Boosted voting for confirming synonymous transliteration

There is no dedicated domestic governing body to set translation standards cross over the Mandarin languages. As a result, synonymous transliterations have been produced in the process of translating foreign documents into local documents. In this article we first present five methods which can be used to determine the similarity of two transliterations. No method can consistently outperform the others on determining whether a pair of transliterations is synonymous. Therefore, we propose a boosted voting scheme to form the expressive hypotheses combination instead of a weak classification algorithm needed in boosting algorithm or a majority voting scheme commonly applied for improving the prediction accuracy. The experimental results show that our boosted voting scheme is effective to confirm synonymous transliterations and outperforms the majority voting scheme in performance for computational phonology.