Homepage > Papers

Mechanisms of semantic information processing

Igor P. Kuznetsov. M. "Nauka", 1978. - 175 p.


Monograph consists of 7 chapters. In first three is proposed the formalism of system construction, that are been based on the knowledge (factographic, logic-informational, analytical and other) is proposed. The trend in their development is connected with their orientation to the end user that does not have special knowledge in the area of computers.

Users must have access to computer information in the convenient forms. For this the organization of the model of subject area, the development of the so-called intellectual interface are required. It shows that in the solution of this problem the task of representation and using the knowledge is very significant. On the basis of the study of the existing systems and anlysing the special features of natural language texts the new formalism of knowleddge representation (language of the extended semantic networks - ESN) is proposed.

Standard semantic networks are consisted from the nodes, which correspond to objects. Nodes are connected by the arcs, which have the markers of relations. However, with the aid of similar networks it is very difficulty to present the complicated information. Therefore in the network we introduce the nodes, which correspond to the names of relations, complex nodes, which correspond to the objects together with their relation, and also special element, called the connection node. The last node is inserted instead arc and connects by one edge to the node-relation, and by other edges - to the node-objects. As result the elementary fragment is formed. Networks ESN is the development of semantic networks in the direction of an increase in the representative possibilities with the preservation the property of gomogenious. In ESN the free location of nodes positions are allowed. As a result following representation becomes possible: when the names of relations play the role of objects and connected by own relations, when the composition of objects is broken on the levels of detalisation, representation of generalized information (with the quantifiers) and others

The fourth and fifth chapters are dedicated to information processing, represented by ESN. For this the method of the comparison of two networks is introduced and is based. It provides the identification of indeterminate components of inputed information. This method is the development of the principle of comparison on the model for the case of gomogenious representation taking into account the possibility of the generalized represantation. At the basis of method we introduce the rules of identification, which make possible to connect nodes and to compare networks in accordance with the laws of logic. In this case in the network the nodes- variables, which correspond to the indeterminate objects, are introduced. In the process of comparison such nodes are indeficated or they put value. According to the given rules any node of the network Kn1 can be identified with the set of nodes from another network - Kn2. The network Kn1 is adequate the network Kn2, if there is such version of replacements of nodes from Kn1 to the identified nodes, with which of Kn1 is obtained Kn2.The matching network Kn1 with Kn2 consist in the search into Kn2 of all parts, which are adequate to Kn1. From each such part are selected the nodes, with which are identified the nodes-variables from Kn1. The first becomes the value of the second.

The problem of identification of the indeterminate components, answer to the demands of factographic information is solved by comparison. The rules of the transformation of networks (prototype of the productions of language DEKL) are introduced. The procedures and the algorithms of solution of the intellectual test problems, used for checking the mental development of the man are examined on this basis: verbal tests (when in the place of points it is necessary to put letter in order to obtain correct words), kriptoarifmetical tasks and so on. Such way we show the possibility to build systems with the sufficiently high coefficient of mental development.

The sixth chapter is dedicated to the task of generalization at the level of homogenous structures - ESN. Method of using ESN for representation of the logic predicate (the first and second order) is considered. Quantifiers are considered as the special relations, which connect the nodes-variables. We introduce metarools of generalization and inductive conclusion that is provided by the construction of ESN with quantifiers. Then we introduce methods and algorithms of their use for the answer to the factographic questions, including to the questions of the generalized nature. The logical validity of these procedures is proved.

Last chapter is dedicated to questions of the use of the introduced means for representation and processing of linguistic knowledge. We consider using the method of comparison for the recognition of the erroneously written words, for entities extaction, for analysing semantic preferences, non-linguistic values and others. This speaks about a sufficient universality of the proposed means and procedures for construction of the intellectual systems of different classes.

Kuznetsov Igor Petrovich (igor-kuz@mtu-net.ru)