the interaction, this time involving the infant and an object of interest. At this stage, infants will occasionally gaze up to the parent while attending to objects. However, they do not make systematic efforts to D ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Microanalysis of Triadic Attention 225 Human Development 2013;56:223–248 DOI: 10.1159/000351313 involve their partners in their object exploration until near the end of their first year [Bakeman & Adamson, 1984]. Summing over years of research, Adamson and Bakeman [1991] describe a ‘‘curious developmental gap’’ (p. 21) between the dyadic states described above and the appearance of true triadic play around the infants’ first birthday. Given that what they consider to be the two components of triadic play – that is, infant-mother interpersonal play and infant-object play – each predominate in earlier periods, it is not clear why infants do not readily or smoothly begin engaging in triadic or joint object play with caregivers. One sort of explanation for the late emergence of triadic play emphasizes a qualitative leap attributable to new conceptual and inferential resources. The explanation stipulates a dawning awareness that other people have ‘‘other minds’’ [Stern, 1985; Tomasello et al., 2005; Trevarthen & Hubley, 1978]. One claim of the work we will be reporting here is that the ‘‘gap,’’ or discontinuity in the behavioral data, almost forces previous researchers to invoke an invisible representational shift to account for the development of triadic attention. Further, we suggest that the behavioral discontinuity is based on past researchers’ choice of units of analyses and on how the resulting component parts that they consider limit their access to the processes involved in development. Our embodied distributed cognitive account of a longitudinal sample of five infants at four, six, nine, and twelve months suggests that these new complex action and exploration patterns can emerge without the need for a conceptual ‘‘sea change.’’ By tracking each of the partners’ access to others’ activities in the world – their motions, words, and affect through space and time – we were able to identify a number of distinct action trajectories wherein activity builds on that seen at earlier sessions. This methodology is a boon to a developmental account in that it focuses on the changing processes of cognition that are visible rather than invisible. In doing so, it provides a basis with which to compare moments of interaction longitudinally across a developing dyad prior to and including the infants’ first birthday [for a similar argument, see Johnson, 2001]. Our analyses lead us to argue that twelve-month complexity is not sudden or discontinuous but a culmination of continuous changes across the first year. Before detailing our approach and our account of triadic development, we first review relevant past research into two categories: macro-level measures and microlevel measures. Macro-Level Analyses The majority of previous studies view the development of triadic attention in terms of “macro”-level changes in mother-infant face-to-face interactions. These can be considered macro both in the timescale at which phenomena of interest are tracked as well as in the unit of analysis. Hsu and Fogel [2003] and Bakeman and Adamson [1984], in historically detailed studies of developing mother-infant interactions, annotated their video with a single dimension – that is, they used a single ‘‘layer’’ of mutually exclusive state variables to classify relatively large time units within the interactions at the level of the dyad. For example, in their classic longitudinal study, Bakeman and Adamson [1984] use a one-dimensional coding scheme that distinguishes between the following six states, each referring to the focus of infants’ attenD ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Human Development 2013;56:223–248 DOI: 10.1159/000351313 226 de Barbaro/Johnson/Deák tion during face-to-face interaction with their mothers: unengaged, onlooking, persons, objects, passive joint, and coordinated joint . Similarly, Fogel’s ‘‘relational system’’ [e.g., Hsu & Fogel, 2003] codes at the level of the attentional coordination of the dyad. From that approach, Fogel notes, for example, whether the attention the dyad members display relative to one another is symmetrical (both attending to the same thing ‘‘actively’’), asymmetrical (one partner viewing the other actively attending to an object, as when a mother might gaze to her infant as he manipulates a toy), or unilateral (one partner engaged in an activity and the other attempting to engage them in a second activity). While these studies reveal systematic changes in attention across development, they do not indicate how such changes occur [for a discussion, see also Fogel & DeKoeyer-Laros, 2007; Forster & Rodriguez, 2006; Johnson, 2001]. Macro-studies indicate the high-level products of the interaction: what the dyad accomplishes during each episode of interaction. The categories are qualitatively different from one another such that we simply do not see anything similar to triadic interactions at earlier months. This leads to behavioral discontinuities between dyads with infants of different ages. In turn, this behavioral discontinuity provides little traction for explaining the age-related shift from one state to another. Distributed and embodied cognition provide an alternative. In conceptualizing the interaction as a rich temporal configuration of component parts, we can observe continuous progress in measures that span the entire age range. By providing continuity across behaviors observed across the first year, we no longer need a discontinuous representational shift to explain the development of triadic attention. Micro-Level Analyses A number of more recent empirical studies track attention at the ‘‘micro’’ level, specifying the particulars of how the mother and the infant attend to one another and to available toys. Micro-level studies can be differentiated from macro-level studies in a number of dimensions, including the timing, specificity and comprehensiveness with which they code attentional behavior. Generally, timing refers to the units of time at which changes in attention are identified. By specificity, we refer to the precision with which attentional behavior is indicated via a particular sensorimotor modality such as gaze. Macrostudies often code a high-level construct of attention akin to the traditional psychological definition of attention as a unitary amodal ‘‘spotlight.’’ This renders attention as an invisible, internal process. Instead, microstudies often specify the particular sensorimotor modality that comes into contact with the targets of attention. Finally, the majority of microstudies of infant activity specify gaze as the single sensorimotor modality by which their subjects attend. However, the utility of coding a wider range of modalities (e.g., hand, mouth, and gaze) and their targets in fine detail is becoming evident. Comprehensiveness refers to the degree to which studies code a variety of sensorimotor behaviors. Below, we review microstudies of attention relevant for our study. Overall, we argue that we need high resolution in timing, specificity, and the comprehensiveness of modalities in order to capture the development of sharing actions between mother and infant. Many studies have shown the critical importance of tracking at the millisecond level when accounting for changes across longitudinal time. The majority of the macD ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Microanalysis of Triadic Attention 227 Human Development 2013;56:223–248 DOI: 10.1159/000351313 ro-level studies described above code at relatively large temporal scales of once every second or even once every three seconds [Bakeman & Adamson, 1984; Cohn & Tronick, 1987]. However, attention to social information can occur at the much finer timescale of tens of milliseconds. This is relevant, for example, in that recent evidence from Yoshida and Smith [2008] suggests that older infants’ gaze to caregivers might shift from longer periods of fixation to shorter but more frequent ‘‘checking-in’’ fixations. More generally, Deák, Krasno, Triesch, Lewis, and Sepeta [in press] found that dyadic attention states between caregivers and infants from three to eleven months of age changed an average of 31.7 times per minute. This suggests that coding intervals of any period longer than 1 s would certainly miss important changes in social attention; even with the largest acceptable interval of 1 s, many other events will be missed. However, many previous efforts to code attention-sharing and triadic attention used coding unit durations of several seconds. Thus, developmental changes of attention distribution within the ongoing social interaction were simply not captured by many past coding schemes. A number of studies have started to specify moment-to-moment changes in the targets of gaze during social interactions. Detailing attention in this way has led to the finding that infants spend relatively little time looking at the face of their caregivers during their interaction with objects. Instead, they spend the majority of time (up to 80%) looking at hands – either their own hands or their caregivers’ hands while those hands are in some sort of contact with the objects [Fiser, Aslin, Lathrop, Rothkopf, & Markant, 2006; Krasno, Deák, Jasso, Lewis, & Triesch, 2007; Yoshida & Smith, 2008]. This holds true for infants of a variety of ages and in naturalistic settings with many potential looking targets [Deák et al., in press]. By combining high-resolution gaze coding with fine-grained temporal analyses, Deák et al. [in press] identified that it is the motion of mom’s hand on a toy, rather than the gaze of the parent, that best predicts infant gaze shifts from one location to another [Deák et al., in press; Yoshida & Smith, 2008]. By precisely coding specific attentional behaviors as they unfold in real time, these studies have identified patterns that contradict the conventional wisdom on early infant attention. However, we know of no studies that detail infant attention in a dyadic context in a comprehensive manner, coding the full range of sensorimotor behaviors by which infants attend to their surroundings. While gaze is the modality most commonly associated with attention, developmentalists have long commented on the attending functions of other sensorimotor modalities. Eleanor Gibson [1988], for example, has written extensively about the importance of the hands to infant sensory exploration: Infants can, for example, rotate an object to view different angles or squeeze it to receive the sensations about its density and internal makeup [see also Streri & Feron, 2005]. Additionally, they can use hands to bring an object to the mouth, another key modality in the first year [Rochat, 1989]. A large body of work from the lab of Holly Ruff has determined that multimodal attention to objects has different physiological and cognitive consequences from simply gazing at an object. Ruff and her colleagues differentiate between ‘‘casual’’ attention which involves gaze-only, and ‘‘exploratory’’ attention, which involves concentrated gaze coordinated with certain types of haptic manipulation such as slow rotating or fingering an object [for a review, see Ruff & Saltarelli, 1993]. In comparison to gaze-only attention, exploratory attention is associated with physiological indices of increased focus such as heart rate deceleration [Lansink & Richards, 1997]. D ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Human Development 2013;56:223–248 DOI: 10.1159/000351313 228 de Barbaro/Johnson/Deák It also diminishes with object familiarity, decreases the likelihood of distraction [Oakes & Tellinghuisen, 1994], and inversely predicts future distractibility [Lawson & Ruff, 2004]. Thus, we anticipate that specifying between these will be important for characterizing attention development across the first year. As a final motivation for a comprehensive tracking of attentional modalities, we note that we saw important differences in the way that infants responded to the mothers’ actions on toys, but only when we considered the infants’ manual activity in addition to their