关键词: 三元组可比语料库, 语言迁移, 自动语言剖析, n-元词串

Abstract: There exists inherent skewed language model in Chinese-English parallel corpus due to the influence of translationese. Obviously, natural language processing systems trained with these corpora, including machine translation and cross-language information retrieval, will inherit the skewed language model, thus seriously degrading the performance of applications. To fix the inherent defaults in parallel corpus, this paper proposes a technical research on building and profiling Chinese-English 3-tuple comparable corpora. The study adopts comparable corpora and automatic language profiling technologies and applies a combined method of statistics and rules for statistical analysis on native English and Chinglish in 3-tuple comparable corpora that consists of native English, Chinglish and standard Chinese. Based on this, automatic extraction technologies, such as n-grams and key clusters, are used in the mining of native-language-based bilingual resources to improve and develop natural language processing applications such as machine translation.

Key words: 3-tuple comparable corpora, language transfer, automatic language profiling, n-grams