Multidimensional Labeling of Gesture in Communication: the M3D Proposal

Abstract

Communication is multimodal in that speakers use not only their voices, but also co-speech gestures to communicate. Recent insights suggest that gestural behavior has a strong association with prosodic structure and that a single gesture can communicate various semantic and pragmatic meanings. This highlights the importance of developing a comprehensive, flexible, and transparent approach to gesture annotation that accounts for multiple dimensions of gesture, including a gesture’s form, prosodic properties, and semantic and pragmatic contributions. To address this need for an increasingly dimensionalized approach to multimodal data annotation, the main goal of this paper is to present and describe a novel labeling system for manual gestures. The MultiModal MultiDimensional (M3D) system consists of an open access package that has been developed in coordination with five different labs working on gesture and its interaction with speech. The package includes a set of reliable annotation guidelines, a validated training program, and two annotated audiovisual corpora that represent over 60 minutes of lecture-style speech.