Workshop Chair: David Traum

Background
There are many different dialogue activities that people (and ultimately virtual people) can engage in. Differences in number of participants, roles and relationships of the participants, task, context, and other factors will lead to very different kinds of dialogues in terms of observable features such as length, words and linguistic constructions used, speech acts performed, and patterns of interaction. A first distinction to make is between universal language principles vs domain-specific rules. "Universal" principles are obviously attractive as they will be applicable beyond a particular domain, however some domain-specific knowledge will always be needed, and domain-specific heuristics may be more accurate, quicker, and/or tractable than general purpose reasoning for domains where they generally hold. In this workshop, we will go beyond this dichotomy between general and domain-specific, and investigate the similarities between sets of domains, which are different from other sets, and rules that generalize over a genre, rather than being limited to a specific domain. For example, responding to a query in an information-seeking domain may be a simple matter of a database retrieval. However, the same query in a negotiation domain may involve an additional calculation of whose benefit it would be to deliver a correct answer, and whether there might be better alternatives (e.g., a self-serving fabrication, a refusal to answer, or bargaining to receive something in return for the answer).

Workshop Goals
The main goal of the workshop will be the investigation of useful methodology for

  • Identifying and classifying dialogue genres, with a focus on the ability to make simplfying assumptions that are not universial, but mainly hold for the genre, allowing opimized processing of dialogues within a genre.
  • developing or utilizing software architectures and algorithms that are well suited to genre-specific processing
  • automatic, on-line detection of genre
  • development of artificial agents that can engage in multi-genre interactions (while exploiting genre-detection and genre-specific processing)

Workshop Format
The format of the workshop will include some talks invited presentations on aspects of dialogue genres, as well as group discussions on the above and related topics. Discussions will involve theoretical and taxonomic issues as well as "hands-on" analysis of dialogue corpora.

Contributions
The workshop will not include archival publication of new papers. To attend the workshop, please see the "to attend" page.

We invite several kinds of contributions to the workshop, including

  • Presentation of Dialogue Genre studies
  • Presentation of data sets of a specific domain or multiple domains
  • Demonstrations of systems that engage in genre-specific interaction or multiple-domains
  • Presentations of genre-specific inference rules, algorithms, or architectures
  • discussion or break-out group topics
  • links to or other bibliographic info on publications concerning dialogue genres
Contact traum@ict.usc.edu