Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
No claims are amended. Claims 1-20 are presented for examination.
Response to Arguments
Applicant arguments filed on 8/26/2026 have been reviewed. Following is the response.
35 U.S.C. §103
Applicant argues “In the Office Action, the Examiner has stated that the feature "generate a prompt to a large language model (LLM) for generating a subject line, the prompt including a chosen list of keywords based on the original list of keywords, wherein the prompt excludes the body of text such that a number of tokens provided to the LLM is reduced relative to the body of text" is disclosed in He. The Examiner asserts that this feature is disclosed by He because He teaches generating a prompt that includes keywords while excluding the body of the text. Specifically, the Office Action relies on He's disclosure that a guiding token sequence may be used as both a keyword input and a decoder prompt, and concludes that this corresponds to the claimed prompt including a chosen list of keywords while excluding the body of the text. Applicant respectfully disagrees. As described in Section 2.1, He expressly models summarization as p(ylx,z), where x is the source document, y is the summary, and z is a keyword sequence. Thus, the generated summary is conditioned on both the source document and the keywords” However, examiner did not rely on section 2.1. Examiner relied on the following section this this limitation - generate a summary from guiding tokens used as prompts and keywords, Table 28-29, Under 4.5; here the prompt does not include the full body of text instead the way to generate summarization is either solely based on guiding tokens defining the keywords or keyword+prompt = prompt , Under 6. Conclusion
Additionally, the p(y|x, z) is a probability distribution
Summary or the gcontrol is conditioned on the source document but when the prompt is being created its using guiding token sequence for each task, where gcontrol can be used to constrain the length of the output, Table 23-24; also the prompt is generated from keyword and prompt from guiding token which does not include full text from article
Applicant further argues “He further explains in Section 2.4 that the keyword sequence is prepended to the source document, separated by a special token, and that the resulting combined sequence is supplied to the model as the input. In other words, the keywords supplement the source document. They do not replace it. He therefore does not teach excluding the body of text from the information provided to the model. However, this portion is irrelevant to what is being relied upon for the claimed invention. Examiner agrees that in He keyword can be derived from the source document, but the prompt excludes source document and just uses the keyword which is similar to applicant own invention. Nowhere do the claims include the concept of keyword not being derived from the body of text.
Applicant further argues “In Section 3.2 of He, the prompt is a short guiding sequence, such as "the purpose of the present invention is," that constrains or guides the form of the generated output. He expressly distinguishes this decoder-side prompt from the keyword input used to guide the encoder. The prompt is therefore not described as a replacement for the source document. The fact that the short guiding sequence does not itself reproduce the source document does not establish that the source document is excluded from the model input. To the contrary, Sections 2.4 and 3.2 clearly disclose that the source document continues to be provided to the model together with the guiding information or prompt.
He's keyword-dropout discussion in Section 2.4 further confirms this distinction. He explains that the automatically extracted keywords can be sufficiently informative that the model may learn to ignore the source document. To address that problem, He randomly drops keywords during training so that the model learns to use the keywords while continuing to carry information from the source document. This disclosure is inconsistent with the Examiner's interpretation. If He excluded the source document and generated the output solely from a keyword-based prompt, there would be no source document for the model to ignore. Instead, He treats continued reliance on the source document as a design objective.
Accordingly, He does not teach or suggest generating a prompt that excludes the body of text so as to reduce the number of tokens provided to the LLM relative to the body of text. He teaches an opposite arrangement in which the source document remains part of the model input, while keywords serve as additional control information.” However, He explicitly mentions that prompt is based on keyword and gcontrol, which is a guiding token or decoder side prompt and hence during the prompt creation the source document is not being directly use which suffice the claim limitation. He gives different ways of generating summarization. Examiner is not saying that in the entire document He is not disclosing the use of source document but the specific portion for e.g. section 4.5 and concluding and table 23-24 is explicitly teaches applicant invention.
Applicant argues “Wu fails to remedy the deficiencies of He. The Office Action relies on Wu for generating a subject line rather than a summary but Wu would not alter the manner in which He's model is conditioned. He expressly teaches a model that requires p(ylx,z), where the source document and keyword sequence are provided together as the model input. Simply substituting a subject line for the generated summary does not remove the source document from that input. Accordingly, even under the proposed combination, the resulting system remains conditioned on both the source document and the keywords, rather than on a prompt that excludes the body of the text.
Nor does Wu supply the missing teaching. Wu teaches generating a subject line from keywords, but does not disclose excluding the body of a source document from the information provided to the language model, nor does it suggest modifying He's document-conditioned generation to omit the source document while retaining the disclosed functionality. Thus, Wu does not cure the deficiency of He.
The Examiner's stated rationale for combining the references likewise does not bridge this gap. The cited motivation of improved efficiency relates to Wu's training methodology and does not explain why one of ordinary skill in the art would have excluded the source document from He's disclosed model input, particularly where He teaches that the source document remains an essential conditioning input during generation.” However, examiner did not rely on Wu for the concept of subject line.
