Capítulo 3. Las Redes Sociales, Transnacionalismo y Migración Tunkaseña
3.1. Enfoques estructuralistas de la Migración Internacional
The research reviewed so far was devoted to building general models of gesture use, i.e., systematic inter-personal patterns of gesture use are incorporated exclusively. What has not yet been considered in these systems is individual or group-specific variation, another line of research which has recently been making headway. In the following, approaches will be reviewed that provide means to customize gesture behavior.
Modulating Expressivity Parameters Based on the observation that gesture use is subject to considerable inter-individual differences, Hartmann et al. (2006) focused on the way individuals differ in the manner and execution of their gestures. Based on perceptual studies, six expressivity parameters of gesture quality were extracted as an intermediate level of behavior parametrization between holistic, qualitative communicative functions such as mood, personality, and emotion on the one hand, and low-level animation parameters like joint angles on the other hand:
– Overall Activation General amount of activity, e.g., passive vs. static or ani- mated vs. engaged.
– Spatial Extent Amplitude of movements. – Temporal Extent Duration of movements.
– Fluidity Smoothness and continuity of overall movement, e.g., smooth versus jerky.
– Power Dynamic properties of the movement, e.g., weak versus strong. – Repetition Tendency to rhythmic repeats of specific movements.
These parameters were applied to the Gesture Engine of the virtual agent GRETA, whose library of known prototype gestures are tagged for communicative function (Hartmann et al., 2002). The parameterization approach allows for a large range of pre-defined gesture variants. See Figure 3.8 for an example of variation of the parameter ‘spatial extent’.
The expressivity parameter approach is driven by a perceptual standpoint, i.e., the question of how expressivity is perceived by humans. To investigate how the six parameters were recognized by human users, Hartmann et al. (2006) employed an evaluation study. It turned out that the recognition of single parameters was best for the dimensions ‘spatial extent’ and ‘temporal extent’, whereas the dimensions ‘repetition’ and ‘overall activation‘ were much less recognizeable.
Recently, Mancini and Pelachaud (2010) implemented the concept of expressivity parameters to create distinctive behavior patterns for ECAs. Their proposed algorithm generates nonverbal behavior for a given communicative intention and emotional
Figure 3.8: Variation of the parameter ‘spatial extent’ in the virtual agent GRETA: The neutral key pose in
the middle is contracted in the left and extended in the right gesture execution. Reprinted from Hartmann et al. (2006).
state, driven by the agent’s general behavior tendency (‘Baseline’) and modulated by dynamic factors such as emotional states, relation with interlocutor, physical constraints, social roles. etc. For each modality (face, gesture, torso, head), the Baseline structure includes a modality preference and specifies values for the six expressivity parameters. The Baseline of a person can be automatically extracted from a video analysis, and then modulated by the agent’s ‘Dynamicline’, which influences the its behavior at two levels: the selection of multimodal signals to display and the specification of the behavior execution quality. In an evaluation study, human subjects were able to discern the achieved distinctiveness in the agent’s behavior.
Rehm et al. (2008) presented another variant which made use of the gestural expressivity parameters to generate culture-specific gestures (CSG for culture-specific generation). Their approach to CSG in embodied agents relies on a multimodal corpus analysis of human interactions in two cultures. The analysis of corpus data focuses on gestural expressivity in terms of the above-mentioned expressivity parameters depending on a speaker’s cultural background. Differences between the cultures can be identified and integrated in a probabilistic model for generating agent behaviors. In a Bayesian network, the culture to be simulated is connected with five dimensions of culture as a middle layer: hierarchy, identity, gender, uncertainty, and orientation. These culture-specific dimensions were then connected with gesture parameters.
This model was applied in an application called ‘cultural mirror’. As a user’s gestural expressivity (measured with a WiiMote) was analyzed and the classification result was set as evidence for the output nodes of the Bayesian network. By diagnostic inference, the user’s cultural background was estimated and this information was then set as evidence to the input nodes of a second network in which an agent’s behavior was parameterized via causal inferences, resulting in behavior congruent to user input.
Providing virtual agents with gesture style In another approach, Ruttkay (2007) aimed at endowing virtual humans with a unique style in order to appear typical of some social or ethnic group. The focus of this work was a markup language to define different aspects of style which were handcrafted to model the behaviour of stereotypic groups rather than individual group members. The markup language GESTYLE allows the generation of speech and accompanying gestures by tagging a text for meaning and declaring a style in which an utterance will be performed. To this end, GESTYLE contained the following set of constructs:
– Character Markup Defines the static characteristics of a character, such as culture, personality, age, sex, and other individual characteristics like handed- ness.
– Situation Markup Specifies the situation by setting dynamical aspects of the speaker (like mood, physical state) and the environment (social relation between interlocutors, characteristics of the addressee etc.)
– Communicative Markup The text to be uttered by the agent is tagged, e.g., for emphasis, get-attention.
– Gesture Markup Specifies the gesturing behavior to be expressed at certain points in time. Specific parameters are provided to modify the characteristics of the motions, e.g., amplitude, motion-manner.2
A set of gesture dictionaries was compiled for particular cultures, social, ethnic groups, and personalities or even for individual subjects. In these dictionaries, one or more gestures are stored to express a particular meaning to offer the alternatives of expressing a communicative function. Each gesture entry is augmented by the probability of using the specific gesture in this role and optional gesture modifying parameters specifying the motion characteristics of the gesture.
The effect of high-level character and situation parameters on the motion charac- teristics is given in terms of low-level gesture parameters. That is, a single gesture is selected from the possible alternatives, prescribed by different gesture dictionaries according to the given character and situation specification. Provided with an interface which translates the GESTYLE representation to the control parameters of an ani- mated player, it is possible to give expression to this specification in the overt behavior of a virtual agent.