... drawn from BC. We then investigate the utility of addingadditional in-domain training data from WSJ. In thebaseline approach, the additional WSJ examples arerandomly selected. With active learning ... with a naive Bayes classifier,the sense s assigned to an example with featuresf1, . . . , fnis chosen so as to maximize:p(s)nj=1p(fj|s)In our domain adaptation study, we start with ... acorpus-based, supervised learning approach. With this approach, one would need to collect a text cor-pus, in which each ambiguous word occurrence isfirst tagged with its correct sense to serve...