Inspired by the goal of modeling the dialog state and the speaker's mental state, moment by moment, we apply Principal Component Analysis to a vector of 76 prosodic features spanning 6 seconds of context. This gives a multidimensional representation of the current state, and we find that word probabilities do vary strongly with several of these dimensions, that the use of this information in a language model gives a 15% reduction in perplexity, and that the dimensions do relate to aspects of mental state and dialog state.
Index Terms: prosody, context, principal component analysis, perplexity, dimensions, dialog activities
Bibliographic reference. Ward, Nigel G. / Vega, Alejandro (2012): "Towards empirical dialog-state modeling and its use in language modeling", In INTERSPEECH-2012, 2314-2317.