(185) previous ~ index ~ next

To: TDT Distrib <tdt-distrib@ldc.upenn.edu>
From: Jonathan Fiscus <jonathan.fiscus@nist.gov>
Subject: Release of October Dry Run index files
Date: Tue, 21 Sep 1999 12:26:47 -0400

Folks,

The executive summary is:
1. the index files are ready for you to get,
2. there's a new version of the evaluation software that
supports the new format of the corpus,
3. I've moved the deadline back for submission of results to
September 30,

The full details are:

1. Index files:

There are two complete sets of index files, one using BBN ASR outputs,
and one using Dragon ASR outputs. You can use one or the other, or both
if you'd like. The index files were generated using the latest release
of the TDT2 corpus, dated September 7, 1999. This corpus, and the fall
evaluation corpus conform to Dave G.'s email of XXXXX.

The first set, (at the URL
'ftp://jaguar.ncsl.nist.gov/tdt3/oct_dryrun/indexes_octdr_tdtrel3.1_DragonRec_v3.tar.Z'),
uese the BBN-BYBLOS English recognition files, (The English .as1
files). The Fall TDT3 evaluation will make use of English ASR
transcripts generated by the BBN-BYBLOS system. Therefore these are the
preferred index files.

The second set, (at the URL
'ftp://jaguar.ncsl.nist.gov/tdt3/oct_dryrun/indexes_octdr_tdtrel3.1_DragonRec_v3.tar.Z'),
uses the Dragon English recognition files, (the English .as0 files).
FYI: TDT3, the corpus for the fall evaluation, will not include Dragon
English recognition files.


2. Evaluation software:

A new version of the evaluation software is available from the URL:
'ftp://jaguar.ncsl.nist.gov/tdt3/TDT3eval_v1.5.tgz'. There are two
changes, a) to support the new format of the corpus, and b) modified
procedures to generate the story link detection index and key files.


3. Dry Run deadlines:

The schedule called for dry run results to be due at NIST by September
27. This day is close, so I've changed it to September 30 to give you
folks three more days. Also, as stated earlier, the index files use the
new corpus format. If you're unable to run your systems using the new
corpus format, please delay your submission until you've successfully
used the new corpus.


This will be a difficult dry run, but it is important and it will be
extremely valuable for all participants to complete.

Jon

--
Jonathan Fiscus
National Inst. of Stds. and Tech.
100 Bureau Dr. Stop 8940
Gaithersburg, MD 20899-8940

Phone: (301) 975-3182
Email: jonathan.fiscus@nist.gov
(185) previous ~ index ~ next

Last updated Wed Sep 22 10:26:05 1999