Authors: Yuyan Luo, Kristy vanMarle, Ashley M. Groh
Categories: Article
Source: Perspectives on psychological science : a journal of the Association for Psychological Science
Authors: Yuyan Luo, Kristy vanMarle, Ashley M. Groh
Meta-analytic evidence indicates that the quality of the attachment relationship that infants establish with the primary caregiver has enduring significance for socioemotional and cognitive outcomes. However, the mechanisms by which early attachment experiences contribute to subsequent development remain underspecified. According to attachment theory, early attachment experiences become embodied in the form of cognitive-affective representations, referred to as internal working models (IWMs), which guide future behavior. Yet, little is known about the cognitive architecture of IWMs in infancy. In this article, we discuss significant advances made in the field of infant cognitive development and propose that leveraging insights from this research has the potential to fundamentally shape our understanding of the cognitive architecture of attachment representations in infancy. We also propose that the integration of attachment research into cognitive research can shed light on the role of early experiences, individual differences, and stability and change in infant cognition, as well as open new routes of investigation in cognitive studies, which will further our understanding about human knowledge. We provide recommendations for future research throughout the article and conclude by using our collaborative research as an example.
Attachment is a psychological bond that infants establish with parents in the first years of life (Bowlby, 1969/1982). The quality of the attachment relationship that infants establish with parents has enduring significance for their socioemotional development (Bowlby, 1973; Sroufe et al., 1990). Specifically, insecure attachment in infancy poses risk for poorer peer relationships, externalizing problems, and (to a lesser extent) internalizing symptoms across the early life course (Deneault et al., 2021; Fearon et al., 2010; Groh, Fearon et al., 2014; Groh et al., 2012; Madigan et al., 2013). Moreover, a growing body of evidence implicates insecure attachment in poorer cognitive abilities, including poorer language competence, cognitive skills, and executive functioning (Deneault et al., 2023; Pallini et al., 2018; Pallini et al., 2019; Van IJzendoorn et al., 1995), skills that are important for school readiness and academic achievement (Blair & Raver, 2015; Blankenship et al., 2019; St Clair-Thompson & Gathercole, 2006; Welsh et al., 2010). Thus, it is critical to understand how infants’ early attachment experiences are embodied and carried forward to impact subsequent development.
According to attachment theory (Bowlby, 1969/1982, 1973, 1980), these processes hinge on infants’ cognitive-affective representations of early experiences with parents, or internal working models (IWMs), which encompass views of the self, others, and the nature of relationships. These representations guide beliefs, feelings, and expectations regarding behavior of the self and others in close relationships, as well as what to expect from the social world. Accordingly, although formed on the basis of infants’ attachment relationships with specific parents, IWMs are thought to generalize and serve as prototypes for other close relationships (e.g., parent-child relationships, friendships, and romantic relationships) throughout development (Groh et al., 2017; Owens et al., 1995; Raby et al., 2015; Roisman et al., 2009; Roisman et al., 2005; Simpson et al., 2007). Further, secure IWMs are characterized by confidence in parent availability, which is thought to enable children’s better management of available cognitive resources (e.g., as opposed to being preoccupied with the attachment relationship) and thus facilitate more flexible learning and exploration of the environment (e.g., more effectively and cooperatively learning from familiar and unfamiliar others) and in turn confer benefits for children’s cognitive development (Deneault et al., 2023; Madigan et al., 2019; Pallini et al., 2018; Pallini et al., 2019; Van IJzendoorn et al., 1995).
Despite the importance of IWMs as an explanatory mechanism of the legacy of early attachment experiences, the content and nature of IWMs was not fully detailed in Bowlby’s writing, perhaps due to the state of knowledge on human cognition at the time. In the intervening years, much theoretical and empirical work has been conducted that elucidates the IWMs construct (Bretherton, 1999; Bretherton & Munholland, 2008; 2016; Grossmann et al., 1999; Waters & Waters, 2006). Significant advances include the development of reliable observational, narrative, and self-report measures of attachment representations in childhood and adulthood (see Crowell et al., 2016; Hesse, 2016; Solomon & George, 2016; Waters & Waters, 2006), and evidence for the role of attachment in shaping the ways in which social information is cognitively processed and responded to emotionally (Cassidy, 1994; Dykas & Cassidy, 2011). These advances, however, have been limited to our understanding of IWMs in the early childhood years and beyond, with virtually no work focusing on infancy (Sherman et al., 2015). One reason for the limited advances in our understanding of IWMs in infancy is that research on infant cognitive development and socioemotional development has remained two largely separate fields of study with little cross communication. Although attachment and infant cognition scholars have turned to both literatures of developmental science to highlight how the consideration of theory, evidence, and methods from the complementary field might advance knowledge in their respective areas (Sherman et al., 2015; Tasimi, 2020), these efforts have been conducted in parallel, effectively keeping the research siloed.
Consistent with calls for developmentally informed theoretical and empirical work aimed at defining the nature and functioning of IWMs in the formative years of attachment development (Thompson et al., 2021), we draw on our combined expertise in attachment and infant cognition to demonstrate how these two fields of research can be integrated to provide novel insights into the nature, content, and construction of attachment representations from their inception in the first years of life, work that is critical to advancing a developmental account of IWMs across the lifespan (Thompson et al., 2021; Thompson et al., 2022). Specifically, we open with an overview of IWMs and evidence for their existence in infancy. Next, we highlight significant advances made in the field of infant cognitive development (e.g., Baillargeon, 2008; Baillargeon et al., 2015; Ting et al., 2020; Xu, 2019), and argue that leveraging insights from this research has the potential to fundamentally shape our understanding of the cognitive architecture of attachment representations in infancy. Although we focus on the contributions of infant cognition to understanding attachment representations, greater cross-communication will also advance theory and research on infant cognition. Thus, we close with a discussion of how the integration of attachment research into infant cognition studies can in turn transform understanding of cognitive development. Throughout this article, we demonstrate how collaborative research can be mutually beneficial, with the aim of fostering greater connection between socioemotional and cognitive developmental science.
A central tenet of attachment theory is that children’s early attachment-relevant experiences inform subsequent interpersonal functioning across the life course. In explaining how early attachments might exert an enduring force on development, Bowlby (1973) adopted from Craik (1943) the concept of working models in which features of the environment, social partners, and experiences are mentally represented and manipulated to generate predictions regarding potential courses of action and their likely outcomes. Applying this concept to attachment relationships, Bowlby proposed
Thus, according to Bowlby, internal working models comprise information regarding the attachment figure’s behavior, the child’s behavior, and the nature of the attachment relationship based on the child’s experiences with the parent. Importantly, these experience-based mental models are used by individuals to gather and interpret information, devise plans, and inform behavior with the attachment figure, as well as other social partners over time (e.g., peers, romantic partners; Bowlby, 1980; Bretherton, 1999; Dykas & Cassidy, 2011).
Main and colleagues (e.g., Main et al., 1985) cogently argued that infant behavior during the Strange Situation Procedure (SSP; Ainsworth et al., 1978) serves as evidence for the presence of IWMs in infancy. During the SSP, infants undergo a series of increasingly stressful separations from, and subsequent reunions with, their parent. The quality of attachment is revealed in the organization of infant behavior around the parent, particularly upon reunion. Whereas infants classified as secure greet their parent upon return and seek him/her out to alleviate any distress elicited by the separation, insecure infants either ignore (i.e., insecure-avoidant) or simultaneously seek and resist (i.e., insecure-resistant) their parent upon reunion. In addition to these “organized” attachment patterns, infants classified as “disorganized” exhibit contradictory and/or fearful behavior in the context of the SSP (Main & Solomon, 1986; 1990). These patterns of behavior during the SSP are thought to reflect infants’ expectations regarding the parent’s availability and responsiveness based on prior experience (Main et al., 1985). Specifically, the parent’s responsiveness to the infant’s attachment signals is thought to inform the infant’s expectations regarding their parent’s availability in times of need, which in turn, informs the infant’s future expectations and behavior. As such, the organization of infant behavior around the parent during the SSP is conceptualized as the behavioral manifestation of infants’ experience-based expectations of parent availability and responsivity.
Although the presence of IWMs of attachment might be intuited from infant behavior during the SSP, the SSP is primarily regarded as a behavioral assessment of the quality of infant attachment to a specific parent, and it is not thought to directly assess the mental representations underlying infants’ IWMs of attachment. Moreover, it can only be used with infants that are at least 12 months of age, since younger infants are not yet able to move themselves through the environment reliably in order to consistently organize their behavior around the parent. Thus, to date, there has been little formal investigation of the nature, content, and development of IWMs in the first years of life (Dykas & Cassidy, 2011; Sherman et al., 2015). This is despite the fact that significant progress has been made in our understanding of the processes by which infants acquire knowledge and construct representations about different aspects of the world. In the following section, we discuss advances in the field of infant cognitive development and their relevance to deepening our understanding of attachment representations in infancy.
The rich field of research on infant cognition has transformed our thinking of early cognitive abilities in the first two years of life, especially before children become proficient language users, and these advances have direct implications for understanding IWMs in infancy^1^. An important theoretical approach to infant cognition suggests that infants construct intuitive causal explanatory frameworks about a limited set of core knowledge domains that reflect critical aspects of the world, including the physical, numerical, psychological, sociomoral, and biological domains, possibly due to our evolutionary history. In each domain, infants’ conceptual understanding is organized around a few basic tenets that define the objects or entities covered by the domain and guide reasoning about them (e.g., Baillargeon, 2008; Baillargeon & Carey, 2012; Baillargeon et al., 2011; Baillargeon et al., 2016; Carey, 2009; Carey & Spelke, 1994; Gelman, 1990; 2002; Gelman & Williams, 1998; Keil, 1995; Leslie, 1994, 1995; Setoh et al., 2013; Spelke, 1994; Spelke & Kinzler, 2007; Ting et al., 2020; Ullman et al., 2017; Wellman & Gelman, 1992; 1997; Wertz, 2019; Woo et al., 2022; Wynn & Bloom, 2013). As detailed below, when applied to attachment, these theoretical, empirical, and methodological advances in the field of infant cognition have the potential to address several fundamental questions pertaining to IWMs in infancy^2^. First, when do infants start the process of forming enduring mental representations of attachment experiences? Second, what is the likely content of attachment representations in infancy? Third, what kinds of evaluations of attachment experiences comprise infant attachment representations? Fourth, how do attachment representations develop during and in the years following infancy?
In formulating the IWMs construct, Bowlby (1969/1982, 1973, 1980) drew from several different theoretical perspectives. Stemming from his training in object relations theory, Bowlby incorporated the idea that infants construct nonconscious mental representations of parents derived from experience (for a review of object relations theory and its relation to attachment, see Ainsworth, 1968). As noted above, Bowlby also adopted the concepts of mental models and controls systems to propose that attachment experiences are represented as mental models of the self, others, and the nature of relationships. Bowlby was also influenced by Piaget’s theory of cognitive development (1954), including Piaget’s work on the developmental progression of the achievement of cognitive milestones (e.g., object permanence), as well as the concepts of assimilation and accommodation to explain stability and change in IMWs over time. Drawing from ethology (especially research by Lorenz [1935] on imprinting), Bowlby proposed that the key function of IWMs is to ensure protection from threat, thereby promoting infant survival (see Bretherton, 1992; 1999; Bretherton & Munholland, 2008). Thus, although a great deal of research on IWMs has been conducted in the preschool years and beyond, according to Bowlby, IWMs may begin to be constructed early in the life course, perhaps as early as the first year. However, exactly when infants begin to construct attachment representations remains unknown.
When the capacity for constructing enduring mental representations emerges has been the focus of much empirical inquiry in cognitive reseach. Piaget (1954) argued that infants rely on sensorimotor experience to build mental representations, such as of physical objects, which appear to break down once objects become hidden, citing as evidence young infants’ failure to search for them. Since the 1970s, however, researchers have challenged Piaget’s claims (e.g., Bower, 1974; Bower et al., 1971), arguing that Piaget underestimated infants’ representational capacities due to an over-reliance on measures of infants’ overt actions such as search behaviors (Piaget, 1954, 1962, 1963). Hence, over the past several decades, researchers have devised alternative methods that measure infants’ visual attention to reveal their representational capacities (e.g., Aslin, 2012; Aslin & McCurray, 2004; Baillargeon, 1987; Baillargeon et al., 1985; Biro et al., 2015; Bornstein, 1985; Brennan et al., 1966; Choi et al., 2023; Cohen, 1976; Fantz, 1964; Fawcett & Liszkowski, 2012; Franchak et al., 2011; Hamlin et al., 2010; Hernik & Broesch, 2019; Hirsh-Pasek & Golinkoff, 1996; Leslie & Keeble, 1987; Libertus & Brannon, 2010; Oakes & Ellis, 2013; Scott & Baillargeon, 2013; Spelke, 1985; Spelke et al., 1995; Tan & Hamlin, 2022; Teller, 1979; Woodward, 1998; for a review, see Margoni et al., 2024).
