1The remarkable ease and frequency with which human infants imitate has led to many claims about the centrality of imitation in development. Imitation has been associated with many developmental functions, from being a precursor to language to promoting bonding between parent and infant. This review focuses on perhaps one of the most intuitively obvious roles of imitation: as a tool for social learning. Imitating others is a powerful means of spreading information within a social group. However, the transfer of information across species members can only be effective if the learner retains the information they have observed and subsequently retrieves that information when the opportunity arises. Imitation has therefore provided an important tool for researchers interested in studying the development of abilities such as representation, memory, and categorisation. In this review we discuss how imitation has been used to explore these areas, and what it has revealed. There is currently some debate as to whether behaviours classed as ‘imitation’ in the developmental literature, to which we are primarily referring, would be termed ‘emulation’ by researchers in other fields. This issue has been discussed in depth in the literature (Want and Harris, 2002; Nielson, this issue). In the present review we will use the term imitation to provide continuity with its use in the articles we are citing, but it is clear that this issue is not yet resolved.
2Imagine you are sitting down to do some work in a quiet place away from any distractions. Suddenly an unfamiliar person, who is sitting nearby, attracts your attention and then pokes his tongue out at you. What would you do? Much like how a child or even an adult might respond in this situation, the newborn baby responds by reproducing the action she has been presented with, her own tongue moving out beyond her lower lip. In the experimental version of this procedure, the “neonatal imitation paradigm” originally developed by Meltzoff and Moore (1977; 1983), the neonate or young infant, is seated facing an adult. The adult attracts the infant’s attention and then demonstrates the target behaviour, typically a facial response such as tongue protrusion or mouth opening, repeatedly for approximately 20-secs. After each demonstration period, the adult produces an expressionless “still-face” providing a response period during which the behaviour of the infant is recorded. In the first few days and months of life, infant humans (e.g., Heimann et al., 1989; Heimann and Schaller, 1985; Meltzoff and Moore, 1977, 1983, 1989, 1992, 1994) and human-reared and mother-reared chimpanzees (Bard, 2005; Myowa, 1996; Myowa-Yamakoshi et al., 2004) produce more tongue protrusions compared to mouth openings in response to an adult’s demonstration of tongue protrusion.
3The phenomenon of neonatal imitation has received considerable attention by researchers, in no small part because it represents a remarkable achievement by the newborn. To reproduce the facial movements of another person means that the newborn is not only copying another person’s action but that she is doing so using her face - a part of her body that she has not seen. Piaget (1962) proposed that this type of imitation should not occur until at least one year of age, much later than when the infant would become able to imitate actions with an observable part of her own body such as imitating the hand or leg movements of another person. Although there is ongoing debate about the existence of neonatal imitation, which has previously been referred to as a “fuzzy phenomenon” (Heimann, 2001), and proposed to be limited only to tongue protrusion (Anisfield, 1996), it is generally accepted that newborn infants have an innate ability to reproduce at least some observed facial behaviours, contrary to Piaget’s theory.
4While the phenomenon of neonatal imitation is generally accepted, debate continues as to the precise nature of the mechanism by which infants are able to reproduce facial behaviours, and whether this is the same type of imitation observed when infants begin to imitate actions with objects at 6 months of age (e.g., Barr et al., 1996; Collie and Hayne, 1999; Herbert et al., 2006). Some researchers have proposed that the infant’s reproduction of facial behaviours simply reflects some type of innate releasing mechanism (Jacobson, 1979; Abravanel and Sigafoos, 1984). That is, ’imitation’ of tongue protrusion occurs when the response to a visual stimulus (the infant’s observation of tongue protrusion) is an automatic motor response (which in this case also happens to be tongue protrusion). Support for this explanation comes from studies showing that infants also appear to increase their rate of tongue protrusions as an oral exploratory response to experiences such as tactile stimulation (Humphrey, 1970), auditory stimulation (Jones, 2006), and visually interesting displays (e.g., Jacobson and Kagen, 1979; Legerstee, 1991). For example, 4-week-old infants in Jones (2006) increased their rate of tongue protrusion during 20-sec intervals of silence immediately after hearing 20-sec intervals of music. These findings suggest that the infant’s apparent imitative responding to the demonstration of tongue protrusion may simply be the same as the response evoked by any other interesting stimulus experienced by very young infants.
5In contrast to an innate releasing mechanism theory, Meltzoff and Moore propose that neonatal imitation occurs through intermodal matching. That is, the infant has an early ability to use the visual information from the adult’s action to guide their own proprioceptive-motor behavioural response. Although once thought to be an ability that emerges only after substantial experience in the world, cross-modal abilities appear to be present within a few days after birth. For example, Streri (2003) demonstrated that 1- to 5-day-old infants given the opportunity to tactilely explore an object (cylinder or prism) in their right hand could subsequently visually recognise the previously experienced stimulus when paired with a novel stimulus. Although the transfer of information from tactile to visual formats in Streri and Gentaz (2003) is different from the transfer of visual information to a proprioceptive-motor behavioural response in neonatal imitation studies, it does suggest that intermodal matching provides a possible mechanism for the infant’s behaviour change in response to experience with a novel stimulus. The finding that infants’ imitative acts become more accurate replications of the target action across successive trials (e.g., Meltzoff and Moore, 1983; 1994; Nagy et al., 2005) also provides further support for the cross-modal matching explanation.
