Study on Automatic Phrase Correspondence by the Analysis of Parallel Text Data between English and Japanese

本研究的第一个目标是日英对比文本数据（平行文本）的收集和数据库的建立。我们有.收集了大量的平行文本，包括《天生神后》、《科学美国人》等。本项目结束后，收集工作将继续进行。第二个目标是平行文本的比较和对应。我们开发了一个算法，以获得句子的句子长度比较的基础上，日语和英语之间的对应关系。从假设日语文本的一个段落和英语文本的一个段落是对应的出发，通过句子的顺序和它们的长度来寻找单个句子的最佳对应。结果相当令人满意。日语和英语句子中短语之间的对应是相当困难的，我们仍然没有找到一个很好的短语对应的算法。第三个目标是开发翻译辅助系统，该系统显示合适的例子翻译的短语，一个翻译不能找到好的翻译。提出了一种近似串匹配算法，从文本数据库中检索合适的实例对。

The first objective of this research is the collection and data base formation of contrastive text data(parallel text)of Japanese and English. We have. collected fairly big volume of parallel texts from "Tensei Jingo", "Scientific American" and some others. The collection will be continued constantly after this project.The second objective is the comparison and correspondence of parallel texts. We have developed an algorithm to get the sentence-by-sentence correspondence based on the length comparison of a sentence between Japanese and English. Starting from the assumption that a paragraph of a Japanese text and a paragraph of an English is corresponded, the best correspondence is looked for individual sentences by the sequencial order of sentences and their lengths. The result was quite satisfactory. The correspondence between phrases in Japanese and English sentences was quite difficult, and still we can not find out a good algorithm for phrasal correspondences.The third objective is the development of translator's assistant system which displays suitable example translations to the phrases which a translator cannot find out good translation. An approximate string-matching algorithm has been deve loped, which retrieves suitable example pairs from the text data base.