Search

Inflection and derivation have been the main ways of creating new words in English. With the development of science and technology, words as such appear faster than ever in scientific literature. Influenced by English, Chinese words with multiple affixes are also becoming a major way of new word creation in scientific literature. By studying the similarities of their original sources, this paper employs a head transduction model in an attempt to automatically translate these words from English to Chinese. With this method we hope to solve the problem of such words usually as unknown words in machine translation systems and build a bilingual lexicon with a richer content.

This paper presents a training method of log-linear model for statistical machine translation based on structural support vector machine. This method is designed to directly optimize parameters with respect to translation quality. By adopting maximum-margin principle of SVM, the MT model can learn from training samples with generalization capability. Experiments are carried out on a hierarchical phrase-based MT system facing Chinese to English translation. Result shows that structural SVM training has the ability of re-ranking the k-best list of MT system according to automatic evaluation criteria BLEU, and it can enhance the average quality of MT system outputs.

Chinese Syntactic Parsing is often divided into two stages, basic model to generate k-best candidate parsing trees and reranking stage to find the best one in these trees, and we focus on the second stage-reranking. Tree kernel has been used in Syntactic parsing in English(Collins 2002) which makes a good improvement in performance. This paper makes tree kernel method practical for Chinese Syntactic Parsing, and improves tree kernel from three points: the first is introducing rules weights rw to distinct different rules, the second is introducing similar rules converging, and the third is introducing a parameter λ to scale the relative importance of tree fragments with their size and solve the peak problem which is a common kernel. Experiments show that these three adjustments to tree kernel increase reranking performance a lot.

This paper presents a new approach to improve the word alignment. Building a bilingual dictionary is one of the main applications for word alignment. However, the research of using the bilingual dictionary to improve the word alignment is not enough. There are two bottlenecks. The first is that large bilingual dictionary is hard to get. The second is that the normal approach of using bilingual dictionary does not make full use of the dictionary. We designed a bootstrapping algorithm to conquer the bottlenecks, achieving a good result.

This paper suggests a methodology which is aimed to translate the Chinese terms into Korean. Our basic idea is to use a method which combines domain-tuned dictionary and target language corpus. The translation system used in our work operated at three steps: (1) tokenization of the term candidate; (2) translation of term candidates using domain-tuned bilingual lexicon; (3) ranking the translation candidate to recommend the best translation according to the frequency of the target word cooccurrence in the corpus. And we have demonstrated the effectiveness of our approach by showing a high degree of accuracy evaluation. Our experiments indicate a significant improvement when measured against un-tuned Chinese-Korean MT system.