A One-Day Workshop at ACL 2010 on Companionable Dialogue Systems
Association for Computational Linguistics (ACL) conference website: http://acl2010.org/
Uppsala, Sweden, July 15, 2010
Important Dates:
April 5, 2010
Submission Deadline, authors to workshop organizers, in ACL format, following instructions on ACL site
May 6, 2010
Notification of acceptances to authors
May 16, 2010
Camera-ready paper due from authors to workshop organizers
Jun 13, 2010
Compiled CDROM and proceedings images due from workshop book chairs to ACL-2010 publication chairs
The current state of dialogue technology has come a long way since the abstract ideas of intelligent computer systems in the beginning of 1950s: dialogue technology now provides interactive service agents, while research explores various aspects of multimodal and multiparty communication so as to improve natural and social aspects of dialogue systems. In this workshop, interest is focussed especially on dialogue systems able to act as Companions, i.e. software agents with advanced human language technology capacities, able to display and recognise emotion and aspects of personality as well as to interact with a user, possibly over a long period, learning about their needs and interests, performing services, entertaining, consoling, and so on. Our focus will be on text and speech aspects of dialogue with Companions, but we shall also be open to discussion of the impact of non-dialogue phenomena (e.g. presence, low-level control and recognition, avatar technology etc.) on dialogue and the interaction of other modalities with dialogue.
Dialogue technology has two main historical sources, both still of relevance today: first, the chatbot tradition going back at least to ELIZA and PARRY and, secondly, the task-driven knowledge-based system back to at least BASEBALL and SHRDLU, all these examples being from the late 1960s. The great chatbots of the past, which bear little relationship to the current rash of Internet products, did have some claim to companionableness of a sort, and e.g. PARRY had explicit emotion parameters of fear and anger that affected its outputs. The chatbots sometimes left initiative with the user (like ELIZA which initiated nothing) and sometimes with the system (like PARRY, who had long paranoid stories to tell if given a chance). The task based systems, however, aimed at efficient task completion with little attention paid to the social or emotional aspects of interaction. The initiative was always with the user and the system was regarded as a tool or servant with no goal other than to answer or carry out a task as efficiently it could.
The two kind of systems gave rise to quite different forms of evaluation as well: the chatbots led to the sophisticated but artificial “Turing Test” environment of the Loebner competition, while the funded and deployed task-systems – of which the best known were the MIT airline reservation systems like PEGASUS and JUPITER – were evaluated in competitions in terms of time and completion of task rates. However, comparison between systems and their performance proved difficult; no generally applicable and agreed evaluation framework or methodology is available for the companionable systems we are interested in this workshop. We shall ask in the workshop whether it is possible to measure companionship, and if so, is it possible to include some aspects of it in the evaluation of dialogue systems?
Although both the chatbot and task-based traditions began as text-only systems, they were able to take advantage of the rapid advances in speech technology, and fuse speech and language research increasingly. However, none of this led to any obvious advance in what is the goal of this workshop: the exploration of research advances in dialogue systems able to act as Companions. It seems clear that both early traditions have much to contribute to the goal of a Companion, and that it cannot be founded exclusively on either alone.
Advances have been expected and achieved in pursuing the overall goal of a Companions in recent years: from the increasing sophistication of ASR and language generation, their integration with NLP and with higher-level issues of emotion and dialogue control; from more sophisticated dialogue management models; from a range of deployable theories of emotion that can be connected directly to text and speech; from the use of new techniques of content extraction in dialogue (such as Information Extraction) and, like every other part of language technology, from the steady advance of machine learning techniques and associated evaluation methods.
In this workshop we want to explore promising new methods to design and evaluate dialogue systems able to act as Companions, as well as to report and review recent advances in a wide range of Companion-related topics. The workshop also aims at being a forum for focussed discussion of what it is to give a convincing and useful illusion of “personality” in a long-term Companion, when that is advantageous and when not and, above all, what is the precise role of language and speech technology in achieving this.
We solicit papers on topics related to the overall notion of companionableness in dialogue systems as defined above, with the emphasis on the Human Language Technology aspects of their function. We seek contributions that deal with issues such as:
- Augmentation of speech recognizers and synthesizers for the detection and provision of emotion, affect and attitude in speech input and output.
- Integration of speech and NLP methods for the extraction of content in companionable systems
- Dialogue management systems that support realistic guidance of dialogue by attitude, empathy, emotion etc.
- Methods and approaches to learn realistic dialogue managers
- Representation and reasoning to support such companionable systems (forms of predicate logic, NL, RDF)
- Applications of companionable dialogue systems for domains such as health, learning, security, care of the vulnerable and lonely etc.
- Novel methods for evaluating (companionable) dialogue systems.
- integration of vision and other non-verbal sensors in dialogue understanding.
- Use of non-verbal behavior of avatars and physical effectors to express dialogue and affective content.
We also look for more theoretical contributions that address questions such as:
- How much and what kind of personality do users want in companionable dialogue systems?
- Are there genuinely new advances in machine learning applicable to dialogue?
- What are the social and ethical consequences of the appearance and deployment of such systems?
- Is there a real tension between task-based approaches and other paradigms?
- Is emotion core or peripheral to such companionable dialogue systems?
- Can IE methods be successfully deployed in the extraction of content in dialogue?
- Have chatbots anything to teach us in the goal of creating companionable systems?
- What is the best metaphor for describing such systems: companions, confidants, intelligent dialogue buddies, guides, escorts, assistants, colleagues?
- Does NL yield the most “affordances” for such companionable systems, as opposed to multi-modal methods?
- How far can such systems offer real company, real relationships in the future?
- How much are politeness and emotion matters of NLP and computational linguistics, rather than some separate discipline or area?
- The perennial problem of the provision of dialogue data and how much is needed for effective machine learning?
- How close are such systems to embodiment in mobile applications?
- Could a Companion be a collective or agency of entities, with separate personalitoes between which a user could choose?
- Need a dialogue Companion be a full conversational participant or sometimes just a “foil” or “echo”, who does not carry the whole conversation entities (thus raising issues of multi-party dialogue acts)?
Invited Speaker:
David Traum of the Institute of Creative Technology, USC will be the invited speaker
Submission Information:
Papers should be submitted via the ACL submission system: http://www.softconf.com/acl2010/CDS/
All submissions are limited to 6 pages (including references) which should be formatted using the ACL 2010 style file that can be found at: http://acl2010.org/authors.html.
As the reviewing will be blind, papers must not include the authors’ names or affiliations. Submissions should be in English and should not have been published previously. If essentially identical papers are submitted to other conferences or workshops as well, this fact must be indicated at submission time.
The submission deadline is 23:59 CET (i.e. time in Paris) on April 5, 2010.