ir讲义20134跨语言信息检索.ppt

Language Model P(e) P(“我是坏人”) v.s. P (“我是好人”) 288,000 v.s. 346,000 Word Reordering in Translation have programming a seen never I language better Word Choice in Translation That is a little banana. (0) That is a small banana. (3) Smoothing P(“吴宏林是好人”)=? P(“吴宏林是”), P(“是好人”) 0 279,000 Let b(y | x) be the probability that word y follows word x. b(y | x) = number-of-occurrences(“xy”) / number-of-occurrences(“x”) P(“吴宏林是好人”) ~ b(吴宏林 | start-of-sentence) * b(是 | 吴宏林) * b(好人 | 是) * b(end-of-sentence | 好人) Translation Model 翻译模型 她 是 二十 岁 的 教师 她 是 二十 岁 教师 教师 She is twenty aged a teacher she is a teacher aged twenty 翻译概率 扭曲度 富余度 Morphology-Based Approaches The similarity is usually due to either a genetic relationship e.g. English night and German nacht or borrowing from one language to another e.g. English sprint and Japanese スプリント In a broad sense, cognates include not only genetically related words and borrowings but also names, numbers, and punctuation. Practically, all parallel corpora contain some kind of cognates. Identify Cognates There are three word similarity measures: Simard’s condition Dice’s coefficient Longest Common Subsequence Ratio Simard’s Condition Simard et al. (1992) proposed a simple condition for detecting probable cognates in French–English bitexts: two words are considered cognates if they are at least four characters long and their first four characters are identical. Dice’s Coefficient Here Dice’s coefficient is defined as the ratio of the number of shared character bi-grams to the total number of bi-grams in both words. For example colour and couleur share three bi-grams co, ou, and ur so their Dice’s coefficient is 6/11. LCSR The Longest Common Subsequence Ratio of two words is computed by dividing the length of their longest common subsequence by the length of the longer word. For example, LCSR ( colour, couleur ) = 5/7 as