Term Translation Pair Alignment Based on a Parallel Corpus of Chinese Historical Classics and Their English Translations

摘要

We introduce an improved Mi-based phrase alignment approach to match Chinese historical terms with their English translations. The proposed approach is motivated by the findings about the inadequacy of popular machine translation systems like Babelfish and Systran in translating Chinese historical terms, and observations about the discrepancies by established translators and Sinologists in handling these historical terms. Modifications to the basic MI alignment approach, including Bridging, Pruning and Head-word Extension, integrate linguistic knowledge of the translation patterns of Chinese historical terms into statistical alignment algorithm. The proposed framework for alignment achieves more than 17 and 10 points increase in percentage respectively above the base linc in two individual corpora and detects a number of valid alternative translations for some top-frequency Chinese historical terms. The discovery also alerts us to the potential impact that different corpus may have for an SMT system from the perspective of readability and efficiency.