Research for Massively Parallel Natural Language Understanding System

点击翻译按钮获取中文摘要

The aim of this research is to develop a realizing method of a massively parallel natural language understanding system, which works on a massively parallel processing system with 10^4 - 10^6 processors. The target language for the system is Japanese.The achieved research results are as follows.・A parallel semantic analysis model, which cooperates with a parallel morphological and syntactic analysis module, is developed. In this model, parameters of semantic priorities are controlled according to 4 situations : scramble, suppress, preserve and activate. A thesaurus is used to control semantic parameters.・A parallel morphological and syntactic analysis system is evaluated on a shared memory type parallel processing machine which has 20 CPUs. About 3 times speed up was achieved.・A semi-automatic case frame acquisition method is developed. The method classifies a set of simple sentences having a same verb by the meaning of the verb by using an electronic dictionary which has more than 400 entries. This method is used to acquire semantic knowledges handled by our parallel semantic analysis system.