By Giuseppe Di Fabbrizio

After the first public announcement in July 2006, the DiSCoH (Spoken Dialogue System for Conference Helpdesk) system has been answering phone calls from all over the world. The automated helpdesk provides information about the first IEEE/ACL Spoken Language Technology Workshop 2006 in Aruba. Many users called the toll free phone number (+1 888 681 5290) and learned about the conference, including paper submission, technical program, information about the venue, accommodation options, workshop registration and costs, etc. More than 500 calls have been received up to date. One of the most popular requests was the paper submission status which was available on DiSCoH a week earlier than the official email notification. Authors were able to find out their paper status over the phone where more than 160 unique paper rankings were promptly presented.

In this first stage of the project, the goal is to collect a spontaneous human-machine conversational corpus in the framework of goal-oriented spoken dialogue systems. The data will include audio files, transcribed utterances, labeled dialogue acts and reward signal annotations for reinforcement learning purposes. Such corpus covers many research interesting topics and is (1) usable for training, evaluating and comparing statistical models, (2) naturally spoken and task oriented, (3) extendible/generalizable, (4) collected using state-of-the-art research and commercial technology, (5) freely available to researchers. The next phase of the project will generalize the data-drive approach to the conference automation task in the context of other international conferences, exploit dialogue ontology learning, as well as  a more comprehensive approach to  interactive question/answering.

The DiSCoH system is based on the most current data-driven approaches to spoken dialogue system and has been developed by researchers at AT&T Labs, Inc. - Research, ICSI (International Computer Science Institute), and the Edinburgh University and it was partially sponsored by NSF, IIS 0624389, and the TALK project, EC IST 507802.  It is physically located in the AT&T Labs - Research building in Florham Park, NJ and connected through an ISDN trunk to a network grade AT&T VoiceTone^® platform.

If you have a Skype Out account, you may call DiSCoH from your desktop for free by clicking on the image below. Just make sure that you have a good quality microphone or preferably a headset.

For more information, contact info [@] discoh [.] org