NTCIR-5 English-Chinese Cross Language Question-Answering Experiments using PIRCS
Kui Lam Kwok, Peter Deng, Norbert Dinstl, Sora Choi · 2005
We participated in the English-Chinese CLQA task with the following procedures. An English question was first classified as to its answer category, and then rendered into Chinese in three ways: raw text translation by MT, extracted entity translation by our web-translation algorithm, and web-assisted question expansion followed by MT and entity web-translation. A combined Chinese question is formed that retrieves the top 100 sentences from the target collection. Candidate Chinese entities are extracted from the sentences and ranked for answer-hood based on a combination of five sources of evidence: