DETAILED ACTION
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 .
Information Disclosure Statement
The information disclosure statement(s) (IDS) submitted on 29 October 2024 is/are being considered by the examiner.
Claim Objections
Claims 1, 12, and 17 objected to because of the following informalities:
Regarding claim 1, and mutatis mutandis claims 12 and 17, the phrase “a contextual unit” at line 9 should read as “a contextual unit of the plurality of contextual units” to clarify the existing antecedent basis for the claim part.
Appropriate correction is required.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-8 and 12-19 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Gurari (WO 2026019889 A1, hereinafter Gurari).
Regarding claim 1, Gurari discloses A method for augmenting an information repository for information retrieval by a large language model (Discloses “systems and methods for LLM-enabled query responses” to “leverage large language models (LLMs) for processing one or more reference sources of information to generate accurate, concise, and interpretable (i.e., transparent in their source) research answers for given queries.”; Gurari, ¶ [0018]), the method comprising: obtaining an information item from an information repository (Discloses receiving “a corpus of source information” where the “source information could be a text, a collection of text, or any suitable source of information”; Gurari, ¶ [0042]); allocating portions of the information item into a plurality of contextual units (The generator 110 can further include “segmentation algorithms for dividing source information into processable segments”; Gurari, ¶ [0042]), wherein each contextual unit of the plurality of contextual units comprises a portion of the information item (The “different segments {the contextual units}” are segments divided from source information {the information item} from the “corpus of source information.”; Gurari, ¶ [0042]); generating, by a first language model, a plurality of content units (“The question-answer pair generator 110 may also include... answer identification algorithms for extracting potential answers from segments, and question generation algorithms for creating questions corresponding to identified answers {content units}”; Gurari, ¶ [0042]) based on the plurality of contextual units, wherein each content unit of the plurality of content units is associated with a contextual unit (The answers are extracted from the segments of the source information and the questions are generated “corresponding to the identified answers.” As such, the answers and generated questions {plurality of content units} are both based on the plurality of segments from the corpus of source information {plurality of contextual units} and each question and answer {each content unit} is associated with at least a respective segment {a contextual unit} of the plurality of segments from which the question and answer is generated.; Gurari, ¶ [0042]); constructing an augmented dataset (“The question-answer pair generator 110 functions to generate a set of question answer pairs from different segments of a corpus of source information” which can be maintained using “a storage engine that manages the physical storage and organization of question-answer pairs” and “an indexing system that creates and maintains various indexes for efficient searching and retrieval {an augmented dataset}.”; Gurari, ¶ [0042], [0046]-[0047]) that includes one or more content units of the plurality of content units (The set of question answer pairs{the plurality of content units} includes one or more question answer pairs {one or more content units}.; Gurari, ¶ [0042], [0046]-[0047]); receiving a query (The system further includes a “query interface 130... through which a request may be made” where the system includes “receiving a query through a query interface S130”; Gurari, ¶ [0048], [0060]); inputting the query and the one or more content units of the plurality of content units into one or more language models (Based on the received query, the system then identifies “candidate question-answer pairs by searching the stored question-answer pairs using the query S140”, which includes processing both the query and the “stored question-answer pairs” where “an LLM may be used for identifying the potential answers from segments in S114, generating questions for potential answers S116, and/or other processes” where other processes includes “LLM or [Retrieval augmented generation (RAG)]-based document search solutions”; Gurari, ¶ [0024], [0060]-[0061]); and obtaining, as output from the one or more language models, a response to the query based on the input query and the one or more content units of the plurality of content units (“generating a response to the query by processing the candidate question-answer pairs S150” as part of a RAG process, where the system uses a large language model alongside retrieved question answer pairs to generate a response to the query.; Gurari, ¶ [0003], [0024]-[0025], [0060]).
