Research on Recognition of Chunks for Reordering in Chinese-English Patent Machine Translation

We focused on when to reorder the Chinese NPs "A+(de)+ B", "A+(de)+ B+(de)+ C" or "A+(de)+ B+(de)+ C+(de)+ D" wherein A, B, C and D are separate and smallest chunks for reordering. By comparing Chinese NPs and their English translations, we analyzed the semantic features of the chunks for reordering and summed three types Boundary Words. Using a rule-based method, we built 53 formalized rules with the Boundary Words and the contexts to identify the chunks. At last, we tested our work in a rule-based MT system, and the experimental results showed that our rule-based method was very efficient.