Determination of Semantic Proximity of Natural Language Terms for Subsequent Neural Network Training

Systems that are controlled using natural language must strictly follow the commands given to them. In any voice command, there is both useful information corresponding to this command and extraneous noise. The actuator must separate the command from the noise. If we take a primitive approach to the process of human survival in the world around us, then it depends on the correct response to events, i.e. reaction to external irritation. This reaction is known as a reflex and is presented in two types, unconditional: breathing, swallowing, sucking, pulling the hand away from the cold and conditional: reading, various skills and abilities. Unconditional ones are embedded in our body from birth, and conditional ones we acquire through learning in the course of life, but despite their differences, they are united by one thing - responding to an event. The paper proposes a solution to this problem by determining the semantic proximity of a command issued in a natural language with a digital representation of its ideal. At the first stage, a natural language simulator is its textual representation.