The ICT statistical machine translation systems for IWSLT 2007

The ICT Statistical Machine Translation Systems for IWSLT 2007 Zhongjun He, Haitao Mi, Yang Liu, Deyi Xiong, Weihua Luo, Yun Huang, Zhixiang Ren, Yajuan Lu, Qun Liu Key Laboratory of Intelligent Information Processing, Institute of Computing Technology Chinese Academy of Sciences No.6 Kexueyuan South Road, Haidian District P.O. Box 2704, Beijing, China, 100080 {zjhe, htmi, yliu, dyxiong, luoweihua, huangyun, renzhixiang, lvyajuan, liuqun}@ict.ac.cn Abstract In this paper, we give an overview of the ICT statistical machine translation systems for the evaluation campaign of the International Workshop on Spoken Language Transla- tion (IWSLT) 2007. In this year’s evaluation, we partici- pated in the Chinese-English transcript translation task, and developed three systems based on different techniques: a for- mally syntax-based system Bruin, an extended phrase-based system Confucius and a linguistically syntax-based system Lynx. We will describe the models of these three systems, and compare their performance in detail. We set Bruin as our primary system, which ranks 2 among the 15 primary results according to the official evaluation results. 1. Introduction This paper describes the statistical machine translation systems of Institute of Computing Technology, Chinese Academy of Sciences, which are used for the evaluation cam- paign of the International Workshop on Spoken Language Translation (IWSLT) 2007. We participated in the Chinese-English transcript trans- lation task and developed three SMT systems based on dif- ferent techniques for this year’s evaluation. The primary sys- tem is Bruin, which is a formally syntax-based system. An- other one is an extended phrase-based system named Confu- cius, and the third one is a linguistically syntax-based system named Lynx. This paper is structured as follows: Section 2 gives an overview of our three SMT systems, Section 3 describes data preparation methods. In Section 4, we will report the experi- ments and results. Finally, Section 5 gives conclusions. 2. Systems Overview Our group focused on statistical machine translation since 2004. During the last three years, we have tried various meth- ods and developed three SMT systems: Bruin (a formally syntax-based system), Confucius (an extended phrase-based system) and Lynx (a linguistically syntax-based system). In this section, we will describe the models and algorithms in detail. 2.1. Bruin Bruin is a formally syntax-based SMT system, which imple- ments the maximum entropy based reordering model on BTG [1] rules. There are three essential elements in Bruin. The first one is a stochastic BTG, whose rules are weighted using differ- ent features in the log-linear form. The second is a MaxEnt- based reordering model predicting the orders between neigh- bor blocks, whose features are automatically learned from bilingual training data. The last one is a CKY-style chart- based decoder with beam search similar to that of Wu [1]. 2.1.1. Translation Model To complete the decoding procedure, three BTG rules are used to derive the translation: A [ ] → (A1, A2) (1) A ⟨ ⟩ → (A1, A2) (2) A → (x, y) (3) The lexical rule (3) is used to translate source phrase y into target phrase x and generate a block A. …

Content truncated to 500 words