A Cascaded Approach to the Optimization of Translation Rules

As far as the rule-based machine translation (RBMT) is concerned, the rule acquisition remains as a bottle-neck problem. This paper proposes a cascaded approach to optimize the rule base, which is automatically acquired from the bilingual corpus. Observing the more risk of errors in the upper layer of the parsing tree, we propose in this paper a method which advocates the optimization of rules by a bottom-up strategy so as to take the advantage of correctness of parsing results near the leaf nodes. The experimental results further prove that such cascaded optimization out-performs the usual practice