•The literature on facial imitation development has largely focused on the visual modality because facial actions are perceptually opaque, and it appeared that visual-domain based routes provide the ...only avenue for development.•Visual-centric, learning-based models of facial imitation development rely on experience in the dyadic context, resulting in implications that have attracted criticism.•When one considers the auditory consequences of facial expressions, it is clear that spectro-temporal features of vocalisations change as a consequence of changes in the mouth cavity shape and lip opening (e.g. when someone smiles).•Applying this facial-expression–acoustic-consequence link to current learning-based models of mimicry results in the possibility of two different mechanisms of audio-centric models of facial imitation development for orolabial actions (involving mouth and lips).•These new models allow for solo, no-dyadic experience to be sufficient for the imitation of many orolabial actions, making this route non-susceptible to many criticisms that have levied against the visuo-centric models.
The question of how young infants learn to imitate others’ facial expressions has been central in developmental psychology for decades. Facial imitation has been argued to constitute a particularly challenging learning task for infants because facial expressions are perceptually opaque: infants cannot see changes in their own facial configuration when they execute a motor program, so how do they learn to match these gestures with those of their interacting partners? Here we argue that this apparent paradox mainly appears if one focuses only on the visual modality, as most existing work in this field has done so far. When considering other modalities, in particular the auditory modality, many facial expressions are not actually perceptually opaque. In fact, every orolabial expression that is accompanied by vocalisations has specific acoustic consequences, which means that it is relatively transparent in the auditory modality. Here, we describe how this relative perceptual transparency can allow infants to accrue experience relevant for orolabial, facial imitation every time they vocalise. We then detail two specific mechanisms that could support facial imitation learning through the auditory modality. First, we review evidence showing that experiencing correlated proprioceptive and auditory feedback when they vocalise – even when they are alone – enables infants to build audio-motor maps that could later support facial imitation of orolabial actions. Second, we show how these maps could also be used by infants to support imitation even for silent, orolabial facial expressions at a later stage. By considering non-visual perceptual domains, this paper expands our understanding of the ontogeny of facial imitation and offers new directions for future investigations.
Conformity--defined here by the fact that an individual displays a particular behavior because it is the most frequent the individual witnessed in others--has long been recognized by social ...psychologists as one of the main categories of social influence. Surprisingly, it is only recently that conformity has become an active topic in animal and comparative biology. As in any new and rapidly growing field, however, definitions, hypotheses, and protocols are diverse, not easy to organize in a coherent way, and sometimes seriously in conflict with one another. Here we pursue greater coherence by reviewing the newer literature on conformity in behavioral ecology and evolutionary biology in light of the foundational work in social psychology. We suggest that the knowledge accumulated in social psychology can be exploited by behavioral ecologists and evolutionary biologists to bring conceptual clarity to the field, avoid some experimental pitfalls, and help design new and challenging experiments. In particular, we propose that the notions of "informational" and "normative" conformity that, until now, have been little recognized in recent literature can resolve some important controversies. In turn, research on animal culture should be of great interest to social scientists, because understanding human culture and human uniqueness requires an evolutionary analysis of our cognitive capacities and their evolutionary origins. Our review suggests excellent opportunities for social and natural scientists to join forces in building an interdisciplinary and integrative approach to the pervasive phenomenon of conformity. (Contains 7 footnotes, 1 tables, and 2 figures.)
•In this paper, we introduce two different underlying mechanisms of over-imitation.•During blanket copying, the necessity of demonstrated actions is not questioned.•When the irrelevance of an action ...is obvious, we deliberately consider whether or not to copy.•We argue that blanked copying is based on type 1 processing, which is less cognitively demanding.•Deliberate considerations whether or not to copy, based on type 2 processing, require more cognitive resources.
Children and adults tend to imitate actions that are causally irrelevant to accomplishing a goal; they “over-imitate.” It has been discussed that humans over-imitate either because of erroneous causal reasoning, meaning that they do not recognize demonstrated actions as being irrelevant, or because of social motivation, for example, because they want to follow a norm or affiliate with the demonstrator. Recent findings give reason to believe that neither of these accounts alone is sufficient to explain this phenomenon. Here, we introduce a dual-process framework that integrates existing explanatory accounts for over-imitation and indicates when which of these accounts is applicable. According to this integrative framework, over-imitation can be based on different underlying motivations. We introduce two distinct processes: First, blanket copying, which is based on type 1 processing that is less cognitively demanding. When children perform blanket copying, they imitate irrelevant actions without questioning their necessity. Second, deliberate considerations whether or not to copy, based on type 2 processing that requires more cognitive resources. When children are aware that some of the demonstrated actions have no functional purpose, they can deliberately consider whether or not to copy. Then, it depends on the children’s goals in a given situation whether they over-imitate or not. Which of these two processes is activated depends on whether children are informed about or recognize the functional irrelevancy of demonstrated actions. Different irrelevant actions vary in how easily they are recognized as being functionally irrelevant. Thus, we propose that some actions rather lead to blanket copying and others to deliberate considerations whether or not to copy. Therefore, we propose a categorization system for irrelevant actions, intended to help unravel contradicting findings on over-imitation.