6The limited time frame during which neonatal imitation occurs has been proposed as an argument against cross-modal matching. By the age of 2-3 months in humans (Maratos, 1982; Abravanel and Sigafoos, 1984; Fontaine, 1984), and 9 weeks in mother-reared chimpanzees (Myowa-Yamakoshi et al., 2004), the infant’s response of reproducing the demonstrated facial actions declines before it reappears again at a later age (Decarie and Ricard, 1996). Meltzoff and Moore (1992) propose, however, that neonatal imitation may simply be replaced by more age-appropriate social behaviours such as smiling and cooing, with these new behavioural responses masking the imitative behaviour. Interestingly, Myowa-Yamakoshi et al., (2004) discuss the existence of a play-face directed towards the human experimenter and produced by the chimpanzees aged 9-16 weeks, which also corresponds with the offset of facial imitation behaviour.
7The predominance of facial behaviour as the focus for neonatal imitation studies likely reflects the desire to challenge Piaget’s theory that the imitation of unseen actions was a late emerging ability. However, the use of facial behaviours creates particular difficulties for imitation research, such as changes in the method of social engagement and responding with age (Meltzoff and Moore, 1992), high frequency baseline levels and the complication of using a socially important stimulus such as a face as a tool in imitation (Nagy et al., 2005). Such difficulties have led Nagy and colleagues to focus their neonatal imitation research on the production of fine motor movement with other body parts. Whilst this research is still in its infancy, it holds much promise for expanding our knowledge of how and why young infants imitate. An early study with index finger movement has already indicated the role that neonatal imitation appears to have in social exchanges and communication between the preverbal neonate and adult. In Nagy and Molnar (2004), neonates aged 3- to 54-hours not only responded to the demonstrated extension of the index finger with a similar response but also, over time, appeared to be attempting to initiate these interchanges. This resulted in what the authors refer to as “dialogues”, overlapping and fast exchanges in gestures between the infant and experimenter. These exchanges were also accompanied by changes in infant heart rate depending on whether the neonate was engaging in imitation or “waiting for a response” (Nagy and Molnar, 2004). Thus, it appears that the newborn infant is both capable and motivated to engage in these early social interactions.
8In addition to investigating the newborn’s ability to engage in early interpersonal communication, neonatal imitation research has also begun to provide the foundations for further investigation into the development of learning and representational abilities. Examining the progressive refinement of imitative acts such as those observed in Nagy et al. (2005; see also Meltzoff and Moore, 1983; 1994) can provide the opportunity to study early learning processes. Infants in these studies produce an initial response to the stimulus but then, over successive trials, modify that response to more accurately reflect the model. Thus the infant appears to be learning across trials and progressively shaping his own response as he repeatedly observes the model. Furthermore, Meltzoff and Moore (1994) have also demonstrated that infants have some ability to remember the demonstrated actions across long delays, with infants aged 6-weeks being able to reproduce an adult’s facial behaviours both immediately and after a 24-hour delay. The results of this highly cited study hint at the early age at which human infants might be able to create and maintain memory representations. Although the range of imitative behaviours that have been studied with neonates remains limited, the research to date provides intriguing clues about the social and cognitive abilities present within a few months of the infant entering the world.
9Studying the ability to imitate simple movements that are within the neonates’ physical capabilities can provide us with insight into the representational capacities of infants at birth. However, the potential of simple movements for studying developmental change in cognitive abilities beyond the neonatal period is restricted by the decrease in responding in facial imitation occurring around 2 months of age, and the rapid changes in fine motor control across early infancy. Furthermore, a focus on body-part movement confounds learning, memory and previous experience because infants can independently perform these behaviours prior to the demonstration, resulting in the opportunity for incremental learning. Studies of cognitive development utilising imitation procedures have therefore primarily focused on imitation of actions-with-objects, where unique actions are performed with novel stimuli.
10In the experimental version of the imitation paradigm, originally developed by Meltzoff (1985, 1988), learning can be directly linked to a single experience with an experimenter and a novel object. In a typical imitation task, the infant observes the experimenter demonstrate a target action, or series of target actions, with the stimuli. After watching the demonstration, the infant is given the stimuli, and the number of target actions reproduced is recorded. Crucially, the target actions produced by infants in these demonstration conditions are compared to the actions produced by infants in a control condition who have not seen a demonstration of the target actions. Learning is inferred if the demonstration group produce a significantly greater number of target actions than the control group produce through spontaneous exploration. This type of imitation paradigm is referred to as actions-with-objects imitation, and is readily exhibited by infants aged between 6- and 24-months of age (e.g., Barr et al., 1996; Collie and Hayne, 1999; Herbert et al., 2006).
11One of the most successful areas of investigation using the actions-with-objects imitation paradigm has been in the domain of early memory development. Inserting a delay interval between when the infant observes the target actions and when he has the opportunity to interact with the stimuli creates a task with memory demands. This procedure is referred to as the “deferred imitation” paradigm. In order for the infant to imitate the target actions after the delay interval, he must remember what he has seen. By manipulating factors such as the duration of the delay interval, and the similarity of the events and stimuli experienced at the time of learning and memory retrieval, we can explore developmental changes in memory capacity, complexity, and flexibility that occur across the infancy period. Research with the deferred imitation paradigm has revealed major changes in retention span, rate of learning, memory for relations between stimuli and contextual details, and the ability to generalise memory retrieval across stimuli and contexts. Each of these developments will be reviewed in turn below. However, to fully appreciate the results of studies with the deferred imitation paradigm it is first important to consider what type of memory deferred imitation reflects.