Regarding claim 2, Gurari discloses, wherein each content unit of the plurality of content units includes at least one of: a question set comprising a plurality of questions; a question-context set comprising a plurality of question-context tuples, wherein each question-context pair of the plurality of question-context tuples includes a contextual unit and a question; a question-answer set comprising a plurality of question-answer tuples, wherein each question-answer pair of the plurality of question-answer tuples includes a question and an answer associated with the question; or a question-context-answer set comprising a plurality of question-context-answer tuples, wherein each question-context-answer tuple of the plurality of question-context-answer tuples includes a contextual unit, a question, and an answer associated with the question (“The question-answer pair generator 110 may also include... answer identification algorithms for extracting potential answers from segments, and question generation algorithms for creating questions corresponding to identified answers” wherein the answer is an answer associated with the question and a contextual unit, and the pair may be viewed as tuple, as the question and answer are associated with the source information and the answer may be a contextual unit from which both the question and answer are derived.; Gurari, ¶ [0026], [0040]-[0042]).
Regarding claim 3, Gurari discloses wherein inputting the one or more content units of the plurality of content units into the one or more language models comprises: generating a similarity metric for each content unit of the one or more content units, wherein the similarity metric includes a measure of similarity between a respective content unit of the one or more content units and the received query (“the method may include computationally evaluating candidate questions for similarity to the query. This can include generating similarity scores or other types of computational characterization of similarity {a measure of similarity...} between the query {the received query} and candidate questions {a respective content unit of the one or more content units}.”; Gurari, ¶ [0103]); and selecting content units based on the similarity metric (“For a question-answer pair to be included in the candidate set of question-answer pairs, the question may be required to satisfy a minimum similarity score threshold to be used for answering the query. “; Gurari, ¶ [0103]).
Regarding claim 4, Gurari discloses wherein the similarity metric is a cosine distance (“The similarity scoring may utilize various metrics including cosine similarity of embedding vectors, semantic distance measures, keyword overlap scores, and weighted combinations of multiple similarity factors.”; Gurari, ¶ [0103]).
Regarding claim 5, Gurari discloses further comprising obtaining a portion of the selected content units, wherein inputting the one or more content units of the plurality of content units into the one or more language models comprises inputting the portion of the selected content units into the one or more language models (“the systems and methods may be used in combination with other LLM or [retrieval augmented generation (RAG)]-based document search solutions” where RAG is known in the art as the process of retrieving source content to act as grounding information for a LLM, where the source content is included as part of the prompt, alongside the query and various constraints, if any, to the LLM for generation of the response. As such, the system generates the similarity scores between the query and candidate questions, where the candidate questions correspond to the selected question answer pair of the plurality of question-answer pairs and “uses source information to pregenerate question-answer pairs, which can then be used at query-time to retrieve and generate accurate answers.” In the context of the RAG embodiment, the retrieved question answer pair is understood to have been added to a prompt for purposes of grounding a query to a LLM.; Gurari, ¶ [0024], [0042], [0103]).
Regarding claim 6, Gurari discloses wherein the first language model and the one or more language models are a same language model (Discloses “The method may involve orchestration of one or more language models within the process,” which includes the possibility that only one large language model is used.; Gurari, ¶ [0061]).
Regarding claim 7, Gurari discloses wherein generating the plurality of content units comprises: for each contextual unit of the plurality of contextual units, providing the contextual unit and the information item to the first language model (“The method may involve orchestration of one or more language models within the process, where an LLM may be used for identifying the potential answers from segments in S114, generating questions for potential answers S116, and/or other processes,” where “processing the corpus comprises processing multiple information segments of the source information S112, identifying potential answers for a given information segment S114,” thus the segments {the contextual units} and the source information are provided, such that the corpus may be processed using the LLM.; Gurari, ¶ [0061]); and obtaining, from the first language model, one or more questions based on the contextual unit and the information item (As performed using the LLM, the LLM both identifies “potential answers for a given information segment S114, and for each potential answer,” and generates “a set of questions that can be answered with the potential answer” based on the segment and the source information.; Gurari, ¶ [0061]).
