Research On The New Technology Of Word Convert In Mobile Intelligent Terminal Based On C/S Structure

The arrival of3G times has greatly promoted the rapid development of mobile intelligentterminal of China. Among basic functions of intelligent terminal, the Input Method plays asignificant role. As the core technology of Input Method, the word conversion technology willdirectly influence the performance of Input Method. People’s higher requirement of accuracy andefficiency of Input makes it urgent to develop a rapid and efficient word conversion technology.Taking the improvement of word conversion of mobile intelligent terminal as the mainresearch objective, this paper has conducted intensive analysis and research of three problems,namely, overlarge language model, syncopation ambiguity of phonetic transcription, and decodingefficiency. Firstly, the statistical language model compression method based on relative entropypruning is put forward, that is, to integrate the concept of relative entropy pruning into languagemodel and use the retaining rate to control the size of language model, thus compressing it withinthe acceptable range of mobile intelligent terminal. Secondly, this paper has conducted intensiveanalysis for syncopation of Chinese phonetic string, defined crossing ambiguous phonetic stringand combined ambiguous phonetic string, and put forward two methods aiming at disambiguationof phonetic string, thus considerably increasing the syncopation accuracy. Thirdly, targeting thelimitation of local client of traditional Input Method, this paper has developed the client-serverinput method mode based on Android platform, that is, to store the compressed binary languagemodel in the client and trigram language model in remote input method server, and design theinteractive mode between client and server, thus improving the conversion accuracy of InputMethod. Lastly, an evaluation program for mobile intelligent terminal conversion has beendeveloped, aiming at the integral evaluation of standard and efficiency of current Input Method.The result shows that the Input Method is competitive in the integral performance of conversion.To sum up, centering on statistical language model compression, ambiguous phonetic stringsyncopation and client-server interaction, this paper has conducted specific research and broughtforward corresponding algorithm or solution, and test its performance through hardware platform.