(225) previous ~ index ~ next
To: tdt-distrib@ldc.upenn.edu
From: James Allan <allan@cs.umass.edu>
Subject: TDT workshop report-out
Date: Tue, 07 Mar 2000 15:45:44 -0500
TDT folk,
First, a quick note to thank everyone for a great meeting in Virginia
last week. The presentations were nicely done, the discussion was
lively, and the communication was great. (The only problem was that
we didn't have more time to go outside and enjoy the weather that [I
hear] was wonderful.)
Second, a quick summary of TDT efforts for the rest of 2000, both to
remind people who were there, and to inform people who were not.
Please take this as an outline of the most likely scenario. Official
and complete information will appear on a Web page at some point.
There should be at most minor changes, though.
- We are now in TDT 2000 (in case you were as unaware as I, TDT-1,
TDT-2, and TDT-3 are the names of corpora, not the names of
meetings; TDT-4 would/will be a new set of data).
- Evaluation index files released mid-September
- Results due to NIST early October
- Evaluation results released late October
- TDT 2000 workshop in ***mid-November***
- The TDT 2000 tasks will use the Oct-Dec data for evaluation. Yes,
that is the same TDT-3 evaluation corpus just used. We hope to
get some additional topics labeled, but we cannot guarantee that
will happen.
- The same five tasks (segmentation, detection, FSD, SLD, and
tracking) will exist and will have evaluation report-outs.
- All tasks will use story boundaries that were generated by
machine, probably using one of the two runs submitted this year.
- Tracking will change to have two prefered versions:
1. Nt=1 with simple data (nwt+ccap, true boundaries, English
sources only)
2. Nt=4 plus Nt_neg=2 (highly similar off-topic stories) with
more complicated data (nwt+asr, Multilingual sources,
machine-generated story boundaries supplied)
- Sites will be strongly encouraged to participate in one of the
tasks (active candidates are tracking and SLD).
- There will probably be a dry run, and probably a meeting
associated with it, but time and location are not yet set.
- Sites are encouraged to explore the issue of topic granularity
(essentially, to explore some fundamental assumptions about the
task). To that end, we hope to get some annotation done.
Please feel free to email me with minor corrections or clarifications.
Please don't bother the list with small corrections since they will
presumably all get fixed on a final Web page.
-- james
(225) previous ~ index ~ next
Last updated Wed May 24 17:18:23 2000