Introduction: The Role of Goal Understanding in Memory Research
Research concerning adult event representation has revealed a remarkable body of evidence regarding the organizing role of goals. Investigations of different repre-sentational forms of events, such as narratives (Mandler & Johnson, 1977), observed action sequences (Lichtenstein & Brewer, 1980), and abstract knowledge of routine behaviours (Schank and Abelson, 1977), all share the basic assumption that adults appear to impose the interpretative framework of goal-directed action on the human behavior they encounter. The main knowledge structures that are based on goals support segmenting of the continuous flow of action sequences into actions with boundaries (like episodes in text reading - Black & Bower, 1979), identifying relevant knowledge (like scripts - Schank and Abelson, 1977), and establishing valuable hierarchical organization in memory (like plans - Lichtenstein & Brewer, 1980). Thus goal-based organization seems to play a primary role in adult event representations.
The problem of developmental continuity in the organization of event represen-tations has initiated numerous studies investigating older children's event repre-sentations. As early as the age of 4, children display event representations that incorporate information about temporal order, causal relations, and goals: their verbal reports on earlier events are skeletal but include only such elements that adults would recount (Nelson & Gruendel, 1986). Thus, children's event memory resembles adult event memory with respect to goal-based organization, even if they use less complex forms of representations (Fivush, Kuebli, & Clubb, 1992; Nelson
& Fivush, 2000). Nevertheless, much less evidence is available for younger children, such as non-verbal toddlers.
Support for early organized event representations comes from an experiment in which irrelevant elements were inserted into novel action sequences. The elicited
114 Ildikó Király
imitation study of Bauer & Mandler (1989), revealed the assumption that causal-enabling relations between event components facilitate organization to entail better retrieval of action sequences. Infants (of 16 and 20 months) could retrieve novel events imitating their components while omitting or displacing their irrelevant components if there were causal relations between their event steps. Causal-enabling relations in an action determine the only meaningful tem-poral order of the event components that lead to the targeted outcome, and as such can support the memory organization of it. Events and actions in the world, however, do not always possess such inherent, enabling temporal organization;
there are events with unbound temporal relations among their goal-relevant com-ponents. For instance, when someone would like to make a cup of cocoa, it is up to the actor's habit whether he or she puts milk or cocoa powder into the cup first, before mixing them to attain the very same tasty drink. According to Bauer and Mandler (1989), the improvement in retrieval of events containing enabling relations in comparison to events lacking such inherent structure was evidenced by superior ordered recall. This suggests that toddlers are sensitive to temporal irreversibility. Furthermore, causal structure as a source of information on temporal organization enhances memory tracing. An alternative explanation for the better "recall" of events containing enabling relations is that planning on the basis of goal-state configuration is enough for the reconstruction of these types of events, since goal information in itself can guide the threading of related event components.
In a further study, to test whether the improved recall performance of events containing enabling relations was due to problem solving rather then retrieval from memory, Bauer Schwade, Wewerka, and Delaney (1992) presented infants (of only 20 months) with the goal-state configuration both of enabling event sequences and of event sequences lacking such structure (the action sequences necessary to reach the goals were not demonstrated). The performance of infants after they had been encouraged to produce the entire event was poor; they rarely demonstrated the target sequences, and they were no more successful with enabling sequences than they were with arbitrary sequences. Somewhat in contradiction to this result, Bauer et al. (1999) showed that 20- and 27-month-old infants were able to use goal-state information to support their planning attempts in the case of novel enabling events, providing evidence on the assumption that goal-state configuration has a central organizational role, though in a problem-solving context.
A possible solution to the above puzzle is raised by a study by Carpenter, Call, and Tomasello (2002). The authors have shown that prior exposure to the end-state or outcome configuration of an action sequence, followed by full modelling of the target action sequence, results in superior performance in imitation of event com-ponents as compared to exposure just to the full modelling of the event. This result confirms that goal information plays a central role in the interpretation and encod-ing of events, and that it is a major factor in the organization of events for later retrieval, as prior information on the end-state, of the event facilitates the monitor-ing of event components.
