| 摘要 |
<P>PROBLEM TO BE SOLVED: To extract parallel translated sentences and parallel translated words from the patent documents of a family of patents which have been applied, based on the same invention content and to construct a translation dictionary and a translation memory for machine translation at a low cost, based on the information thus extracted. <P>SOLUTION: A Japanese patent document 11 and its corresponding United States patent document 12 are made to be a translation corpus 1. The sentence division of the patent documents corresponding to each other, which are selected from the translation corpus 1, is applied by respective punctuation symbols, reference numbers are extracted from the sentences obtained by sentence division, a parallel translated sentence pair 3 is extracted by the cooccurrence of the reference numbers thus extracted, and a noun, located before the same reference number as that of the translated sentence pair extracted, is made to be a parallel translated word pair 5. The parallel translated word pairs 5 are accumulated in a parallel translation dictionary for special fields, based on IPCs. <P>COPYRIGHT: (C)2005,JPO&NCIPI |
