Overview of Sigdial 2009 Conference
Svetlana Stoyanchev
SLTC Newsletter, October 2009
September 11 - 12, Queen Mary College hosted Sigdial in London, for the first time, as a conference. This year Sigdial had 125 participants and 104 submissions. There were 24 oral and 24 poster presentations, 3 demos, and two invited speeches. This article gives an overview of the topics, invited talks, and awards presented at the conference.
Overview
The conference gathered dialogue researchers from all over the world: USA, Canada, Netherlands, Germany, UK, France, Japan, and others. The majority of papers presented this year were from North America.
Topics featured at the conference included work on spoken dialogue systems as well as analysis and modeling of human conversations. The majority of papers focused on spoken dialogue systems addressing the problem of speech recognition and user utterance interpretation. A few papers on dialogue systems focused on information presentation, generation, or giving useful feedback to improve system performance (including Paksima et. al).
One of the Sigdial focus areas this year was research on meetings, such as detection of noteworthy utterances, summarizing, and facilitating virtual meeting participants (including the papers of Akker et al, Niekrasz and Moore, Banerjee and Rudnicky, Bui et.al). Another focus area was multiparty and multimodal communication, such as a system assistant to virtual meeting participants or determining when a virtual meeting participant is being addressed. Research involving tutoring systems also had a significant presence at the conference (including Dzikovska et al, Rotaru and Litman, Dohsaka et al.).
The best paper award was presented to Dan Bohus from Microsoft for the work on a multi-modal system that predicts user engagement in a multi-party situation. The system plays the role of a virtual secretary in a lobby of a Microsoft building. Using video and audio input, image processing for detection of faces and their orientation, the system predicts whether a user should be addressed by a virtual secretary. You can view a video of the system in-action .
The best student paper award was presented to Luciana Benotti from Universite Henru Poincare, France for her paper on the clarification potential of instructions. In her work Luciana analysed clarification questions asked in a human-human task-oriented dialogue.
Invited Speakers
The invited speakers this year were Psychology professor Janet Bavelas from the University of Victoria and a Computer Science professor Yorick Wilks from the University of Shefield.
Janet Bavelas discussed the uniqueness of dialogue. She raised an interesting question 'what is dialogue?'. How much feedback from the audience/addressees does a speaker need in order to classify speech as a “dialogue”? In her experiments Dr. Bavelas uses three scenarios with different levels of audience feedback. The two dialogue scenarios involved a face-to-face communication with full multimodal feedback and a telephone communication with only audio channel feedback. The monologue scenario involved speech into a dictaphone with no addressee feedback. Dr. Bavela's experimental results show that gestures, figurative language, facial displays, and direct quotations are unique to dialogue situations with or without video feedback.
Professor Yorick Wilks described a companion project , a 4-year project that has completed its second year. A companion has speech recognition, speech synthesis, and internet connectivity capabilities. A companion can be a system that learns and records your experiences throughout your life. A companion has a capability to learn from pictures by verbal interaction with a user and build a user's illustrated family map. A companion may be used to entertain a lonely elderly person by pulling information from the web, news, or TV shows. A companion can be a positive motivator for a healthy lifestyle, such as good diet and exercise. A sample conversation with an automatic system companion can be viewed on Youtube. Professor Yorick described challenges and achievements of the ongoing Companion initiative.
Questions about the evaluation of spoken dialogue systems were, as always, a present topic at Sigdial. Researchers at Carnegie Melon announced an open Spoken Dialogue Challenge that could serve as an evaluating mechanism for dialogue system technologies. The Dialogue Challenge invites researchers to participate in building and evaluating dialogue systems in the bus information domain, providing information such as schedules and route information for the city of Pittsburgh's Port Authority Transit (PAT) buses. The researchers are given an option to build their own system or to use and modify the existing Let's Go bus information system. By taking advantage of the existing system, the researchers may save time on system building and would be able to focus on their specific research question. See more details on the Dialogue Challenge wiki page
Business Meeting Announcements
Participants made several suggestions for innovations in the future Sigdial conferences. One of the suggestions was to introduce mentoring/guidance to the authors who help foreign authors. Another idea proposed expanding future conferences to 3 days.
A new journal on Dialogue and Discourse was announced at the business meeting.
Sigdial's web page has moved to a new domain: http://www.sigdial.org. The new website is designed to be interactive and to facilitate discussion between researchers. Everyone is invited to register, participate in discussions on the topics of Corpora, Tools and Methodologies , Discourse Processing and Dialogue Systems, Semantic and Pragmatic Modeling, or to add new discussion topics.
For references, please see the online Sigdial proceedings.
Svetlana Stoyanchev is a research associate at the Open University, Computing Department. Her interests are dialogue systems, adaptation in dialogue, conversion of text to dialogue for presentation. Email: s.stoyanchev@open.ac.uk
Add A Comment