Applicant further argues “He'284 likewise fails to remedy the deficiencies of He and Wu. The Examiner relies on He'284 for modifying an original set of keywords to obtain a chosen set of keywords. He'284 teaches generating summaries conditioned on both the source document and the keyword sequence. For example, He'284 expressly teaches prepending the generated keyword sequence to the source document before providing the combined sequence to the summarization model (Fig. 4).” Same reasons as in He is applied here.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
And
KSR, 550 U.S. at 418, 82 USPQ2d at 1396. Exemplary rationales that may support a conclusion of obviousness include:
(A) Combining prior art elements according to known methods to yield predictable results;
(B) Simple substitution of one known element for another to obtain predictable results;
(C) Use of known technique to improve similar devices (methods, or products) in the same way;
(D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results;
(E) "Obvious to try" – choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success;
(F) Known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of ordinary skill in the art;
(G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention.
See MPEP § 2143 for a discussion of the rationales listed above along with examples illustrating how the cited rationales may be used to support a finding of obviousness. See also MPEP § 2144 - § 2144.09 for additional guidance regarding support for obviousness determination.
Claims 1-8, 10-18 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over He (CTRLSUM: Towards Generic Controllable Text Summarization - 2022.emnlp-main.396.pdf) and further in view of Wu ( US 20240143941) and further in view of He ( US 20220067284 ) hereinafter He’284
Regarding claim 1, He teach a system comprising: a processing unit configured to execute instructions to cause the system to: tokenize a body of text into numerical tokens ( tokenize the document, 2.2; also A.1 – tokenization) ; generate embedding vectors representing the tokens, the embedding vectors capturing semantic relationships between tokens (BERT based training at inference and/or utilize any sequence to sequence model for the extraction, Under 2.2-2.3; BERT is a well-known model to capture the sematic relationship between token) ; input the embedding vectors into a trained machine learning model to generate an original list of keywords summarizing the body of text based on the semantic relationships (keyword based on the guiding token and keyword +prompt for summarization, Under 3.2; utilize any sequence to sequence model for keyword extraction, Under 2.2; where CTRLSUM is our fine-tuned version of the pretrained BARTLARGE model (Lewis et al., 2020)., Under 4.1- BART intherently captures the sematic relationships ) generate a prompt to a large language model (LLM) for generating a subject line, the prompt including a list of keywords based on the original list of keywords ( we design a guiding token sequence for each task, which is used as both the keyword input and the decoder prompt., Under 3.2, Table 28-29; while prompt and keyword can be viewed as unified “prompt” which are both guiding tokens with different usage., Page 5883), wherein the prompt excludes the body of text such that a number of tokens provided to the LLM is reduced relative to the body of text ( gcontrol can be used to constrain the length of the output, Table 23-24; also the prompt is generated from keyword and prompt from guiding token which does not include full text from article) ; and obtain, from the LLM responsive to the prompt, at least one generated subject line corresponding to the body of text ( generate a summary from guiding tokens used as prompts and keywords, Table 28-29, Under 4.5; here the prompt does not include the full body of text instead the way to generate summarization is either solely based on guiding tokens defining the keywords or keyword+prompt = prompt , Under 6. Conclusion)
While, He's body of text is an article and summary can be viewed as a subject line since the process is same if the article is a email body of text, one line summary with the length control factor can be taken as a subject line but He does not teach the concept of email and a prompt generating a subject line
However, Wu teaches the concept of email and the subject line and the prompt generating a subject line ( using a generative model to generate a subject line using keyword, Para 0068; just using keywords and NER, Para 0068; generative the subject line, Fig 3, Fig 7)
He already has a process of generating keywords based on the document/article/body of text and based on those keywords generate a summary of a particular length. Wu has an improved way of generating subject line using the keyword and it would have been obvious to POSITA having the teachings of He to further modify with the concept Wu to using He concept to generate a subject line in place of summary to provide improved efficiency in the subject generation system (Para 0034, Wu)
He modified by Wu does not explicitly teach a chosen list of keywords based on the original list of keywords
However He’284 teaches a chosen list of keywords based on the original list of keywords (the user 150 can also edit the customized keywords, which allows for more flexible customized summarization without the user manually editing the summary directly, user input of a control token sequence and/or one or more control parameters relating to a characteristic of the summary to be generated may be received to modify the set of keywords into a customized set of keywords, e.g., via the control center 232 in FIG. 2. During inference time, the user 150 may provide different configurations of control tokens 232 reflecting keyword control to entity and length of the summary. Para 0023, 0045-0046)
It would have been obvious to a POSITA having the teachings of He and Wu to further incorporate the concept of He’284, thereby allows for more flexible customized summarization without the user manually editing the summary directly ( Para 0023, He’284)
