Using parallel texts to improve recall in IE.
M.McGee Wood, Susannah J. Lydon, Valentin Tablan, Diana Maynard, Hamish Cunningham · Research Explorer (The University of Manchester) · 2004
Our experiments show that applying known IE techniques to independent parallel texts describing the same information and merging the results brings significant improvements in performance. Recall summed over six botanical descriptions of several plant species is more than triple the average for each text individually. We analyse these results, and describe the processing techniques used. This leads us to the concept of “greedy extraction”, where the information template is not frozen before processing, but can incorporate new information found in the text(s).