Learning a Motor Grammar of Iconic Gestures

Amir SadeghipourCITEC - Sociable Agents Group, Bielefeld, NRW, Germany
Stefan KoppCITEC - Sociable Agents Group, Bielefeld, NRW, Germany

Abstract

In this paper, we present a computational investigation into the compositionality of iconic gestures by trying to learn a motor grammar. We propose a grammar formalism that learns (1) the salient, invariant features of single movement segments (motor primitives) and (2) the hierarchical organization of these segments in complex gesturing. The formalism is applied to a dataset of natural iconic gestures. The extracted structure reveals compositional patterns of iconic gesturing.

Files

Learning a Motor Grammar of Iconic Gestures (3.3 MB)



Back to Table of Contents