Abstract: Translation example base is the main knowledge source of example-based machine translation system. In this paper, a shallow parsing information based approach is proposed to extract translation examples. First, translation units of source and target language sentences are segmented respectively according to shallow parsing information. Then, guided by word alignment result, an statistical model is used to align translation units between source and target translation units, and thus translation examples are extracted. Experiment result shows that the proposed method achieves satisfying result in both direct evaluation of example base and indirect evaluation by EBMT system.