VOICE COMMUNICATION IN NATURAL LANGUAGE BETWEEN HUMAN AND DEVICE Russian patent published in 2016 - IPC G10L15/22 

FIELD: information technology.

SUBSTANCE: invention relates to human-machine interaction, specifically voice communication in natural language between human and device. Voice input processing and formation of a response to natural language includes steps of receiving voice input and input voice signal; processed input voice signal to obtain detected input text line, standardised recognised input text line, detecting lexemes, named objects and expression of time in detected input text line. Method then includes morphological analysis with resolution ambiguities in terms of main forms of words and detection of morphological criteria and parsing dependencies for detected input text line by statistical parsing for formation of tree relationships. In order to support flexibility of word order, statistical parsing algorithms based on graphs with treatment nonprojective trees are used. Further, method includes performing statistical analysis of dependency tree for detection of key concepts and their dependencies to generate syntax tree, converting syntax tree into SQL-query, sending SQL-query to database and receiving data relating to SQL-query, converting received data into text response string in a natural language and synthesizing voice signal in accordance with response character string response.

EFFECT: technical result is to provide a handle user voice input to morphologically rich language without computationally expensive operations related to large number of applicable rules.

9 cl, 3 dwg