As we move towards systems that are more plausibly conversational in nature we are faced with questions and issues that concern how to manage a dialogue. The last 30 years has seen a range of approaches to dialogue systems from the simplicity of Eliza (Weizenbaum 1966) ---originally a network but which we would probably now manage with a few simple finite state rules--- on up to high functionality multi-modal systems. As we look at the design of dialogue systems over many years, we can ask if the underlying components are now basically agreed and the rest is notational variant or taste in logic? If we are in that state, it may be only a sign of maturity, as in Information Extraction (IE) when Hobbs argued successfully about 1992 that all IE systems were then basically isomorphic.
On one hand, the notions of adaptation and learning have become important issues when working with spoken interactive systems, and on the other hand, machine learning research has matured so as to provide tools and techniques for system designers and developers to build adaptive and learning systems. The goal of the workshop is to provide a forum for discussing how these aspects can be combined in spoken dialogue systems, addressing especially such questions as what kind of adaptation and learning is necessary, desirable, and possible for speech interface systems, and how the new learning techniques can help in achieving these goals. An important issue is also to clarify the notion of user-centered design when building flexible and adaptive systems: as it is not only the user who should adapt to the (limits of the) system properties, the question that system builders and designers must address is where and when the system needs to adapt to the user. User modelling is thus an essential part of interaction management in intelligent systems, and one of the workshop aims is to investigate the limits and relation of the user models to interaction models in general. Furthermore, there are various types of architectures and frameworks that have been proposed to accommodate adaptation and learning aspects into spoken language interfaces, and the workshop aims to share experience, as well as successful and unsuccessful solutions.
The workshop will address the unique requirements that the adaptive and learning view-points pose for dialogue research, interface design and system development. It offers a platform for discussion of Dialogue Systems - a topic that is becoming increasingly prominent in the field of Natural Language Processing, and in particular, it will focus on two related issues:
1) The adaptation and learning in intelligent interactive systems and
the techniques and applications that are needed to realize such
systems.
2) The approaches to Dialogue Management.
In particular, whether Dialogue Management Systems (DMS) are
now generic and if not whether there are real differences in approach.
We solicit papers from a number of research areas, including but not limited to:
- How can adaptive techniques be used in spoken dialogue systems?
- How effective is learning for Dialogue systems?
- What kind of limitations and challenges are there for applying machine
learning research in practical dialogue systems?
- Is Dialogue Management language independent?
- What kind of representations and type of information are necessary for
adaptation?
- What kind of requirements are there for the architecture and general
development framework of adaptive systems?
- Can we always effectively distinguish the roles of dialogue management
components and general architectures for dialogue and language processing?
- Are all dialogue systems the same underneath the surface or are there
fundamental differences to approaching Dialogue Management?
- Are there any specific requirements for adaptation in mobile
environment?
- What is the role of user modelling and integration of user models in
dialogue systems?
- What are future challenges for adaptive and learning systems?
The following thematic sessions are planned: Adaptation and learning in spoken dialogue systems Styles of dialogue management: are they really different?
The papers must be no longer than 8 pages, including title page, examples, references, etc. In addition to this, two additional pages are allowed as an appendix which may include extended example discourses or dialogues, algorithms, graphical representations, etc.
The papers should be sent electronically to the e-mail address: eacl-w6@dcs.shef.ac.uk
The paper format is the same as used in EACL submission. Style files are available at Style files Papers must be submitted in pdf format.
The title page (no separate title page is needed) should include the following information:
-Title
-Authors' names, affiliations, and email addresses;
-Preferred thematic session, if any.
-Abstract (short summary up to 5 lines);
Yorick Wilks (co-chair)
University of Sheffield
Computer Science Dept.
Regent Court
Sheffield S1 4DP
UK
email: yorick@dcs.shef.ac.uk
Björn Gambäck
SICS, Swedish Institute of Computer Science AB
Box 1263
SE - 164 29 Kista, Sweden
email: gamback@sics.se
Bill Black
UMIST
Computation Department
P.O. Box 88
Sackville Street
Manchester M60 1QD
UK
email: wjb@co.umsit.ac.uk
Roberta Catizone
University of Sheffield
Computer Science Dept.
Regent Court
Sheffield S1 4DP
UK
email: roberta@dcs.shef.ac.uk