US 12,093,659 B2
Text generation with customizable style
Nan Duan, Beijing (CN); Ming Zhou, Beijing (CN); and Yaobo Liang, Beijing (CN)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Appl. No. 17/291,772
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
PCT Filed Dec. 6, 2019, PCT No. PCT/US2019/064801
§ 371(c)(1), (2) Date May 6, 2021,
PCT Pub. No. WO2020/139536, PCT Pub. Date Jul. 2, 2020.
Claims priority of application No. 201811585240.7 (CN), filed on Dec. 24, 2018.
Prior Publication US 2022/0027577 A1, Jan. 27, 2022
Int. Cl. G06F 40/42 (2020.01); G06F 40/30 (2020.01)
CPC G06F 40/42 (2020.01) [G06F 40/30 (2020.01)] 13 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
receiving a first natural language text;
converting the first natural language text into a second natural language text by a text generation model, the second natural language text at least partly reflecting a meaning of the first natural language text and comprising a style distinguishable from the first natural language text, the text generation model comprising a modifiable parameter, wherein the modifiable parameter is applied to the first natural language text when converting the first natural language text into the second natural language text and converting the first natural language text into the second natural language text comprises:
extracting a first feature vector from the first natural language text;
converting the first feature vector into a second feature vector by a planning function with the modifiable parameter by:
applying the first feature vector to a neural network to provide a state sequence; and
providing the state sequence to the planning function to generate the second feature vector, wherein the second feature vector is defined by a first parameter value of the modifiable parameter and represents a style distinguishable from the first natural language text; and
generating, based on the second feature vector, the second natural language text;
receiving an update to the modifiable parameter via a graphical user interface element that adjusts an output of the conversion; and
in response to receiving a modification to the parameter, converting the first natural language text into a third natural language text by the text generation model with the updated modifiable parameter, the third language text at least partly reflecting the meaning of the first natural language text and comprising a style distinguishable from both the first natural language text and the second natural language text.