Multimodal Japanese corpus of multi-party conversation on two different topic types

We collected a multimodal corpus of three-party conversation in Japanese (mother tongue) and in English (second language) to compare the basic characteristics of conversation, such as utterance duration and silence, for two topic types: free-flowing and task-oriented. The corpus contains video and audio data of eighty conversations (four hundred eighty minutes in total) with speech duration, upper body motion, and eye gaze activity labels. This article reports on the corpus collecting process and the results of preliminary analyses.