12Many researchers have argued that memory is not a unitary system, but instead comprises multiple systems subserved by different neural networks that emerge at different rates across development (e.g., Bachevalier, 1990; Meltzoff, 1995; Nelson, 1995). Memory systems have been roughly divided into declarative and non-declarative forms (for review, see Squire, 2004). Declarative memory is memory for facts and events, and is so-called because it has often been identified by the level of consciousness that accompanies encoding and retrieval. That is, declarative memories are frequently referred to as memories that can be stated or ‘declared’. This memory system is assumed to be more mature than other forms of memory, emerging later in development. Non-declarative memories are memories for habit and skills acquired through recent or repeated exposure to a stimulus. The term non-declarative memory is an umbrella term covering many different types of memory types that are thought to emerge early in development.. The division of memory into these two categories was primarily established to explain the dichotomous performance of patients with temporal lobe amnesia on memory tasks. Temporal lobe amnesiacs perform normally on non-declarative memory tasks such as simple delay conditioning and mirror learning, but perform poorly on declarative memory tasks (for review see Squire, 2004, Eichenbaum, 1997). Such findings suggest that the temporal lobe of the brain is crucial to declarative memory.
13Defining declarative memory through criteria of consciousness is possible with adults but is not applicable to nonverbal populations. Identifying the medial temporal lobe (including, crucially, the hippocampal region) as the brain region most closely associated with declarative memory has facilitated explorations in animal populations through the use of lesion studies (e.g. Eichenbaum, 1997, 2004; Eichenbaum and Bunsey, 1995). However, current brain imaging techniques do not allow us to map the development of the hippocampus in sufficient detail in human infants. Developing tasks that tap declarative memory is therefore an important way to explore the development of both declarative memory and the hippocampal region in infancy. As memory performance in the deferred imitation paradigm is based on a brief observation of a set of target actions, without prior motor practice or trial-and-error learning, many researchers have argued that this task provides a measure of declarative memory (e.g., Barr and Hayne, 1996; Hayne, 2004; Meltzoff, 1995). The finding that patients with temporal lobe amnesia or developmental amnesia associated with a reduction in hippocampal volume also have difficulty with an adult-appropriate version of the deferred imitation task provides support for the idea that deferred imitation tasks are a non-verbal measure of declarative memory (McDonough et al., 1995; Adlam et al., 2005). Furthermore, the pattern of retention exhibited in the deferred imitation paradigm conforms closely to that seen in declarative memory paradigms. For example, the infant’s memory representation contains detailed and specific information and the duration of retention is influenced by factors such as the age of participants and the retention interval over which it is assessed (for review see Hayne, 2004; Rovee-Collier et al., 2001). What has the last twenty years of deferred imitation research been able to reveal about how declarative memory develops in infancy? Key findings from this paradigm about the emergence of declarative memory, changes in memory processing across early development, and the flexibility of memory retrieval are reviewed below.
14The age at which declarative memory comes ‘online’ has been a hotly debated topic in the literature for almost half a century. The observation that most adults have no verbally accessible memories prior to the age of 2 or 3 years (a phenomenon termed ‘childhood amnesia’) has led some researchers to propose that this shift to verbally accessible memory reflects the onset of declarative memory (e.g. Bachevalier, 1990). Indeed, whilst some structures in the hippocampal formation are mature at birth, others are not fully functional until the 3rd or 4th year of life (Seress, 2001). However, establishing a ‘starting point’ for declarative memory requires direct studies of memory operation in infants using suitable paradigms, and the deferred imitation task is an ideal candidate.
15Piaget (1962) proposed that the emergence of deferred imitation at 18- to 24-months of age was the hallmark of mental representation, an observation that is consistent with the idea that declarative memory may emerge later in development than non-declarative memory. Prior to this stage, Piaget postulated that the infant did not have the cognitive structures necessary to store information and as such the infant was limited to reproducing actions in the here-and-now. With the emergence of deferred imitation, the infant demonstrated that he or she was no longer constrained to a specific time or place and could apply his or her knowledge when confronted with new situations and tasks.
16Subsequent research with deferred imitation has revealed that many early researchers, including Piaget, underestimated the abilities of young infants to form and maintain memory representations. When motor demands are sufficiently reduced, 6-month-old infants can remember a sequence of actions for 24 hours (Barr et al., 1996; Learmonth et al., 2004) and provided their memory is prolonged with a series of reminders, 3-month-old infants also show evidence of learning in deferred imitation tasks (Campanella and Rovee-Collier, 2005). Memory development in infancy has more recently been characterized as a continuous process of change than a discontinuous shift that reflects the onset of a declarative memory system (Hayne, 2004; Rovee-Collier et al., 2001). As a result of this, research with the deferred imitation paradigm has moved beyond establishing the age of onset of an “advanced” memory system to establishing the parameters of memory which are changing during the infancy period.
17Older infants learn information faster than younger infants. In a commonly used imitation task developed by Barr et al. (1996), infants are shown a sequence of actions with a hand-held glove puppet. The infant observes the experimenter remove a mitten from the puppet’s hand, shake the mitten, and replace the mitten. At 12-months of age, seeing these target actions demonstrated three times in succession is sufficient to support memory retrieval after a 24-hour retention interval (Barr et al., 1996). At 6-months of age, however, infants only show evidence of retention over a 24-hour delay if they have seen the actions six times in a single session. The fact that 6-month-olds fail to imitate after 24-hours when they see the actions demonstrated three times does not reflect a complete failure of encoding – when tested immediately they show similar levels of memory to the older infants (Barr et al., 1996). Younger infants appear to require more exposure time to create a representation that is strong enough to persist over the longer retention interval, implying that speed of encoding increases across the first year of life.
