(041) previous ~ index ~ next
To: Paul van Mulbregt <paulvm@dragonsys.com>
From: George Doddington <doddington@msn.com>
Subject: Re: Contrasts involving Segmentation
Date: Mon, 08 Mar 1999 12:38:29 -0800
> 1. At the end of the PI meeting on Tuesday, it seemed to be decided that
> the Tracking with No Boundaries would be done on just the ASR data.
Tracking will be done in TDT3 the same way as in TDT2, namely on all sources
together. This will hold both for boundaries GIVEN and NOT_GIVEN. It is
important to process all sources simultaneously because source interaction
is a major research issue. Results will be scored both overall and, as was
done in TDT2, conditional on just the audio sources alone. This conditional
evaluation avoids confounding the effect of averaging audio source errors
with newswire errors.
--------
> 2. The same for detection? Namely one run on ASR with known boundaries
> compared to one run on ASR with automatically generated boundaries.
Detection will also be done in TDT3 the same as in TDT2. Again, all
sources are to be processed together, and results will be scored both
overall and conditional on source type.
--------
> 3. For the segmentation contrast involving transcripts (FCDH vs CCAP),
> I'd like to suggest that this be done on a subset of broadcasts for which
> FDCH and CCAP are both available. In the Dec 1998 year eval, 2 CCAP shows
> were substituted for missing FDCH transcripts, and I'm not convinced that it
> was necessary.
Only a single manual transcription will be used for each audio source in
TDT3. The difference between FDCH and CC is too small to warrant further
study. Since CC is the primary operational transcription mode, CC is the
TDT3 choice for manual transcription. (Where CC doesn't exist, however,
FDCH transcription will be used, because the company is more reliable
and the cost is the same as for CC.)
--
George Doddington in CA: doddington@nist.gov or 925/631-6628
(041) previous ~ index ~ next
Last updated Thu May 13 09:28:18 1999