Novosibirsk State Technical University, Novosibirsk, email: avg@vt.cs.nstu.ru

The implementation of the undestanding of natural language (NL) by computer is one of principal problems solved in area of computer science, called usually "“artificial Intelligence". And now most popular task solved in this area is searching of text documents by query on natural language. In this paper one technology is described for this task.

In a basis of performance are put the following fundamental principles:

• Orientation to a recognition of semantics with minimum usage of knowledge about syntax of the language,
• Creation of hierarhies from concepts with horizontal (associative) links between nodes of these hierarhies as result of processing of documents,
• Recognition of words and word collocations on maximum resembling with usage of neural algorithms.

Last two principles were formulated by autor and are used by him at creation of the programming system of robots on the natural language [1].

The knowledge base (KB) represents a semantic net of the frames. The frames are such named some conditionally, because ones have slots as only links to another frames-nodes of semantic net. Frame may be terminal frame (connected with word in dictionary or place in document), frame-concept, complex frame (including some words) and frame-header (class of texts). The neural algorithms are used for recognition of frames-concepts, complex frames and frame-header. So in some cases the frame may be visibled as neuron.

Conditionally in the knowledge base it is possible to select a constant component circumscribing data domain, and variable part circumscribing contents of documents. The constant part is formed as a result of of preteaching of system. The variable part is formed as a result of document processing under the control of a constant component.

Now the software for demonstration of this technology [2] is developed and available for downloding from http://insycom.chat.ru.

This software consisting of two programs:

• The program for creation and debugging of the knowledge base about contents of documents,
• The program for documents retrieval on sense by end user.

The developed technology is planned to use for development of Documents Managment System. This investigation in part is supported by grant of Ministry of Education of Russia Federation.

REFERENCES

[2] A.V.Gavrilov. The architecture of the Software for Recognition and Searching of Documents

by Natural Language Query. - Int. Conf. KDS-2001, S.-Peterburg, 2001.- V.1, Pp. 124-130.

(Russian).