DETAILED ACTION
This Office Action is responsive to amendments and arguments filed on July 1st, 2026. Claims 1-2, 4-9, 11-16 and 19-20 are amended, claims 1-20 are pending and have been examined; hence, this action is made FINAL.
Any previous objections/rejections not mentioned in this Office Action have been withdrawn by the Examiner.
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 .
Priority
Applicant’s claim for the benefit of a prior-filed application 63/538747, filed September 15th, 2023, under 35 U.S.C. 119(e) is acknowledged. Accordingly, the current application is granted the benefit of the earlier filing date.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on September 27th, 2024; July 1st, 2025; August 26th, 2025; and December 5th, 2025 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Response to Arguments and Amendments
Regarding rejections made under 35 U.SC. 101, Applicant argues that "the claims as amended do impose a meaningful limit on any alleged abstract idea, and improve the functioning of the computing system and overall field of natural language processing," (starting on page 11 of Remarks). Applicant goes on to argue, "Returning references in this fashion is more accurate, faster, and requires fewer computing resources than doing so with the LLM itself. (Spec. par. 76). Furthermore, LLMs are extremely costly in terms of time and computational resources, so using the embedding based techniques rather than the LLM to return references also decreases latency and spares computational resources. (Spec. par. 107)."
Applicant’s arguments are persuasive; the claimed invention describes an improvement to the functioning of a computer system used to provide answers and responses to queries with relevant citations, as described in Specification paragraphs [0073]-[0076]. Accordingly, the rejections under 35 U.S.C. 101 are withdrawn.
Regarding rejections made under 35 U.S.C. 103 in view of Raimondo and Ma, Applicant argues that "Ma's architecture does not generate references without use of language models. To verify it's extracted text spans and append citations, Ma forces the text through a secondary, resource-heavy multi-model pipeline," and further, "Because Ma requires cascading language model layers and trained neural network fusion classifiers to achieve sentence alignment, the proposed combination cannot yield the streamlined, embedding-to-embedding direct verification framework defined by Claim 1,” (starting on page 17 of Remarks).
Applicant’s argument is not persuasive. While the teachings of Ma describe the use of multiple LLMs in question answering, attention is brought to the limitations that claim an improvement of “generating the references without use of the response LLM;” in the claims, the response LLM provides an answer to a query, and then that LLM is not used to create citations. In Ma’s architecture, “a second LLM” is invoked to compare a query to candidate answers, thereby processing the citations in a component separate from the one called to provide the answers. Further, the teachings of Raimondo do not rely on an LLM (a model that may be used to generate answers) but models trained specifically for semantic comparisons. Because the claims do not use “the response LLM” rather than “any LLM,” the teachings of Raimondo and Ma can be obviously combined to teach the claimed method. Accordingly, the rejections under 35 U.S.C. 103 are maintained.
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.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication 2025/0190455 to Raimondo (hereinafter, "Raimondo") in view of China Invention Application 116340467 to Ma et al. (hereinafter, "Ma").
Regarding claims 1, 8 and 15, Raimondo teaches a method, system and computer-readable medium comprising: accessing, by a computing system provided as a cloud computing service, a text portion comprising a retrieved document chunk from a knowledge base for question answering tasks stored in a context and memory store (paragraph [0131], "In one or more embodiments, the semantic representation generator 320 has access to the plurality of text objects 242 stored in the database 240 including one or more of: a plurality of documents, a plurality of paragraphs, a plurality of sentences, and a plurality of words," and paragraph [0034], "In the context of the present specification, a 'database' is any structured collection of data, irrespective of its particular structure, the database management software, or the computer hardware on which the data is stored, implemented or otherwise rendered available for use. A database may reside on the same hardware as the process that stores or makes use of the information stored in the database or it may reside on separate hardware, such as a dedicated server or plurality of servers.");
identifying, by the computing system, a plurality of sentences in the text portion (paragraph [0131], "In one or more embodiments, the semantic representation generator 320 has access to the plurality of text objects 242 stored in the database 240 including one or more of: a plurality of documents, a plurality of paragraphs, a plurality of sentences, and a plurality of words.");
embedding, by an embedding model of the computing system, each of the plurality of sentences in the text portion to generate a respective plurality of text sentence embeddings (paragraph [0140], "The semantic representation generator 320 generates, for a given sentence 414, a respective sentence representation 434. The respective sentence representation 434 is indicative of semantic features of the text content of the given sentence 414 and thus captures at least a portion of its meaning.");
comparing, by the computing system, each of the plurality of response embeddings to each of the plurality of text sentence embeddings to generate a similarity score indicating a similarity between the retrieved document chunk and the LLM response for each sentence embedding-response embedding pair (paragraph [0161], "The semantic representation comparator 360 uses one or more of the set of semantic comparison ML models 270 to determine a semantic similarity between two text objects. It is contemplated that the semantic representation comparator 360 may use topological similarity and/or statistical similarity techniques to determine the semantic similarity score between two text objects, in addition to the set of semantic comparison ML models 270.");
