A semi-automatic extraction of the SERB in machine translation based on SL

Machine translation based on semantic language (SL) should include a large scale semantic element representation base (SERB), which needs to be extracted from corpus automatically or at least semi-automatically. This paper presents a practical and efficient semi-automatic method to build a SERB from parallel corpus. This method processes the basic characters in both Chinese sentences and English sentences directly, instead of using Chinese word segmentation. First, a preliminary SERB is built. Then some semantic elements (SEs) are picked up by SER pattern match algorithm from the SERB and some of them are pruned by SE pruning algorithm. Last, the SEs are reconstructed to build some SE trees and new SEs and the SEs whose parameter vector categories need to be modified are put forward to users to examine and the correct SEs are appended to the SERB. Thus the SERB is built up. The correctness of the SEs in SERB can be guaranteed.