Multimodal Corpus Collection for the Design of User-Programmable Robots

Abstract

In order to design a robot able to learn from its user, the example of card game instruction is investigated. A detailed description of a setup for the collection of a human-to-human multimodal instruction corpus is given. This setup uses touch screens and will also provide a base for the human-robot instruction interface to be designed. Preliminary results on learning dialogues are given and issues of corpus transcription, annotation and analysis are discussed. 1