US 12,333,267 B2
Text generation model generating device, text generation model, and text generating device
Hosei Matsuoka, Chiyoda-ku (JP)
Assigned to NTT DOCOMO, INC., Chiyoda-ku (JP)
Appl. No. 18/252,140
Filed by NTT DOCOMO, INC., Chiyoda-ku (JP)
PCT Filed Oct. 20, 2021, PCT No. PCT/JP2021/038829
§ 371(c)(1), (2) Date May 8, 2023,
PCT Pub. No. WO2022/102364, PCT Pub. Date May 19, 2022.
Claims priority of application No. 2020-189333 (JP), filed on Nov. 13, 2020.
Prior Publication US 2024/0303445 A1, Sep. 12, 2024
Int. Cl. G06F 40/58 (2020.01)
CPC G06F 40/58 (2020.01) 13 Claims
OG exemplary drawing
 
1. A text generation model generating device generating a text generation model generating output text of a second language different from a first language in accordance with input of input text of the first language by machine learning,
wherein the text generation model is an encoder decoder model that includes a neural network and is configured using an encoder and a decoder,
wherein learning data used for the machine learning of the text generation model includes first data, a context, and second data,
the first data including an array of a plurality of words composing the input text,
the second data including an array of a plurality of words composing the output text corresponding to the input text, and
the context including one or more words of the second language relating to the second data,
the text generation model generating device comprises circuitry configured to:
input the first data to the encoder in accordance with an arrangement order of words;
input the context, a start symbol that is a predetermined symbol indicating start of output of the output text, and words composing the second data to the decoder in accordance with an arrangement order;
update weighting coefficients configuring the encoder and the decoder on the basis of an error for each word between an array of words output from the decoder in a later stage after input of the start symbol and an array of words included in the second data; and
output the text generation model in which the weighting coefficients are updated by the circuitry.