Regarding claim 2, He’284 as above in claim 1, teach , wherein the processing unit is further configured to provide the original list of keywords to a user device, wherein the user device is configured to present the original list of keywords as suggestions (control section present user with keyword user can edit the keyword or modify the keyword, Para 0045-0046, Fug 1b, He’284)
Regarding claim 3,He’284 as above in claim 3, wherein the processing unit is further configured to receive the chosen list of keywords from the user device ( user can edit or modify the keyword, Para 0023, 0045-0046, Fig 1)
Regarding claim 4, He’284 as above in claim 4, teach wherein the chosen list of keywords is the original list of keywords or includes at least one of the keywords from the original list of keywords ( the keywords provide a generic interface to control multiple aspects of summaries, which allows the user to optionally rely on automatically extracted keywords, user provided keywords, or a combination of both, Para 0026)
Regarding claim 5,He modified by Wu and He’284 as above in claim 1, teach :provide the at least one generated subject line to a user device for presentation; receive a revised list of keywords, based on the chosen list of keywords, from the user device; generate another prompt to the LLM for generating another subject line, the prompt including the revised list of keywords, wherein the prompt does not include the body of text; and obtain, from the LLM, another at least one generated subject line ( ( Fig 2, Subject Line– user can change the keyword) ; generate another prompt to the LLM for generating another subject line, the prompt including the revised list of keywords, wherein the prompt does not include the body of text; and obtain, from the LLM, another at least one generated subject line (user can change the keyword and the updated subject will be generated, Fig 2, Wu; user can edit the keyword, Para 0023, He’284)
Regarding claim 6, Wu as above in claim 5, teaches wherein the processing unit is further configured to :save the at least one generated subject line in memory; and provide the at least one generated subject line and the other at least one generated subject line to the user device for presentation ( Fig 9-12, Wu multiple subject lines)
Regarding claim 7, Wu as above in claim 5, wherein the processing unit is further configured to: provide the chosen list of keywords with the at least one generated subject line to the user device for presentation ( Fig 9, Wu; user can edit the keyword, Para 0023-0026, He’284)
Regarding claim 8, He modified by Wu and He’284 as above in claim 1, teach wherein the processing unit is further configured to: provide the at least one generated subject line to a user device for presentation; provide feedback options for the at least one generated subject line to the user device for presentation; and receive feedback from the user device ( user can modify the keywords and/or blacklist etc., and generate an updated subject, Fig 9; customize keyword to generate customized summary and keyword can be modified, Para 0023-0026, 0045-0046, He’284)
Regarding claim 10, He as above in claim 1, teach wherein the trained machine learning model is a text summarizer that uses natural language processing (NLP) techniques ( CTRLsum, Abstract)
Regarding claim 11, rejections analogous to claim 1, are applicable.
Regarding claim 12, rejections analogous to claim 2, are applicable.
Regarding claim 13, rejections analogous to claim 3, are applicable.
Regarding claim 14, rejections analogous to claim 4, are applicable.
Regarding claim 15, rejections analogous to claim 5, are applicable.
Regarding claim 16, rejections analogous to claim 6, are applicable.
Regarding claim 17, rejections analogous to claim 7, are applicable.
Regarding claim 18, rejections analogous to claim 8, are applicable.
Regarding claim 20, rejections analogous to claim 1, are applicable.
Claim 9 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over and further in view of He (CTRLSUM: Towards Generic Controllable Text Summarization - 2022.emnlp-main.396.pdf) and further in view of Wu ( US 20240143941) and further inview of He ( US 20220067284 ) hereinafter He’284 and Agrawal ( US 20240312087)
Regarding claim 9, He modified by Wu and He’284 does not teach wherein the prompt further includes one or more of historical email data, business information, and user demographic information associated with a user.
In the same field of endeavor Agrawal teaches wherein the prompt further includes one or more of historical email data, business information, and user demographic information associated with a user (For instance, to generate a subject line for an email on a certain occasion, the following sentence prompt is provided to a generative model: “generate one short, exciting email subject line from {brand} on {occasion}.” Each generative model of the content generation apparatus 110 is fine-tuned with multiple (e.g., 100) examples, Para 0036)
It would have been obvious having the teachings of He and Wu and H’284 to further include the concept of Agrawal to make the subject line more consistent with brands/organizations ( Abstract, Agrawal)
Regarding claim 19, arguments analogous to claim 9, are applicable.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Mohammad (US 20250139382)
Mishra (US 20250094711) teaches obtain an original list of keywords based on a body of text using a trained machine learning model (obtain a keyword/prompt word for text segments, Para 0147, 0154)
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Richa Sonifrank whose telephone number is (571)272-5357. The examiner can normally be reached M-T 7AM - 5:30PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Phan Hai can be reached at (571)272-6338. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Richa Sonifrank/ Primary Examiner, Art Unit 2654