Although looking-time measures have become a staple of infant cognition research, there remain debates regarding their use. For instance, some researchers still consider overt actions as the gold standard indicator of infant knowledge (e.g., Buttelmann et al., 2009; Engelmann & Tomasello, 2019; Mueller & Overton, 1998; for a relevant commentary, see Surian & Margoni, 2020). Others raise issues with the conclusions drawn from looking-time tasks and suggest low-level alternative explanations for the results, as opposed to infant conceptual knowledge (e.g., Aslin & Fiser, 2005; Bogartz et al., 1997; Haith, 1998; Heyes, 2014; Paulus, 2022; for rebuttals, see Margoni et al., 2024; Stahl & Kibbe, 2022). In contrast to these views, proponents of looking-time measures suggest that these measures can be used with very young infants, different from action tasks that require certain motor skills, and that meaningful control conditions, which are often included in looking-time tasks, together with experimental manipulations, aid in ruling out low-level interpretations (Margoni et al., 2024). The following discussions on infant cognition are aligned with this perspective. Moreover, while we focus on evidence for infant cognitive capacities from looking-time measures, we also include corroborating evidence from action tasks.
Two widely used looking time measures include preferential looking (PL) and violation of expectation (VOE). In PL studies, infants are presented with two stimuli simultaneously. Longer looking at one of the two stimuli demonstrates not only that infants differentiate the stimuli, but also that they prefer one stimulus over the other. VOE studies employ either a within-subject design or a between-subjects design. In VOE studies with a within-subject design, infants are presented with two events in alternation, one that accords with expectations (the expected event) and one that violates expectations (the unexpected event). In VOE studies with a between-subjects design, infants watch the expected or unexpected event only. In both VOE study designs, infants’ reliably longer looking at unexpected than at expected events reveals their expectations because infants naturally closely examine things that are unexpected (Baillargeon et al., 2015; Margoni et al., 2024). In recent years, technology innovations such as eye trackers also allow researchers to gather even finer-grained data when analyzing infant visual attention, e.g., pinpointing exactly where infants are looking while they attend to a display or measuring pupil dilation to indicate exactly when a detection of violation occurs during an event (e.g., Aslin, 2007; Hepach & Westerman, 2016; Laeng et al., 2012; Zhang & Emerson, 2020). In the following sections, most of looking-time infant cognitive tasks reviewed use the VOE method, and we will specify when the PL or eye-tracking method is used instead.
When leveraging looking-time methods, infants’ capacity for mental representation has been demonstrated beginning as early as the first months of life. Specifically, in terms of mental representations of physical objects, it has been shown in VOE tasks that infants as young as 2.5 months look reliably longer at events depicting violations of object permanence, such as when objects magically disappear or occupy space already occupied by other objects, than at events consistent with object permanence (e.g., Aguiar & Baillargeon, 1999; Baillargeon et al., 1985; Luo et al., 2003; Spelke et al., 1992; Spelke et al., 1995; Wang et al., 2004; Wang et al., 2005; for relevant debates, see e.g., Baillargeon, 2004; Bremner & Mareschal, 2004; Hood, 2004; Leslie, 2004). In the psychological domain, by at least 5–6 months of age, infants can keep track of which objects in a scene others (henceforth agents, i.e., entities that can control their behavior, human or nonhuman) can or cannot perceive (e.g., one or two objects) and what segments of an event agents have or have not witnessed (e.g., the displacement of an object) and use this information to make sense of their behavior, even when the agents’ perspectives differ from their own (e.g., Choi et al., 2022; Hamlin et al., 2013; Kim & Song, 2015; Liszkowski et al., 2007; 2008; Luo & Baillargeon, 2007; Luo & Johnson, 2009; Meristo & Surian, 2013; Scott & Baillargeon, 2009; Southgate et al., 2010; Southgate & Vernetti, 2014; Ting et al., 2019; 2021; Tomasello & Haberl, 2003; Vouloumanos et al., 2014; for reviews, see Baillargeon et al., 2016; Scott et al., 2022; for different views, see e.g., Butterworth & Jarrett, 1991; Gergely & Csibra, 2003; Tomasello, 1999; Woodward, 2005). In the sociomoral domain (usually about social situations involving two or more agents), infants prefer individuals who were helpful towards others in a previous context, as assessed via PL tasks as early as 3 months and as assessed via their own choices by 6 months (Hamlin et al., 2007; Hamlin et al., 2010). In VOE tasks, they also demonstrate an expectation for individuals to act towards others in accordance with moral principles such as fairness (e.g., Buyukozer Dawkins et al., 2019; Schmidt & Sommerville, 2011; Sloane et al., 2012; Ziv & Sommerville, 2017; for reviews, see Baillargeon et al., 2015; Hamlin, 2013b; Ting et al., 2020; Woo et al., 2022; for views focusing on the roles of socialization processes and experiences with peers and other people in early moral cognition, see e.g., Dahl, 2019; Eisenberg et al., 2006; Smetana, 2006; Turiel, 2015).
Taken together, the wealth of research on infant cognition made possible largely due to the use of looking-time measures has revealed the presence of representational capacities in infants within the first months of life, long before children become proficient language users. In light of such evidence that infants can use their representational capacities to learn about various core aspects of the world, it stands to reason that infants may also begin to build mental representations of their attachment experiences during the first year of life. Thus, because the vast majority of investigations of the content and nature of IWMs are conducted after the acquisition of language, a potentially critical developmental period in the construction of IWMs of attachment has been missed. To advance understanding of the development of IWMs across the life course, empirical investigation of attachment representations must start much earlier than is current practice, beginning in infancy.
An enduring question in attachment research concerns the contents of IWMs across development (Thompson et al., 2021; Thompson et al., 2022). Indeed, the diversity of perspectives on the contents of IWMs was recently captured in a compilation of chapters organized around the issue of the nature and function of IWMs (Cassidy, 2021; Girme & Overall, 2021; Oppenheim & Koren-Karie, 2021; Thompson, 2021; Waters et al., 2021). Despite differences in perspective, a common feature of accounts of IWMs is that such models include information about attachment figure(s) derived from experience. Given the centrality of attachments to parents in the early life course, IWMs in infancy likely contain information about a parent’s behavior in attachment-relevant contexts, such as a parent’s responsiveness to the infant’s attachment signals, that the infant uses to inform expectations about the parent’s behavior in future contexts. Below we draw on theory and evidence from the field of infant cognition to demonstrate how such work might inform understanding of how infants construct mental representations of their attachment experiences, with a specific focus on representations of parental responsiveness to attachment signals.
A critical first step in infants’ capacity to mentally represent parental responsiveness to their signals is the ability to detect contingencies between their signals and the parent’s behavior. Evidence from infant work suggests that infants are well equipped to detect contingencies from a very young age. Even from birth, infants naturally expect correspondences in input from different perceptual modalities (e.g., a match between visual and tactile information or a match between a sight and a sound; Kaye & Bower, 1994; Kuhl & Meltzoff, 1982; 1984; Meltzoff & Borton, 1979; Spelke, 1976; Streri & Spelke, 1988; Walker et al., 2010). These cross-modal correspondences can also be formed on the basis of abstract numerical values (e.g., an audio sequence of 18 syllables and a visual display of 18 objects; Izard et al., 2009; Starkey et al., 1983). In addition, young infants can quickly construct novel contingencies, for example, by pairing their actions with changes in the environment (e.g., infants quickly learn that kicking their leg causes a mobile hanging above their crib to jiggle, through a ribbon tied connecting their ankle to the mobile, and remember these contingencies for a relatively long period; Rovee-Collier, 1997; 1999). Importantly, they can distinguish the causal roles in such novel contingencies (e.g., a red brick collides into a green brick and causes it to move off; Kominsky et al., 2017; Leslie & Keeble, 1987). This sensitivity to contingency also extends to infants’ interactions with their social partners. For instance, from 2–4 months of age, infants act differently when their behaviors (e.g., smiles and vocalizations) are contingently responded to either by their mother or a female stranger (the interaction partners’ behaviors match in form and are separated within 1 second) in face-to-face interactions, compared to when such contingency is disrupted (e.g., as in the still-face procedure or when infants watch the replay of a previous interaction; Beebe et al., 2016; Bigelow & Rochat, 2006; Mesman et al., 2009; Murray & Trevarthen, 1985; Tamis-LeMonda et al., 2001; Tronick et al., 1978). It has been suggested that social contingency detection allows infants to better understand the effect of their own actions on the external social world (e.g., Bigelow & DeCoste, 2003; Gergely & Watson, 1999; Haith & Benson, 1998; Tarabulsy et al., 1996). Thus, taken together, evidence indicates that starting from early in life, infants detect contingencies in their observations of, actions on, and interactions with the physical and the social aspects of their environment, and that they extract causality from these contingencies. Moreover, once learned, infant knowledge of contingencies persists for a period of time (e.g., Rovee-Collier & Hayne, 2000). As such, it is reasonable to expect that from an early age, infants may track contingencies between their attachment cues (e.g., crying) and the parent’s responses (e.g., providing comfort), register the cause-and-effect relations (e.g., when I cry, my parent responds by comforting me), and carry these representations with them over time.
Further, extant evidence suggests that infants should be able to detect patterns in contingencies between their parent’s responsiveness to their attachment signals over time (e.g., my parent always, inconsistently, or rarely responds to my crying). This is because it is widely accepted in the cognition literature that infants, even newborns, detect regularities, such as frequencies or probabilities of occurrences, in accumulated experiences (e.g., Aslin et al., 1998; Bulf et al., 2011; Denison et al., 2013; Kochukhova & Gredebäck, 2007; Saffran et al., 1996; Teinonen et al., 2009; Xu & Garcia, 2008; for reviews, see e.g., Aslin & Newport, 2012; Saffran & Kirkham, 2018). A well-known example is that 8-month-olds are shown to segment “words” from a stream of continuous “syllables” in an artificial language (e.g., “bidakupadotigolabubidaku …”) based only on transitional probabilities (e.g., within the word “baby,” “ba” is always followed by “by” and hence with a transitional probability of 100% between the two syllables). Specifically, the “syllable” stream consisted of 4 nonsense “words” (e.g., “bidaku”) presented in random order and was designed such that within-word probabilities of syllable pairs were 100% (e.g., “bida”) while between-words probabilities were 33% (“kupa”). Even after 2 min of listening to the stream, infants seemed to have extracted the “words” from the stream, as they spent significantly more time listening to novel items presented during test trials (e.g., “dabiku” with transitional probabilities of 0 because both “dabi” and “biku” were new syllable pairs) compared to familiar ones (the “word” “bidaku”). This has been interpreted as evidence that infants have the capacity to segment the continuous auditory stream into words based only on statistical patterns in the input (Aslin et al., 1998; Saffran et al., 1996). This early-emerging sensitivity to statistical patterns has been viewed as a domain-general learning mechanism applied to all data or experiences received (e.g., Perfors et al., 2011; Xu, 2019), which almost certainly includes attachment experiences (Dweck, 2017; Sherman et al., 2015; Tasimi, 2020). Indeed, in social contingency detection, from 2 months of age, infants have been found to be sensitive to their mother’s contingent responsiveness level (e.g., in a 5-min face-to-face interaction, how often the mother smiles in contingent response to the infant’s smile compared to how often the mother smiles in the whole segment; Bigelow, 1998; Bigelow & Rochat, 2006). It is also possible that infants may organize the patterns they detect in parent responses to formulate rules, as they have been shown to be adept at learning new causal rules from just a few meaningful examples (e.g., Baillargeon & Carey, 2012; Baillargeon & DeJong, 2017; Wang, 2019; Wang & Baillargeon, 2008; Wang & Kohne, 2007). Applying this to attachment-relevant scenarios, for instance, when an infant falls, is hurt, or frightened by a new person or a sound, the parent might respond quickly. However, when the infant becomes frustrated in the course of play, the parent might delay responding to the fussy infant for a short while to see if the infant might independently resolve the problem. From such experiences, infants may categorize these responses and apply causal meaning to them (e.g., In times of need, my parent responds to my signal).
