Symmetric Trends: Optimal Local Context Window in Chinese Word Sense Disambiguation

Word Sense Disambiguation (WSD) is a task of classification to identify sense of ambiguous word. Most systems choose symmetric local context window on empirical grounds, that is, the distance from the ambiguous word to both sides of the window is same, such as [-1, +1]. Why? In this paper, we do an experiment on Senseval-3 Chinese data set. Use cross-validation method to identify the optimal local context window for word features and part of speech (POS) features separately. The results show that in the same window scope, the smaller the gap, the lower the error, and the better the WSD performance. In other words, the window that tends to symmetric will get better performance.