In this paper, we present three models distinguished by combining different language features, for the task of cross domain Chinese-English sentence alignment. The separating capability of each single and combination of these features are compared, with further discussion of influences on performance. For each feature we introduce a polarity parameter to measure the distinction between aligned pairs and other candidate pairs. Finally, thresholds are set to polarities to further enhance performance precision (from 98.00% to 99.69%).
展开▼
