Study on OOV Translation Mining from Parallel Corpora and the Web
Jianmin Yao · Jiangnan daxue xuebao. Ziran kexue ban · 2010
This paper presents an approach to translate OOV through the search engine and to mine the translation of OOV from local parallel corpora extracted from bilingual web pages.An improved Frequency Change Measurement which combines adjacent information method was used to generate MLUs(Multi-Lexical Units) and an approach using multi-features including a frequency-distance model and a transliteration model to select the correct translation.Besides,a mining system using a Maximum Entropy(ME) Classifier combines word overlap feature,word alignment feature and location feature to automatically discover parallel corpora from bilingual web pages.A comparison of the performance of the web mining OOV translation and the local parallel corpora based OOV translation was made.The experimental results show that the Top 10 inclusion of web achieves 94.6%,and the Top 10 of parallel corpora is 37.5%.