Another Sys Called Qanda

2 Principal contact: This year our primary goal was to improve on the performance of our TREC-8 system. In addition to improving the system directly, we worked on a number of tools to aid our development. We continued our work on a tool for automatic scoring of system responses, a “judge” program. We designed a tool for doing regression testing of question answering systems. We developed a measure of candidate confusability which measures the effectiveness of a set of features for reducing the choices that a ranking system has to make: a coarse form of perplexity. Finally, we performed preliminary work on a method for generating supervised training data.