Importantly, beyond simply tracking the parent’s (patterns of) responsiveness, evidence from the field of infant cognition suggests that infants might also represent such information in conceptual terms. In particular, infants’ knowledge in the psychological domain provides useful insights. In this core domain, infants attempt to interpret and predict agents’ behavior in terms of psychological causes, such as intentions, dispositions (e.g., preferences), perceptions, and beliefs (for reviews, see Baillargeon et al., 2016; Scott et al., 2022). For instance, when infants as young as 3–5 months of age are presented with events in which a human agent consistently reaches for and grasps one of two objects, object-A but not object-B, they seem to attribute to her an intention of choosing object-A, in other words, a preference for A over B. When the two objects’ positions are switched, infants demonstrate in VOE tasks an expectation that she will act in accordance with her intent or preference and choose A again and they look reliably longer when the agent chooses B instead (e.g., Choi et al., 2018; Gerson & Woodward, 2014; Sommerville et al., 2005; Woodward, 1998; for a review, see Choi & Luo, 2023). Indeed, there have been a wealth of studies documenting infants’ cognitive capacities to attribute intentions to agents as underlying reasons for their actions (for reviews, see Baillargeon et al., 2015; Hunnius & Bekkering, 2014; Luo & Baillargeon, 2010; Robson & Kuhlmeier, 2016; Uithol & Paulus, 2014; Woodward et al., 2009). Therefore, as infants keep track of parent responses (e.g., my parent responds to my signal), they also might attempt to make sense of their parent’s behavior in terms of intentions (e.g., my parent intends to soothe me).
Moreover, as infants engage in intentional reading of parent behavior, they may grasp that a parent can achieve the same goal via different means, for example, an understanding that providing a pacifier or picking up and rocking the infant are both behaviors intended to soothe the infant’s distress. This is because young infants are found to make use of information on equifinality in intentional reading. For example, they seem to understand that when an agent looks at or pokes at object-A, but not object-B, from different angles, or when an agent approaches object-A via different movement paths, the agent intends to choose A again (e.g., Biro & Leslie, 2007; Hernik & Southgate, 2012; Johnson, Ok, et al., 2007; Luo, 2011). Hence, infant attachment representations may include similar information about parents can behave in different ways in response to the same type of attachment signals.
In addition to intentional reading, it is critical for infants to consider the broader context, including caregiver and contextual constraints, in making sense of the parent’s behavior. Focusing first on caregiver constraints, although a parent may typically respond to an infant’s attachment signals, in daily life it is inevitable that the parent sometimes cannot respond in a timely manner to infant signals in her/his usual way, for instance, because the parent is occupied elsewhere (e.g., when driving) or tries but fails to respond (e.g., when accidentally stumbling on a toy on the way to respond). To still attribute to the parent an intent to respond in these situations, infants need to be able to consider contextual limitations on others’ behavior, as well as people’s failed attempts in their goal pursuits. Cognitive work has made clear that at least by about 8 months of age, infants become such skillful intention readers. For instance, they seem to examine reasons for an agent’s choice of goal-directed actions within given contexts. In an imitation study (Gergely et al., 2002), based on Meltzoff (1988), 14-month-old infants were first presented with a human agent who used her head, rather than her hands, to turn on a box and causing it to illuminate. After a 1-week delay, infants used their hands to illuminate the box if they had seen that the agent’s hands were occupied during the initial demonstration (e.g., she was holding a blanket to keep warm), suggesting that infants seemed to have recognized the restrictions to her using her hands to operate the lightbox. However, if infants had seen that the agent’s hands were free during demonstration, they more faithfully copied the agent’s action and used their head to touch the lightbox, possibly due to their innate proclivity to learn from others (Csibra & Gergely, 2011; Hernik & Csibra, 2009), hence accepting the agent’s unusual way of achieving the goal of illuminating the lightbox. Similar results on the effect of situational restrictions on imitation performances have been reported in 12-month-old infants (Zmyj et al., 2009). Further, in a VOE study (Luo, 2010), based on a task devised by Woodward (2003), 8-month-old infants appeared to interpret an agent’s staring at object-A but not object-B as guided by her preference, when situational restrictions accounted for why the agent did not reach for her preferred she could only stare at the chosen object through a small window, or because her hands were occupied with holding a sippy cup. However, when the agent stared at the chosen object through a large window, or when her hands were visibly free, infants failed to attribute to her a preference, in line with findings from Woodward (2003), as though they could not understand why she did not use her hand to reach for the object. Moreover, infants are found to “read” agents’ intentions even when agents’ actions fail to produce desired outcomes. When a human agent tries to pull apart a dumbbell, but her hands keep slipping off, or she attempts but fails to obtain an out-of-reach toy, infants in the second year of life reenact her intended action or help her achieve the goal themselves in action tasks (e.g., Dunfield & Kuhlmeier, 2010; Meltzoff, 1995; Warneken & Tomasello, 2006; 2007). Infants at 9, 12, and 18 months were even found to react relatively patiently when an agent failed to pass them a toy because she accidently dropped it or was unable to obtain it herself (Behne et al., 2005). Similarly, infants at 8 months of age were found to choose a failed helper (who intended but failed to help an agent achieve its goal) over a failed hinderer (who hindered the agent, but the agent’s goal was still attained), further demonstrating infants’ intentional reading skills in making social decisions (Hamlin, 2013a). Taken together, these findings suggest that infants take into consideration constraints on others’ behavior and failed attempts in interpreting goal-directed behavior. Thus, infants might also take into consideration caregiver constraints (e.g., limitations on behaviors available, failed attempts) when interpretating their caregiver’s attachment-relevant behavior.
To make sense of attachment-relevant experiences, infants must also consider contextual constraints, including infants’ context-specific needs that lead to possible changes in parent behavior. For example, although chest-to-chest soothing is considered an optimal way to calm and comfort an infant when the infant is hurt or scared, when the infant is simply bored, such a response is unnecessary. Instead, the parent’s attempts to engage with the infant when he/she is bored would be viewed as more appropriate within the given context (Ainsworth et al., 1971; 1974; NICHD Early Child Care Research Network; Woodhouse et al., 2020). There has been ample evidence from cognitive tasks showing that infants expect agents to adapt to contextual changes in pursuit of goals. For example, in experiments using computer-animated events (Csibra et al., 1999; Gergely et al., 1995), 9- and 12-month-old infants were habituated^3^ to an event in which two circles on the opposite sides of a tall barrier, one large and one small, first engaged in a turn-taking interaction involving expansion-contraction, and then the small circle jumped over the barrier to stop against the large circle so that the two again took turns expanding and contracting. In the VOE test events, the barrier was removed, and the small circle either jumped as before (old-path event) or went in a straight line (new-path event) to go to the large circle. Infants looked reliably longer at the old-path than at the new-path event, suggesting that they had interpreted the small circle’s actions during habituation as directed toward the goal of approaching the large circle. They expected the small circle to act efficiently in light of the change in the context (i.e., the removal of the barrier) and thus found the old-path event unexpected. These results have been extended to younger infants, even 3-month-olds, in experimental scenarios involving a human agent’s actions (e.g., a person moving around or reaching over a barrier to approach or retrieve an object behind the barrier is expected by infants to change her actions and directly approach the object once the barrier is removed; Brandone & Wellman, 2009; Kamewari et al., 2005; Liu et al., 2019). Further, in an eye-tracking study, infants at 9 months of age were found to expect a human agent to give a toy to an agent who could not reach for it because of a barrier blocking its way, but not to another agent who had relatively easy access to the toy, hence pointing to infants’ considerations of others’ situationinduced needs (Köster et al., 2016). Therefore, in attachment-relevant experiences, it seems plausible for infants to also expect parents to tailor their responses according to contextual constraints, including contextual changes (e.g., akin to the presence/absence of a barrier in the studies just mentioned) and context-specific infant states and needs (e.g., crying versus being bored).
Taken together, evidence from the infant cognition literature suggests that infants have the capacity to build attachment representations that likely contain the consistency (i.e., contingencies and regularities) of parent responses to infant attachment signals, intentional reading of parent behavior, and considerations of caregiver and contextual constraints on parent responses that contribute to expectations of parental behavior in attachment-relevant contexts (see Figure 1). These cognitive building blocks of attachment representations are likely informed by infants’ conceptual knowledge^4^, especially their psychological understandings of people’s behavior. That said, given the distance between the evidence reviewed and parent-child attachment relationships, such claims warrant further empirical attention, an issue we return to below.
In addition to representing the attachment figure’s responsiveness in times of need, attachment representations are hypothesized to involve views of others (e.g., views of the parent as available, responsive, and trustworthy; Bowlby, 1973). Specifically, based on a history of experiencing sensitive and responsive caregiving, secure individuals are thought to construct attachment representations in which others are viewed as loving, competent, and available. In contrast, based on a history of insensitive responding in which attachment signals were ignored or responded to inconsistently, insecure individuals are thought to construct attachment representations in which others are viewed as unloving, incompetent, and/or unavailable (Bowlby, 1973; Bretherton & Munholland, 2008). Infant attachment representations would therefore not only include enduring mental representations of context-specific patterns and interpretations of parent responses from a young age (e.g., a parent acts with an intent to soothe within given contexts) for which they derive expectations, but also differential evaluations of parents’ behavior (e.g., the parent’s intent is positive, or the parent is loving or trustworthy), which in turn inform future infant behavior.
How do infants form evaluations of their parents? Here we draw on the infant cognition literature to speculate on three, not necessarily mutually exclusive, possibilities. First, infants might simply view their parent positively if the parent responds or at least tries to respond in times of need but negatively otherwise. Support for this possibility comes from groundbreaking work on early sociomoral cognition which reveals that from as young as 3 months, infants possess social evaluation skills to apply meanings to others’ behavior (e.g., Hamlin & Wynn, 2011; Hamlin et al., 2010). They distinguish agents who behave positively (e.g., helping others achieve their goals) from those who behave negatively (e.g., hindering others from goal attainment; Hamlin et al., 2007; Hamlin et al., 2011; Kuhlmeier et al., 2003). For instance, after watching a live puppet show in which agent-1 tried to climb a hill and was helped by agent-2 (i.e., pushed up) or hindered by agent-3 (i.e., pushed down), infants chose the helper (agent-2) over the hinderer (agent-3) when presented with the two agents (Hamlin et al., 2007), perhaps guided by their positive evaluation of the helper (note, however, that this is an active area of research with ongoing debates, see e.g., Benton & Lapan, 2022; Powell & Spelke, 2018; Salvadori et al., 2015; Scarf et al., 2012; Schlingloff et al., 2020 for alternative explanations or mixed replication attempts; see Hamlin, 2023; Hamlin et al., 2012; Margoni & Surian, 2018; Woo et al., 2022 for rebuttals, and Lucca et al., 2021 for a collaborative replication study).
The second possibility is that after repeated attachment-relevant experiences with their parent, infants might gradually come to internalize the parent’s behavior as the social norm, even when the parent is usually unresponsive, which leads infants to expect others to act in accordance with this norm. Support for this possibility comes from studies using the lightbox task mentioned above (e.g., Gergely et al., 2002; Meltzoff, 1988). In this task, during demonstration, an agent uses her head to illuminate the box with her hands available. Infants at 12–18 months of age are found to imitate her head touch actions when they are presented with the lightbox (e.g., Altinok et al., 2022; Buttelmann et al., 2013; Chen & Waxman, 2013; Király et al., 2013; Paulus et al., 2011; Pinkham & Jaswal, 2011; Poulin-Dubois et al., 2011; Zmyj et al., 2009; Zmyj et al., 2010). One explanation is that the agent’s demonstrations likely convey the message to infants that “this is how we typically act on this object,” leading to infants’ decisions to imitate her (e.g., Gergely & Csibra, 2006). Hence, these findings reveal infants’ willingness to accept as norms others’ behavior, even odd behavior, and use it to inform their own actions on novel objects. In a similar vein, infants might also accept as norms their parent’s behavior (e.g., this is how a parent behaves) and use it to inform their interactions with the parent and their expectations of others’ behavior (e.g., my parent is responsive, thus others’ parent will be responsive too).
