In software requirements capturing meeting, it is important to find out interesting items of clients. There is a method of finding out interesting items by knowing their structure of unconsciousness. There are some indications of unconsciousness. For example, there are some mistakes in speaking. To interpret mistakes in speaking seems to need a high technic. We aim at the method of not interpreting the mistakes or corrects in speaking. We experimented the relationship between interests and corrects in speaking. We found the case that the speaker was interested in the corrected word. We made a requirements capturing method to find the topics in the next meeting extracting the corrected words.

View full abstract

It is well known that direct parsing of a long Japanese sentence, including many conjunctive clauses, is extremely difficult. Therefore, it is preferable to segment such a sentence into shorter, simpler ones prior to parsing. Some methods for sentence segmentation have been reported so far. However, because those conventional methods are based on handmade segmentation patterns or rules, they have problems in keeping consistency of the patterns, and in deciding the optimal order of applying those rules. This paper proposes a new method of sentence segmentation using a decision tree, which acquires optimal segmentation patterns and the optimal order of their application automatically from a corpus, taking both linguistic phenomena and their occurrence frequencies into account. Generation and evaluation of a decision tree for sentence segmentation were conducted on an EDR corpus. For 400 evaluation sentences, precision and recall were both 84%, and the percentage of correctly segmented sentences was 77%. It was also confirmed that pruning reduces the tree size significantly without deteriorating the performance.

View full abstract

The referential properties of noun phrases are useful for article generation in Japanese-English machine translation and in anaphora resolution in Japanese same noun phrases and are generally classified into generic noun phrases, definite noun phrases and indefinite noun phrases. In the previous work, an estimation of referential properties was done by developing rules that used clue words. If two or more rules were in conflict with each other, the category having the maximum total score given by the rules was selected as the desired category. The score given by each rule was established by hand, so the manpower cost was high. This paper describes a machine learning method that reduces the amount of manpower needed to adjust these scores.

View full abstract

It is often useful to arrange words in their meanings' order that is obtained by using a thesaurus. In this paper, we introduce a method of arranging words semantically, and show the relationship between this method and some types of dictionaries and thesauruses. We also examine an ideal dictionary that could be used for future natural language processing. Finally, we describe three main examples of this method.

View full abstract