Symmetric is Not the Optimal Local Context Window in Chinese Word Sense Disambiguation

Word sense disambiguation (WSD) is a task of classification, where the local context is the basic features to identify the sense of ambiguous word. Most systems choose optimal local context window on empirical grounds, which is usually symmetric, the distance from the ambiguous word to both sides of the window is same, such as [-1, +1] or [-2, +2]. Is symmetric window better than asymmetric window? In this paper, we take Senseval-3 Chinese data set as example. First find the optimal window estimated by cross-validation using only the training set, which is a symmetric window. Then, perform a WSD evaluation on the test data using this symmetric window for comparison with other classical symmetric window. The results show that asymmetric is better than symmetric window, and symmetric window is not always the best option.