A third possibility is that infants might hold initial expectations of their parent based on their conceptual knowledge in the sociomoral domain and evaluate their parents’ behavior accordingly. Within the infant cognition literature, the nature of such sociomoral knowledge remains a matter of lively debate (e.g., Hamlin, 2023; Spelke, 2022; Woo et al., 2022). Nevertheless, according to a principle-based account, infants expect agents to act toward one another possibly based on four principles that stem from an initial sketch of human morality (Baillargeon et al., 2015; Buyukozer Dawkins et al., 2020; Graham et al., 2013; Ting et al., 2020). These principles fairness (e.g., Bian et al., 2018; Buyukozer Dawkins et al., 2019; Meristo et al., 2016; Sloan et al., 2012; Ziv & Sommerville, 2017); harm avoidance (i.e., individuals within a moral circle, e.g., human agents, but not those of different circles, should avoid causing significant harm to others; Hamlin, 2015; Kanakogi et al., 2017; Ting & Baillargeon, 2021; Ting et al., 2019); ingroup support (i.e., individuals within one social group, e.g., those who speak the same language, should help and support one another and show preference for and loyalty to ingroup members; Bian & Baillargeon, 2022; Buttelmann et al., 2013; Howard et al., 2015; Jin & Baillargeon, 2017; Kinzler et al., 2007; Powell & Spelke, 2013; Spokes & Spelke, 2017); and authority (relevant to the dominance hierarchy within a group and the ensuing obligations of leaders and followers; e.g., Gazes et al., 2017; Margoni et al., 2018; Mascaro & Csibra, 2012; Pun et al., 2016; Stavans & Baillargeon, 2019; Thomsen et al., 2011). It has also been reported that infants’ preferences and choices are not only directed toward ingroup members and helpful characters (e.g., Hamlin et al., 2007; Kinzler et al., 2007) but also fair agents (e.g., an agent who distributes resources equally between two identical recipients; Burns & Sommerville, 2014; Geraci & Surian, 2011; Lucca et al., 2018). In short, infants appear to gravitate toward those who behave in accordance with moral principles and avoid those who do not (Ting et al., 2020).
Of these moral principles, we focus on the principles of harm avoidance and ingroup support that lead to infant expectations of helping (vs. ignoring or harming) and comforting (vs. ignoring) others because these interpersonal expectations are particularly relevant to attachment representations (e.g., baby cries or falls, attachment figure provides comfort and/or aid; Bretherton, 1999; Posada & Waters, 2018; Waters & Waters, 2006). With regards to helping expectations, the available evidence suggests that infants in the second year of life expect ingroup members to help one another (Ting et al., 2020). For instance, when presented with three human agents, infants seem to expect Agent1 to help Agent2, and not to ignore her, if they comprise a group marked with minimal cues (i.e., they each say, “I am a bem!” and “I am a bem too!”)^5^ separate from Agent3 (i.e., she says, “I am a tig!”). When group memberships are unclear (Agents1 and 2 each say, “I saw a bem!” and “I saw a bem too!”), infants no longer expect Agent1 to help Agent2 (Jin & Baillargeon, 2017). When Agents 1 and 2 belong to different groups, infants even tolerate Agent1 dropping a toy that Agent2 needs on the floor, presumably because it is considered a minor hinderance, as opposed to significant harm (Ting et al., 2019). Hence, if infants in the first year of life were also found to expect ingroup members as morally obliged to help and support (vs. ignore) one another, and/or people not to cause significant harm to each other (because supposedly all humans belong to one moral circle), then similar expectations might hold true within parent-infant attachment relationships.
Although evidence for helping expectations in the first few months of life is currently limited (although see Putrich et al., 2023), Jin and colleagues (2018) reported pivotal results on early comforting expectations. They found that at 4, 8, and 12 months of age, infants expect an adult woman in videotaped events to comfort, as opposed to ignore, a crying baby. Thus, in a situation in which an adult is alone with a baby, even without any information given about the relationship between the two, the adult is expected to comfort a baby who signals distress. Two recent studies (Thomas, Saxe, et al., 2022; Thomas, Woo, et al., 2022) examined possible cues to a dyad’s relationship and yielded converging evidence (by measuring which of two targets infants looked to and longer at, as if expecting them to respond). In one study (Thomas, Woo, et al., 2022), an adult woman and a puppet on video engaged in saliva-sharing (e.g., food-sharing or the woman putting a finger from her mouth to the puppet’s mouth), which possibly denotes close relationships. Infants aged 8.5 to 10 months seemed to expect the dyad to respond to each other if one of them was in distress. Infants at 12 months of age were also found to expect a puppet who called out their own name (thus demonstrating some familiarity with their family) to respond when their own parent was shown to be in distress on video (Thomas, Saxe, et al., 2022). These expectations about comforting (vs. ignoring) a crying baby and about who should respond to someone in distress can be interpreted as stemming from the moral principles of harm avoidance (that is, if letting someone cry is considered significant harm) and/or ingroup support (Jin et al., 2018). In light of such evidence, it is possible that sociomoral principles might form the basis from which infants evaluate their parents’ responses to their attachment signals (e.g., crying). For example, in the case of a parent responding in unhelpful, unsupportive, or even abusive ways to their infant’s attachment signals, the infant is confronted with repeated violations of early-emerging moral expectations, which might lead them to view their parent as failing to fulfill their moral obligations (Spokes & Spelke, 2017) or even lacking in moral character (Ting & Baillargeon, 2021). If so, such representations might help to explain why early adversity within infants’ early attachment relationships with parents poses enduring risk for developmental maladjustment (Jin et al., 2018).
In this section, we have drawn on research in the sociomoral domain of cognition. Importantly, this is a relatively new area of research in the field of cognitive development, and what is known about infant cognition in this domain is constantly evolving and debated. Thus, although speculative, each of the frameworks reviewed above offers intriguing possibilities of how infants might evaluate their parent’s behavior within attachment relationships. Specifically, infants might use their knowledge about the social world to evaluate their parent’s behavior such that they view their parent in a positive or negative light, as behaving in a normalized way, and/or as acting consistently or inconsistently with the moral principles of being helpful and supportive to others. These attributive evaluations of their parents might also become part of infant attachment representations, in conjunction with how infants track and make sense of the consistency and sensitivity of their parent’s responses within given contexts (see Figure 1). Importantly, although each of these possible evaluative frameworks is consistent with attachment theory, each holds different implications for understanding what knowledge is derived from parent-child attachment experiences, becomes mentally represented, and is carried forward overtime. Indeed, evaluations derived from attachment experiences that inform views of others as likely to violate moral principles versus views of others as potentially negative interaction partners, might have differing implications for children’s (mal)adjustment. Thus, research not only on the content of attachment representations but also on the evaluations of such content is critical to understanding the mechanistic role of IWMs in the legacy of early attachments.
Bowlby theorized that IWMs serve as the key explanatory mechanism by which attachment experiences are carried forward across the life course. Yet, there remains no unifying theory of the development of IWMs. To be sure, there is a wealth of research on IWMs in the years following language acquisition. However, a lifespan developmental theory of IWMs requires an understanding of the origins of attachment representations. As such, we have reviewed the significant theoretical, empirical, and methodological advances that have contributed to our understanding of the rich mental world of infants, and the relevance of such work for informing the nature, content, and construction of attachment representations in infancy. In addition to extending research on IWMs down to infants, there is need to consider how representations constructed in infancy develop beyond the first few years of life. In this section, we draw on theory and research on cognitive development in the early life course to propose potential ways in which attachment representations develop beyond the infancy period, with a particular focus on connecting research on attachment representations prior to and following language acquisition.
In describing the development of IWMs in the early life course, Bowlby (1969/1982) proposed “Starting, we may suppose, towards the end of his first year, and probably especially actively during his second and third when he acquires the powerful and extraordinary gift of language, a child is busy constructing working models of how the physical world may be expected to behave, how his mother and other significant persons may be expected to behave, how he himself may be expected to behave, and how each interacts with all the others.”(p. 354) Extending Bowlby’s theorizing regarding the early construction of IWMs, we have drawn on infant cognition research to propose that infants begin to construct attachment representations earlier in the first year of life than Bowlby originally proposed. Beyond infancy, however, Bowlby emphasized that the construction of IWMs is intimately tied to language acquisition. As children begin to acquire language and in the years beyond, two representational constructs that have received a great deal of attention from attachment scholars and that will be the focus here include event representations and autobiographical memories (Hesse, 2016; Main et al., 1985; Waters & Roisman, 2019; Waters & Waters, 2006; see Figure 1), representational constructs that are informed in part by communication processes, such as conversations between parents and children about past events (Nelson & Fivush, 2004; Oppenheim & Koren-Karie, 2021; Posada & Waters, 2018).
Bretherton (1985; 1990) drew on cognitive research regarding event representations to propose that event schemas (Mandler, 1979; 1983) or scripts (Schank & Abelson, 1977) might serve as the building blocks of IWMs. Event schemas are dynamic generalized event representations constructed from repeated similar experiences that summarize commonalities (e.g., main character[s], causal chain[s] of events, resolution[s]) across a class of events (e.g., going to a park, falling down, and getting hurt) and provide a script for how events typically unfold (Nelson, 1986; Schank, 1982). Applying this work to attachment, Bretherton (1985) proposed that “…event schemas developed in interaction with specific persons are also the raw material from which young children construct internal (affective/cognitive) working models of the self and of significant others, including attachment figures” (p. 32). Waters and colleagues (Waters & Rodriques-Doolabh, 2001; Waters & Waters, 2006) extended these ideas by proposing that attachment-relevant experiences of secure base support in childhood are represented as secure base scripts. Specifically, a history of consistent secure base support in which attachment needs are recognized and effectively responded to inform a well-developed secure base script characterized by understanding that attachment figures may be relied on in times of need and will provide competent support to navigate problems. The Attachment Script Assessment (Waters & Rodrigues-Doolabh, 2004) was developed to assess secure base scripts, and research indicates that children as young as 36 months show evidence of script-like representations of attachment (Posada et al., 2018).
With further development, such representations become autobiographical in nature, and by adulthood, individuals construct elaborate autobiographical narratives that contain specific memories of salient past events and their impact on the individual and their relationships. Waters and colleagues (2017) noted that autobiographical histories share many similarities with the narrative and reflective content tapped by the Adult Attachment Interview (AAI; Main et al., 1985; Main et al., 2003–2008)—the most well-researched narrative measure of adult attachment representations. In the AAI, individuals are tasked with producing autobiographical narratives about their attachment-relevant experiences, which are then evaluated for coherence, or the capacity to develop internally consistent narratives about attachment experiences without becoming emotionally overwhelmed (Main et al., 2003–2008). Importantly, aligning with the development of event and autobiographical representations, evidence suggests that the construction of a secure base script informs the development of coherent autobiographical attachment representations (Waters et al., 2017).
According to theorizing regarding the development of event and autobiographical representations, these representational constructs have their origins in infancy (Nelson & Fivush, 2004). For example, research suggests that infants demonstrate the presence of non-verbal scripts, indicating that they can encode sequences of events and that such representations are maintained overtime (e.g., Bauer & Thal, 1990; Bauer & Wewerka, 1995; Lucariello et al., 1986; Nelson, 1986). With development, new functions and capacities emerge, including language, that build on prior representational capacities to bring about more complex and advanced representations, including event representations and autobiographical memory. Applying this work to attachment, it is possible that the development of IWMs begins in infancy with the construction of attachment cognitions (including attachment-relevant expectations and evaluations). In connecting these early-emerging attachment cognitions to subsequent representational constructs, as discussed above, a key feature of the attachment cognitions proposed here is contingency, and thus, early emerging attachment representations are likely characterized by a temporal-causal structure (e.g., If I cry, my parent responds), which is an important feature of later developing event and autobiographical representations (Nelson & Fivush, 2004). With development, children may begin to construct more complex attachment-relevant scripts. Initially, such scripts are likely non-verbal. Indeed, extending Main and colleagues’ (1985) theorizing that infant behavior in the SSP reflects the presence of IWMs, the organization of infant behavior around the attachment figure in the SSP might be interpreted as reflecting the presence of a non-verbal attachment script that contains information about the temporal ordering of how the infant behaves (e.g., whether and how to signal attachment needs) and how the parent behaves (e.g., whether and how responsive they will be) in response to an attachment-relevant problem (e.g., separation from parent). With the acquisition of language, attachment-relevant scripts might become verbal and informed by not only attachment-relevant experiences, but also conversations with parents (Lu et al., 2018; Posada et al., 2018; Posada & Waters, 2018). As children’s representational capacities further develop, individuals may come to construct attachment representations that are autobiographical in nature (Waters et al., 2017; see Figure 1 for conceptual model).
