Word accent and emotion

In this paper, we address the question whether prosodically/linguistically prominent syllables carrying the word accent (stressed syllables) are better indicators for emotional marking than unstressed syllables. To this aim, we use a large spontaneous database with children interacting with Sony's Aibo robot, annotated with word-based emotion labels, large acoustic-prosodic feature vectors, and support vector machines as classifiers. It turns out that, in most of the cases, stressed syllables are better emotion markers than unstressed syllables. Moreover, we discuss specific phenomena such as vocatives and other constellations, to be modelled in future studies.