Abstract Tom Gurion October 2019

From IMC wiki
Jump to: navigation, search

Title: Comparing models of head nods for speakers and listeners


This talk presents a study that compares theories of head nods to head movement data in natural dialogues. Pairs of participants conversed freely while fitted with movement caption system. Their head movement was analyzed for head nods and these were compared to predictions by models inspired by backchannels and mimicry theories, as well as predicted listeners response to speech disfluencies. The results suggest that speakers nod more frequently than listeners. The models that are designed specifically for listeners (backchannels and disfluency models) perform better for listeners than for speakers, while mimicry based models don't differentiate between speakers and listeners. Overall, the models explain only a small portion the data, suggesting that other models (or theories) might explain most of the observed head nods.