Home | People | Projects | Publications | Resources

The Multiparty Discourse Group

DAMSL | DAT | Selected papers

The Multiparty Discourse Group, a multi-site group of which the University of Rochester is a member, is part of the Discourse Resource Initiative (DRI). To quote from that page: "The Discourse Resource Initiative project is an effort to assemble discourse resources in support of discourse research." The goal of the Multiparty Discourse Group is to devise a common high-level framework of dialog acts. Groups could devise subtypes of the agreed upon ontology allowing flexibility while the common super types facilitate sharing of data and comparative studies across corpora. The Multiparty Discourse Group has met at the three DRI meetings. Below is the draft framework (before the third workshop) and an accompanying annotation tool (dat). Following these resources is information on the three meetings of the Multiparty Discourse Group.

Latest Version of Annotation Resources

Installing dat:

Dat requires Perl 5.003 (or higher), the Perl Tk package, and the Perl FileDialog widget. We are currently running dat with Perl 5.004_04 and Perl Tk 402.003. The pointers below correspond to this setup.

First meeting at UPENN (1996)

The second and third workshops involved a draft framework called DAMSL (Dialog Act Markup in Several Layers) primarily aimed at task-oriented dialogs. All workshops involved pre-workshop "homeworks" where participants tried annotating dialogs with various tag-sets. An annotation tool (dat) was developed for use in a DAMSL tagging project at the University of Rochester. dat was used in the homeworks for the second and third workshops.

Second meeting at Schloss Dagstuhl (1997)

Third meeting at Chiba University and Kazusa Academic Park (1998)

A new draft dialog act framework should be forthcoming as a result of this third meeting. Disagreements in coding homeworks led many participants to state that actual coding schemes complying to this standard should have mutually exclusive or low dimension tag sets unlike the current draft of this standard which has 15 independent dimensions of tags (an utterance could be tagged with a label from each of these dimensions). It may be difficult to train annotators to use a 15 dimension framework and data analysis becomes more difficult. You should keep this in mind if you chose to use the dat tool.

For More Information

Here are a list of online papers detailing the efforts of three groups to work within the current draft of the Multiparty group's dialog act standard. For more information about annotation in general (labeling speech acts, co-reference, syntax, transcription, etc.) see the LDC's annotation page as well as the MATE Project's page.

Last change: 25 Jan 2001 by mcore

Click here to send in comments or questions.