SYSTEMS AND METHODS FOR TRANSLATING NATURAL LANGUAGE SENTENCES INTO DATABASE QUERIES Russian patent published in 2023 - IPC G06F16/2452 G06F40/40 

FIELD: automatic translation.

SUBSTANCE: invention relates to a method, systems and a computer-readable medium for automatic translation from a natural language into an artificial machine-readable language. The method’s first step is execution of an artificial language (AL) encoder and a decoder associated with the AL encoder, wherein the AL encoder is configured to receive the first input array containing an artificial language representation of the AL input sentence and, in response, generate the first internal array, wherein the decoder is configured to receive the first internal array and, in response, produce the first output array containing a representation of the first artificial language AL output sentence; in response to providing the first input array to the encoder AL, the first similarity score indicating a degree of similarity between the AL input sentence and the first AL output sentence is determined; the first set of decoder parameters is adjusted in accordance with the first similarity score to improve the match between the AL encoder input and the decoder output; it is determined whether the first stage training completion condition is satisfied; in response to determining whether the learning completion condition of the first stage is satisfied, if the learning completion condition of the first stage is satisfied, a second stage is executed, which is execution of the natural language (NL) encoder configured to receive the second input array containing a representation of the NL input sentence formulated in natural language, and, in response, outputting the second internal array to the decoder; the second output array generated by the decoder in response to receiving the second internal array is determined, the second output array containing a representation of the second AL output sentence formulated in an artificial language; a second similarity score indicating a degree of similarity between the second AL output sentence and the target AL sentence containing the artificial language translation of the NL input sentence is determined; and the second set of NL encoder parameters is adjusted in accordance with the second similarity score to improve the match between the decoder output and target output representing respective artificial language translations of the input received by the NL encoder.

EFFECT: increasing the accuracy of translation into an artificial machine-readable language through the use of two stages of training on different types of data.

14 cl, 14 dwg