词性标注（国外英文资料）.doc

词性标注（国外英文资料） In the part of speech tag set have been identified, and every word in the dictionary are determined on the basis of part of speech, to an input word string converted to the corresponding process called part-of-speech tagging of part of speech tag string. The problem of speech labeling needs to be solved How to determine the meaning of the word in context. Guess the word for an unlogged word Multi-category words influence on syntactic analysis: although the proportion of multi-category words in the vocabulary is not very high, but because they appear higher percentage, therefore directly influence for syntax analysis. Lexical notation: Probability method The lexical annotation method based on the hidden markov model The method of machine learning rules A transformation based error driven word annotation method Consider the problem of lexical annotation from a statistical model A given word string W = w1 w2... Wn, if T = t1t2... Tn is a string of words for W. The word W is defined as the process of asking for T in a given W and a word table with a lexical annotation. Suppose that W has multiple possible lexical tags, T1, T2,... Ti, the lexical annotation of W is to find the maximum conjugations of P (T | W) in the context of known W. 3T = argmax P (T | W) Such as word string the/this / / report/editor / / / in some words have more than one part of speech tag (multi-category words), so the words on the corresponding part-of-speech tagging chain has more than one. The total mark result is equal to the number of words of each word, which is 4 times 1 times 1 times 2 times 2 times 2 times 3, which is 96. The task of the lexical labelling is to find the most likely word of the word string T in multiple possibilities. The corresponding lexical string in the previous example is PRVNVMQ For an etymological tagging system, the most likely to be considered the most likely to be the most likely to be the right one, may be wrong. To make it convenient, do the follow