IP Library › Granted Patent US 12,265,782
Granted Patent B2
US 12,265,782 · App. 18/525,763 · Granted Apr 1, 2025

Transformer model architecture for readability

Inventors: Jing Wang (Atlanta, GA); John Matthew Mastin (Atlanta, GA); Sowmyanka Andalam (Cumming, GA); Piyasa Molly Paul (Decatur, GA); Dallas Leigh Taylor (Atlanta, GA); Andres Castro (Atlanta, GA)
Assignee: Intuit Inc.
G06F40/151G06F40/166G06F40/253G06F40/284
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,265,782
App. No.
18/525,763
Granted
Apr 1, 2025
Kind
B2
Abstract

A method including detecting, in a written electronic communication, an input sentence satisfying a readability metric threshold. The method also includes transforming, by a sentence transformer model, the input sentence to output suggested sentences. The method also includes evaluating the suggested sentences along a set of acceptability criteria. The method also includes determining, based on the evaluating, that the set of acceptability criteria is satisfied. The method also includes modifying, based on determining that the set of acceptability criteria is satisfied, the written electronic communication with the suggested sentences to obtain a modified written electronic communication. The method also includes returning the modified written electronic communication.

Claims (65)

1. A method comprising:

detecting, in a written electronic communication, a first input sentence satisfying a readability metric threshold;

transforming, by a sentence transformer model, the first input sentence to output a plurality of suggested sentences;

evaluating the plurality of suggested sentences along a set of acceptability criteria;

determining, based on the evaluating, that the set of acceptability criteria is satisfied;

modifying, based on determining that the set of acceptability criteria is satisfied, the written electronic communication with the plurality of suggested sentences to obtain a modified written electronic communication; and

returning the modified written electronic communication.

2. The method of claim 1 , wherein each of the plurality of suggested sentences contain fewer words that the first input sentence.

3. The method of claim 1 , further comprising:

displaying on a display device, responsive to determining that the set of acceptability criteria is satisfied, an alert window indicating that the first input sentence may be shortened.

4. The method of claim 3 , further comprising:

displaying a widget in the alert window, wherein the widget is configured to be selected by a user to accept the modified written electronic communication,

wherein modifying of the written electronic communication is performed responsive to selection of the widget by the user.

5. The method of claim 1 , wherein the plurality of suggested sentences together have a new semantic meaning similar to the written electronic communication.

6. The method of claim 1 , wherein, as part of modifying the written electronic communication, the method further comprises:

excluding a set of excluded words from the modified written electronic communication.

7. The method of claim 1 , wherein evaluating comprises:

transforming, by a vector embedding model, the plurality of suggested sentences to a suggested set vector embedding;

transforming, by the vector embedding model, the first input sentence to an input sentence vector embedding; and

calculating a similarity metric between the suggested set vector embedding and the input sentence vector embedding,

wherein determining that the acceptability criteria is satisfied is based at least in part on the similarity metric satisfying a similarity threshold.

8. The method of claim 1 , wherein evaluating comprises:

executing a natural language tokenizer on the plurality of suggested sentences to obtain tokenizer output; and

determining a number of sentences in the tokenizer output,

wherein determining that the acceptability criteria is satisfied is based at least in part on the number of sentences being greater than one.

9. The method of claim 1 , wherein evaluating comprises:

calculating a first number of bias terms in the plurality of suggested sentences; and

calculating a second number of bias terms in the first input sentence,

wherein determining that the acceptability criteria is satisfied is based at least in part on the first number being less than the second number.

10. The method of claim 1 , wherein evaluating comprises:

executing a grammar model on the plurality of suggested sentences to obtain a first grammar score; and

executing the grammar model on the first input sentence to obtain a second grammar score,

wherein determining that the acceptability criteria is satisfied is based at least in part on the first grammar score being greater than the second grammar score.

11. The method of claim 1 , further comprising:

detecting, in the written electronic communication, a second input sentence satisfying the readability metric threshold;

processing, by the sentence transformer model responsive to the second input sentence satisfying the readability metric threshold, the second input sentence to output a second suggested set of sentences;

evaluating the second suggested set of sentences along the set of acceptability criteria; and

determining that the second suggested set of sentences fails to satisfy the set of acceptability criteria; and

presenting a readability metric alert window advising of the second input sentence satisfying the readability metric threshold, the readability metric alert window presented without the second suggested set of sentences.

