Dialogue Moves and Relations Between Them

The notion of dialogue move (dialogue act) originates with Austin's ans Searle's speech act theory, according to which communication is not only transmitting information but also acting, i.e., changing the world. Many of these changes are to the mental states of conversation participants or the state (or context) of dialogue [Traum1999]. Dialogue moves/acts are used as labels for particular combinations of changes that result from interpreting utterances. Various taxonomies of dialogue moves/acts have been proposed, and used for dialogue annotation and modelling, e.g., the DAMSL taxonomy and its variant for the Switchboard spoken dialogue corpus SWBD-DAMSL, the Verbmobil taxonomy, the BE&E taxonomy for tutorial dialogues, etc.

In human-human conversation, one can observe patterns of typical responses, and even differentiate betweem preferred (expected) and dispreferred (unexpected) responses. For example, a question expects an answer (but what counts as an answer?), a statement expects acceptance (but what if the party can't agree?), etc. How do we systematically dicover and model the appropriate ways to respond to given inputs? And how does a system deliver a dispreferred (unexpected) response in a polite/natural/cooperative way?

Reading: [Traum1999]