Regarding claim 8, Gurari discloses wherein generating the plurality of content units comprises generating a question and an answer based on the contextual unit and the information item for each contextual unit of the plurality of contextual units (Discloses “processing a corpus of source information using a large language model to generate a set of question-answer pairs S110 may include: processing multiple information segments of the corpus of source information S112 {based on the contextual unit and the information item...}; identifying potential answers for a given information segment S114; and for each potential answer,” which corresponds to each contextual unit and associated information item, of the plurality of contextual units, from which the answer is derived, “generate a set of questions S116”; Gurari, ¶ [0077]).
Regarding claim 12, Gurari discloses A processing system (Discloses “systems and methods for LLM-enabled query responses” to “leverage large language models (LLMs) for processing one or more reference sources of information to generate accurate, concise, and interpretable (i.e., transparent in their source) research answers for given queries.”; Gurari, ¶ [0018]), comprising: a memory comprising computer-executable instructions; and a processor configured to execute the computer-executable instructions (The system may be implemented using “one or more computer- readable mediums (e.g., non-transitory computer-readable mediums) storing instructions that, when executed by the one or more computer processors, cause a computing platform to perform operations comprising those of the system or method described herein”; Gurari, ¶ [0160]) and cause the processing system to: obtain an information item from an information repository (Discloses receiving “a corpus of source information” where the “source information could be a text, a collection of text, or any suitable source of information”; Gurari, ¶ [0042]); allocate portions of the information item into a plurality of contextual units (The generator 110 can further include “segmentation algorithms for dividing source information into processable segments”; Gurari, ¶ [0042]), wherein each contextual unit of the plurality of contextual units comprises a portion of the information item (The “different segments {the contextual units}” are segments divided from source information {the information item} from the “corpus of source information.”; Gurari, ¶ [0042]); generate, by a first language model, a plurality of content units (“The question-answer pair generator 110 may also include... answer identification algorithms for extracting potential answers from segments, and question generation algorithms for creating questions corresponding to identified answers {content units}”; Gurari, ¶ [0042]) based on the plurality of contextual units, wherein each content unit of the plurality of content units is associated with a contextual unit (The answers are extracted from the segments of the source information and the questions are generated “corresponding to the identified answers.” As such, the answers and generated questions {plurality of content units} are both based on the plurality of segments from the corpus of source information {plurality of contextual units} and each question and answer {each content unit} is associated with at least a respective segment {a contextual unit} of the plurality of segments from which the question and answer is generated.; Gurari, ¶ [0042]); construct an augmented dataset (“The question-answer pair generator 110 functions to generate a set of question answer pairs from different segments of a corpus of source information” which can be maintained using “a storage engine that manages the physical storage and organization of question-answer pairs” and “an indexing system that creates and maintains various indexes for efficient searching and retrieval {an augmented dataset}.”; Gurari, ¶ [0042], [0046]-[0047]) that includes one or more content units of the plurality of content units (The set of question answer pairs{the plurality of content units} includes one or more question answer pairs {one or more content units}.; Gurari, ¶ [0042], [0046]-[0047]); receive a query (The system further includes a “query interface 130... through which a request may be made” where the system includes “receiving a query through a query interface S130”; Gurari, ¶ [0048], [0060]); input the query and the one or more content units of the plurality of content units into one or more language models (Based on the received query, the system then identifies “candidate question-answer pairs by searching the stored question-answer pairs using the query S140”, which includes processing both the query and the “stored question-answer pairs” where “an LLM may be used for identifying the potential answers from segments in S114, generating questions for potential answers S116, and/or other processes” where other processes includes “LLM or [Retrieval augmented generation (RAG)]-based document search solutions”; Gurari, ¶ [0024], [0060]-[0061]); and obtain, as output from the one or more language models, a response to the query based on the input query and the one or more content units of the plurality of content units (“generating a response to the query by processing the candidate question-answer pairs S150” as part of a RAG process, where the system uses a large language model alongside retrieved question answer pairs to generate a response to the query.; Gurari, ¶ [0003], [0024]-[0025], [0060]).