12. The method of claim 1 , further comprising:

presenting a readability metric alert window advising of the first input sentence satisfying the readability metric threshold, the readability metric alert window comprising the plurality of suggested sentences.

13. A system comprising:

a data repository storing a written electronic communication comprising an input sentence; and

a computer processor, in communication with the data repository, for executing:

a sentence transformer model configured to:

process the input sentence to output a plurality of suggested sentences,

an evaluation process configured to:

detect that the input sentence satisfies a readability metric threshold, evaluate the plurality of suggested sentences along a set of acceptability criteria,

determine, based on evaluating, that the set of acceptability criteria is satisfied, and

modifying, based on determining that the set of acceptability criteria is satisfied, the written electronic communication with the plurality of suggested sentences to obtain a modified written electronic communication; and

a graphical user interface in communication with the computer processor and configured to:

present, responsive to the set of acceptability criteria being satisfied, the plurality of suggested sentences.

14. The system of claim 13 , further comprising:

a network connection configured to transmit the modified written electronic communication to a recipient.

15. The system of claim 13 , wherein each of the plurality of suggested sentences contain fewer words that the input sentence.

16. The system of claim 13 , further comprising:

a display device,

wherein the computer processor is further programmed, responsive to determining that the set of acceptability criteria is satisfied, to display on the display device an alert window indicating that the input sentence may be shortened.

17. The system of claim 16 , wherein the computer processor is further programmed to display a widget in the alert window, wherein:

the widget is configured to be selected by a user to accept the modified written electronic communication; and

modifying of the written electronic communication is performed responsive to selection of the widget by the user.

18. The system of claim 13 , wherein the plurality of suggested sentences together have a new semantic meaning similar to the written electronic communication.

19. The system of claim 13 , wherein, as part of modifying the written electronic communication, the evaluation process is further configured to exclude a set of excluded words from the modified written electronic communication.

20. The system of claim 13 , further comprising:

wherein the computer processor is further programmed to edit, in a window of the graphical user interface, the written electronic communication.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Feb 24, 2025
From: WANG, JING; MASTIN, JOHN MATTHEW, JR.; ANDALAM, SOWMYANKA; PAUL, PIYASA MOLLY; TAYLOR, DALLAS LEIGH; CASTRO, ANDRES
To: INTUIT INC.
Reel/Frame 070309/0748 →
Continuity (2)
Continuation 18104258 · Jan 31, 2023
Related Publication 20240256759A1 · Aug 1, 2024
References Cited (20)
US 11258734B1 · Shevchenko · 2022 [cited by examiner]
US 20050267735A1 · Kharrat · 2005 [cited by examiner]
US 20130096909A1 · Brun · 2013 [cited by examiner]
US 20150310571A1 · Brav · 2015 [cited by examiner]
US 20160103875A1 · Zupancic · 2016 [cited by examiner]
US 20170193093A1 · Byron · 2017 [cited by examiner]
US 20170277781A1 · Deolalikar · 2017 [cited by examiner]
US 20200142504A1 · Yu · 2020 [cited by examiner]
US 20200394361A1 · Parikh · 2020 [cited by examiner]
US 20210149933A1 · Chang · 2021 [cited by examiner]
US 20220030110A1 · Khafizov · 2022 [cited by examiner]
US 20220083725A1 · Pande · 2022 [cited by examiner]
US 20220114340A1 · Graeser · 2022 [cited by examiner]
US 20220188514A1 · Thota · 2022 [cited by examiner]
US 20220293271A1 · Chang · 2022 [cited by examiner]
US 20230153546A1 · Peleg · 2023 [cited by examiner]
US 20230177878A1 · Sekar · 2023 [cited by examiner]
Kaychak, D., “Text Simplification Data Sets”, https://os.pomona.edu/˜dkauchak/simplification, accessed on Jan. 23, 2023, 1 page. [cited by applicant]
Gao, T., et al., “SimCSE: Simple Contrastive Learning of Sentence Embeddings”, Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, Nov. 7-11, 2011, 17 pages. [cited by applicant]
Raffel., C., et al., “Exploring the Limits of Transfer Learning with a Unified Text-to-Text transformer”, Journal of Machine Learning Research, Jan. 20, 2020, 67 pages. [cited by applicant]