Estimating phrase pair relevance for translation model pruning
Matthias Eck, Stephan Vogel, Alex Waibel · KITopen · 2007
We present pruning strategies for translation models that are based on estimating the relevance of phrase pairs. We apply the overall translation system to a set of data and collect a number of statistics for each phrase pair. Using these statistics in various scoring terms we are able to significantly outperform baseline pruning methods and we can show that the number of phrase pairs can be reduced by up to 80% without significantly affecting the overall system performance.