Long sentence partitioning using top-down analysis for machine translation

Long sentence processing is an important part for English-Chinese machine translation systems. The system performance is directly affected by the correctness of long sentence processing. A basic thought for processing a long sentence is to partition it into short sub-sentences and to merge the sub-translations for the whole translation. In this paper, a rule-based top-down partitioning algorithm is provided. The rules are inducted from sentence patterns and use regular expressions as main part. Firstly, the algorithm reduces some sentence components to shorten the sentence; then coordinate sub-sentences are recognized and partitioned; finally, clauses within sub-sentences are processed. Experiment shows an approximate 85% accuracy and an over 90% recall rate of the algorithm.