18The length of time infants can remember information for increases across development, even when initial levels of encoding are equated. Whilst 6-month-old infants remember the puppet task for 24-hours (Barr et al., 2001), 12-month-olds remember for up to a week, and 18-month-olds remember for up to 2 weeks (Hayne and Campbell, 1997. Declarative memory during the first year of life. Unpublished. Paper presented at the Society for Neuroscience, New Orleans, LA). In a related task, which 18-month-old infants also remember for 2 weeks, 24-month-old infants can remember the target actions for 3 months (Herbert and Hayne, 2000). In both of these studies, strength of encoding was equated by testing additional groups of infants immediately and ensuring that levels of initial encoding and spontaneous baseline levels of performance were comparable across age groups. Age-related differences are, therefore, observed in the duration over which a memory representation can be maintained.
19The duration of retention observed in these developmental comparisons does not, however, represent the absolute duration over which a memory can be maintained at a particular age. Different tasks result in different estimates of retention. For example, although 6-month-old infants exhibit 24-hour retention in the puppet task, the same age group exhibit immediate imitation only, and no retention at 24-hours, with a stimuli that requires the infant to push a recessed button on a toy (Herbert et al., 2006). Similarly, the duration of retention can be prolonged in the puppet task by providing the infant with providing additional brief exposure to the stimuli during the retention interval. For example, Barr et al. (2005) provided 6-month-old infants with the opportunity to observe the puppet stimulus again, but not the actions, after 24 hours. Providing this type of passive reminder at the end of the normal retention interval extended 6-month-old infants’ retention to 10 days. Repeated reminders prolonged 6-month-olds retention for 2-and-a-half months, a duration of retention which is comparable to the retention span of a 2-year-old. It appears possible to increase or decrease the duration over which a memory can be maintained for infants of any age. Across ages, however, there is a developmental increase in the duration over which the same event will be remembered for.
20Creating associations between different events can also modify the retention interval. For example, memory for an imitation task can be retrieved over substantial delays if it is associated with another task that is typically remembered for longer, such as an operant conditioning task. For example, Barr et al. (2001) associated the puppet task with an operant train task where 6-month-old infants learn to press a lever to make a toy train move around a track, a task which can be remembered for two weeks. If infants were allowed to retrieve the memory of the train task first, infants imitated the target actions in the puppet task after 2 weeks, the same duration over which the train task is remembered. Thus, infants had formed an association between the two tasks that allowed memory for the train task to cue memory in the deferred imitation paradigm. This implies that memory for a motor skill, acquired through repeated practice, may facilitate memory for passively observed actions.
21Exploring the relationship between observational learning and motor experience within a single task such as imitation may increase our understanding of the way in which different types of learning and memory are integrated in the brain. The nature of the link between learning through practice and learning through observation has been explored directly with a range of imitation tasks. Barr and Hayne (1996) showed 18-month-old infants a sequence of actions performed with a set of stimuli. For example, a ball, a cup and a lid could be put together to form a rattle in a three-step sequence. Infants in the practice condition were then allowed to reproduce the actions immediately after the demonstration. Having the opportunity to practice had no effect on memory for the target actions at the 1 week test. Meltzoff (1995) also found no effect of practice on imitation of a set of actions over a 2- or 4-month delay with 14- and 16-month-old infants. However, neither of these studies examined whether practice could prolong retention beyond the interval over which it is typically exhibited by infants who only observe the target actions. Furthermore, practice may have more pronounced effects when retrieval is tested across a change in context or cue (e.g. Bauer and Dow, 1994; Barr and Hayne, 2000; Learmonth et al., 2004), an issue that will be discussed in the next section.
22Although infants may appear to have poor memory abilities, imitation research reveals that this does not reflect a failure to encode information. Rather, memory retrieval in early infancy is constrained by elements of the test setting. Changes in physical or social context between encoding and retrieval can interfere with the infant’s ability to retrieve their knowledge. For example, 6-month-olds who learn the puppet task in one location, such as at home, fail to retrieve this information when they are tested in a different location, such as the lab (Hayne et al., 2000). Changes in social context also affect 6-month-old infants’ memory retrieval. If the experimenter who demonstrated the target actions is not present at the test session, retrieval fails even after a 10-minute delay (Jones and Herbert, unpublished data. The role of context in learning: Deferred imitation in 6- and 9-month-old human infants.), which is well within the 6-month-olds retention span for this task. Such research demonstrates that young infants encode a wide range of cues in their memory representations that exert strong control over retrieval. With age these constraints are gradually lifted. By 9-months, infants can retrieve their knowledge across a change in experimenter after a 10-minute delay (Jones and Herbert, 2008), and by 12-months, infants can imitate across a change in location after a 24-hour delay (Hayne et al., 2000). Research with deferred imitation tasks has thus demonstrated that contextual flexibility increases over the first year of life.
23Further insight into the factors controlling memory retrieval in infancy can be gained by exploring the effect of manipulating the learning experience. For example, physical practice appears to facilitate contextual flexibility in infancy. When given the opportunity to reproduce the target actions immediately after the demonstration, 9-, 12- and 18-month-old infants’ memory retrieval at a 24-hour test is no longer affected by a change in experimenter (Learrmoth et al., 2004). In contrast, physical practice does not facilitate the performance of 6-month-old infants, who fail to show evidence of retention with the novel experimenter. Physical experience may therefore facilitate flexibility across changes in social context, but there are also age-related changes in the extent to which it is successful.