In a task that required infants to represent relations between temporally sepa-rated actions and their converging structure, in which multiple actions served to enable a single outcome, Travis (1997) was able to show that 24-month-olds were capable of representing and imitating elements of an event in relation to its goal-based hierarchical structure. In particular, in the case of an event with embedded goal-irrelevant steps (in which two otherwise independent actions enabled a third action), infants grouped actions related to a common goal temporally, and repro-duced goal-relevant action more than goal-irrelevant actions. The results clearly prove that 2-year-olds are able to represent converging causal structure, which is a characteristic of goal-directed action organization.
The importance of the early availability of goal-based organization can be appre-hended in the fast parsing and identification of event types, and probably in forming general event representations. Despite the fact that there is empirical evidence that goal information plays a central role in the organization of event representations in the first years too, in the domain of memory development the notion of teleologi-cal stance is not widely appreciated. Teleologiteleologi-cal stance - an early interpretative schema for action coding proposed by Gergely and Csibra (1998) - is a convenient frame for guiding the perception and encoding of relevant, adequate ("real") com-ponents of events, even after only one brief exposure to them. The significance of this model lies in its power to clarify the central role of goal information in action representations through describing the inference structure and basic mechanism of interpretation as mediated and triggered by the rationality principle.
The aim of the present chapter is to introduce the consequences and experimen-tal implementation of the model of teleological stance in the domain of imitative learning and memory development.
Early Interpretative Scheme for Action Understanding:
The Teleological Stance
The theory of the teleological stance is based on the results of a series of habituation studies (Csibra, Biro, Koós, & Gergely, 2003; Csibra, Gergely, Bíró, Koós, &
Brockbank 1999; Gergely, Nádasdy, Csibra, & Biro, 1995) that demonstrated goal attribution in 9- and 12-month-olds. These studies pointed out that by at least 9 months of age infants can (a) attribute goals to observed actions; (b) do so even if the agents are unfamiliar abstract entities that lack human features; (c) evaluate the relative efficiency of the goal approach in relation to the situational constraints on actions; and (d), if the relevant environmental constraints change, expect the agent to modify or change its means action adaptively to achieve efficient goal attainment in the new situation (Csibra et al., 1999, Csibra et al., 2003; Gergely et al, 1995).
Findings since 2000 confirmed that even 6-month-olds are able to interpret an ongoing action within the frame of the teleological stance: at this age infants are willing to attribute goals to humans and human-like robots (Kamewari, Kato,
116 Ildikó Király
Kanda, Ishiguro, & Hiraki, 2005) and to any kind of inanimate object if it appears to be able to vary its goal approach (Csibra, forthcoming).
To account for these findings, Csibra and Gergely proposed that infants are equipped with an abstract and domain-specific action interpretation system, the teleological stance (Csibra & Gergely, 1998; Gergely & Csibra, 1998, 2003). Briefly, the teleological stance is a representational system that relates three kinds of ele-ments in a specific type of (teleological) explanatory structure: (a) action: the observed behaviour, (b) goal: the consequent change of state in the world, and (c) situational constraints: the relevant aspects of the situation that constrain actions leading to the goal. An essential component of the teleological stance is the "prin-ciple of rational action". This prin"prin-ciple is responsible for (a) driving inferences about goal-directed actions and, at the same time, for (b) providing criteria of well-formedness for teleological action interpretations. The importance of the rationality principle is rooted in the piece of evidence that it can guide the selection of goal-related (in contrast to ungoal-related), or goal-relevant (in contrast to goal-irrelevant) acts, as it can guide the online assessment of the ongoing action sequence. The mechanism of continuous evaluation by the rationality principle allows us to predict the outcome of an ongoing action just by assuming that it is a "direct way toward"
an end-state or outcome or (in this case obviously) a goal.
Teleological Stance and Imitation: The Selective Interpretative Nature of Imitative Learning in Human Infants
In the domain of memory development, Meltzoff (1988) has demonstrated that infants are able to re-enact — that is, retrieve novel actions - after a one-week delay;
in other words, infants are able very early on imitatively to learn novel means actions by way of observing others. In the most impressive task of the above men-tioned study, 14-month-olds watched as a human model leaned forward from the waist and touched the top panel of a light box with her forehead, thereby illuminat-ing it. A week later, 67 per cent of the infants re-enacted the novel "head action,"
while none performed it in a base-line control group for whom the action was not demonstrated. This result was an obvious indicator of long-term memory retention in infants for a specific event. Alternatively, this result seemed unexpected from the point of view of the 1-year-old's teleological stance (Csibra and Gergely, 1998), since, on the grounds of this model, one would have expected that in this task infants, as rational agents, should have performed the most efficient goal-directed action available to them (using their hand to contact the light box), instead of imitating the unique, but less efficient, "head action."
