In most public speaking datasets, judgements are given after watching the entire performance, or on thin slices randomly selected from the presentations, without focusing on the temporal location of these slices. This does not allow to investigate how people’s judgements develop over time during presentations. This contrasts with primacy and recency theories, which suggest that some moments of the speech could be more salient than others and contribute disproportionately to the perception of the speaker’s performance. To provide novel insights on this phenomenon, we present the 3MT_French dataset. It contains a set of public speaking annotations collected on a crowd-sourcing platform through a novel annotation scheme and protocol. Global evaluation, persuasiveness, perceived self-confidence of the speaker and audience engagement were annotated on different time windows (i.e., the beginning, middle or end of the presentation, or the full video). This new resource will be useful to researchers working on public speaking assessment and training. It will allow to fine-tune the analysis of presentations under a novel perspective relying on socio-cognitive theories rarely studied before in this context, such as first impressions and primacy and recency theories. An exploratory correlation analysis on the annotations provided in the dataset suggests that the early moments of a presentation have a stronger impact on the judgements.
11 November 2023 in Journal on Multimodal User Interfaces
Transactive memory system (TMS) is a team emergent state representing the knowledge of each member about “who knows what” in a team performing a joint task. We present a study to show how the three TMS dimensions Credibility, Specialisation, Coordination, can be modelled as a linear combination of the nonverbal multimodal features displayed by the team performing the joint task. Results indicate that, to some extent, the three dimensions of TMS can be expressed as a linear combination of nonverbal multimodal features. Moreover, the higher the number of modalities (audio, movement, spatial), the better the modelling. Results could be used in future work to design human-centered computing applications able to automatically estimate TMS from teams’ behavioural patterns, to provide feedback and help teams’ interactions.
The Proteus effect describes how users of virtual environments adjust their attitudes to match stereotypes associated with their avatar's appearance. While numerous studies have demonstrated this phenomenon's reliability, its underlying processes remain poorly understood. This work investigates deindividuation's hypothesized but unproven role within the Proteus effect. Deindividuated individuals tend to follow situational norms rather than personal ones. Therefore, together with high embodiment and identification processes, deindividuation may lead to a stronger Proteus effect. We present two experimental studies. First, we demonstrated the emergence of the Proteus effect in a real-world academic context: engineering students got better scores in a statistical task when embodying Albert Einstein's avatar compared to a control one. In the second study, we tested the role of deindividuation by manipulating participants' exposure to different identity cues during the task. While we could not find a significant effect of deindividuation on the participants' performance, our results highlight an unexpected pattern, with embodiment as a negative predictor and identification as a positive predictor of performance. These results open avenues for further research on the processes involved in the Proteus effect, particularly those focused on the relation between the avatar and the nature of the task to be performed. All supplemental materials are available at https://osf.io/au3wk/.
There is a growing need for public speaking skills in both professional and private life. With this background, our research project’s long-term aims are to develop tools that can analyse public speeches and provide helpful feedback. The impact of audio and visual characteristics on the automatic analysis of speech quality has been widely explored in the existing literature. However, only a few studies have focused on textual features. In response to this shortcoming, this paper investigates the importance of textual content for the automatic analysis of public speaking. We created an open-source Python library of textual features and integrated them as inputs of simple machine-learning models for automatic public-speaking analysis, and persuasiveness prediction, in particular. The best result (accuracy of ) is obtained using a logistic regression. We then evaluated the impact of these features on persuasiveness prediction using both correlation analysis and Explainable AI methods. This evaluation was conducted on the French data set 3MT_French, including student performances in the "Ma Thèse en 180 Secondes" competition.