Importantly, whether this model of the development of attachment representations is best characterized by conceptual, qualitative change or elaborative, continuous change remains an open question. In fact, just how the early cognitive representations investigated in infancy connect to the later emerging, often verbal, representations investigated following language acquisition is a topic of substantial debate in the cognitive development literature. Focusing first on the view of developmental discontinuity, according to proponents of this perspective, learning and experiences, especially language and symbolic learning, bring about qualitative changes in different domains of knowledge so that initial infant knowledge is considered “proto-conceptual” but develops into genuine concepts over time (e.g., Carey, 2009; Xu, 2019). An example of this view from the psychological domain is the two-system account (Apperly & Butterfill, 2009; Butterfill & Apperly, 2013; Low et al., 2016; Low & Watts, 2013) proposed to explain the discrepancy found between infants’ and children’s performances in false-belief tasks. The understanding of false beliefs (i.e., to realize that mental states are representations, not copies, of reality, and hence can be false) is usually examined in a variety of traditional tasks (e.g., the Sally-Anne task; Wellman et al., 2001; Wimmer & Perner, 1983). Children younger than 4 years of age typically fail this task, which has been interpreted to suggest that they do not yet understand that individuals can hold false beliefs. In the past two decades, however, nontraditional tasks, including VOE and PL tasks, have been used to examine early false-belief understanding and yield positive results from 6-month-old infants to 3-year-old children (e.g., Onishi & Baillargeon, 2005; for reviews, see Scott & Baillargeon, 2017; Scott et al., 2022; for a recent debate on replications of these reports, see e.g., Baillargeon et al., 2018; Powell et al., 2018). To explain this discrepancy in findings between the traditional and nontraditional tasks, the two-system account suggests that an early cognitive system affords belief-like or minimal false-belief understandings to allow infants’ and young children’s success in nontraditional tasks, while a different system, coupled with more advanced language skills, emerges around age 4 and enables children’s success in traditional false-belief tasks. As such, the two-system account illustrates how conceptual understandings can be distinct during and after infancy, how the two systems may exist in parallel in adulthood, and how they may become engaged under different cognitive loads (e.g., the early system is thought to be more efficient and faster than the later system).
Turning next to developmental continuity, this perspective is best captured by core knowledge theory, which suggests that core understandings (e.g., the notion of object permanence), while emerging early and becoming enriched even during infancy to allow more and more accurate predictions of physical event outcomes (Baillargeon, 2008), should remain invariant through adulthood (Spelke, 2022). Citing false-belief understandings as an example, proponents of developmental continuity have cogently argued against the two-system account and instead suggested that early belief understandings are qualitatively similar to later understandings, and that the traditional and the nontraditional false-belief tasks place different demands on children’s cognitive processing resources, leading to the patterns of successes and failures at different ages (e.g., Scott et al., 2022). In support of this view, when the task demands are alleviated, even 2.5-year-olds are found to succeed in modified versions of traditional false belief tasks (Setoh et al., 2016; for a relevant debate, see Rubio-Fernández et al., 2017; Scott et al., 2017; for a replication, see Grosso et al., 2019). Further, neuroimaging studies provide support for developmental continuity in infants’ and adults’ false-belief understandings and little evidence of a discontinuity as children begin to pass the traditional false-belief tasks (e.g., Bardi et al., 2017; Hyde et al., 2015; Hyde et al., 2018; Richardson et al., 2018).
Taken together, how the development of attachment representations is best conceptualized remains an open question. Aligning with the view of developmental discontinuity, according to Nelson and Fivush’s (2004) developmental account of autobiographical memory discussed above, representational constructs, such as autobiographical memory, are thought to gradually emerge as a new level of complexity through the interaction of structures existing at simpler levels. Thus, one possibility is that with development, infant attachment cognitions in combination with other conceptual and linguistic advances bring about a new level of complexity in the nature of attachment representations, such that the early and the late attachment representations coexist and are distinct but still interrelated constructs. Alternatively, across domains of cognition, there is increasing evidence supporting the view of developmental continuity. For example, in addition to the evidence reviewed above from research on false belief understanding, in the domain of numerical cognition, research indicates that core representational systems for objects and approximate magnitudes that exist in infancy become elaborated over time and with experience, and that early representational systems are ultimately integrated with other cognitive networks once children acquire culturally relevant skills (e.g., counting procedure) or symbolic systems (e.g., verbal count list, visual numeral form, such as the Arabic numerals), as reflected by neural connections, to enable exact and precise representations of numbers (e.g., Dehaene, 1992; Hyde, 2011; 2021; Mou et al., 2023). Thus, alternatively, the development of attachment representations might be best conceptualized as a process of elaboration and integration, such that the later developing attachment representations are overlaid on prior representational constructs.
Regardless of which model best characterizes change in attachment representations, as emphasized throughout this article, only by investigating attachment representations from their inception can a developmental account of IWMs be achieved. Toward that end, below we discuss initial pioneering research examining attachment cognitions in infancy and discuss future directions necessary to advance understanding of attachment representations in infancy and beyond.
To date, a limited number of reports have probed attachment cognitions in infancy, and this work has mainly focused on the potential content of infant attachment representations^6^. Specifically, research has examined infants’ representations about adults’ responsiveness to infant attachment signals. In a pioneering investigation by Johnson and colleagues (2007; 2010), they examined infants’ expectations in a separation scenario in which an “adult” (depicted by a large oval) left a “baby” (depicted by a small oval), who then started crying. After being habituated to this separation event, infants then saw the “adult” return to the crying “baby,” or ignore the “baby” by continuing to move away in alternating test trials. The results suggest that by 12–16 months, the infants did not show an expectation for the adult to respond by returning to the crying baby. However, distinct patterns emerged when infants were grouped by their attachment whereas secure infants (n = 18) looked reliably longer when the adult failed to do so, reflective of an expectation that the adult would respond to the crying baby, insecure infants (n = 15) did not, suggesting they did not hold such an expectation. This work thus provides the first evidence for a link between infant attachment quality and infants’ mental representations about how adults are expected to behave in response to infant distress. Although ground-breaking, it is important to note that this initial report included a small sample of infants (N = 33) and has not been directly replicated.
Extending Johnson et al.’s (2007; 2010) research, Biro and colleagues (2015) created animated scenarios (each lasting 20 s), which depicted a separation between an “adult” and an “baby” and the “adult” being responsive or unresponsive to the “baby,” to examine how 12-month-old infants (N = 60) allocated their visual attention during these animations via eye tracking^7^. Infants were found to perform differently during the response part of the animations as a function of maternal Infants with more (v. less) sensitive mothers looked significantly longer at the “adult” than at the “baby,” regardless of whether the “adult” was responsive. Biro and colleagues (2015) interpreted these results as partially supportive of the findings of Johnson and colleagues (2007; 2010), by suggesting that infants’ history of maternal sensitivity led them to seek out attachment figures in times of need. It is worth noting, however, that they did not find an anticipated association between infant attachment security and overall looking time to the fixed-length animations. This null effect of infant attachment security was also reported in a subsequent investigations from Biro and colleagues (2017) using similar animations, thus giving rise to the possibility that the procedural changes from Johnson et al. (2007; 2010) might not be sufficiently inducive to infants’ differential looking times to a separation scenario based on attachment security.
Jin and colleagues (2018) have also examined infant expectations of adult responding to infant distress in a comforting scenario. As reviewed above, findings revealed that by 4 months, infants (N = 18) expect an unfamiliar adult to comfort, and not ignore, a crying baby. This expectation was also demonstrated among different samples of 8- and 12-month-old infants. Importantly, such expectations were specific to infant crying—a potent infant attachment distress cue (Bowlby, 1969/1982), as findings indicated that infants did not exhibit a similar expectation of adult responsiveness to infant laughter—a positively-valanced, non-threating signal. Given the consistency of findings across the first year of life, it might be concluded that expectations of adult responsiveness to distress is normative and relatively unchanged over the first year of life. However, examination of the effect sizes reveals that the magnitude of this effect at 12 months (d = .39) is smaller than that at 4 months (d = .78). Such evidence might alternatively suggest that infant expectations of responsiveness to distress, in fact, diminish with age, in that infants might have an early emerging expectation of adult responsiveness to infant attachment-relevant distress cues that overtime is shaped by infant attachment experiences, such that secure, but not insecure, infants continue to hold such expectations, as reported by Johnson and colleagues (2010). Importantly, however, drawing such conclusions is premature because the changes in the magnitude of the effect sizes might alternatively be explained by task and sample differences in the three age groups examined. Moreover, Jin and colleagues did not include assessments of infant attachment quality or caregiving experience. Thus, it remains unknown whether infant comforting expectations are related to infant attachment.
Together, this initial research (Biro et al., 2015; Jin et al., 2018; Johnson et al., 2010) has demonstrated the feasibility of investigating attachment cognitions from as early as 4 months and yielded intriguing evidence linking infant attachment representations to individual differences in attachment security and caregiving experiences following infants’ first birthday. Although pioneering, these initial research efforts employ different attachment scenarios (separation, comforting) and different looking-time measures, and thus raise more questions than they do provide answers. Below, we suggest important directions for future research that can help provide a developmental account of attachment representations in infancy.
As a first step, future research must consistently employ attachment-relevant assessments, which requires greater collaboration among infant cognition and attachment scholars. This need is especially evident from the review of extant evidence above. Specifically, unlike in the initial work by Johnson and colleagues (2007; 2010), Jin and colleagues (2018) did not include assessment of infant attachment quality. Given that the comforting scenario involves adult responsiveness to an infant attachment distress cue, it is possible that the paradigm taps infant attachment cognitions. However, it is also possible that the paradigm taps infants’ principle-based moral understanding of harm avoidance and/or ingroup support because, for example, ignoring a crying baby might be construed by infants as causing harm and/or failing to provide support to an ingroup member. Indeed, even at the 12-month assessment—the age at which infant attachment quality was demonstrated by Johnson and colleagues (2007; 2010) to modulate attachment cognitions in a separation scenario—a large percentage of infants (73%) in Jin et al. (2018) demonstrated the expectation that the adult would respond to the distressed infant in the comforting scenario, which is well above the representation of secure infants in the population (~52%; Madigan et al., 2023). Thus, it remains unknown whether comforting expectations stem from infants’ understandings of moral principles of harm avoidance and/or ingroup support, or from infants’ own attachment history and hence specific to their attachment expectations. These alternatives are important to consider because they have direct implications for our understanding of the content and development of IWMs in infancy. For example, if attachment history is tied to infant expectations of responsiveness to distress in separation, but not comforting, scenarios, it would suggest that certain elements of infant attachment cognitions are constructed independently of infant moral understandings, and that infant moral understandings are relatively untouched by early attachment-relevant experiences. However, this possibility seems unlikely given meta-analytic evidence demonstrating the significance of early attachment quality for children’s morally relevant behaviors (e.g., prosociality; Deneault, Hammond, et al., 2023). Alternatively, if infant attachment history is tied to infant expectations of responsiveness to distress in both the separation and comforting scenarios, this might suggest a connection between infant attachment and moral cognitions in line with the above interpretation of extant evidence that perhaps all infants initially hold the expectation of adult responsiveness to infant distress, as part of the first sketch of human morality, that overtime becomes modified by attachment experiences.
In addition to determining cognitive components included in infant attachment representations, the bounds of attachment in predicting variation in infant cognitions must also be tested to examine the specificity of associations between infant cognition and attachment, an issue regularly considered in the literature on the predictive significance of attachment (e.g., Belsky & Cassidy, 1995; Sroufe, 2016; Thompson, 1997; 2016). As an example, the core-knowledge theory in the cognitive literature suggests that as infants’ knowledge in a core domain becomes enriched over time through learning, the initial understanding likely stays relatively untouched by experiences because such core understanding is essential to our conceptual understanding of structure of the world (e.g., Baillargeon, 2008; Baillargeon & Carey, 2012; Carey, 2009; Gelman, 2002; Keil, 1995; Leslie, 1994; Spelke, 1994; Wellman & Gelman, 1997). In other words, tasks tapping infant core knowledge should yield similar results across ages, regardless of individual experiences. An impressive body of research identifies domain-specific core knowledge held by infants at different ages, including object continuity (i.e., objects move continuously and retain their properties) in the physical domain and number discrimination in the numerical domain (e.g., Aguiar & Baillargeon, 2003; Feigenson et al., 2002; Feigenson et al., 2004; Hespos & Baillargeon, 2001a; 2001b; Izard et al., 2009; Libertus & Brannon, 2010; Luo & Baillargeon, 2005; Mou & Luo, 2017; Spelke et al., 1992; Wang & Baillargeon, 2005; 2006; Xu & Spelke, 2000). In the psychological and sociomoral domains, infants’ ability to attribute preferences to agents based on consistent choices and their preferences toward prosocial others or ingroup members, respectively, have also been reported throughout infancy (e.g., Bian & Baillargeon, 2022; Hamlin et al., 2007; Woodward, 1998; for reviews, see Choi & Luo, 2023; Scott et al., 2022; Woo et al., 2022). Thus, inclusion of control assessments of infant cognitions in these core domains will allow for empirical tests of the specificity of attachment-cognition associations by examining whether infant attachment is exclusively associated with variation in attachment cognitions.