Regarding claim 13, Gurari discloses wherein each content unit of the plurality of content units includes at least one of: a question set comprising a plurality of questions; a question-context set comprising a plurality of question-context tuples, wherein each question-context pair of the plurality of question-context tuples includes a contextual unit and a question; a question-answer set comprising a plurality of question-answer tuples, wherein each question-answer pair of the plurality of question-answer tuples includes a question and an answer associated with the question; or a question-context-answer set comprising a plurality of question-context-answer tuples, wherein each question-context-answer tuple of the plurality of question-context-answer tuples includes a contextual unit, a question, and an answer associated with the question (“The question-answer pair generator 110 may also include... answer identification algorithms for extracting potential answers from segments, and question generation algorithms for creating questions corresponding to identified answers” wherein the answer is an answer associated with the question and a contextual unit, and the pair may be viewed as tuple, as the question and answer are associated with the source information and the answer may be a contextual unit from which both the question and answer are derived.; Gurari, ¶ [0026], [0040]-[0042]).
Regarding claim 14, Gurari discloses wherein to input the one or more content units of the plurality of content units into the one or more language models comprises to: generate a similarity metric for each content unit of the one or more content units, wherein the similarity metric includes a measure of similarity between a respective content unit of the one or more content units and the received query (“the method may include computationally evaluating candidate questions for similarity to the query. This can include generating similarity scores or other types of computational characterization of similarity {a measure of similarity...} between the query {the received query} and candidate questions {a respective content unit of the one or more content units}.”; Gurari, ¶ [0103]); and select content units based on the similarity metric (“For a question-answer pair to be included in the candidate set of question-answer pairs, the question may be required to satisfy a minimum similarity score threshold to be used for answering the query. “; Gurari, ¶ [0103]).
Regarding claim 15, Gurari discloses wherein the similarity metric is a cosine distance (“The similarity scoring may utilize various metrics including cosine similarity of embedding vectors, semantic distance measures, keyword overlap scores, and weighted combinations of multiple similarity factors.”; Gurari, ¶ [0103]).
Regarding claim 16, Gurari discloses wherein to generate the plurality of content units comprises to: for each contextual unit of the plurality of contextual units, provide the contextual unit and the information item to the first language model (“The method may involve orchestration of one or more language models within the process, where an LLM may be used for identifying the potential answers from segments in S114, generating questions for potential answers S116, and/or other processes,” where “processing the corpus comprises processing multiple information segments of the source information S112, identifying potential answers for a given information segment S114,” thus the segments {the contextual units} and the source information are provided, such that the corpus may be processed using the LLM.; Gurari, ¶ [0061]); and obtain, from the first language model, one or more questions based on the contextual unit and the information item (As performed using the LLM, the LLM both identifies “potential answers for a given information segment S114, and for each potential answer,” and generates “a set of questions that can be answered with the potential answer” based on the segment and the source information.; Gurari, ¶ [0061]).
