US 12,450,430 B2
Automated identification of sentence concreteness and concreteness conversion
Xiao Xia Mao, Shanghai (CN); Jialei Ma, Shanghai (CN); Nan Nan Li, Shanghai (CN); Min Huang, Shanghai (CN); and Xiao Feng Ji, Shanghai (CN)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed on Mar. 15, 2023, as Appl. No. 18/184,016.
Prior Publication US 2024/0311561 A1, Sep. 19, 2024
Int. Cl. G06F 40/253 (2020.01); G06F 40/166 (2020.01); G06F 40/30 (2020.01)
CPC G06F 40/253 (2020.01) [G06F 40/166 (2020.01); G06F 40/30 (2020.01)] 20 Claims
OG exemplary drawing
 
1. A method comprising:
training a machine learning model to generate revised sentences with revised concreteness levels, wherein the training comprises:
submitting a revised sentence generated by the machine learning model and another real sentence into a classifier, the other real sentence having a target concreteness level,
the classifier comparing the revised sentence to the other real sentence and, in response, producing a similarity classification,
submitting the revised sentence generated by the machine learning model and another fake sentence into a discriminator,
the discriminator comparing the revised sentence and the fake sentence and, in response, producing a real/fake classification, and
updating parameters of the machine learning model based on the similarity classification and based on the real/fake classification;
building a concreteness-level classification tree via a computer performing concreteness level iterative continuous dichotomy of multiple sentences of a sample text corpus to determine a total number of concreteness levels corresponding to a respective semantic category;
determining, via the computer performing natural language processing, a concreteness level of a first individual sentence of a text corpus, wherein the determining comprises the computer passing the first individual sentence through the concreteness-level classification tree;
generating, via the computer and based on the natural language processing, a proposed change of the first individual sentence, the first individual sentence with the proposed change having a modified concreteness level and preserving a general meaning of the first individual sentence, wherein the generating the proposed change of the first individual sentence comprises inputting the first individual sentence and a target concreteness level into the trained machine learning model and, in response, receiving the proposed change as output of the trained machine learning model; and
transmitting, via the computer, the proposed change for presentation of the proposed change.