Including control assessments would also prove useful in ruling out alternative interpretations of extant findings. Specifically, although findings from Johnson and colleagues (2007; 2010) have been interpreted to suggest that infants with an insecure attachment fail to demonstrate an expectation of responsiveness to infant distress in a separation scenario, such findings might be accounted for by alternative interpretations. For example, it has been suggested that insensitive caregiving predictive of insecure attachment might contribute to infant unease in a testing environment, resulting in inconsistent performance in some cognitive tasks (Bornstein et al., 2011; Bornstein et al., 2012; Hohenberger et al., 2012; Jin et al., 2018). For instance, Hohenberger et al. (2012) reported that 10-month-olds with more sensitive mothers detected violations in a physical reasoning task, based on Kotovsky & Baillargeon (1994; 1998), but not those with less sensitive mothers. In line with this evidence, Bornstein and colleagues (2011; 2012) also reported different results in facial expression discrimination and object recognition from infants with depressed versus non-depressed mothers. To rule out the possibility that insecure infants’ performance is due to unease in the testing environment, future research should include cognitive control tasks to determine whether insecure infants fail to succeed in cognitive tasks, more generally, or attachment cognitive tasks, specifically. Such evidence would ultimately provide stronger support for the interpretation that infant performance in attachment cognition tasks reflects their representations of attachment.
Throughout this article, we have drawn on the infant cognition literature to propose that the construction of attachment representations begins early, perhaps during the first months of life. However, the earliest age at which valid measures of infant attachment quality can be administered is 12 months. Thus, there is need to consider other assessments that allow for an evaluation of attachment-relevant experiences prior to infants’ first birthday. Given sensitive caregiving (i.e., prompt, appropriate responding to infant signals; Ainsworth et al., 1978) is the aspect of the caregiving environment theorized to determine infant attachment quality (Ainsworth et al., 1978) and evidence from experimental and meta-analytic research documenting a robust link between sensitive caregiving and infant attachment security (Bakermans-Kranenburg et al., 2003; De Wolff & van IJzendoorn, 1997; Madigan et al., 2024), employing assessments of parental sensitivity presents one solution. In contrast to infant attachment assessments, caregiving quality can be assessed from birth. Thus, employing caregiving assessments would facilitate the investigation of infant attachment cognitions as a function of attachment-relevant experiences in the first year of life.
Importantly, however, there remains an environmental origins gap in our understanding of the determinants of infant attachment quality (Groh et al., in press). Specifically, despite evidence from genetically informed research suggesting that individual differences in infant attachment quality are largely explained by variation in the environment (shared and non-shared; Bakermans-Kranenburg et al., 2004; Bokhorst et al., 2003; Fearon et al., 2006; Luijk et al., 2011; Roisman et al., 2013; Roisman & Fraley, 2008), the specific aspects of the caregiving environment that contribute to infant attachment quality remain underspecified. Indeed, the magnitude of the association between parental sensitivity and infant attachment security is modest (r = .25; Madigan et al., 2024). One way to address this issue is to employ longitudinal research designs (see below) that follow infants over the first year of life to allow for inclusion of assessments of sensitivity and infant attachment quality to determine whether they yield converging evidence. Research might also include other caregiving behaviors that have been tied to infant attachment quality to better reflect infants’ early attachment-relevant caregiving experiences (e.g., autonomy support, intrusiveness, detachment, secure base provision, and anomalous behaviors; Cassidy & Berlin, 1994; Isabell & Belsky, 1991; Madigan et al., 2006; Swanson et al., 2000; Whipple et al., 2011). For example, Whipple and colleagues (2011) have demonstrated that inclusion of multiple caregiving behaviors (e.g., sensitivity and autonomy support) explains greater variance in infant attachment security beyond the variance explained by sensitivity alone. Similarly, Woodhouse and colleagues (2020) found that observations of mothers’ secure base provision (i.e., degree to which a caregiver is able to meet an infant’s needs on both sides of the attachment-exploration continuum [e.g., fully soothing a crying infant chest-to-chest and allowing the exploring infant to experience calm, regulated states]) was strongly associated with infant attachment security (r = .41) over and above sensitivity.
Additionally, the extant evidence on attachment cognitions has been cross-sectional in nature. Thus, the precise timing of when infants begin to construct attachment representations and how such representations develop over the first years of life remains unknown. Longitudinal investigations employing multiple assessments of attachment cognitions and caregiving experiences that follow infants throughout the first years of life are needed to examine the construction of attachment cognitions over time as a function of attachment-relevant experiences. This approach also provides a more rigorous test of the expected developmental patterning of associations, by allowing for an investigation of whether infants’ attachment-relevant caregiving experiences contribute to subsequent attachment cognitions, as would be expected. In addition, investigations should include assessments of infant attachment quality to examine whether the culmination of the process of constructing attachment cognitions by the end of the first year informs the quality of attachment relationship infants form with their parent(s).
Importantly, as noted above, initial studies examining attachment-relevant cognitions comprised small samples. Future research requires larger samples than those employed in previous studies (e.g., N = 30–35 in each one of three experiments of Johnson et al., 2010; N = 60 in Biro et al., 2015). Although such sample sizes produce sufficient power to examine normative trends typically investigated in infant cognition research, they are underpowered for research on individual differences generally, and attachment specifically, that typically produce smaller effect sizes (e.g., r = .20; Schuengel et al., 2021), posing threat to reproducibility. Adequately powered samples are also critical if researchers are interested in specific patterns of insecure attachment that have low base rates (e.g., resistant attachment; Madigan et al., 2023; Mesman et al., 2016; Van IJzendoorn & Kroonenberg, 1988; Van IJzendoorn & Sagi-Schwartz, 2008). Relevant to this issue, Johnson et al. (2010) reported promising trends regarding looking-time differences between insecure-avoidant and insecure-resistant infants. While insecure-avoidant infants (n = 8) tended to look longer at the responsive event (in which the “adult” returned to the crying “baby”) than at the unresponsive event (in which the “adult” ignored the “baby”), insecure-resistant infants (n = 7) did not. Larger, sufficiently powered samples can therefore help determine whether insecure-avoidant infants, in contrast to secure infants, indeed expect adults to ignore crying babies, consistent with their own caregiving history of being ignored or rejected, whereas insecure-resistant infants hold no expectations, consistent with a history in which their attachment signals are responded to inconsistently (Benoit, 2004; Cassidy, 1994). Further, such large samples will allow researchers to examine the attachment-relevant expectations of disorganized infants, a pattern of attachment rarely probed in the few studies conducted to date.
In the initial research, there has also been a confounding of attachment scenarios (comforting or separation) with the nature of video stimuli employed, with some work employing unfamiliar people to infants (Jin et al., 2018) and other work employing computer-animated geometric shapes (Biro et al., 2015; Johnson et al., 2007; Johnson et al., 2010). It is worth noting that the use of computer-animated events involving geometric shapes to simulate social interactions is a common approach in many infant cognition studies (e.g., Csibra et al., 1999; Gergely et al., 1995; Kuhlmeier et al., 2003; Liu et al., 2017; Premack & Premack, 1997; Surian & Geraci, 2012), and recent analyses on the use of simplified or anthropomorphic stimuli including animated shapes and puppets in cognitive research suggest that infants and young children tend to conceptualize them in similar ways as real people (Kominsky et al., 2022; Yu & Wellman, 2022). Moreover, the use of animated shapes might be viewed as a strength. Specifically, instead of examining infants’ relationship-specific expectations (e.g., about their own parents, which would entail using stimuli representing or at least resembling these parents, including matching hair color and skin color), the use of geometric shapes points to the abstraction of infant relationship-specific attachment experiences to generalized expectations regarding attachment-relevant scenarios. Nevertheless, it is important for future research to use realistic and simplified stimuli including real people, puppets, and animated shapes in both comforting and separation scenarios to demonstrate the robustness of the findings.
Finally, moving beyond the questions raised by the initial research on infant attachment cognitions (Biro et al., 2015; Jin et al., 2018; Johnson et al., 2010), we propose a few future directions to further probe the cognitive underpinnings of infant attachment. For instance, in addition to considering the overall level of parental responsiveness, future research should consider the consistency of care the parent provides the infant. As discussed earlier, infants use statistical patterns in accumulated experiences to draw inductive inferences (e.g., Aslin et al., 1998; Gopnik et al., 2004; Perfors et al., 2011; Saffran et al., 1996). Given that it is impossible for parents to be 100% responsive (or unresponsive), an infant might statistically track the responsiveness of a specific parent to inform attachment representations over time. To keep matters simple, we focus on ratios (or probabilities) here. A parent might respond sensitively most of the time (e.g., 75%), leading the infant to interpret the parent’s behavior as consistent and reliable (e.g., My parent responds to my attachment signals), and expect similar reliable responding from that parent in the future (e.g., If I signal an attachment need, my parent will be responsive). This statistical inference might in turn contribute to the formation of a secure attachment representation. By contrast, if a parent responds inconsistently, the accumulation of such experiences (e.g., 25% of responsivity) might lead to the formation of insecure attachment representations. Importantly, this process might also be informed by infants’ own characteristics (e.g., some infants are more tolerant of inconsistencies than others, possibly due to temperamental differences, for example, by making them differentially susceptible to caregiving experiences; Belsky, 1997; Ellis & Boyce, 2008; Van IJzendoorn & Bakermans-Kranenburg, 2012). In cognitive work, researchers have started to examine the role of various ratios of consistent/inconsistent occurrences (e.g., 3 to 1) in infants’ conceptual understandings (e.g., Duh & Wang, 2019; Luo et al., 2017; Steckler et al., 2017). Thus, investigations on parent responsiveness ratios, coupled with infants’ own caregiving history, would yield an enriched understanding of the extent to which attachment representations are robust to changes in caregiving and determine the amount of change necessary to update attachment representations in the early life course. Such research has the potential to not only advance our basic science understanding of IWMs, but also to inform intervention efforts aimed at improving the quality of parent-child attachment (Berlin et al., 2008; Dozier & Bernard, 2017; Juffer et al., 2017; Woodhouse et al., 2018), which in turn can have positive effects on later social, emotional, and cognitive outcomes.
To deepen our understanding of infant attachment representations, future research must also extend beyond the previous focus on infant expectations regarding adult responsiveness to distress. There are a number of ways in which such work might be extended, and here we highlight two intriguing possibilities. First, preferences for social partners are another key component of IWMs that has been linked with attachment variation in adulthood, with less known in infancy (Dykas & Cassidy, 2011; Frazier et al., 1996; Latty-Mann & Davis, 1996). Thus, future work might enrich understanding of representations of others in attachment relationships by extending to a focus on infant attachment-relevant preferences, which might in turn shed light on infants’ evaluations of attachment experiences, as previously described. In cognitive work, infant preferences can be demonstrated by measuring their manual choices in action tasks or preferential looking in PL tasks. As reviewed above, infant preferences for prosocial individuals and those behaving consistently with moral principles have been shown through their choices between different characters, e.g., choosing a helper over a hinderer in Hamlin et al. (2007) (for reviews, see Baillargeon et al., 2015; Ting et al., 2020), which are interpreted as evidence for their social evaluation skills. Thus, infant choices between individuals based on their differential sensitivity and responsiveness to attachment signals (e.g., crying) can similarly reveal their evaluations of caregiving behaviors. Indeed, Biro (2023) found that after watching computer-animated separation scenarios similar to those in Biro et al. (2015), 12-month-olds showed their preferences and chose a responsive “adult” who returned to a crying “baby” over an unresponsive one who ignored the “baby.” Additionally, the looking-time PL measure (e.g., Halberda, 2003; Hamlin & Wynn, 2011; Libertus & Brannon, 2010; Luchkina et al., 2018) offers an ideal tool to examine infant preferences for caregivers from a young age. For instance, infants as young as 3 months prefer to look at a helper over a hinderer before they can reliably reach to indicate their choice (Hamlin et al., 2010). Similarly, PL studies can be used to determine whether infants hold preferences for responsive characters, presumably guided by their favorable evaluations of attachment experiences, or they may prefer a character that behaves similarly to their own parent, responsive or unresponsive, presumably guided by their view of their own attachment experiences as normalized. Together, the combination of attachment-relevant measures with both manual choice and looking-time (VOE and PL) tasks can reveal how infants’ attachment-relevant experiences relate to the content of their attachment representations, including evaluations of attachment experiences, that likely comprise their conceptual understandings of parents.
