詞對應與轉換辭典的優化

Abstract

在本論文中,我們提出一個對應中英語料庫中之中文詞與英文詞的方法,此 方法首先引用一個斷詞及詞性標註程序做中文斷詞,然後檢查辭典中所記 錄的中文詞之翻譯是否對應到英文句中之英文詞,以建立定位點,根據這些 定位點,我們可以利用相鄰詞的對應或上下文之訊息將中文裡尚未對應的 詞加以對應,這些新的對應,經過篩選後,可當做新的翻譯並加入於轉換辭 典內以優化辭典,這個優化過的辭典可再被用於詞對應模組,以建立更多的 定位點,進而得到更正確的詞對應。 Research in bilingual corpus become important in machine translation. In this thesis, we describe a method for word alignment of Chinese-English corpora. This method first segments Chinese sentences by applying a segmentation and tagging process. It then establishes the anchor points by aligning those Chinese words whose translations are recorded in our dictionary to the words appearing in corresponding English sentence. According to these anchor points, we can align those un-aligned words by using adjacent word's alignment or contextual information. Some criteria are introduced for choosing the new translation from these new alignments. These new translations will be added into our transfer dictionary to refine this dictionary. The refined transfer dictionary can then be used into word alignment module to eatablish more anchor points and obtain more accurate alignment results.