{Are fillers helpful to fulfill a flawless classification of Dialog Acts?

MOHAMED BEN LASSOUAD

Abstract: The ability to classify Dialog Acts (DA) from a conversation can revolutionize our understanding of conversations and enable bots to respond appropriately. In language learning, it is essential to recognize the fillers and expressions used by speakers to buy time while they think about what to say next and how to articulate it in order to sound fluent. All languages, including sign languages, have their own unique fillers, and spotting them seems straightforward for a learning algorithm using deterministic rules. However, can we predict Dialog Acts by analyzing neighboring utterances of fillers? Our research suggests that the benefit gained from this approach is minimal and that utterances expressed after the fillers are more informative than those before the fillers.