A combination of neural and semantic networks in natural language processing

The architecture of learned software for searching of semantics in text documents is proposed. In a basis of performance and the recognition of NL semantics the following fundamental principles are proposed: 1. Orientation to a recognition of semantics with minimum usage of knowledge about syntax of the language, 2. Creation of hierarchies from concepts with horizontal (associative) links between nodes of these hierarchies as result of processing of text documents, 3. Recognition of words and collocations on maximum similar with usage of neural algorithms. The main algorithms of learning of software and searching of documents are considered. Also the features of learning (creation of knowledge base) of proposed software are analyzed. Now research prototype of software with this architecture is implemented.