patterns of gaze. In the study by Deák et al. [in press] described above, all infants throughout the study period (from three to eleven months) showed similar patterns of shifting gaze towards a toy following maternal manipulation. In our own analyses (described below), we have found that there are indeed longitudinal changes from four to twelve months in infants’ responses to maternal bids. However, these changes are in the nature of infants’ multimodal contact with the object and how that unfolds over timescales of five to thirty seconds, rather than solely on their gaze immediately following the bid. Embodied affect is another important dimension for characterizing the developing social attention in the dyad [Adamson & Bakeman, 1991]. Affect is studied via general arousal levels as well as facial expressions, especially relative to similar activity in the other. That affect and attention have an important relationship in infancy is clear from previous work showing, for example, that infants use gaze aversion and negative affect to regulate overarousal [e.g., Field, 1981]. Additionally, affect has been shown to have its own developmental course in coordination with other components of attention. For instance, by twelve months, infants have developed patterns of affect that are precisely timed with actions on an object and associated with the gaze to the mother [Eckerman, Whately, & McGehee, 1979]. Additionally, affect from the mother is important for the development of attention and later learning, perhaps via the social reinforcement that the infant is receiving for particular actions [Dodici, Dra per, & Peterson, 2003]. In our study, we further observe that affect in the infants shifted from being attached to immediate events – the mom’s smiling face, a toy tossed in his/ her lap – to occurring within a larger routine – at the denouement of a game, at the recognition of an imitation, at the accomplishment of (or frustration with) a task, etc. Summarizing, the current micro-level literature shows the relevance of the microanalysis of affect and attention, including the targets of gaze of both participants, as well as their manual actions. However, contemporary work is still too piecemeal to account for the emergence of triadic attention at twelve months. For this, we need a distributed analysis that situates analyses of embodied attention in the social interaction as we detail below. Our Approach: Reconceptualizing Interaction Our definition of attention for this work is based on theories of distributed and embodied cognition wherein activity of the body is considered cognitive activity [Clark, 2008; Hutchins, 1995]. Perception is not a passive process in which sensory information is displayed on the retinas. Instead, infants move and adjust their bodies in real time in order to identify and explore dynamic features of their environment [Gibson, 1988; Noë, 2004; Suchman, 1987]. Furthermore, infants do not just sense wavelengths of light, but they actively seek out differences in their surroundings via many types of sensory receptors. D ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Microanalysis of Triadic Attention 229 Human Development 2013;56:223–248 DOI: 10.1159/000351313 This leads us to define attention as an effortful change in the sensory access of one individual to objects or other individuals [Johnson & Karin-D’Arcy, 2006]. Consistent with the neurobiological organization of human sensorimotor systems, we distinguish different types of sensory access including visual, oral, and haptic. Thus, changes in the targets of gaze as well as in manual and oral contact with toys and partner were all regarded as important in characterizing the attention of our participants. Furthermore, we distinguished between reaching towards a toy, grasping a toy, and manipulating a toy since each of these motor activities differs in the type and timing of the tactile and proprioceptive access that they afford. Above we reviewed a number of benefits to operationalizing attention in this way. Moreover, by tracking dynamics of sensorimotor modalities as individuals gain access to targets, attention becomes a process that unfolds over moment-to-moment time. Studying the changing process of how infants’ modalities become organized to attend to their caregivers and objects in the world around them provides a new window into the development of infants’ attention. For example, we can characterize the dynamics of sensorimotor modalities as they become coordinated with a toy: How many modalities are on the toy; what order do they get there; how long does gaze remain on the toy given concurrent maternal elaboration, given concurrent infant manual elaboration, or given maternal elaboration of another toy? To give an example, our qualitative results show that gaze typically leads relative to other modalities in infants of all ages but that younger infants (four months) maintain gaze contact with the toy for the full duration that any modality is in contact with that toy, whereas older infants (six to nine months) may begin to look away once the hand has made contact with the toy, or even look away during the reach (at nine and twelve months). Thus, while reaching is visually guided at all ages, it depends on a decreasing level of gaze involvement as infants come to decouple their sensory modalities [see also Bushnell, 1985]. In this way, by documenting the microdynamics of how infants organize their sensorimotor modalities over developmental time, we can observe the variety of cognitive changes involved in the emergence of triadic attention. Drawing from distributed cognition, we embed this analysis of embodied, multimodal attending in a triadic context of the mother-infant-object. From a distributed perspective [Forster, 2002; Hutchins, 1995; Johnson, 2010], the focus of research is not just on the elements of a system but on their configuration. A distributed account of ontogeny, then, is one of configural change. Given, for example, the multimodal data generated by embodied analyses, we can observe a set of elements that reconfigure as the infant ages. Many of these elements (such