Sub-parametric features of head movements and gaze conveying epistemicity: a study on French Sign Language (LSF) and French co-speech gestures. ; : Mouvements de la tête et du regard qui expriment l'épistémologie en LSF et dans les gestes ; : Movimenti della testa e dello sguardo che esprimono epistemologia nella LSF e nei gesti

International audience ; The LexIKHuM project [1] aims at developing a human-machine communication system based on a kinesthetic lexicon inspired by natural human gestural communication: co-speech gestures [CSG] and sign languages [SL]. As we would like to enable the AI machine to communicate its degree of certainty with regard to the message delivered, we studied the kinesthetic properties of epistemic gestures.Previous studies have identified some markers of the epistemic gesture, such as rapid head nods to express certainty (in German and Turkish SL; in French, English and Catalan CSG [2 3 4 5]) or slow head tilts to express uncertainty. However, these studies have only looked at the epistemic gesture as a whole and are focused on SL or CSG. In the LexIKHuM project, we analyze the sub-parametric properties of these gestures, searching for common features in SL and CSG. We assume that (a) some properties of the movement carried in the 3 degrees of freedom [DoF] of the head are involved in the construction of epistemic gestures in SL and CSG (such as the amplitude of the flexion/extension movement or the duration of the abduction/adduction movement of the neck); and (b) that the direction of gaze associated with the head movements is also relevant to express epistemicity. To investigate these hypotheses, we examined head movements and gaze positions in epistemic gestures in DEGELS [6], a parallel corpus of French CSG and French SL (LSF).We followed a protocol inspired by the prosodic analysis of SL and kinesiological studies [7 8]. We manually (a) identified the epistemic sequences of that corpus (with inter-annotator agreement) using ELAN; (b) transcribed head and gaze movements using the Typannot transcription system [9] on a sample of 40 epistemic sequences; and semi-automatically (c) generated the measures of the head movements in the different DoFs with AlphaPose software (see an example in Figure 1).Our preliminary studies reveal that (a) gestures formed in a context of certainty show a greater amplitude ...