We study the word alignment between the Chinese problems and the English answers for the English-Chinese translation tests at the high school level in Taiwan. After applying a dictionary-based approach, we attempted to take advantage of the information about co-occurrence of unaligned words to find more translation pairs. We explored five scoring methods that were previously proposed in the literature. We ran our experiments with more than 17,000 test items, and checked the aligned word pairs that were assigned high scores. Experimental results showed that we could improve the F measure for the alignment task from 76.9% to 83.7% with the best performing scoring method.
Proceedings of the 22nd Conference on Computational Linguistics and Speech Processing, ROCLING 2010