数据平滑技术综述

数据平滑是统计语言建模的关键技术,它不仅可以改进语言模型的性能,还可以提高语音识别、文字识别等应用领域的系统识别率,不同的数据平滑方法之间的对比应在各种不同规模的训练集上操作。各种平滑算法中,以Good-Turing估计、线性插值平滑、Katz’s回退式平滑最为典型和常用。该文对各种数据平滑方法进行了经验性对比,并讨论了影响这些数据平滑方法性能的有关因素。

Data Smoothing Technology Summary

Data smoothing is the key technology of statistical language modeling,It not only can improve the performance of language modeling,it Can also improve speech recognition and Application areas such as language identification system recognition rate.Different data smoothing method should be at the contrast between the different scale of operation on the training set.A variety of smoothing algorithms,To Good-Turing estimate,linear interpolation smoothing,Katz's back-off-type is most typical and commonly used s...