MATE Deliverable D1.1

Supported Coding Schemes

(Carnegie Mellon University, Language Technology Institute; Universität Karlsruhe)

Coding book:
Coding book not public available.

Number of annotators:
Between 4 and 10. (linguists, computational linguists, computer scientists)

Number of annotated dialogues:
Lots of test sets. Each test set consists of roughly 100 utterances.

Evaluations of scheme:
The correct word to state classification rate is 89%.

("Statistical Analysis of Dialogue Structure"; Ye-Yi Wang, Alex Waibel)

"scores from different judges may vary by as much as 10 percentage points" in Donna Gates et al. : End-to-End Evaluation in JANUS: A Speech-to-Speech translation system in: E. Maierr, M. Mast, S. LuperFoy (Eds.):Lecture Notes in Artificial Intelligence 1236 Dialogue Processing in Spoken Language Systems, Springer-Verlag Berlin Heidelberg, 1997

Underlying task:
Appointment scheduling

List of phenomena annotated:
Related to C-STAR phenomena.

[nicety] (Hello Dr. Noah)
[nicety] (Hi Tor)
[suggest-meeting] (let's set up a meeting for a couple of hours)
[temporal] (in the next two weeks)
[your-availability] (when's good for you)
[interject] (let's see)
[suggest-time] (how about Friday the second in the morning)
[my-unavailability] (I'm busy that morning)

Markup language:
Own format.

Existence of annotation tools:
Annotation by hand.

Janus system

Contact person:
Lori Levin (Lori_Levin@alexis.boltz.cs.cmu.edu)

Last Modificatiuon: 27.8.1998 by Marion Klein