generating, by the computing system, references comprising a subset of the plurality of sentences based on the similarity score for each sentence embedding-response embedding pair, thereby generating the references without use of the response LLM (paragraph [0168], "In some embodiments, to compute the semantic similarity between two text objects, the semantic representation comparator 360 may compute the semantic similarity score between the representation components of the two text objects. As a non-limiting example, to compare the paragraph representation 432 and the other paragraph representation 442, the semantic representation comparator 360 may compare at least some of the sentences and word representations of the sentences and words in the paragraphs 412, 422, which may be used to determine the semantic similarity score between the paragraph representation 432 and the other paragraph representation 442. The semantic representation comparator 360 provides an indication of the components used to determine the semantic similarity score, which may be used as most relevant keywords and passages displayed together with the paragraphs 412, 422."); and
outputting, by the computing system, the references with the response to the query (paragraph [0186], "The semantic search interface generator 380 is configured to inter alia: (i) receive a search request including an indication of a text object; (ii) receive, based on the search request, a set of semantically similar text object; and (iii) generate the interactive semantic search interface 225 comprising the set of semantically similar text objects to be provided in response to the search request.").
Raimondo does not explicitly teach “providing, by the computing system to a response large language model (LLM), the text portion or a derivative thereof and a query,” “receiving, by the computing system from the language model, a response to the query based on the text portion,” “identifying, by the computing system, a plurality of sentences in the response,” or “embedding, by the computing system, the plurality of sentences in the response to generate a plurality of response embeddings,” and thus, Ma is introduced.
Ma teaches providing, by the computing system to a response large language model (LLM), the text portion or a derivative thereof and a query (page 7, "receiving the embodiment, respectively combining the query text with the candidate reference text fragment into text pair, then processing the text pair by invoking the second language model, to obtain the matching score between the query text and a plurality of candidate reference text fragments, matching score represents the similarity between the semanteme of the text, the higher the matching score is, the semantic is closer.");
receiving, by the computing system from the response LLM, a response to the query based on the text portion (page 8, "After determining the optimum matching segment, then it is necessary to determine which part of the text segment from optimum matching segment belongs to the answer text segment… In step 1035, the part between the initial element and the end optimum the matching segment, as the answer text segment matched with the text query");
identifying, by the computing system, a plurality of sentences in the response (page 8, "each text search result is finally screened to obtain an answer text segment, then still need to further screen from a plurality of answer text segment. The specific screening process is as follows: determining the relevance of query text respectively with a plurality of answer text segments, the correlation of the maximum value corresponding to the answer text segment, determining the reference text of the text query."); and
embedding, by the computing system, the plurality of sentences in the response to generate a plurality of response embeddings (page 8, "In step 1041, the query text, answer text segment, and each element in the title text as a conventional mark, the conventional mark of each element is spliced into the second mark sequence," and page 9, "In step 1042, the second mark sequence is embedded processing to obtain the embedded feature vector of the second mark sequence.").
Raimondo and Ma are considered analogous because they are each concerned with information retrieval and question answering. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified the semantic comparison of Raimondo with the cooperative models of Ma for the purpose of improving question answering efficiency. Given that all the claimed elements were known in the prior art, one skilled in the art could have combined the elements by known methods with no change in their respective functions, and the combination would have yielded nothing more than predictable results.
Regarding claims 2, 9 and 16, Raimondo further teaches a method, system and computer-readable medium wherein comparing the plurality of response embeddings to the plurality of text sentence embeddings comprises computing the similarity scores based on cosine similarity (paragraph [0096], "In one or more embodiments, the semantic similarity score may be calculated using a distance between two semantic representations in a semantic space. As a non-limiting example, the semantic similarity may be calculated using a cosine similarity between two representations.").
Regarding claims 3, 10 and 17, Raimondo further teaches a method, system and computer-readable medium comprising comparing, by the computing system, each of the similarity scores to a predetermined threshold value (paragraph [0180], "In one or more embodiments, the semantic representation retriever 340 determines a threshold dynamically based on at least one of a number of text objects, the type of text objects, and the respective similarity scores of the text objects. In one or more other embodiments, the threshold may be predetermined."); and
selecting, by the computing system, the subset of the plurality of sentences based on whether the similarity scores meet or exceed the threshold value (paragraph [0181], "The semantic representation retriever 340 determines a set of semantically similar text objects based on the threshold. In one or more embodiments, the semantic representation retriever 340 selects the set of relevant text objects from the set of potentially semantically similar text objects based on a similarity score threshold. As a non-limiting example, the semantic representation retriever 340 may select text objects associated with semantic similarity score that is equal to or above 0.8 on a scale of 1.").
