微引擎流水线机器翻译系统结构（Microengine assembly line machine translation system structure）.docVIP

微引擎流水线机器翻译系统结构（Microengine assembly line machine translation system structure） Microengine assembly line machine translation system structure liu Institute of computational linguistics, Peking University The institute of computing technology of the Chinese academy of sciences liuqun@ Abstract: this article from the knowledge representation, knowledge acquisition, knowledge, using three levels for rules in the process of machine translation systems development methods and statistical methods are summarized and analysis, and introduces us the micro engine assembly line of machine translation system structure. Keywords: machine translation, hybrid method, multi-engine strategy, micro-engine assembly line structure 1. The machine translation method combining the rules and statistics The rules and methods of natural language processing and machine translation studies (or rationalism and empiricism) are two main approaches. We think that this formulation is too general, and here we are going to do some sort of categorization of some of the existing research methods from knowledge representation and knowledge acquisition. From the perspective of knowledge representation, existing methods can be classified into the following categories: 1. Rules: symbolic rules are an intuitive representation of knowledge. For linguists, the expression is very convenient, the granularity can be thick and thin, and the flexibility is great. But the general knowledge of the rules is usually followed by a binary logic, and the system is less robust; 2. Data: in various statistical models, knowledge is manifested through data; And the actual data type knowledge is not necessarily statistical knowledge, such as fuzzy set membership is also a data type knowledge; 3. Rule + data: this is a form of mixed knowledge representation, typical of the probabilistic grammar of various forms, with probability information or confidence information for each rule; 4. Corpus: in most cases, the knowledge contai