Moreover, research to date has focused on infant cognitions pertaining to adult responsivity to infant attachment signals. In addition to encompassing representations of others in relationships, IWMs are thought to include representations of the self in relationships (Bowlby, 1973). Thus, research on infant representations of the self in relationships is also necessary. With regards to this, it is conceivable that the contents of infant representations of parents’ behavior detailed above also characterize how infants represent themselves in attachment relationships. For example, in addition to representing expectations and preferences regarding the parent’s responsiveness to their attachment signals, infants might also represent expectations and preferences regarding their own behavior in attachment contexts (e.g., whether and how to signal attachment needs). This could be easily examined by adapting existing attachment scenarios employed in the initial research. For example, instead of manipulating adult responsiveness in the separation scenario, infant behavior might be manipulated, such that the infant might seek out or ignore the adult, or the infant might be soothed by the adults’ return or persist in crying. By examining infants’ expectations and preferences in such scenarios, greater insight would be gained into conceptual understandings of infants within attachment relationships.
Throughout this section, we have focused on how research might clarify the construction of attachment representations in infancy. It is also important to extend this work beyond infancy. Meta-analytic evidence indicates that parent-child attachment quality has modest, yet enduring predictive significance for children’s subsequent social and cognitive development. Indeed, in terms of social development, early security with mothers and fathers as assessed via observational assessment of attachment (e.g., SSP) is associated with better social competence (mothers: r = .19, r = .14; Groh, Fearon, et al., 2014) and prosocial behavior (mothers r = .19, r = .11; Deneault, Hammond, et al., 2023), as well as fewer externalizing (mothers: r = .15, r = .18; Deneault et al., 2021; Fearon et al., 2010) and internalizing (mothers: r = .08, r = .09; Deneault et al., 2021; Groh et al., 2012; Madigan et al., 2013) problems. In terms of cognitive development, early security with mothers assessed via observations is associated with better language skills (r = .16; Deneault, Duschinsky, et al., 2023), executive functioning (mothers: r = .19; Pallini et al., 2018), and cognitive skills (r = .17; Deneault, Duschinsky, et al., 2023; too few studies reported on observed attachment quality with fathers to be included in these meta-analyses). As stated in the opening of the article, the key mechanism invoked in explaining such long-term associations is IWMs. However, given the lack of tools for evaluating infant IWMs, this explanatory framework has been left untested in the early life course. Given the mechanistic role IWMs are expected to play in the legacy of attachment, important next steps will be to examine the role of infant attachment cognitions in predicting subsequent social and cognitive outcomes and in mediating between attachment history (e.g., attachment-relevant caregiving experiences, attachment quality) and development in these domains.
Moreover, as noted above, it is important to situate research on IWMs in infancy within a broader developmental framework of IWMs across the life course. Future research should thus examine the connection between infant attachment cognitions and later developing representational constructs in the years following language acquisition. Importantly, the infant attachment cognitive constructs discussed here and other early emerging attachment representations, such as secure base scripts, can be assessed simultaneously in early childhood. For example, children’s secure base scripts have been assessed as early as 36 months (Posada et al., 2018) and infant cognition tasks can be modified for older children (Barrett et al., 2013; He et al., 2011; Scott & Fisher, 2012; Yuan & Fisher, 2009), facilitating the investigation of links between infant attachment cognitions and children’s attachment scripts. Critically, such research should include multiple assessments of attachment representations over time to examine not only how infant attachment cognitions are associated with other aspects of attachment representations, but also the patterning of associations over time, as such work will provide critical insight into the development of attachment representations over the early life course. Indeed, such an approach has provided evidence that secure base script knowledge assessed via the ASA informs adults’ subsequent coherence in the AAI (but not vice versa; Waters et al., 2017), providing support for the claim that the construction of a secure base script supports the development of a coherent autobiographical representation of attachment experiences and advancing our understanding of the developmental progression of attachment representations.
Although we have primarily focused on the ways in which integrating infant cognition and attachment research might advance understanding of IWMs of attachment in infancy, we would be remiss if we did not also highlight the ways in which cross-communication and collaboration among attachment and infant cognition scholars might also advance research on infant cognitive development. Specifically, integral components of attachment theory and research (1) the role of early experience in shaping development, (2) the emphasis on individual differences, and (3) the consideration of developmental continuity and change in life course trajectories. These factors are well-appreciated, but not always incorporated in infant cognition research. We argue that considering such factors would prove useful in a deeper understanding of cognition in the early life course. Further, references to attachment research could help introduce new lines of investigations into cognitive work. Below, we highlight how the integration of attachment and infant cognition research can be mutually informative by discussing ways in which attachment research might provide insights into various topics in cognitive development.
One theoretical approach to infant cognition, as represented by the core knowledge theory, aims to uncover the initial state of human cognition. A key premise of such research, as discussed earlier, is that infants might start with a primitive understanding of several core domains, possibly due to our evolutionary history (Baillargeon, 2008; Carey, 2009; Gelman, 2002; Keil, 1995; Leslie, 1994; Spelke, 1994; Wellman & Gelman, 1997). Although the role that learning and experience play in knowledge acquisition is emphasized (e.g., Baillargeon & DeJong, 2017; Gopnik & Wellman, 2012; Xu, 2019), the contribution of infants’ social experiences to their cognition remains underexplored. Indeed, careful considerations of the influence from the social world, particularly the home environment in which infants spend a large portion of time, in infant cognition research is critical for rigorously evaluating claims regarding the origins and development of early knowledge. In line with this argument, Tasimi (2020) suggested that although experiences with parents might not be expected to hold much bearing on infants’ physical understanding (e.g., all physical objects, including all people, in different corners of the world, exist and move continuously in time and space), it remains an open question whether infants with different caregiving histories build different mental representations of the social world (for interesting discussions on how attachment might serve the evolutionary function for infants to construct and elaborate social cognitive knowledge, see Fonagy et al., 2007). Thus, programmatically examining aspects of infant cognition that are susceptible to variation in early social experiences, including caregiving experiences, and those that are not, is critical for the field of infant cognition.
The need for such research is pressing given the recently growing literature on infants’ sociomoral understandings (e.g., Bian & Baillargeon, 2022; Choi & Luo, 2015; Pun et al., 2021; for reviews, see Ting et al., 2020; Woo et al., 2022). As discussed earlier, ensuing debate aside, infants evaluate others positively or negatively based on social interactions (Hamlin, 2013b). They also seem to grasp at least four moral principles that likely stem from a “first draft” of human morality, or a moral core, that denotes how others should and should not act toward one another (Baillargeon et al., 2015; Hamlin, 2013b; Ting et al., 2020). This interpretation, however, does not yet include considerations of early social experiences, including caregiving experiences, as others have pointed out (e.g., Jin et al., 2018; Tasimi, 2020). Indeed, given that infants are more likely to form secure than insecure attachments with parents (e.g., Ainsworth et al., 1978; Madigan et al., 2023; Mesman et al., 2016; Van IJzendoorn & Kroonenberg, 1988; Van IJzendoorn & Sagi-Schwartz, 2008), an alternative interpretation of these sociomoral understandings is that they are derived from the supportive caregiving securely attached infants have experienced in everyday life, which manifest in these infants’ performances in cognitive tasks. For example, infants with a secure attachment might prefer others who act positively to social interaction partners because they have a caregiving history in which their social cues have typically been responded to positively. This possibility poses a challenge to interpretations of findings from a core knowledge theoretical lens. Thus, an important direction for future research is to examine whether and how infants’ interpersonal understandings might change as a function of their everyday experiences, demonstrating to what extent social experiences might modify the core knowledge, for instance, in the sociomoral domain.
Another particularly ripe aspect of social cognition that would benefit from consideration of infants’ attachment-relevant experiences is trust. Trust – the belief in a person to do a promised or expected action (Hardin, 2002) – serves important purposes in children’s lives (Erikson, 1963). Cognitive developmental research has examined how infants and young children trust (or distrust) unrelated and unfamiliar others as sources of knowledge (i.e., epistemic trust; for reviews, see e.g., Harris, 2012; Koenig & Sabbagh, 2013; Mills, 2013; Poulin-Dubois & Brosseau-Liard, 2016; Sobel & Kushnir, 2013) and for social support, reassurance, or to keep a promise (i.e., social trust; Markson & Luo, 2020). To learn information and knowledge from others, infants may start out with a default tendency to trust, which is viewed as adaptive early in life (e.g., to lay the foundation for language acquisition; Baier, 1986; Coady, 1992; Jaswal & Kondrad, 2016). As young children acquire conceptual knowledge, they can make inferences about others’ accuracy and reliability and thus become selective in who they trust (Sabbagh et al., 2017; Sobel & Kushnir, 2013). Indeed, the large body of research on infants’ and young children’s epistemic trust suggests that others’ knowledge level or reliability in the information they provide is paramount (e.g., Begus & Southgate, 2012; Brooker & Poulin-Dubois, 2013; Buttelmann & Zmyj, 2020; Crivello et al., 2018; Dautriche et al., 2021; Koenig et al., 2004; Koenig & Woodward, 2010; Kuzyk et al., 2020; Luchkina et al., 2018; Luo et al., 2022; Poulin-Dubois et al., 2011; Schütte et al., 2020; Scofield & Behrend, 2008; Zmyj et al., 2010; for reviews, see Baillargeon et al., 2016; Sabbagh et al., 2017; Shafto et al., 2012; Sobel & Kushnir, 2013).
Markson and Luo (2020) also specified social trust as a type of interpersonal trust (with familiar people in close relationships) that is generalized to unfamiliar and unrelated individuals (Rotter, 1967, 1980), as well as understandings acquired through experience that people generally behave according to social norms, to achieve a certain level of predictability or reliability (Schmidt & Rakoczy, 2018; Schmidt et al., 2012). To date, the famous “marshmallow” task used to assess children’s delay-of-gratification behavior (e.g., Mischel et al., 1989) has been repurposed to study social trust. In the task, preschoolers are given a marshmallow by an experimenter and told that they can eat it immediately or wait until the experimenter returns, in which case they will get a second marshmallow. Mischel and colleagues suggested that in addition to needing self-control skills to keep from eating the first marshmallow, children must also trust the experimenter in order to be willing to wait for an unknown period of time for the experimenter to return and fulfill her promise (Mischel, 1961; Mischel & Ebbesen, 1970). Recent work supports this notion by showing that if children previously witnessed the experimenter lying to another individual or breaking a promise, they were less likely to wait in the marshmallow task, suggesting less trust in the experimenter and/or the experimental setting (Kidd et al., 2013; Michaelson & Munakata, 2016).
Given the centrality of trust to attachment relationships (Bowlby, 1969/1982, 1980; Campbell & Stanton, 2019; Collins & Feeny, 2004; Dagan et al., 2021; Mikulincer, 1998; Simpson, 2007), it is intriguing to consider whether epistemic and social trust have connections to parent-child relationships. Indeed, by integrating attachment research, questions regarding the origins of trust, the potential role of experience in shaping the development of trust, and the differential contribution of early parent-child relationships to epistemic versus social trust might be explored. Initial evidence supports the utility of pursuing such research. Regarding epistemic trust, in one study, preschool-aged children’s trust in claims made by their mother or an unfamiliar experimenter was found to be related to the quality of the mother-infant attachment (Corriveau et al., 2009). Children who had been classified as securely attached as infants were flexible, trusting their mother’s claims when presented with novel or ambiguous stimuli, but relying on the stranger’s claims when the stranger was possibly correct about the stimuli. Insecure-avoidant children, however, were least likely to trust their mothers’ claims, and insecure-resistant children were most likely to do so, even when their mother was possibly wrong. On the other hand, although no studies to date have examined links between attachment and social trust, research suggests that the environment might also contribute to variation in social trust. For example, findings indicate that children living in poverty usually are less likely to wait for the experimenter’s return in the marshmallow task, possibly because of the environmental unpredictability they have experienced in everyday life leading to little or no trust in the unfamiliar experimenter and/or environment (Sturge-Apple et al., 2016; Watts et al., 2018). In light of this promising initial evidence, extending this research to further examine links between infant attachment experiences and children’s epistemic and social trust could uncover important processes through which individuals’ experiences inside and outside a home environment and their relationships with familiar and unfamiliar others become interconnected over time (for discussions on how infants might use conditional probabilities in parents’ responses to inform their cognitive understanding of environmental predictability, see Frankenhuis et al., 2013).
