Language Modeling with N-grams (Speech and Language Processing)
2017-03-10 19:20
453 查看
语言模型
计算词序列(words sequences)概率的模型称为语言模型(LMs),词序列(w1,w2,...,wn)的概率为:P(w1n) = P(w1)p(w2|w1)P(w3|w1w2)...P(wn|w1n-1)
Bigram model
二元模型的前提是Markov假设(一个词的概率只依赖于其前面一个词),值为前一个词下的条件概率,不再是前面词序列下的条件概率。P(wn|w1n-1) => P(wn|wn-1)
N-gram model
N元模型词概率设为前N-1个词下的条件概率P(wn|w1n-1) => P(wn|wn-(N-1)n-1)
计算实例
计算下二元模型的词序列概率。下图展示了一个语料库里各词出现次数
下图展示了二元词序列的出现次数及其各词概率
如(i want)词序列出现827次,i出现2533次,P(want|i) = 827/2533 = 0.33
log概率
通常概率计算转换为log概率,避免概率相乘过小溢出。存储的时候只记录log和,需要原始概率时再进行转换。P1P2P3P4=e(lnP1 + lnP2 + lnP3 + lnP4)
概率大小就存储为lnP1 + lnP2 + lnP3 + lnP4
参考
http://web.stanford.edu/~jurafsky/slp3/相关文章推荐
- 《Speech and Language Processing》第三版手稿读书笔记——00(Daniel Jurafsky,James H. Martin)
- Speech and Language Processing (2nd Ed.): Updates
- Speech and Language Processing, NLP 处理
- 《Speech and Language Processing》读书笔记之信息抽取IE
- (zhuan) Speech and Natural Language Processing
- An Introduction to Language Processing with Perl and Prolog: An Outline of Theories, Implementation,
- 论文笔记——Creating Speech and Language DataWith Amazon’s Mechanical Turk
- Signals and Systems with MATLAB Computing and Simulink Modeling, Fourth Edition
- Getting Started with Kinect and Processing
- Theano-Deep Learning Tutorials 笔记:Modeling and generating sequences of polyphonic music with the RNN
- How to build Multi-Language Web Sites with ASP.NET 2.0 and VS.Net 2005
- processing ALV WITH subtotal and total items
- <Natural Language Processing with Python>学习笔记二
- 用Python进行自然语言处理-1. Language Processing and Python
- Playing Around with Methods Overloading, C-language and Operators (1)
- 哥伦比亚大学Coursera课程Natural Language Processing:Quiz 2: covers material from weeks 3 and 4
- ASP.NET 2.0 Language Swithcer and Theme Swicher 多语言转换和多样式主题转换
- ASP.NET 2.0 Language Swithcer and Theme Swicher 多语言转换和多样式主题转换
- Natural language processing: Deep Neural networks with multitask learning
- ASP.NET 2.0 Language Swithcer and Theme Swicher 多语言转换和多样式主题转换