Regarding claim 17, Gurari discloses A method for augmenting an information repository for information retrieval by a large language model (Discloses “systems and methods for LLM-enabled query responses” to “leverage large language models (LLMs) for processing one or more reference sources of information to generate accurate, concise, and interpretable (i.e., transparent in their source) research answers for given queries.”; Gurari, ¶ [0018]), the method comprising: receiving a query (The system further includes a “query interface 130... through which a request may be made” where the system includes “receiving a query through a query interface S130”; Gurari, ¶ [0048], [0060]); obtaining an information item from an information repository based on the query (“Processing of a corpus may be substantially performed in advance of any received query,” where an optional embodiment indicates that both options (i.e., either “performed in advance of…” the query or “performed after receipt of…” the query) are contemplated. The system further describes the processing of untrusted source content in response to a validation query (e.g., “on-demand verification during query processing” based on “a validation query input”).; Gurari, ¶ [0052], [0056]-[0057], [0065], [0068]); allocating portions of the information item into a plurality of contextual units (The generator 110 can further include “segmentation algorithms for dividing source information into processable segments”; Gurari, ¶ [0042]), wherein each contextual unit of the plurality of contextual units comprises a portion of the information item (The “different segments {the contextual units}” are segments divided from source information {the information item} from the “corpus of source information.”; Gurari, ¶ [0042]); generating, by a first language model, a plurality of content units (“The question-answer pair generator 110 may also include... answer identification algorithms for extracting potential answers from segments, and question generation algorithms for creating questions corresponding to identified answers {content units}”; Gurari, ¶ [0042]) based on the plurality of contextual units, wherein each content unit of the plurality of content units is associated with a contextual unit (The answers are extracted from the segments of the source information and the questions are generated “corresponding to the identified answers.” As such, the answers and generated questions {plurality of content units} are both based on the plurality of segments from the corpus of source information {plurality of contextual units} and each question and answer {each content unit} is associated with at least a respective segment {a contextual unit} of the plurality of segments from which the question and answer is generated.; Gurari, ¶ [0042]); constructing an augmented dataset (“The question-answer pair generator 110 functions to generate a set of question answer pairs from different segments of a corpus of source information” which can be maintained using “a storage engine that manages the physical storage and organization of question-answer pairs” and “an indexing system that creates and maintains various indexes for efficient searching and retrieval {an augmented dataset}.”; Gurari, ¶ [0042], [0046]-[0047]) that includes one or more content units of the plurality of content units (The set of question answer pairs{the plurality of content units} includes one or more question answer pairs {one or more content units}.; Gurari, ¶ [0042], [0046]-[0047]); inputting the query and the one or more content units of the plurality of content units into one or more language models (Based on the received query, the system then identifies “candidate question-answer pairs by searching the stored question-answer pairs using the query S140”, which includes processing both the query and the “stored question-answer pairs” where “an LLM may be used for identifying the potential answers from segments in S114, generating questions for potential answers S116, and/or other processes” where other processes includes “LLM or [Retrieval augmented generation (RAG)]-based document search solutions”; Gurari, ¶ [0024], [0060]-[0061]); and obtaining, as output from the one or more language models, a response to the query based on the input query and the one or more content units of the plurality of content units (“generating a response to the query by processing the candidate question-answer pairs S150” as part of a RAG process, where the system uses a large language model alongside retrieved question answer pairs to generate a response to the query.; Gurari, ¶ [0003], [0024]-[0025], [0060]).
Regarding claim 18, Gurari discloses wherein each content unit of the plurality of content units includes at least one of: a question set comprising a plurality of questions; a question-context set comprising a plurality of question-context tuples, wherein each question-context pair of the plurality of question-context tuples includes a contextual unit and a question; a question-answer set comprising a plurality of question-answer tuples, wherein each question-answer pair of the plurality of question-answer tuples includes a question and an answer associated with the question; or a question-context-answer set comprising a plurality of question-context-answer tuples, wherein each question-context-answer tuple of the plurality of question-context-answer tuples includes a contextual unit, a question, and an answer associated with the question (“The question-answer pair generator 110 may also include... answer identification algorithms for extracting potential answers from segments, and question generation algorithms for creating questions corresponding to identified answers” wherein the answer is an answer associated with the question and a contextual unit, and the pair may be viewed as tuple, as the question and answer are associated with the source information and the answer may be a contextual unit from which both the question and answer are derived.; Gurari, ¶ [0026], [0040]-[0042]).
Regarding claim 19, Gurari discloses wherein inputting the one or more content units of the plurality of content units into the one or more language models comprises: generating a similarity metric for each content unit of the one or more content units, wherein the similarity metric includes a measure of similarity between a respective content unit of the one or more content units and the received query (“the method may include computationally evaluating candidate questions for similarity to the query. This can include generating similarity scores or other types of computational characterization of similarity {a measure of similarity...} between the query {the received query} and candidate questions {a respective content unit of the one or more content units}.”; Gurari, ¶ [0103]); and selecting content units based on the similarity metric (“For a question-answer pair to be included in the candidate set of question-answer pairs, the question may be required to satisfy a minimum similarity score threshold to be used for answering the query. “; Gurari, ¶ [0103]).
