对数线性翻译模型的判别式训练综述  

A Survey of Discriminative Training for the Log-linear Translation Model

作　　者：刘乐茂[1] 赵铁军[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2013年第6期14-17,共4页Intelligent Computer and Applications

基　　金：国家自然科学基金(61173073;61100093;61073130;61272384);国家高技术研究发展计划(863)(2011AA01A207)

摘　　要：目前几乎所有的统计机器翻译系统都采用对数线性模型建模.判别式训练是基于对数线性翻译系统的一个重要组成部分,其任务就是优化对数线性模型的参数。到现在为止,有很多判别式训练方法可以用来训练翻译模型权重。从似然函数、错误率函数和可扩展方法三个方面,系统地阐述并分析了这些训练方法,旨在让更多的研究者更好地了解判别式训练方法的发展现状、为判别式训练的进一步发展起到推动作用。同时,还就判别式训练提出了两个值得进一步探讨的问题。Nowadays, almost all statistical machine systems are modeled by the log- linear model. Discriminative training is an important component in a statistical machine system and its task is to optimize a weight for a log- linear translation model. Until now, there have been many discriminative training methods. This paper systematically reviews and analyzes them from three aspects consisting of the log likely - hood function, the error rate function and scalable methods, and it is hoped to be helpful for researchers to understand the current situation of discriminative training and further development of discriminative training. Meanwhile, this paper proposes two questions to be explored in future as well.

关 键 词：统计机器翻译 对数线性模型 判别式训练 

分 类 号：TP391.2[自动化与计算机技术—计算机应用技术]