Learning to classify the feedback function of head movements in a Danish Corpus of first encounters

Abstract

This paper deals with the automatic classification of feedback by head movement in the Danish NOMCO corpus, a collection of dyadic interactions in which speakers get to know each other for the first time. The results show that by using a combination of features related to the shape of head movements and facial expressions, together with features of the words these gestures are related with, good results (an F-score of 0.72) are achieved in distinguishing head movements used to express feedback from those that serve a different communicative function. Moreover, the distinction between feedback give and feedback elicit, can also be learnt with very good accuracy (an F-score of 0.913), although this result should be taken with caution due to the fact that one of the behaviours is much more dominant than the other in the corpus.