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 9-11 and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Gurari as applied to claims 1 and 17 above, and further in view of Musoles (US 2021/0294781 A1, hereinafter Musoles).
Regarding claim 9, the rejection of claim 1 is incorporated. Gurari disclose all of the elements of the current invention as stated above. However, Gurari fails to expressly recite wherein allocating portions of the information item into the plurality of contextual units comprises applying an n-gram based contextual unit segmentation procedure to the information item, wherein each contextual unit of the plurality of contextual units comprises a sequence of n consecutive sentences or words from the information item.
Musoles teaches “natural language processing for cross-context natural language model generation.” (Musoles, ¶ [0002]). Regarding claim 9, Musoles teaches wherein allocating portions of the information item into the plurality of contextual units comprises applying an n-gram based contextual unit segmentation procedure to the information item, wherein each contextual unit of the plurality of contextual units comprises a sequence of n consecutive sentences or words from the information item (“Operations of the process 300 may begin at block 304. In some embodiments, the process 300 may include obtaining a corpus of text... [and] may include documents from various sources, where the text in the documents may be organized as a single text block or be separated into multiple sections of the document” where the documents in the corpus “may be separated into n-grams, where an n-gram may include a sequence of n items from text” and “where the items may include phonemes, syllables, letters, words, symbols, base pairs, or the like”; Musoles, ¶ [0057])..
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the generated question answer pairs for retrieval augmented generation (RAG) of Gurari to incorporate the teachings of Musoles to include wherein allocating portions of the information item into the plurality of contextual units comprises applying an n-gram based contextual unit segmentation procedure to the information item, wherein each contextual unit of the plurality of contextual units comprises a sequence of n consecutive sentences or words from the information item. Gurari discloses a system for using segments of documents in the generation of question answer pairs, such as for responding to a query for a RAG configured LLM. However, while Gurari discloses dividing source information into processable segments, Gurari relies on broad disclosures of utilizing natural language processing techniques for proper segmentation of the received documents. It is well understood that LLMs have a limited maximum token size for an input and limited context windows for maintaining attention. Musoles describes an n-gram based contextual unit segmentation procedure that partitions text into defined chunk sizes. It would have been obvious to a person having ordinary skill in the art at the time of filing to implement the segmentation engine of Gurari using the n-gram based segmentation procedure described in Musoles, as this would mathematically ensure that the resulting segments never exceed the strict token limits of the downstream language model, thereby preventing context window overflow while efficiently preparing large documents for processing, as recognized in light of Musoles. (Musoles, ¶ [0068]-[0070]).
Regarding claim 10, the rejection of claim 9 is incorporated. Gurari disclose all of the elements of the current invention as stated above. However, Gurari fails to expressly recite wherein applying the n-gram based contextual unit segmentation procedure comprises dividing the information item into contextual units based on a value of a positive integer n.
The relevance of Musoles is described above with relation to claim 9. Regarding claim 10, Musoles teaches wherein applying the n-gram based contextual unit segmentation procedure comprises dividing the information item into contextual units based on a value of a positive integer n (The “documents in the corpus may be separated into n-grams, where an n-gram may include a sequence of n items from text, where “n” represents an integer,” and where the integer is counting a number of items which only exist as positive integers.; Musoles, ¶ [0057]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the generated question answer pairs for retrieval augmented generation (RAG) of Gurari to incorporate the teachings of Musoles to include wherein applying the n-gram based contextual unit segmentation procedure comprises dividing the information item into contextual units based on a value of a positive integer n. Gurari discloses a system for using segments of documents in the generation of question answer pairs, such as for responding to a query for a RAG configured LLM. However, while Gurari discloses dividing source information into processable segments, Gurari relies on broad disclosures of utilizing natural language processing techniques for proper segmentation of the received documents. It is well understood that LLMs have a limited maximum token size for an input and limited context windows for maintaining attention. Musoles describes an n-gram based contextual unit segmentation procedure that partitions text into defined chunk sizes. It would have been obvious to a person having ordinary skill in the art at the time of filing to implement the segmentation engine of Gurari using the n-gram based segmentation procedure described in Musoles, as this would mathematically ensure that the resulting segments never exceed the strict token limits of the downstream language model, thereby preventing context window overflow while efficiently preparing large documents for processing, as recognized in light of Musoles. (Musoles, ¶ [0068]-[0070]).