24An event memory contains both the individual actions that make up the event, and the order in which they occurred. Memory for order information within an event is particularly central to the issue of understanding the type of memory a particular task is measuring. Recently, the view that the hippocampus is specialized for processing the relationships between items in a memory trace has become increasingly popular (for review see Eichenbaum, 2004; Eichenbaum and Cohen, 2001). Thus, declarative memory may be crucially relational in nature, comprising a network of associations of elements of an event and the relationships between them (Eichenbaum, 1997). As such, measuring infants’ recall of action order in the deferred imitation paradigm may provide a sensitive index of the development of declarative (or relational) memory.
25To explore the issue of whether infants’ exhibit ordered memory recall, retention can be measured in two ways in the deferred imitation task. First, the infant’s reproduction of the individual target actions is scored. Second, the order in which pairs of target actions are performed can be examined (e.g. Carver and Bauer, 1999; Bauer et al., 2000; Mandler and McDonough, 1995). It is possible that infants could score highly on amount of information remembered, but not produce any actions in the correct temporal order. Measures of order recall reveal developmental changes in the fidelity with which infants remember the order of an action sequence. At 6-months, 25% of infants tested by Barr and colleagues retained some order information (Barr et al., 1996). By 9 months, Carver and Bauer (1999) find that roughly 50% of infants show memory for order, and by 14-months the proportion has risen to 78% (Bauer et al., 2000). Thus, while young infants show some evidence of retrieving order information, there are substantial individual differences and the majority of infants do not show order recall until the second year of life.
26A lack of order recall does not imply that an infant is insensitive to the underlying structure of events. This can be explored more deeply by varying the structure of the action sequence shown to infants, and considering how this impacts on learning and retrieval. The actions in a sequence can have one of two underlying structures: enabling or arbitrary. In an enabling sequence, the individual actions need to be performed in a certain order to reach the desired goal. For example, a ball must be placed inside a rattle for it to produce a sound when shaken. In an arbitrary sequence the actions can be performed in any order and the goal can still be obtained. For example, the order in which eyes are placed on the face, the ears are revealed, or food is placed in the animal’s mouth is unimportant to the final endpoint of creating a face with a toy. By comparing performance with enabling and arbitrary sequences, infants’ sensitivity to the relations between events can be explored.
27When performance is compared on the arbitrary and enabling sequences, infants typically show superior memory for both the individual actions and the order of their presentation in the enabling sequences (Barr and Hayne, 1996; Bauer and Mandler, 1989, 1992; Bauer and Thal, 1990, Mandler and McDonough, 1995). For example, when 11-month-old infants were tested after a 20-sec delay the individual actions from both types of event sequence were well recalled, while after 24 hours infants only showed evidence of recall for the enabling sequences (Mandler and McDonough, 1995). Recall of order information was only seen for the enabling sequences. Further, memory for the enabling sequences was surprisingly strong: infants also show some evidence of recall after a 3 month delay. This difference in recall between enabling and arbitrary sequences appears to be present across development – similar distinctions have been found for 16-month-olds (Bauer and Mandler, 1992) and 20-month-olds (Bauer, 1992). Indeed, in a study starting with 16-month-old infants, Wenner and Bauer (1999) found that it was not until 20-months of age that infants showed above-chance recall of arbitrarily ordered events. This research suggests that infants can use the structure of an event sequence to scaffold retrieval. It may be that retrieving the first action in an enabling sequence provides a strong cue to the retrieval of the subsequent steps. If this is the case, giving infants this scaffolding at test may improve their retrieval of a series of unrelated actions.
28Ordered recall is typically studied in older infants using a sequence of actions with a single stimulus. These multi-step sequences are too complex for the abilities of young infants. It is possible, however, to study order recall in young infants by using a series of single step stimuli. In typical imitation studies, the stimuli are represented to the infants in the same order in which the infant saw them during the demonstration session (Meltzoff, 1985; Meltzoff, 1988). If infants process the order in which the actions occur, replicating the order at test may provide them with strong retrieval cues. If this is the case, changing the order of presentation may affect retrieval. In contrast, if young infants do not encode the order of an arbitrarily presented sequence of toys, they should be unaffected by the order in which toys are presented at test. Recently, Knopf and colleagues explored the effect of presenting the toys in a different order with 10- and 11-month-old infants (Knopf et al., 2006). Infants observed four objects, each of which had a simple action associated with it. For example, one object was a drum with a large red button on the top and the action was to press the button. Each object was shown individually and the target action was demonstrated. After a 30-minute delay, the objects were presented to the infant in either the same order as during the demonstration session, in the reverse order, or in a mixed order. Memory recall was observed only when the order of presentation was the same at demonstration and test. By 10-months of age, infants appear to be sensitive to the order of presentation of even arbitrarily related events.
29Research on ordered recall, therefore, suggests that by late in the first year of life infants are encoding the relationships between events, and using this information to cue retrieval. However, this ability undergoes substantial development, and recall of the order of arbitrarily structured sequences is not apparent until late in the second year (Wenner and Bauer, 1999). This timeframe fits with the development of other abilities that are proposed to be hippocampally dependent. Between 18- and 24-months there are striking developments in infants’ ability to process relations between objects, representing multiple spatial locations, and recalling a single spatial location over a long delay, abilities that can all be linked to hippocampal functioning (Sluzenski et al., 2004). The end of the second year may thus represent an important period of development for hippocampally-related skills, including declarative memory.
30In summary, research with deferred imitation paradigms provides substantial insight into the developing brain and the changes occurring in memory in the first two years of life. Early investigations indicated that the development of declarative memory is more appropriately characterised as a continuous process of change than as a discrete step. Although there are dramatic changes in retention across infancy, such as the six-fold increase in the duration of retention between 18- and 24-months of age, there is no evidence of different memory systems being responsible for memory at different ages. Instead, changing the task demands, such as providing longer learning periods, or forming associations between events, affects the duration over which events can be maintained within an age bracket. In the next section, we turn to an area of investigation that is closely related to the study of memory development– the use of imitation tasks to explore categorisation in infancy.
31How do infants structure their knowledge? As adults, we often organise representations into groups of related elements that are termed categories. Categories increase processing efficiency because they specify and delimit the range of situations to which a particular piece of knowledge might be applicable. For example, if we are told that a teal is an exemplar of the category ‘bird’, we might infer that a teal has wings and can fly. Studying the development of categorisation can reveal how infants structure their knowledge as they acquire it, and may inform debates on the nature of categories in the mature brain.
32One domain in which categories are particularly useful is in helping us use our prior knowledge to guide our future behaviour. When faced with a new car, for example, we quickly work out how to use it from past experience with other members of the car category. When infants acquire new knowledge such as learning a new skill, it is important for them to be able to apply that skill in any relevant situation. This process can be facilitated by the infant forming a category of all the things to which that skill can be applied. Piaget (1962) described two processes through which infants structure their knowledge. As a skill is learned, new objects on which that skill can be performed are assimilated into the relevant category. At the same time, the skill undergoes accommodation, to make it applicable to the new objects. For example, the infant first learns to suck the mother’s nipple. As they develop, they assimilate other objects (such as a pacifier) into the category of things that can be sucked. Simultaneously, the sucking reflex changes to accommodate a pacifier. Thus, infants build up a flexible and organised knowledge base. Piaget’s theories imply that categories are formed gradually through the infancy period, and that they are constructed through the infant acting on the world around them.
33When infants see and imitate an action demonstrated on an object, they are learning about the way in which the object can be used. However, other related objects may have similar functions. Infants will benefit most from observational learning if they can apply their learned knowledge to other objects in the same functional category. Imitation paradigms can thus be used to explore how infants learn to place two objects in the same category, by investigating the conditions under which they will imitate a sequence of actions on a similar but not identical object. In this section, we will briefly review imitation research which has explored the development of categorisation in infancy. In the imitation literature the process of imitating a set of actions with a novel object has often been termed generalisation. We will thus begin by addressing the relationship between categorisation and generalisation in imitation tasks.
34In the generalisation version of the deferred imitation task, infants are given test stimuli which are perceptually different to the demonstration stimuli, but afford the same actions. If the infant performs the target actions with the new stimuli, researchers conclude that the infant has generalised across the stimulus change (e.g. Hayne et al., 2000). Generalisation can result from one of two processes. First, infants may have forgotten or failed to encode details of the original stimulus. Thus, when presented with a novel but similar stimulus, they may fail to discriminate the two and imitate as they would have done with an identical toy. Alternatively, infants may realise that the two stimuli are different, but group them together in a category of objects that afford the demonstrated function. In this latter case, the infant generalises within the category they have formed.
35How can we determine whether generalisation stems from forgetting or categorisation? There are three lines of evidence to suggest that generalisation of imitation in infancy is not simply the result of forgetting the details of a stimulus. These areas of research also illuminate the ways in which research with the deferred imitation paradigm can tell us about categorisation in infancy. In the following section, we review these three areas, discussing how they relate to both categorisation and generalisation and what they reveal about the development of these processes in infancy. First, we consider how generalisation changes through development, highlighting the importance of perceptual similarity to early categorisation. Second, we consider generalisation from memory: how does the interval over which retention is assessed affect categorisation? Third, we end with a discussion of research demonstrating that generalisation co-exists with recall of specific detail in infancy, supporting the theory that generalisation may result in part from categorisation skills.
36If generalisation resulted from forgetting, we would expect that younger infants, who show faster forgetting, would thus generalise more readily than older infants. In contrast, memory in early infancy is very specific, and it becomes less specific with age. For example, 6-month-old infants remember the puppet task after a delay of 10 minutes or 24 hours when the same puppet is used for both demonstration and test sessions. However, when they are tested with a puppet that differs in either colour or form from the original, they show no evidence of memory (Hayne et al., 2000). By 12-months of age, infants can generalise across a change in colour when tested after a 10-minute delay, and by 18-months, infants can generalise to a puppet that differs in both colour and form (Hayne et al., 2000). Age-related increases in the flexibility of memory retrieval do not support the theory that generalisation results from forgetting of stimulus attributes.
37Increases in generalisation appear to reflect, at least in part, the development of categorisation rather than forgetting. Over the first two years of life, infants tested in the puppet task learn to form categories that include the objects that afford the puppet’s target actions. At 6-months, this category only includes puppets that are nearly identical to the demonstration puppet. By 12-months, infants include puppets of the same general form as the demonstration puppet, and by 18-months infants categorise any puppet with a reasonable resemblance to the demonstration puppet as an object on which the same actions can be performed. This task has also been used to explore developmental change in the features infants use to form categories. For example, 12-month-old infants can generalise across a change in the colour of the puppet, but not its form (Hayne et al., 2000). This may be because at this age, infants group the puppets together by their shape, rather than by their colour. Form is often a more useful cue to function than is colour, and infants may begin to identify this between 6- and 12-months of age.
