In this paper, a high speed multi-language machine translation approach based on pruning on tree representations of semantic elements is proposed. This is the multi-language machine translation with the following several characteristics: Chinese segmentation before translation into another languages is not necessary, and the translation time is O(L) rather than general O(LN), where L is the length of text, N is the number of semantic elements (i.e. number of language patterns) in SER-base, even if N is hundreds of thousands or millions.