as touch, eye contact, visually tracking a proffered object, etc.) are common to all ages. However, at each age, these elements organize relative to each other and to new behaviors (like ‘‘reach’’ or ‘‘stack’’) in a distinctive way. By characterizing such changes in organization, the distributed approach can help provide a coherent account of the transitions involved in the development of triadic attention. The distributed approach is well adapted not only for data that are multimodal but also for those that are multiparty. In the development of triadic attention, there are many critical relationships such as eye contact, imitation, altering the other’s access to the toy, etc. that cannot be specified by the behavior of one subject alone. For example, when a mother ‘‘presents’’ a toy, the criteria for scoring that event include both the mother’s grasp and extension of the object as well as the infant’s available line of sight to that object. Taking interaction as the unit of analysis is a key characteristic of the disD ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Human Development 2013;56:223–248 DOI: 10.1159/000351313 230 de Barbaro/Johnson/Deák tributed approach. From this view, the infant’s ontogenetic challenge is not to perform particular, prespecified actions but to adapt to the conditions embodied by the mother’s activity, which in turn are adjusted in response to the infant’s current behaviors. Furthermore, we can describe pronounced individual differences in the mother’s tendency to act contingently with her infant that help shape how and when the infant engages. Thus, rather than assigning binary performance scores (e.g., whether the infant does or does not ‘‘attend’’ at a given age), distributed analyses produce ‘‘profiles of participation’’ [Forster & Rodriguez, 2006] that reflect the coregulation of activity within a dyadic interaction. Assessed longitudinally, such analyses can reveal the developmental course of changes in mother-infant-object coordination. The study of cognition from this approach also highlights a tenet of the distributed cognition approach: Cognitive events are multiscalar – i.e., unfolding simultaneously at the micro-, macroand historic-developmental timescales [Hutchins, 1995]. To produce a coherent distributed account of cognitive development, information must be collected at all these timescales. At the microsecond scale, we observe shifts of gaze, facial expression (e.g., gleeful smiles), and hand movement. At the macroscale, a particular look or grasp is positioned within an ongoing routine where, for example, it may repeat (as in peek-a-boo), or change, or organize with other events. At the historic-developmental timescale, the dyad’s long-term experience with such routines (e.g., a playful father’s tendency to initiate exciting games) comes into play. In the study reported here, a monthly six-minute sample of free play was recorded from infant-mother dyads at home. For the current analyses, we observed episodes from when the infant was four, six, nine and twelve months of age. By tracking the details of these interactions across months, we can capture relevant changes at both the microand macro-levels. Of course, making inferences about these historic shifts requires interpolation and induction. We do not know how often and at what ages a given mother-infant dyad has played, for example, ‘‘peek-a-boo’’ or ‘‘build-up/knockdown’’ games. Nonetheless, some historic relations are directly observable, by comparing configural change across months. For example, we can describe long-term changes in the extent and nature of scaffolding by the mom by observing how her bids for the infant to attend to objects vary over developmental time. That is, the level and placement of motor activity by the mom clearly changes as the infant ages, from elaborate movements and expressions repeatedly directed at the infant’s immediate frontal space in the earlier sessions to only a distal object touch and glance at the infant by twelve months. Together, assessments at these different timescales enable us not only to see cognition in action but to watch it develop as well. While a multimodal, multiparty, multiscalar account can become very complex very quickly, its grounding in the embodied activity provides a straightforward approach that can reveal both significant pattern shifts and unexpected continuities. At each age, we will describe the same set of variables and participatory event types. These include the type and number of attentional modalities directed to each target, the order and rate of modality-specific transitions between targets, the contingencies between the participants’ activities, and the type and timing of accompanying affect. We can thereby document how the organization of these elements changes as triadic attention develops. Each description will also include how ‘‘maternal bids’’ in which the mother directs the infant’s attention to a different object are coordinated at each age. As we shall see, these interactions and their constituent microbehavioral elements configure differently over time as they become organized into routines that are D ow nl oa de d by : U ni v. o f C al ifo rn ia S an D ie go 13 2. 23 9. 1. 23 1 9/ 2/ 20 13 5 :5 9: 13 P M Microanalysis of Triadic Attention 231 Human Development 2013;56:223–248 DOI: 10.1159/000351313 increasingly prolonged, attentionally divided, and eventually embedded. Through this approach, we can observe how constraints on the infant’s motor development – for example, decoupling the hands to simultaneously contact multiple objects – shape how the infant responds to maternal toy bids. In this way, we argue, an embodied developmental change (i.e., manual action behaviors) directly feeds into developmental changes in distributed, dyadic-participatory changes. This provides an illustration of how this approach, using embodied and distributed analysis to parse complex social interactions at multiple temporal and behavioral units, can address the cognitive and behavioral complexity that emerges in very elaborate social interactions such as infants’ triadic interactions with caregivers.