基于扩张矩阵理论的汉语韵律短语分析.pdfVIP

26 1 Vol . 26 No. 1 2003 1 CH INESE JOU RNA L O F COM PUT ERS J an . 200 3 谌卫军 林福宗 李建民 张 钹 ( 100084) ( 100084) , . , , , , ; , ; . , , . ; ; ; T P 18 Chinese Prosodic Phrasing Based on Extension Matrix Theory CH EN WeiJun L IN FuZong L I JianMin ZHA N G Bo ( Dep ar tment of Comp uter S cience and T echnology , Tsinghua Univ ersity , Beij ing 100084) (S tate K ey L a oratory of Intelligent T echnology and Systems , Tsinghua Univ ersity , Beij ing 100084) Abstract T his pap er presents a new induct ive learning algorithm based on the ex tension matrix t heo ry , and uses it t o solve t he prosodic phrasing problem for Chinese T ex ttoSpeech systems. Firstly, t he ext en sion matrix t heory for rule indu ct ion is brief ly out lined . Secondly, aut hors propose a novel def ini t ion of t he consistency of a rule and of a set of posit ive examples, and reveal t heir relat ionship using a theorem the set of posit ive examples covered by a con sist ent rule is also con sist ent. Based on t his con clusion, a new induct ive learning algorithm is presented by dividing the posit ive examples of a specif ic class in a given ex ample set int o consistent groups and adopt ing a simple strateg y t o find a conjunct ive rule for each group which covers all the g roup! s positive examples and none of the negat ive ex amples, the algorit hm finds a set of consist ent rules in the form of variablevalu ed logic. T hirdly, authors col lect 9 37 s