SPOKEN DOCUMENT RETRIEVAL FOR TREC-7 AT CAMBRIDGE UNIVERSITY
S.E. Johnson , P. Jourlin, G.L. Moore, K. Sparck Jones & P.C. Woodland
January 1999
This paper presents work done at Cambridge University, on the TREC-7 Spoken Document Retrieval (SDR) Track. The broadcast news audio was transcribed using a 2-pass gender-dependent HTK speech recogniser which ran at 50 times real time and gave an overall word error rate of 24.8\%, the lowest in the track. The Okapi-based retrieval engine used in TREC-6 by the City/Cambridge University collaboration was supplemented by improving the stop-list, adding a bad-spelling mapper and stemmer exceptions list, adding word-pair information, integrating part-of-speech weighting on query terms and including some pre-search statistical expansion. The final system gave an average precision of 0.4817 on the reference and 0.4509 on the automatic transcription, with the R-precision being 0.4603 and 0.4330 respectively.
The paper also presents results on a new set of 60 queries with assessments for the TREC-6 test document data used for development purposes, and analyses the relationship between recognition accuracy, as defined by a pre-processed term error rate, and retrieval performance for both sets of data.
If you have difficulty viewing files that end '.gz',
which are gzip compressed, then you may be able to find
tools to uncompress them at the gzip
web site.
If you have difficulty viewing files that are in PostScript, (ending
'.ps' or '.ps.gz'), then you may be able to
find tools to view them at
the gsview
web site.
We have attempted to provide automatically generated PDF copies of documents for which only PostScript versions have previously been available. These are clearly marked in the database - due to the nature of the automatic conversion process, they are likely to be badly aliased when viewed at default resolution on screen by acroread.