基于概率的语法检查.ppt

基于概率的语法检查 中国外语教育研究中心 梁茂成 3 4 5 2 1 主要内容 Grammaticality is gradient POS tag sequences from NS corpora Methodology Results Hands-on work Grammaticality is gradient 1 Grammaticality is not dichotomous (Keller 2000). Chomsky (1975) himself acknowledges that grammaticality can come in varying degrees: “There is little doubt that speakers can fairly consistently order new utterances, never previously heard, with respect to their degree of ‘belongingness’ to the language” (p. 132). Grammaticality is gradient 1 That linguistic structures are not always equally (un)grammatical is in fact reflected in linguists’ own annotations (e.g., [?], ?, ??, *, **) Grammaticality is gradient 1 1. Not to be invited is sad. 2. (?)To have not had a happy childhood is a prerequisite for writing novels. 3. ? To be not invited is sad. 4. * To get not arrested under such circumstances is a miracle. (Haegemann Guéron, 1999, p. 308) Grammaticality is gradient 1 Sentences are acceptable or unacceptable only relative to each other (Featherston 2005). Gradient judgments in grammaticality judgment tests POS tag sequences from NS corpora 2 It_PPH1 is_VBZ not_XX easy_JJ to_TO examine_VVI each_DD1 of_IO these_DD2 cases_NN2 ._. PPH1 VBZ XX JJ TO VVI DD1 IO DD2 NN2 . PPH1 VBZ XX JJ TO VVI DD1 IO DD2 NN2 . POS tag sequences from NS corpora 2 POS tag sequences from NS corpora: PPH1 VBZ XX JJ TO VBZ XX JJ TO VVI XX JJ TO VVI DD1 JJ TO VVI DD1 IO TO VVI DD1 IO DD2 VVI DD1 IO DD2 NN2 Matches in a grammatical sentence 2 It_PPH1 is_VBZ not_XX difficult_JJ to_TO complete_VVI each_DD1 of_IO those_DD2 tasks_NN2 ._. It_PPH1 is_VBZ not_XX difficult_JJ to_TO complete_VVI each_DD1 of_IO those_DD2 tasks_NN2 ._. It_PPH1 is_VBZ not_XX difficult_JJ to_TO complete_VVI each_DD1 of_IO those_DD2 tasks_NN2 ._. It_PPH1 is_VBZ not_XX difficult_JJ to_TO complete_VVI each_DD1 of_IO those_DD2 tasks_NN2 ._. It_PPH1 is_VBZ not_XX difficult_JJ to_TO complete_VVI each_DD1 of_IO those_DD2 tasks_NN2 ._. It_PPH1 is_