To clarify this situation, Gergely, Bekkering, and Király (2002) performed a modified version of Meltzoff' s task (1988). They hypothesized that "if infants noticed that the demonstrator declined to use her hands despite the fact that they were free, they may have inferred that the head action must offer some advantage in turning on the light. They therefore used the same action themselves in the same
situation" (Gergely et al., 2002, p. 755). To test this idea, Gergely and colleagues tested two groups of 14-month-olds varying the situational constraints of the model.
In the "hands-occupied" condition, the model's hands were visibly occupied: she pretended to be chilly and wrapped a blanket around her shoulders, holding it with both hands while performing the "head action." In the "hands-free" condition, however, after wrapping the blanket around her shoulders, the model placed her hands onto the table, so that they were visibly free, before demonstrating the "head action."
When the model's hands were occupied, 14-month-olds were less likely to imitate the "head action" (21 percent). Instead, they illuminated the box by touching it with their hand, performing the simpler, and equally effective, emulative response available to them, but not to the model. In contrast, when the model's hands were free, but she still used her head to illuminate the box, 69 per cent of 14-month-olds imitated her "head action" (p < 0.02) (this result is a replication of Meltzoff 's results of 1988). So, differential imitation in the two conditions suggests that imitative learning is not an automatic "copying" process invoked by identification with the human actor, nor is it due to automatic behavioral "copying" of the modeled action.
Rather, imitative learning is a selective interpretative process that involves the evalu-ation of the revalu-ationality of the means in relevalu-ation to the situevalu-ational constraints of the actor. Thus, re-enactment of the novel means takes place only if (a) the action is judged as rational given the situational constraints of the model, and (b) the action is judged as more rational than other available alternatives given the situational constraints of the infant him or herself.
With their recent model of human pedagogy, Gergely and Csibra (2005) shed new light on the constraints of teleological action interpretation. Their theory's main argument is
that Mother Nature's "trick" to make fast and efficient learning of complex - and, for the learner, cognitively "opaque" - cultural knowledge possible was to have humans evolve specialized cognitive resources that form a dedicated interpersonal system of mutual design in which one is predisposed to "teach" and to "learn" new and relevant cultural informa-tion to (and from) conspecifics, (pp. 471-472)
A fundamental statement of their argument is that expert humans who possess cultural knowledge are disposed not only to use, but also ostensively to manifest, their knowledge to inexpert conspecifics, and inexpert conspecifics are specially receptive to ostensive communicative manifestations of others.
Briefly, the model of Human pedagogy outlines three major constituents that serve pedagogical knowledge transfer. First, there is a design specification that an expert conspecific (a "teacher") ostensively communicates her cultural knowledge by manifesting it for the novice (the "learner") with the help of referential cues (such as eye contact, turn-taking contingency). Secondly, because of her special kind of receptivity, the learner is predisposed to interpret the teacher's ostensive-communicative cues that accompany her knowledge manifestation as evidence
that the manifestation will convey new and relevant cultural information for her.
As a consequence, this allows fast learning of the communicated content without any further need to test its relevance independently. Thirdly, the built-in presump-tion of relevance of pedagogically communicated knowledge manifestapresump-tions also enables the acquisition of knowledge contents that are arbitrary, conventional, and causally/functionally non-transparent, which stand for many forms of cultural knowledge.
The selective imitational findings of the Gergely et al. (2002) study is a nice example of how pedagogy operates: how infants infer differentially in two conditions what is new and relevant information for them (see also the argument of Gergely &
Csibra, 2005). In the "hands-occupied" condition of the head-on-box study, the novel outcome, including the presented property of the object (illuminability-upon-contact), is the new information, so it is going to be retained in memory and repro-duced through action. Taking the teleological stance in this case, infants can infer that, given the physical constraints of the actor (hands occupied), touching the box with her forehead is justified as a sensible and efficient means to the goal, as the physical-causal efficiency of the "head action" is cognitively "transparent" here.
