A smoothing algorithm for the task adaption chinese trigram model电子书.pdfVIP

Proceedings of ICSP ‘98 A Smoothing Algorithm for the Task Adaptation Chinese Trigram Model* .liang Minghii Yuan Baozong Lin Biqin Tang Xiaofang (Institute of Information Science,Northern Jiaotong University. Beijing 100044,P.R.China) . i h t r ; i c t : The paper main solve above two problems. In I. Introduction the piper :I (‘hinese Trigram model o f task adaptation ability :\I present C‘hincsc N - g ~ i n iniotlcl i h \\ i d c q ” ” ~ I i i d iii ;ire het up. ,izerogram to trigram probability statistics speech recognition and n;itural 1angii;ige ~iiiclcrstiiiiiliii~.hiit I I I C inform:ition base o f 1994 “People Daily” are built, it made model has t\\o prohlems: the one h;inJ \\lien tlic xopc 01’ c o i - l ~ i r IIW o f the success esperience o f Hhlhl in speech recognition, is not enough large. tlic most tie-in \\orcis or abutmen[ \\ aril\ ant1 atloptecl Baurn-W’elch algorithm for optimum of the frcclueiic\ 01‘ appcur;incc ih \ er!. little in twining corpiis. c\ e n weights. Each weigh stands for correlation statistic reliability ultimate \\ ithout appearance. I n this it c~iiihespiirsc iI;i[;i of these models. The probability statistics matrix smooth prohlem. bring about kno\\ ledge shortage. This sitiiation al’l’~-ct\ algorithm o f the parameter space was carried on, in order to the performancc of post-processing ncg;rti\el! . and thcreli)rc. \\ c offset tlie niatris sparse d