IMPROVED RE-RANKING MODEL FOR TRANSLATION HYPOTHESIS USING EXTREME LEARNING MACHINE by Wen Yuan Thesis Supervisor: Associate Professor, Chi Man Vong Master of Science in Software Engineering In statistical machine translation (SMT), re-ranking of (possibly infinite number of) randomly generated translation hypotheses is one of the essential components in determining the quality of translation result. In this work, an improved re-ranking model for translation hypothesis using ELM classification is proposed to select the good but rare hypotheses to ease the burden of computation. Compared with the performance of baseline system the performance of the proposed model with ELM classification can raise up to 5.8% in IWSLT 2014 Chinese to English corpus, compared with a state-ofthe-art proposed method, we have a promotion of 0.373 which is better than theirs 0.346 in WMT corpus. While its training time is about 80 times faster than traditional reranking method based on ELM regression. Index Terms: Re-ranking, Extreme Learning Machine, Statistical Machine Translation, Translation Hypothesis