Frames

It seems to me that theoretical studies conducted in the field of artificial intelligence (AI) and psychology as a whole were not general enough to explain either from a practical point of view or phenomenologically the efficiency of human thinking. The main structural elements that form the foundation for the deployment of the processes of perception, storage of information, thinking and the development of linguistic forms of communication should be larger and have a clearer structure; their actual and procedural content should be more closely linked to each other in order to be able to explain the phenomenon of power and "speed" of human thinking.

A similar point of view is held by specialists of scientific centers working on solving problems of artificial intelligence. It is well known, for example, the proposal of A.Newell and G.Saymon (1972) to solve the problem of representation in terms of "problem spaces" or the proposal of S. Peipert and the author of these lines (M.Minsky, S. Peypert, 1972) to break the whole body of information, necessary for the AI ​​system, on the "microworld". The same views take a different form in the works of well-known theorists such as R. Schenk (1973), R. Abelson (1973) and D. Norman (1973), who use larger structures to study the mechanisms of understanding natural language. This reflects the desire of scientists to go beyond the research of purely behaviouristic and formal-logical directions and to abandon attempts to solve the problem of representation using sets of separate simple data structures. (Behaviourism (from English. Behavior - behavior) is one of the directions in psychology, which is based on the statement that behavior is the subject of psychology, not consciousness. The main task of psychologists, according to the founder of behaviorism, J. Watson (1914), is the establishment of objectively observable relationships in accordance with the well-known "stimulus - reaction" scheme and reducing to them all the concepts of internal mental processes (see MG Yaroshevsky, 1976).)

This paper attempts to tie together the results of some of the above studies and create a unified and coherent theory. Its shortcomings are noted, since more questions are put here than answers are given to them.

The starting point for this theory is the fact that a person, trying to learn a new situation for himself or take a new look at already familiar things, selects from his memory a certain data structure (image), which we call a frame, so that by changing in it the individual details make it suitable for understanding a wider class of phenomena or processes.

A frame is a data structure to represent a stereotyped situation. Each frame is associated with information of different types. One part of it indicates how this frame should be used, the other - which is supposed to entail its implementation, and the third - what should be done if these expectations are not confirmed.

A frame can be thought of as a network consisting of nodes and links between them. The "upper levels" of the frame are clearly defined, since they are formed by such concepts that are always fair to the intended situation. At lower levels there are many special vertex terminals or "cells" that must be filled with characteristic examples or data.

Each terminal can set the conditions that must be met by its tasks. Simple conditions are defined by markers, for example, in the form of a requirement that a terminal should be a subject, or an item of suitable size, or a pointer to a subframe of a certain type. (Subframes, frames, and superframes are hierarchically ordered elements that form frame systems.) More complex conditions define the relationship between concepts included in various terminal vertices.

Groups of semantically close frames are combined into a frame system. The results of significant actions are represented as transformations between the frames of the system. This makes it possible to model such concepts as attention and value of information, to make some types of calculations more economical, and also to show the efficiency of using frames in AI systems.

For visual perception of images of the frame system are used as follows: different frames correspond to different positions of the observer analyzing the same scene, and the transformations between them reflect the results of moving the observer from one place to another. For systems of other types, the differences between frames may correspond to the results of performing any actions, certain causal relationships between objects of the external world, or different points of view on the same issues. Some of the same terminals can be part of several frames of the system - this is one of the central moments of the theory, which allows coordination of information from various sources.

The theory of frames is largely due to the possibility of using expectations and other types of assumptions in it. The terminals of the frame in their usual state are filled with so-called "tasks of absence" or in advance prepared values, that is, with information about details (particulars) that need not be present in any particular situation. The link between the absence tasks and their terminals is not rigid and unchanged, so they can easily be replaced by other information that is more appropriate to the current situation. Tasks of absence can, therefore, fulfill the role of variables, serve for argumentation with the help of examples (which often makes use of logical quantifiers unnecessary), provide general information and describe the most probable cases, indicate ways of conducting useful generalizations, etc.

Frame systems are connected, in turn, by a network of information retrieval. If the proposed frame cannot be adapted to the real situation, that is, if it is not possible to find such tasks of the terminals that meet the conditions of the corresponding markers, the information search network allows you to select a frame that is more suitable for the situation. Such structures make it possible to use various methods of presenting information in frame systems, which is of particular importance for the development of mechanisms for understanding.

After selecting the frame in the process of negotiation, the terminals are assigned such values ​​that satisfy all the conditions of the corresponding markers. The progress of the reconciliation process is partly controlled by information related to the frame itself (including indications of how to respond to unforeseen circumstances) and to a large extent the experience of solving similar or similar tasks. If the coordination of external data with terminal markers is unsatisfactory, the information obtained on its basis can be successfully applied when choosing an alternative frame.

Note that the schemes proposed in this paper are imperfect in many respects. First, some options for presenting information are discussed irrespective of the processes in which they should be used. Sometimes only the descriptions of the properties with which certain structures should be provided are given. Markers and tasks of the terminals are considered as if their connections and connections with larger structural units are known, which is not yet true.