Abstraction Method of Words Based on Relationships between Subjects, Objects and Predicates

The study [1] found that relationships between the target words and their neighbor words are useful to decrease ambiguity of predicate. However, sentences usually contain many concrete words. Obviously, it is impossible to have the whole relationships between all words. We expect that we will be able to reduce the number of relationships necessary to express all of relationships between words, if we use abstract words instead of their concrete words. Therefore, the purpose of our study is to propose the method to abstract nouns based on relationships between the subject, the object and the predicate in an input sentence. The method extracted words, which is a subject, an object and a predicate from an input sentence, and abstracted the subject and the object based on the relationship with their predicate. In our study, we used a topic map to keep relationships between words. I made topic maps from Wikipedia, GoiTaikei, EIJIRO and Nihongo Dai-thesaurus. We conducted an experiment to evaluate our method. As a result, we obtained 93 proper assignment of "human" tag to words out of 100 words. Forty six words without "human" tag were abstracted and 42 words were properly abstracted.