DRI Discourse Structure In Dialogue
Homework 2 Verbmobil materials
General Information
<![CDATA[
The second homework consists of doing CGU/IU analyses for one
VERBMOBIL and one MAPTASK dialogue. However, the MAPTASK dialogue
must be distributed at a later date. We apologize for being 4 days
late with the distribution. Since we only have one dialogue to code
for now, we'd like to have the VERBMOBIL codings due April 27, 1998 as
scheduled. (We will allow extra time for MAPTASK). Email
chn@research.att.com about any scheduling difficulties.
A few important notes:
- Please email all HW2 related codings, questions, criticisms etc.
to both traum@cs.umd.edu and chn@research.att.com.
- FYI, the VERBMOBIL and MAPTASK dialogues are being coded
by all three subgroups (coreference, speech act/utterace labels
("backward/forward-looking functions"), discourse structure).
The purpose is to facilitate inter-group discussion and try
to relate the various subgroup efforts.
- For HW2, the coding procedure will be slightly modified: please do
the CGU analysis first. Then, in contrast to HW1, you will use a
different file of pre-ordained CGUs (verbmobil.fixed.cgus) as the
basis for IU analysis. To repeat: do *NOT* use your CGU coding for IU
analysis. The reason is to enable more comparable analysis of IU
coding reliability on the HW2 dialogues by using fixed CGU units to
start. ***Of course, you should not look at the verbmobil.fixed.cgus
file until you are done with your own CGU analysis and are ready to
proceed with IU analysis***. [Any methodological reflections on this
we would appreciate hearing. Thanks to the many who volunteered ideas
in the first round.]
- ABOUT VERBMOBIL: [written by Mark Core, B/F group]
"The Verbmobil project is a long term effort to develop a mobile
translation system for spontaneous speech in face-to-face
situations. The current domain of focus is scheduling business
meetings. To support this goal, some English human-human dialogs were
collected in this domain. Dialog r148c is one of these dialogs. In
r148c, the two speakers try to establish a time and place for a
meeting. The speakers are affiliated with a college of some kind as
one speaker asks if they want to meet on campus or not and they
mention a building called "Cyert Hall". The sound files provided
correspond to turns (a continuous piece of text with the same speaker)
so when playing an utterance's speech you will hear the whole turn."
=================================
EXTRA NOTES ON VERBMOBIL DIALOGUE [from Norbert Reithinger]
=================================
"These dialogues were recorded in the USA, so I do not know the exact
conditions. The standard data collection setup in Verbmobil phase 1
were these data were recorded is as follows: two persons in one room,
looking at each other, having calendars marked with dates,
instructions about possible dates, and the topics they should mention.
They have no [push-to-talk] button are were asked to avoid
crosstalk. They did not rehearse, but sometimes speakers recorded more
than one dialogue."
Three VERBMOBIL files should be sent to chn@research.att.com *and*
traum@cs.umd.edu (tarred and uuencoded):
Files to turn in:
1. verbmobil.cgu.<coder's_login>
2. verbmobil.iu.<coder's_login>
3. misc.<coder's_login>
[How to email uuncoded tar files in UNIX:
> tar -cvf <login>.hw2.tar <file1> <file2> <file3>
> uuencode <login>.hw2.tar <login>.hw2.tar >tmp
> Mail chn@research.att.com,traum@cs.umd.edu <tmp
]
Send any and all questions to both chn@research.att.com and
traum@cs.umd.edu. Please do *NOT* mail questions on coding choices
for the homework dialogues to the whole group, so as to not bias other
coders. We are reviewing the HW1 data, comments and criticisms and
will report back as soon as we can. We would like to post everyone's
codings and misc comments in some form before May 17. If you would
like to edit yours in any way or otherwise object, email
chn@research.att.com please.
Happy tagging!
Christine and David
]]>