38Perceptual similarity is an important determinant of categorisation in infancy. Through the first two years of life, infants become able to generalise their knowledge to objects that are increasingly different from the stimuli they previously encountered. This development may be one of the factors underlying the development of pretense, which typically occurs between 18- and 24-months of age (Leslie, 1987; Piaget, 1962). In pretend play, infants perform actions from memory on an object that is not typically used for that purpose. For example, they may pretend a banana is a telephone by using it as a receiver which they talk into. Pretense requires children to generalise their knowledge of an action sequence to a new situation, a similar situation to that used in deferred imitation paradigms that are designed to tap generalisation.
39For young children, pretense is governed by the degree of perceptual similarity between the real and pretend object. In Striano et al. (2001), 2- to 3-year-old children observed the experimenter performing actions with toys, and were then allowed to play with the toys. All age groups imitated more of the target actions with toys that were replicas of those typically used to perform those actions (such as a toy car) than with inanimate natural objects (such as a twig), or instrumental objects (such as a pair of scissors). Indeed, the youngest children only showed evidence of imitation with the replica objects. Thus, the perceptual similarity of the test objects to the objects traditionally used to perform that function was an important variable for the children. Research with deferred imitation paradigms, both from the generalisation and pretend play literatures, thus supports the contention that perceptual similarity is an important cue to categorisation in the early years.
40If generalisation results from forgetting, infants should exhibit greater levels of generalisation after longer retention intervals. Imitation research in fact reveals the opposite. Although 12-month-old infants generalise across a change in the colour of the puppet after a 10-minute delay, after a 24-hour delay they only show evidence of memory with the demonstration puppet (Hayne et al., 2000). Similar delay effects on generalisation are seen with older infants. For example, Herbert and Hayne (2000) show that when 18-month-old infants are shown a sequence of actions (like making a rattle), they imitate the target actions with either the original toys, or with a novel but related set when tested immediately. However, when tested after a 24-hour delay, infants only reproduce the target actions with the exact objects they had seen during the initial demonstration session. The strength of the memory trace created in the demonstration session therefore appears to affect the likelihood that novel stimuli will be placed in the same category. After short delays, when the memory representation is still very active, there is an increased likelihood that the novel stimuli will be grouped together as members of the same category. After 24 hours, when the memory representation is becoming weaker, infants may begin to categorise the new stimuli as completely unrelated. Consistent with this argument, research with other paradigms has demonstrated that categorisation is time-dependant (for review see Rovee-Collier, 1995). Alternatively, it may be that contextual factors are important. The infant may categorise the two puppet stimuli as belonging to the same group through encountering them both in the same context. Context is likely to change gradually over time (whether through changes in the infant’s internal state, or through changes in the physical appearance of the testing room). Thus, over longer retention intervals, the context of testing may change more than over short retention intervals (for a theory of forgetting in terms of context change, see Bouton et al., 1999a, b; but also see Riccio et al.,1999). The context in which testing occurs is likely to change more from one day to the next than over an interval of 10 minutes. Indeed, at 14-months generalising to a toy of a novel size and colour is impeded by a shift in context between demonstration and test (Barnat et al., 1996). Furthermore, giving 12-month-old infants a distinctive, unique context for the demonstration and test sessions facilitates their ability to generalise after a 10-minute delay (Jones and Herbert, 2008).
41Decreases in generalisation over increasing retention intervals may thus reflect changes in the way infants categorise objects. This may be driven by changes in the strength of the representation of the stimulus itself, or of the context in which it appears. Both explanations imply that in order to place two objects in the same category, the representations of both must be active at the same time. However, it could be argued that categorisation is not required here. If the infant can use the presence of the new puppet to retrieve the memory of the original puppet, they should also remember the target actions. They may thus perform the target actions on the novel puppet not because they have formed a category and are generalising across it, but simply because their memory for the target actions has been cued. Indeed, categorisation training does not facilitate generalisation to a novel puppet in 12-month-old infants (Jones and Herbert, 2008). In this study, infants were shown the target actions with two different puppets, and then tested with a third. This manipulation was designed to facilitate category formation, but performance by this group of infants was no better than that of infants who only saw one puppet during the demonstration session. It is possible that this reflects the fact that generalisation does not rely on categorisation processes.
42This can be partially addressed by considering whether it is possible for generalisation and specific recall to co-exist. Showing that they do co-exist would provide evidence that generalisation does not result from either forgetting or indiscriminate application of the target actions. A recent study by Lechuga et al. (2001) has demonstrated that generalisation can co-occur with recall of specific features of the toys used in the demonstration session. This contradicts explanations of generalisation that rely on forgetting. One week after the demonstration, 25-month-old infants were able to imitate the target actions with both the same stimuli used in the demonstration and with an analogous set of novel stimuli. In a follow-up experiment, infants were allowed to choose between two stimulus sets during the test session. When given a choice between the analogous novel stimuli and a set of unrelated distracters, infants treated the analogous stimuli as if they were the same as the original demonstration set. However, when given the choice between the stimuli they had seen one week previously, and the analogous set, infants treated the analogous stimuli like unrelated distracters. This demonstrates that despite robust memory for details of the toys used during the demonstration session, 25-month-old infants still generalise those actions to novel toys.
43In summary, there is good evidence to believe that studies of generalisation can provide insight into categorisation processes in infancy. Such research reveals that for young infants, and infants tested after long retention intervals, categorisation is heavily dependent on physical similarity. Further insight comes from examining the results of studies that aim to facilitate generalisation. First, we can explore what kinds of information help infants to generalise their knowledge. Second, we can consider the implications this has for theories of categorisation and generalisation in infancy. The final section addresses both these points, using two examples from the imitation literature which have been consistently identified as facilitating generalisation: the provision of verbal cues, and physical experience with the target objects.
