Abstract: Recently confusion network decoding showed a better performance in combining outputs from multiple machine translation (MT) systems. However, overcoming different word orders presented in multiple MT systems during hypothesis alignment still remains to be the biggest challenge to confusion-network-based MT system combination. The previous alignment methods do not consider the information about semantics. In order to improve the system performance, we introduce word sense disambiguation (WSD) into confusion network alignment. Meanwhile, the selection of skeleton is taken through sentence similarity score, and the sentence similarity is computed by the largest bipartite graph matching algorithm. In order to combine WSD based on WordNet with our system, the experiments showed that the result using revised translation error rate (TER) algorithms is better than classic TER system combination.