藏语N-gram语言模型中平滑技术的研究仁青吉（甘肃民族师范学院 藏区

Research smoothing technique of Tibetan N-gram language model <br>Renqing Ji <br>(Tibetan Intangible Cultural Heritage Key Laboratory of Gansu Normal University for Nationalities, Gansu cooperation 747000) <br>[Abstract] article build Srilm modeling platform in the Linux environment, then corpus sub-block processing, and establishing a language model and counting with a N-gram count and N-gram, by using a smoothing algorithm to some perplexity level tested, these last few perplexity value of the comparison and analysis of data, a summary for the current corpus under optimal smoothing method and locale.

Study on smoothing technology in Tibetan N-gram language model<br>Ren Qingji<br>(Gansu National Normal College Tibetan Non-Relic Key Laboratory, Gansu Cooperation 747000)<br>The article builds Srilm modeling platform in the Linux environment, then blocks the materials, and uses N-gram count and N-gram to count and establish the language model, and tests the puzzle with several smoothing algorithms. Finally, the numerical comparison and data analysis of these puzzles are analyzed, and a optimal smoothing method applicable to the current language and language environment is summarized.

Research on smoothing technology in Tibetan n-gram language model<br>Ren Qing Ji<br>(Key Laboratory of Tibetan intangible cultural heritage, Gansu Normal University for nationalities, Gansu cooperation 747000)<br>[Abstract] in this paper, srilm modeling platform is built in Linux environment, and then the corpus is processed in blocks. N-gram count and n-gram are used to count and build the language model, and several smoothing algorithms are used to test the degree of confusion. Finally, the values of these degrees of confusion are compared and analyzed, and an optimal one for the current corpus and language environment is concluded Smoothing method<br>