2000 Volume 14 Issue 3
Published: 15 June 2000
-
Select all|
-
Review2000, 14(3): 1-8.With their high occurrence rates in argumentative Chinese texts , discourse markers play a significant role in the automatic processing of these kinds of Chinese texts ,such as automatic summarization. This paper reports on an effort in applying machine learning to identify discourse markers in Chinese. We have processed 80 Chinese texts from which we have selected subsets for data training and data testing. We used C4.5 in our experiments and obtained accuracies of the order of 80%. We also interpret and analyze our experimental results in the linguistic perspective.
-
Review2000, 14(3): 9-15.The research of Psycholinguistics and cognition development have proved that there was a independent inductive learning phase in the early course of language development . The start point of this paper is the order of language development ,the acquisition process of formal semantics and its representation base are regarded as one pace of a speech information processing model which has the consistent mechanism to achieve language understanding and its production. In this paper a case based machine learning system is discussed. In order to get the formal semantics of words structure ,the definition of operation semantics is adopted and a discovery learning algorithm based on referring was designed ,which can refine the semantics with the increment of sentence samples.
-
Review2000, 14(3): 16-24,36.This paper presents an automatic identifying system of Chinese names. The system makes use of a large amount of statistical data ,which are extracted from real name library and real text corpus ,to enhance its identifying performance. The testing sample ,including 100 articles ,are extracted from the People's Daily 1994 News Corpus. The experiment shows that the recall and the precision can both reach above 90%.
-
Review2000, 14(3): 25-29.In this paper ,a classical term weighting method — IDF (Inverse Document Frequency) is discussed detailedly. The two important classification algorithms — Bayes Judge Rule and VSM (Vector Space Model) are summarized. Furthermore , the way of how to combine term weighting methods with two classification algorithm is also provided in the paper.
-
Review2000, 14(3): 30-36.This paper puts forward a post-processing approach based on words intermatching. It takes full advantage of the relationship between words ,with words innermatching、words intermatching and Markov Model. Each reinforces another. As a result ,the combination produces the ability of error correcting and the stability of post-processing.
-
Review2000, 14(3): 37-41.This paper presents a recognition method of similar Chinese handwriting by support vector machine. This method can be used to small scale recognition , like artificial neural networks , but it has stronger generalization ability because the support vector machine theory is based on the minimization principle to structure risk. Because the algorithm of support vector machine is a convex quadratic optimization problem , the local optimal solution is certainly the global optimal one. This paper presents a sample to recognize three groups of similar Chinese handwritings with the algorithm of support vector machine and good results are obtained.
-
Review2000, 14(3): 42-45,59.The rule of peeling off edge tonal element in syllable concatenation within a sound foot ,presented in "edge tonal element peeling theory" ,can be used to explain the stress principle on two-syllable and three-syllable words and the phenomenon of tone sandhi. In this paper a new rule of peeling off tonal element in sound foot concatenation is further developed. It is effective to explain the stress principle and the edge tonal element peeling principle on four-seven syllable words successfully as well as two-syllable and three-syllable words. The use of our developed rule can successfully improve the naturalness in Chinese Text-to-Speech Synthesis.
-
Review2000, 14(3): 45-53.Syntax parsing is not a kind of symbol inference ,but a kind of entity inference. Adding semantic information during syntax parsing is an effective method to implement entity inference. The syntax parser ,introduced by this paper ,have two features. One feature is that to use clear-up-rule of multi-parts of speech based words to improve the efficiency of parser ,the other featureis that to use static and dynamic syntax-semantic properties of words to limit the too strong power of syntax rules. The parser gain pretty good effect due to those features.
-
Review2000, 14(3): 54-59.X Window provided Chinese character only with bitmap font , and supported only GB2312 - 80 Chinese character set ,not meet the need of people for CJK unified character set . The paper lays emphasis on providing with CJK unified character set Chinese TrueType font supporting on X Window. This paper outlines the X Server architecture ,presents in detail the implementation algorithm of Chinese font file renderer.
-
Review2000, 14(3): 60-65.This paper first proves that when the Zipf's law is used to describe the total distribution of Chinese Character frequency ,some fitting frequence accumulations take obvious errors no matter how elaborately the parameters a and c are selected ,and then presents a method to overcome the shortcoming by using the Zipf's law to describe only the tail of the distribution.