Infant cognition research has largely focused on understanding normative development at the group level. Studies are typically aimed at evaluating the cognitive capacity of infants, as a group, in specific domains of cognition, and examining the earliest age at which such capacities emerge. Critically, however, researchers have examined sociocultural variation in cognition (e.g., Koenig & Sabbagh, 2013; Rogoff et al., 2017; Rogoff et al., 2011; Tamis-LeMonda et al., 2014; Wilks et al., 2019). There have also been an increasing number of reports on individual differences in physical, numerical, and social cognitions (e.g., Libertus & Brannon, 2010; Perez & Feigenson, 2021; Sodian et al., 2016; Tan et al., 2018; Thoermer et al., 2013). In contrast, the vast majority of research on attachment is concerned with individual differences in the quality of children’s attachment relationships and their developmental sequelae. We argue that collaboration among infant cognition and attachment scholars would bring greater consideration of individual differences in infant cognition and advance the field in new and exciting directions.
As an example of how a consideration of individual differences might yield richer insight into infant cognitive development, researchers have begun to examine links between infants’ helping expectations and their own helping behavior. In the first year of life, infants appear to expect an individual to help an agent in need as opposed to one who does not need help (e.g., Köster et al., 2016). They themselves, however, do not reliably help others until the second year of life (e.g., Dunfield & Kuhlmeier, 2010; Warneken & Tomasello, 2006). In fact, about half of infants at 16 months of age would not spontaneously help an experimenter in need, e.g., by obtaining a toy out of her reach (Köster et al., 2019). Including measures of 16-month-olds’ motor skills, as well as their willingness to interact with the experimenter (e.g., by taking turns rolling a ball back and forth with her), revealed that these individual differences predicted infant helping infants with better motor and social skills were more likely to help (Köster et al., 2019). Similarly, Sommerville and colleagues (2018) reported that when asked to bring a heavy block to an experimenter for her task of building a block tower, 18-month-old infants’ walking experiences predicted how likely they were to help the experimenter. Thus, while on average, infants as a group may demonstrate their conceptual understanding (e.g., helping expectations), individual characteristics contribute to their differential tendencies to act based on such understanding.
A similar approach might be applied by integrating infant attachment and cognition research. For instance, in addition to examining links between attachment quality and infants’ comforting expectations (Jin et al., 2018), research might examine how such expectations relate to infants’ own comforting behavior (e.g., to respond to others’ distress), including a consideration of the interplay of factors that could vary across individuals, such as emotional understanding and reactivity (e.g., Brownell, 2013; Dunfield, 2014; Eisenberg et al., 2006; Hastings et al., 2006; Vaish & Tomasello, 2014). By focusing on how infants’ conceptual understanding manifests in their overt actions, this multi-factor approach could provide a more enriched understanding of cognitive development. As another example, this approach could also shed light on individual differences in infants’ knowledge acquisition processes. As discussed earlier, infants’ statistical learning capacities may help them build attachment representations based on acquired data (e.g., parents’ responsiveness ratios). A multi-factor model could thus be experimentally manipulated in a variety of ways. For instance, with known infant and parent characteristics, along with parent responsiveness ratios controlled, the variations in the parent-infant attachment patterns might reveal individual differences in infants’ statistical learning capacities.
A core question guiding attachment research concerns the continuity of the quality of attachment over the life course and mechanisms of change. As such, attachment scholars regularly employ longitudinal approaches to studying attachment variation over time. This approach has proven instrumental in estimating the degree to which attachment security is stable across the life course, as well as identifying mechanisms that contribute to lawful change in attachment quality (Booth-Laforce et al., 2014; Fraley, 2002; Groh, Roisman, et al., 2014; Pinquart et al., 2013; Waters et al., 2000). Given the emphasis on normative development, cognitive studies, on the other hand, typically employ cross-sectional designs. However, deeper understanding of the developmental processes by which infant cognitive development unfolds might be gained by the greater use of longitudinal designs in infant cognition research.
As a useful example, we point to interesting findings from research on language acquisition that has employed a longitudinal approach. As word learning is essentially a process through which linguistic labels are linked with certain aspects of the world (e.g., nouns with objects), it is believed that cognitive understanding (e.g., categorization) serves as the basis of semantic development, and that language helps to enrich cognition (e.g., Bowerman et al., 2001; Casasola et al., 2009; Gentner, 2016; Hespos & Spelke, 2004). Guided by these theoretical beliefs, the bi-directional links between language and object categorization in infancy have been extensively studied (for a review, see Perszyk & Waxman, 2018). For instance, hearing naming sentences “Look at the toma! Do you see the toma?” helps 6-month-olds form a category (e.g., of dinosaurs) that encompassed different items (Fulkerson & Waxman, 2007). In turn, infants with better object categorization skills at 12 months of age are found to have larger vocabularies concurrently and six months later (Ferguson et al., 2015), a finding rendered discoverable only by using a longitudinal design. Longitudinal research on language and categorization in infancy offers illustrations of how such approaches may advance understanding of cognitive development, which has been put into increasing use in cognitive research (e.g., Blankenship et al., 2019; Brooks & Meltzoff, 2015; Fernald et al., 2006; Libertus & Brannon, 2010; Perez & Feigenson, 2021; Rose et al., 2012; Sodian et al., 2016; Starr et al., 2013; Tan et al., 2018; Yamaguchi et al., 2009).
Infant sociomoral knowledge is another domain that stands to benefit from applications of longitudinal designs. Here, we highlight the example regarding infant interpersonal expectations of being supportive and helpful of others (Ting et al., 2020). As discussed earlier, research to date has provided evidence for at least two types of infant expectations, those regarding adults comforting a crying baby (Jin et al., 2018) and those about adults in one social group helping one another (Jin & Baillargeon, 2017). Further, research has demonstrated that such expectations manifest in early (e.g., prior to 6 months) and late (e.g., around or after the first birthday) infancy (Jin & Baillargeon, 2017; Jin et al., 2018; Ting et al., 2019). Hence, a longitudinal approach to examine infant helping as well as comforting expectations, in conjunction with experiences such as attachment-relevant experiences would be beneficial. For example, findings might reveal that attachment-relevant experiences shape comforting expectations, but not helping expectations. Such evidence would thus clarify how certain experiences come to revise some aspects of the “first draft” of human morality (Graham et al., 2013) while leaving other sections intact, thereby contributing to the literature on moral cognition.
In addition to extending current lines of inquiry, the integration of attachment and infant cognition theory and research has the potential to spawn novel research questions, in turn fundamentally advancing our understanding of infant cognition. Here, we offer a few examples but note that these suggestions are not meant to be exhaustive of all possible new directions.
A wealth of genetically-informed research points to the environmental contributions to individual differences in infant attachment security (Bokhorst et al., 2003; Fearon et al., 2006; Luijk et al., 2011; Roisman et al., 2013; Roisman & Fraley, 2008). In particular, sensitive caregiving in which the parent promptly and appropriately responds to infant attachment signals and needs is a robust, yet modest predictor of infant attachment security (Bakermans-Kranenburg et al., 2003; DeWolff & Van IJzendoorn, 1997; Madigan et al., 2023; Verhage et al., 2016). An underlying assumption of research on the caregiving origins of infant attachment is that infants, regardless of their caregiving experiences, can detect timing (e.g., how quickly the parent responds to infant signals, considering the parent’s current position) and appropriateness (e.g., the parent’s response was effective at addressing the infant’s need) of parent responses from a young age, in order for them to build attachment representations. We discussed earlier how infants might consider a variety of constraints on parent behavior to assess the appropriateness of parent responses in attachment-relevant situations, mainly based on their psychological understandings. With regards to timing, to our knowledge, although infants in the first year of life can discriminate between two different speeds (Möhring, Libertus, et al., 2012) and two different durations (vanMarle & Wynn, 2006), and seem to realize that an object that moved behind a screen should not immediately reappear from the other side of the screen (Wilcox & Schweinle, 2003), an understanding of the interplay between time, speed and distance may not emerge until the second year of life (Möhring, Cacchione et al., 2012), based on cognitive research. However, attachment theory and research suggest that such sensitivity to the timeliness of parent responses could be present from a much younger age and play a part in infants’ attachment-building processes. For instance, if a parent’s response is always delayed following an infant’s distress signals, even when the parent is close by, the infant might view the parent as insensitive and in turn form an insecure attachment with the parent based on these repeated experiences. Therefore, combining work on attachment building with cognitive research can help achieve a deeper understanding of infants’ conceptual understanding of time and duration.
Another important feature of IWMs is that although they are constructed from attachment-relevant experiences with parents, over time they are expected to generalize and inform subsequent close relationships, such as friendships and romantic relationships (Girme & Overall, 2021; Groh et al., 2017; Owens et al., 1995; Roisman et al., 2009; Roisman et al., 2005; Simpson et al., 2007). Meta-analytic evidence indicates that infant attachment security is associated in enduring ways with better social competence and fewer externalizing behavioral problems (which are typically manifested in peer contexts) into adolescence (Deneault et al., 2021; Fearon et al., 2010; Groh, Fearon, et al., 2014), and IWMs are thought to be one of the key mechanisms by which infants’ attachment experiences contribute to subsequent positive and negative peer behaviors. Integrating attachment work into cognitive research could thus give rise to new studies on infant understanding of peer relationships. To this end, we suggest a similar approach to that of exploring the links between attachment quality and expectations for adult-child relationships (e.g., Johnson et al., 2010) can be used to examine the significance of attachment quality, expectations, and preferences for peer relationships, assessed via VOE, PL, and choice tasks, as well as to investigate the role of individual differences in infant peer expectations and preferences for their subsequent peer interactions. Relatedly, the body of work on infant knowledge in the sociomoral domain has revealed early concepts of social roles and relationships, e.g., being a leader of a group, ingroup vs. outgroup, or allies vs. foes (e.g., Bian & Baillargeon, 2022; Bian et al., 2018; Jin & Baillargeon, 2017; Liberman et al., 2014; Margoni et al., 2018; Mascaro & Csibra, 2012; Powell & Spelke, 2013; Pun et al., 2016; 2021; Rhodes et al., 2015; Spokes & Spelke, 2017; Stavans & Baillargeon, 2019; Ting et al., 2019; Thomas, Saxe, et al., 2022; Thomas, Woo, et al., 2022; Thomsen et al., 2011). It is possible that infants’ conceptualization of these social roles and relationships may bear the mark of attachment representations concerning the earliest interpersonal relationship in an infant’s life. For example, infants’ characterization of ingroup/outgroup might be partly based on how similarly or differently an individual looks or behaves compared with their attachment figures. Addressing these questions will thus add to the increasingly rich understanding of the development of infant cognition.
Taken together, we propose that recent advances in the infant cognition literature can expand methods used in attachment research and address theoretical issues regarding the content and development of IWMs of attachment. We also propose several ways in which the integration of attachment research into infant cognition research might extend current inquiry into early cognitive development and open new lines of investigations. To address these important questions, we call for increased collaboration among socioemotional and cognitive development scholars. As a first step, together with our colleagues, we have initiated collaborative research (Bhoyar et al., 2021; Jin et al., 2018; Putrich et al., 2024a; 2024b) that aims to examine the cognitive architecture of infant attachment by focusing on the development of infant expectations for behavior within close relationships, as well as infant preferences for particular social interaction partners, as a function of infant attachment experiences and in relation to attachment-relevant socioemotional and cognitive outcomes. As detailed throughout this article, such work represents a first step, and we hope this exposition will spark similar collaborative endeavors to achieve a more comprehensive understanding of the fascinating first years of human life.