Classifiers are often required to output not only a classification result but also the probability of the classification. We focus on the decision list classifier which has successfully been applied to a wide variety of NLP tasks. We propose methods based on Bayesian learning to calculate the reliability of contextual evidences in decision lists, which enables decision lists to output theoretically well-founded probabilities. Experimental results obtained using Senseval-1 data set show that our

more »

... methods enable decision lists to output probabilities appropriately reflecting their reliabilities and improve the classification performance of the decision list algorithm.