Incremental Dialogue Processing
David Schlangen · 2011
Spoken language unfolds in time, and is understood and generated in a continuous process: when I speak spontaneously, I don’t plan full sentences which I then merely ‘read out’, and you don’t have to wait for me to finish my utterance before you can start to think about and react to it. This may seem obvious, and yet many branches of linguistics, for different reasons, abstract away from these continuous processes and focus on the sentence as their unit of analysis. In this talk, I will briefly review the evidence for incremental processing, and will then describe a model of such processing that we have recently developed (Schlangen & Skantze, EACL 2009 / Dialogue & Discourse 2011), and two implementations of the model in example dialogue systems (Skantze & Schlangen, EACL 2009; Bus, Baumann & Schlangen, SIGdial 2010), and discuss what we have learned from these implementations.