![](data:image/png;base64,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)
5.3 噪声信道模型 159
5.3 噪声信道模型
在5.2节中,我们实现了一个简单的基于词的统计机器翻译模型,内容涉及建模、
训练和解码。但是,还有很多问题还没有进行深入讨论,比如,如何处理空翻译?如
何对调序问题进行建模?如何用更严密的数学模型描述翻译过程?如何对更加复杂
的统计模型进行训练?等等。针对以上问题,本节将系统地介绍 IBM 统计机器翻译
模型。作为经典的机器翻译模型,对 IBM 模型的学习将有助于对自然语言处理问题
建立系统化建模思想,特别是对问题的数学描述方法将会成为理解本书后续内容的
基础工具。
首先,重新思考一下人类进行翻译的过程。对于给定的源语句 s,人不会像计算
机一样尝试很多的可能,而是快速准确地翻译出一个或者少数几个正确的译文。在
人看来,除了正确的译文外,其他的翻译都是不正确的,或者说除了少数的译文人甚
至都不会考虑太多其他的可能性。但是,在统计机器翻译的世界里,没有译文是不
可能的。换句话说,对于源语言句子 s,所有目标语词串 t 都是可能的译文,只是可
能性大小不同。这个思想可以通过统计模型实现:每对 (s,t) 都有一个概率值 P (t|s)
来描述 s 翻译为 t 的好与坏(图5.11)。
s
b
t
正确翻译
s
t
1
t
2
t
3
t
4
P (t
1
|s) = 0.1
P (t
2
|s) = 0.2
P (t
3
|s) = 0.3
P (t
4
|s) = 0.1
(a)人的翻译候选空间
(b)机器的翻译候选空间
图 5.11 不同翻译候选空间的对比:人(左)vs 机器翻译(右)
IBM 模型也是建立在如上统计模型之上。具体来说,IBM 模型的基础是噪声信道
模型(Noise Channel Model),它是由 Shannon 在上世纪 40 年代末提出来的
[241]
,并于
上世纪 80 年代应用在语言识别领域,后来又被 Brown 等人用于统计机器翻译中
[9, 10]
。
在噪声信道模型中,目标语言句子 t(信源)被看作是由源语言句子 s(信宿)经
过一个有噪声的信道得到的。如果知道了 s 和信道的性质,可以通过 P (t|s) 得到信
源的信息,这个过程如图5.12所示。
s
t
噪声信道
信宿 信源
图 5.12 噪声信道模型
举个例子,对于汉译英的翻译任务,英语句子 t 可以被看作是汉语句子 s 加入
噪声通过信道后得到的结果。换句话说,汉语句子经过噪声-信道传输时发生了变化,
在信道的输出端呈现为英语句子。于是需要根据观察到的汉语特征,通过概率 P (t|s)