Unlike other animals, human children will copy all of an adult's goal-directed actions, including ones that are clearly unnecessary for achieving the demonstrated goal. Here we highlight how social ...affiliation is key to this species-specific behavior. Preschoolers watched 2 adults retrieve a toy from a novel apparatus. One adult included irrelevant actions in her demonstration; the other only used actions causally related to opening the apparatus. After both adults took turns demonstrating, 1 left the test room, and the remaining adult gave the apparatus to the child. Children reproduced the irrelevant actions only when given the apparatus by the adult who had demonstrated them, even though the departed adult's actions emphasized how unnecessary these redundant actions were. Our results highlight the specialized skills for participating in cultural groups that have evolved in humans and provide insight into why finding such high fidelity copying in other animals has proven elusive. (Contains 1 table and 1 footnote.)
Recent research suggests that we can simultaneously represent the actions of multiple agents in our motor system. However, it is unclear exactly how concurrently observed actions are represented. ...Here, we tested two competing hypotheses. According to the independence hypothesis, concurrently observed actions are represented as independent actions. According to the compound hypothesis, they are instead integrated, whenever possible, into compound actions. In Experiment 1 (N = 32), we first show that the standard imitation-inhibition task with a single hand can be extended to measure automatic imitation of compound actions. In Experiments 2-5 (NTotal = 368), we then investigated the representation of concurrently observed actions by further extending this task to include two hands. The results showed that two hands performing two different actions (e.g., one hand lifts index finger, one hand lifts middle finger) produced an effect similar to that of both hands performing just one of those actions (e.g., both hands lift index finger) but different from that of both hands performing both actions together (i.e., a compound action; lift both index and middle finger). This indicates that concurrently observed actions are coded as independent actions in the motor system.
Public Significance Statement
This study shows that concurrently observed actions are not integrated into compound actions, suggesting that the actions of different agents are processed independently in the motor system.
Recent Advances in Robot Learning from Demonstration Ravichandar, Harish; Polydoros, Athanasios S; Chernova, Sonia ...
Annual review of control, robotics, and autonomous systems,
05/2020, Volume:
3, Issue:
1
Journal Article
Peer reviewed
Open access
In the context of robotics and automation, learning from demonstration (LfD) is the paradigm in which robots acquire new skills by learning to imitate an expert. The choice of LfD over other robot ...learning methods is compelling when ideal behavior can be neither easily scripted (as is done in traditional robot programming) nor easily defined as an optimization problem, but can be demonstrated. While there have been multiple surveys of this field in the past, there is a need for a new one given the considerable growth in the number of publications in recent years. This review aims to provide an overview of the collection of machine-learning methods used to enable a robot to learn from and imitate a teacher. We focus on recent advancements in the field and present an updated taxonomy and characterization of existing methods. We also discuss mature and emerging application areas for LfD and highlight the significant challenges that remain to be overcome both in theory and in practice.
Social norms are communally agreed upon, morally significant behavioral standards that are, at least in part, responsible for uniquely human forms of cooperation and social organization. This article ...summarizes evidence demonstrating that ritual and ritualized behaviors are essential to the transmission and reinforcement of social norms. Ritualized behaviors reliably signal an intentional mental state giving credibility to verbal expressions while emotionally binding people to each other and group-based values. Early ritualized infant-caregiver interactions and the family routines and rituals that emerge from them are primary mechanisms for transmitting social norms vertically from parent to offspring, while adult community rituals are a primary mechanism by which norms are reinforced horizontally within the community.
Research has shown that people automatically imitate others and that this tendency is stronger when the other person is a human compared with a nonhuman agent. However, a controversial question is ...whether automatic imitation is also modulated by whether people believe the other person is a human. Although early research supported this hypothesis, not all studies reached the same conclusion and a recent meta-analysis found that there is currently neither evidence in favor nor against an influence of animacy beliefs on automatic imitation. One of the most prominent studies supporting such an influence is the study by Liepelt and Brass (2010), who found that automatic imitation was stronger when participants believed an ambiguous, gloved hand to be human, as opposed to wooden. In this registered report, we provide a high-powered replication of this study (N = 199). In contrast to Liepelt and Brass (2010), we did not find an effect of animacy beliefs on automatic imitation. However, we did find a correlation between automatic imitation and perceived self-other similarity. Together, these results suggest that the gloved hand procedure does not reliably influence automatic imitation, but interindividual differences in perceived similarity do.
In sequential decision-making, imitation learning (IL) trains a policy efficiently by mimicking expert demonstrations. Various imitation methods were proposed and empirically evaluated, meanwhile, ...their theoretical understandings need further studies, among which the compounding error in long-horizon decisions is a major issue. In this paper, we first analyze the value gap between the expert policy and imitated policies by two imitation methods, behavioral cloning (BC) and generative adversarial imitation. The results support that generative adversarial imitation can reduce the compounding error compared to BC. Furthermore, we establish the lower bounds of IL under two settings, suggesting the significance of environment interactions in IL. By considering the environment transition model as a dual agent, IL can also be used to learn the environment model. Therefore, based on the bounds of imitating policies, we further analyze the performance of imitating environments. The results show that environment models can be more effectively imitated by generative adversarial imitation than BC. Particularly, we obtain a policy evaluation error that is linear with the effective planning horizon w.r.t. the model bias, suggesting a novel application of adversarial imitation for model-based reinforcement learning (MBRL). We hope these results could inspire future advances in IL and MBRL.