44Language may have particular relevance to studies of categorisation. When learning words, infants must learn to group objects together in order to apply the same verbal label to all members of the group. Giving the same label to two different objects may thus lead infants to place those objects in the same category. In Herbert and Hayne (2000), 18- and 24-month-old infants were given a verbal label for a set of stimuli during the demonstration. At the 24-hour test, infants were prompted with the same verbal label they had heard during the demonstration session. For 24-month-olds, being given a verbal label facilitated generalisation performance such that infants performed the same number of actions with both the original and novel toy sets. However, at 18-months the verbal label had no effect. While 18-month-olds showed good evidence of memory when tested with the demonstration stimuli, when they were given a novel stimulus set no evidence of generalisation was seen, irrespective of whether a verbal label was provided.
45Why could the 18-month-old infants tested in Herbert and Hayne (2000) not exploit the verbal retrieval cues provided? Between the ages of 18- and 24-months language acquisition is rapid (Fensen et al., 1984; Nelson, 1996), so limited verbal ability may prevent 18-month-olds from using the labels as a cue to categorisation. Alternatively, it may be that verbal labels are only effective when infants show some spontaneous generalisation. In Herbert and Hayne’s study, the 24-month-old infants showed some evidence of generalisation without the verbal label, whereas the 18-month-olds did not. It may be that infants need some idea of the correspondence between the toy sets before they can exploit the additional information provided to them by a verbal label.
46Indeed, research from the pretend play literature provides evidence that the ability to use language to promote generalisation develops gradually, and is in part dependent on children’s ability to spontaneously see the connection between objects. Elder and Pederson (1978) explored the development of pretend play in 2 ½-, 3- and 3 ½-year-old children. In each trial, children were given an object and asked to pretend that it was something else. In the ‘Similar’ condition, the objects they were given were perceptually similar to the object they were pretending to use. In the ‘Dissimilar’ condition, the objects used for pretense looked different and had a different function from the real object. For example, in one trial the real object was a comb. In the Similar condition, children were given a flat piece of wood, and in the Dissimilar condition they were given a rubber ball to use as a comb. In the ‘No Object’ condition, children were asked to pretend with no object support. The 2 ½-year-old children performed best in the ‘Similar’ condition, where the appearance of the objects concorded with the verbal instructions the children were given. Interestingly, they performed better in the ‘No object’ condition than in the Dissimilar condition, where the verbal instructions contradicted the children’s previous behavioural experience with the objects they were given. Thus, in the youngest age group their ability to use verbal information to generalise a sequence of actions to a novel toy depended in part on the perceptual appearance of the toy used. In contrast, the 3 ½- year-olds performed equally well in each condition, showing increased sophistication in using verbal information to control behaviour.
47Practicing the target actions in a deferred imitation task can also facilitate performance of those actions on a novel set of toys. While 18-month-old infants can spontaneously generalise across small changes in the stimuli used in deferred imitation tasks (Hayne et al., 1997), larger changes preclude retrieval (Hayne et al., 1997, Experiment 4; Herbert and Hayne, 2000). However, when given the opportunity to practice the target actions prior to the generalisation test, 18-month-olds can generalise across large changes in the toys used (Bauer and Dow, 1994; Barr and Hayne, 2000). Practice can also be effective with younger infants. Learmonth et al. (2004) used the puppet task to explore generalisation of imitation in 6- to 12-month-old infants. When 9-.and 12-month-old infants were allowed to practice the actions immediately after they were demonstrated, they were subsequently able to perform the actions on a novel puppet after a 24-hour delay. In contrast, infants who were not given the opportunity to imitate immediately showed no evidence of generalisation. As with verbal labels, there are age-related changes in the facilitative effect of physical practice: 6-month-old infants could not generalise to the novel puppet after 24 hours, even when they were allowed to practice immediately.
48In deferred imitation paradigms, verbal labels and physical experience facilitate generalisation. This concurs with results from other paradigms that show that object names and object functions serve as important cues to categories for infants. For example, 12- and 13-month-old infants increase their attention to perpetual similarities between objects when they are labelled with a novel name (Waxman and Markow, 1995). Similarly, agent-produced motion (a simple object function) increases attention to other perceptible object features (Booth, 2001). In a familiarisation and forced-choice paradigm, 14- and 18-month-old infants form categories more readily when given functional information and verbal labels than in their absence (Booth and Waxman, 2002). Research with the deferred imitation task thus identifies similar parameters to other paradigms in studies of categorisation, and provides a way to study these variables that is closely related to how infants typically learn.
49In conclusion, the last twenty years of research with the deferred imitation paradigm has provided considerable insight into developmental changes in memory during the infancy period. In less than a two year period of development there are dramatic increases in the infant’s ability to maintain and retrieve their memory representations over the long-term and to use this knowledge in a flexible manner when details of the context or stimulus have changed. The imitation paradigm is by no means the only task suitable for investigating changes in cognitive development or brain maturation across the infancy period. For example, recent advances in electrophysiological methods such as EEG and optical imaging hold considerable promise for understanding developmental change when paired together with our current behavioural techniques. Observational learning, however, remains an ecologically valid measure of the infant’s learning and memory processes. Infants learn 1-2 new behaviours every day simply by observing those around them (Barr and Hayne, 2003). Social species make considerable use of this learning method throughout the lifespan, whenever new situations are encountered and there are others around who are more knowledgeable that can be learnt from. Infants in particular, are highly motivated to engage in this type of learning even when the person they are observing is a complete stranger. It remains to be determined whether the same findings will be observed when the demonstrator is highly familiar to the participant, or how learning might occur when there are cross-species learning opportunities.