Résumé
The complexity of studying co-verbal manual gestures lies in the challenge of formalizing its occurrences descriptively and functionally. Attempts to cataloggestural schemas and roles have so far proven unsatisfactory, as they often rely on heterogeneous or study-dependent criteria. Annotating those aspects of gestures is time-consuming, complex—particularly due to the dynamic aspects involved—and prone to approximations, making it challenging to accurately correlate these diverse features to define gestures. To address these limitations, we conducted a multidisciplinary study aimed at (1) identifying necessary and sufficient aspects for precisely characterizing gestural features and (2) automating the qualification of these aspects to ensure more objective descriptions and facilitate annotation processes. This work is part of the SYNCOGEST ANR project (2025-2029) which aims to improve gesture generative systems by incorporating formal and semantic annotations into the training of neural networks. This presentation addresses the preliminary results of this study with a focus on the formal aspects we identified, highlighting six complementary modalitiesdeemed necessary and distinctive: manuality, trajectory, location, hand configuration, speed, and size. Except for manuality, which remains a methodological challenge, all these aspects have been automated or are in the process of being automated. Motion capture data enables the application of various technological approaches to determine these elements, including vector calculations, angular analysis, position point detection, body measurements, and other relevant methods. While preliminary distinctions can be established within each aspect, our ongoing analysis of observable combinations seeks to build reliable and distinctive formal classes. These classes are designed to (1) facilite the study of relationships between gestures and speech and (2) enhance gesture recognition and generation in neural systems.