Automatic ordinal prediction of children’s participation in social play settings using nonverbal vocal features

In social play, young children can exhibit different types of participation. Some children are engaged with other children in the play activity while others are just looking. In this study, we investigated methods to automatically predict the children’s levels of participation in play settings using nonverbal vocal features (which have been shown to correlate with related phenomena such as engagement and dominance). Rather than labelling the level of participation in an absolute manner, as has frequently been done in previous related studies, we designed an annotation scheme that takes the order of children’s participation levels into account. Taking full advantage of the ordered annotations, we explored the use of SVM-based ordinal learning, i.e., ordinal regression and ranking, and compared these to a rule-based ranking and a classification method. We found promising performances for the ordinal methods. Particularly, the ranking method demonstrated the most robust performance against the large variation of children and their interactions.