Figures & data
FIGURE 1 Example gestures from different speakers, each referring to the same u-shaped building.
TABLE 1 Example Transcript from the Corpus
TABLE 2 Coding Scheme for Gestures and Their Discourse Context
FIGURE 2 Example of intersubjective variability: gesture handedness in placing gestures across five speakers (N = 70).
FIGURE 3 General structure of a GNetIc decision network. Gesture production choices are considered either probabilistically (chance nodes drawn as ovals) or rule-based (decision nodes drawn as rectangles). Each choice depends on a number of contextual variables. The links are either learned from corpus data (dotted lines) or defined in a set of if-then rules (solid lines).
FIGURE 4 Schematic of the gesture formulation process (see text for details).
FIGURE 5 Examples from different speakers each describing a round window of a church (first row) and GNetIc-generated behavior simulating those speakers in the same initial situation (second row).
FIGURE 6 Network structures obtained with different algorithms (columns) for different data sets (rows). See Table 2 for the reading of the variables and their value sets.
TABLE 3 Prediction Accuracy of Single Features for the Networks Learned with Different Structure Learning Algorithms from Individual and Combined Speaker Data, Respectively
FIGURE 7 Network structures obtained with NPC algorithm for two different speakers (left, right): link strength corresponds to significance level.
