A Multiobjective Simulated Annealing based Approach for Feature Selection in Anaphora Resolution

In this paper we propose a multiobjective simulated annealing based technique for anaphora resolution. There is no generally accepted metric for measuring the performance of anaphora resolution systems, and the existing metrics-MUC, B 3 , CEAF, Blanc, among others-tend to reward significantly different behaviors. Systems optimized according to one metric tend to perform poorly with respect to other ones, making it very difficult to compare anaphora resolution systems, as clearly shown by the results of the SEMEVAL 2010 Task 1 on the Multilingual Coreference Resolution. One solution would be to find a single completely satisfactory metric, but its not clear whether this is possible and at any rate it is not going to happen any time soon. An alternative is to optimize models according to multiple metrics simultaneously. In this paper, we propose a multiobjective simulated annealing based technique to solve the feature selection problem of anaphora resolution by optimizing multiple objective functions. Experimental results show that the proposed approach performs superior in comparison to the previously developed multiobjective genetic algorithm based feature selection technique.