Regarding claims 4, 11 and 18, Raimondo further teaches a method, system and computer-readable medium wherein accessing the text portion comprises accessing a document and splitting the document into a plurality of text portions including the document chunk (paragraph [0086], "In one or more embodiments, a given semantic representation ML model may use techniques such as bag-of-words, tokenization, stop words removal, stemming, lemmatization, morphological segmentation, latent Dirichlet allocation (LDA), and the like to generate a semantic representation of a text object," paragraph [0127], "The purpose of the semantic representation generator 320 is to process a text object, e.g., document, paragraph, sentence, word, or a combination thereof so as to generate a machine understandable representation of its meaning in a semantic space, such that at least semantic representations of the same type of text objects (i.e., document-document, paragraph-paragraph, sentence-sentence, word-word) may be compared to each other and the semantic similarity (or dissimilarity) of the text objects may be quantified via a semantic similarity score," and paragraph [0128], "The semantic representation generator 320 uses one or more of the set of semantic representation ML models 260 to generate each of the document representation 430, 440, the paragraph representation 432, 442, the sentence representations 434, 444 and the word representations 436,446.").
Regarding claims 5 and 12, Raimondo further teaches a method and system wherein accessing the text portion further comprises selecting and retrieving the document chunk (paragraph [0172], "The semantic representation retriever 340 determines a set of potentially relevant or potentially semantically similar text objects from the plurality of text objects 242 stored in the database 240. The set of potentially similar text objects includes at least a portion of the plurality of text objects 242 stored in the database 240.").
Regarding claims 6, 13 and 19, Raimondo further teaches a method, system and computer-readable medium comprising causing display of the response on a graphical user interface, wherein, upon detecting user interaction with the graphical user interface to request reference information, the computing system causes display of the references (paragraph [0070], "The search engine server 220 is configured to inter alia: (i) provide an interactive semantic search interface 225 and a search engine functionality; (ii) receive, from the client device 210 at the interactive semantic search interface 225, a search request, the search request including an indication of a text object including one or more of: a document, a paragraph, a sentence, and a word; (iii) receive, based on the search request, an indication of a semantic representation of the search request; (iv) receive, based on the indication of the semantic representation of the search request, a set of potentially similar text objects each associated with a respective semantic similarity score; (v) determine, based on the respective semantic similarity scores, a set of semantically similar text objects; and (vi) provide, for display on the client device 210, an interactive semantic search interface 225 comprising the set of semantically similar text objects.").
Regarding claims 7, 14 and 20, Raimondo further teaches a method and computer-readable medium wherein the references are displayed in a pop-up window over the displayed response (paragraph [0202], "FIG. 5 shows a first interactive semantic search interface 500 with an input keyword 510 as the input text object. The first interactive semantic search interface 500 comprises a set of semantically similar keywords 520, a set of semantically similar documents 530 and a set of semantically similar sentences 540 having been determined to be semantically similar to the input keyword 510 by the semantic search system 300. The user 215 may select a semantically similar text object such as one of the set of semantically similar keywords 520, the set of semantically similar documents 530 and the set of semantically similar sentences 540 to display more details about how the semantic similarity of the text object was determined, such as the respective similarity score (not shown), the respective components text objects in the text object (not shown), and the like.").
Conclusion
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.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
U.S. Patent 10,430,445 to Crouch et al.
U.S. Patent 10,997,221 to Chakraborty et al.
U.S. Patent 11,074,284 Cunico et al.
U.S. Patent 11,741,139 to Zhou et al.
U.S. Patent 12,505,136 to Khosla et al.
U.S. Patent Application Publication 2021/0133264 to Bobbarjung et al.
U.S. Patent Application Publication 2022/0405484 to Kanchibhotla et al.
U.S. Patent Application Publication 2023/0267267 to Sulka.
U.S. Patent Application Publication 2024/0403341 to Berglund et al.
European Patent Application 3910493 to Yuzhen et al.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SEAN T SMITH whose telephone number is (571)272-6643. The examiner can normally be reached Monday - Friday 8:00am - 5:00pm.
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, PIERRE-LOUIS DESIR can be reached at (571) 272-7799. 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.
/SEAN THOMAS SMITH/Examiner, Art Unit 2659
/BHAVESH M MEHTA/Supervisory Patent Examiner, Art Unit 2656
Prosecution Insights