In the "hands-free" condition the situation is different. The goal state involving the newly experienced affordance of the box is new information here, too, so it will be reproduced. In contrast, when setting up a teleological interpretation as to what particular action would constitute the most rational/efficient means to the goal under these situational constraints, given the fact that the actor's hands were free, the infant must have identified the available "hand action" as the most efficient means to perform. Unexpectedly, however, the demonstrator chose not to use her free hands, but performed the unusual "head action" instead. This contrastive choice marked the "head-action" as new and relevant information that the ostensive-communicative manifestation conveyed. As a result, both the new goal and the new means were retained and imitated.
The head-on-box study (Gergely et al., 2002) from this perspective confirms that pedagogical cues are necessary factors for imitative learning in human infants, although (1) there is interpretative selectivity guiding what aspect of the modeled behavior will be imitatively learned, and (2) this is directed by the implicit assump-tions of the infant's "pedagogical stance" - namely, that the observed individual is about to manifest "for" them some significant aspect of cultural knowledge that will be new and relevant.
Relevance-Guided Selective Imitation:
Verbal Labels Serving Human Pedagogy
Regarding the domain of memory development, it is a normal and common feature of imitational paradigms that the target event is presented in a rich ostensive context, comprising communicative-referential speech acts and overt verbal instruc-tions (for example, "Look, I'll show you something!") before the target action is
demonstrated. Such speech acts or short verbal instructions (labels) could enrich any situation in a Gricean sense, letting the observer perceive the actor's/speaker's intent of presenting something new and relevant. These assumptions are directly analogous, if not identical, to the Gricean pragmatic assumptions of ostensive com-munication, as made obvious by Sperber and Wilson (1986). From a slightly dif-ferent perspective, however, pedagogy is a primary adaptation for cultural learning that is not necessarily conscious but a cue-dependent fast-learning attitude, and not a specialized module dedicated to the recovery of the speaker's intent in linguistic communication, an assumption that has evolved later as a sub-module of human theory of mind (Sperber and Wilson, 2002). Mentalistic terms are not necessary for conceiving the understanding of relevance. Speech acts are ready to convey the intention of the speaker. At the same time, however, these verbal acts appear as part of the external situation: they convey intents or goals through setting up in advance a possible end-state and thus highlighting a possible goal as part of the external situation. This alternative goal can either correspond to or mismatch with the outcome/goal of the ongoing action sequence. Thus verbal labels can enhance the understanding of the ongoing action in terms of its goal: if the verbal act is in line with a specific end-state achieved in the following action (consistent with its
"intention-in-action"), it helps the encoding of the relevant steps of the event sequence; if the uttered goal or intent and the goal of the action sequence do not overlap completely, it can alter the encoding of what is relevant in the situation by setting up a goal hierarchy.
Verbal instructions, as part of the communicative referential context (the exter-nal situation), can serve as overt articulation {manifesto) of what to learn (what is relevant) in the situation. For instance, a closer look at our example "Look, I'll show you something!" reveals that this verbal act can imply that manifestation of new and relevant information in general is a goal of the agent. The main purpose of a further study was to examine this supposition. To do this, we manipulated and controlled the verbal labels used in the experimental situation. Our aim was to investigate the role of verbal labels in highlighting the relevant features of the ongoing actions that are "to be encoded." Our hypothesis was that verbal labels pragmatically referring to the manifestation of new and relevant information bring about imitative learning of cognitively opaque subevents, while verbal labels that accord with the presented event sequence and end-state (that refer to a script) result
Verbal instructions, as part of the communicative referential context (the exter-nal situation), can serve as overt articulation {manifesto) of what to learn (what is relevant) in the situation. For instance, a closer look at our example "Look, I'll show you something!" reveals that this verbal act can imply that manifestation of new and relevant information in general is a goal of the agent. The main purpose of a further study was to examine this supposition. To do this, we manipulated and controlled the verbal labels used in the experimental situation. Our aim was to investigate the role of verbal labels in highlighting the relevant features of the ongoing actions that are "to be encoded." Our hypothesis was that verbal labels pragmatically referring to the manifestation of new and relevant information bring about imitative learning of cognitively opaque subevents, while verbal labels that accord with the presented event sequence and end-state (that refer to a script) result