Gossip is More than Just Story Telling Topic Modelling and Quantitative Analysis on a Spontaneous Speech Corpus

Gossip is one of the most widespread human activities with multiple functions such as enhancing human cooperation, establishing social order, information sharing, norm enhancing or stress reduction. Gossip has been analyzed mostly by qualitative or survey methods. In this paper, we describe a quantitative approach to identify gossip in a large corpus containing spontaneous talk with LDA topic modeling and quantitative analysis. We aim to identify gossip and its characteristics to analyze its topics, the verbal and non-verbal emotions that were used during gossiping, and other non-textual data such as the number of speakers and the number of persons present during the gossiping events. We also analyze the topics to distinguish gossiping and storytelling by dividing gossip and non-gossip texts in our large spontaneous speech corpora.