We are happy to announce that we have released the dataset for dialogue breakdown detection challenge 3 (DBDC3). https://dbd-challenge.github.io/dbdc3/
You can download the data at: https://dbd-challenge.github.io/dbdc3/data/
The dataset contains human-machine dialogues in both English and Japanese: 615 English dialogues and 150 Japanese dialogues. All system utterances are annotated with dialogue breakdown labels by 30 annotators.
The data can be useful to those working on dialogue systems as well as those who want to analyze psychological aspects of dialogue; e.g., how users perceive human-machine interaction.
If you have any questions, please contact us at: dbdc3-organizers at googlegroups.com
DBDC3 organizers Ryuichiro Higashinaka, Kotaro Funakoshi, Michimasa Inaba Yuiko Tsunomori, Tetsuro Takahashi, Nobuhiro Kaji
*Acknowledgments* We thank our sponsors, Denso IT Laboratories, Nxtremer, Honda Research Institute Japan, and NTT DOCOMO, Inc., for supporting our data collection. We also thank the Japanese Society for Artificial Intelligence (JSAI) for supporting the event.
-- Ryuichiro Higashinaka NTT Media Intelligence Laboratories. NTT Corp. 1-1 Hikarinooka, Yokosuka, 239-0847 Japan. phone: +81-46-859-2027 fax: +81-46-855-1054