Regarding claim 11, the rejection of claim 10 is incorporated. Gurari disclose all of the elements of the current invention as stated above. However, Gurari fails to expressly recite further comprising generating multiple sets of contextual units by varying the value of n, wherein each set of contextual units corresponds to a different n-gram size.
The relevance of Musoles is described above with relation to claim 9. Regarding claim 11, Musoles teaches further comprising generating multiple sets of contextual units by varying the value of n, wherein each set of contextual units corresponds to a different n-gram size (“Each document in a corpus of text or a subset of documents in the corpus of text may store or otherwise be associated with a set of metadata tags indicating a domain of the document” and “Documents in the corpus may be separated into n-grams, where … different models may use different items as the base element for an n-gram” and “a first n-gram does not need to include the same number of items as a second n-gram.” As understood by the office, the indication that the value of n and item type may differ between the first n-gram and the second n-gram indicates that generated n-gram sets may have different values.; Musoles, ¶ [0057], [0060])..
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the generated question answer pairs for retrieval augmented generation (RAG) of Gurari to incorporate the teachings of Musoles to include further comprising generating multiple sets of contextual units by varying the value of n, wherein each set of contextual units corresponds to a different n-gram size. Gurari discloses a system for using segments of documents in the generation of question answer pairs, such as for responding to a query for a RAG configured LLM. However, while Gurari discloses dividing source information into processable segments, Gurari relies on broad disclosures of utilizing natural language processing techniques for proper segmentation of the received documents. It is well understood that LLMs have a limited maximum token size for an input and limited context windows for maintaining attention. Musoles describes an n-gram based contextual unit segmentation procedure that partitions text into defined chunk sizes. It would have been obvious to a person having ordinary skill in the art at the time of filing to implement the segmentation engine of Gurari using the n-gram based segmentation procedure described in Musoles, as this would mathematically ensure that the resulting segments never exceed the strict token limits of the downstream language model, thereby preventing context window overflow while efficiently preparing large documents for processing, as recognized in light of Musoles. (Musoles, ¶ [0068]-[0070]).
Regarding claim 20, the rejection of claim 17 is incorporated. Gurari disclose all of the elements of the current invention as stated above. However, Gurari fails to expressly recite wherein allocating portions of the information item into the plurality of contextual units comprises applying an n-gram based contextual unit segmentation procedure to the information item, wherein each contextual unit of the plurality of contextual units comprises a sequence of n consecutive sentences or words from the information item.
The relevance of Musoles is described above with relation to claim 9. Regarding claim 20, Musoles teaches wherein allocating portions of the information item into the plurality of contextual units comprises applying an n-gram based contextual unit segmentation procedure to the information item, wherein each contextual unit of the plurality of contextual units comprises a sequence of n consecutive sentences or words from the information item (“Operations of the process 300 may begin at block 304. In some embodiments, the process 300 may include obtaining a corpus of text... [and] may include documents from various sources, where the text in the documents may be organized as a single text block or be separated into multiple sections of the document” where the documents in the corpus “may be separated into n-grams, where an n-gram may include a sequence of n items from text”, and “where the items may include phonemes, syllables, letters, words, symbols, base pairs, or the like”; Musoles, ¶ [0057]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the generated question answer pairs for retrieval augmented generation (RAG) of Gurari to incorporate the teachings of Musoles to include wherein allocating portions of the information item into the plurality of contextual units comprises applying an n-gram based contextual unit segmentation procedure to the information item, wherein each contextual unit of the plurality of contextual units comprises a sequence of n consecutive sentences or words from the information item. Gurari discloses a system for using segments of documents in the generation of question answer pairs, such as for responding to a query for a RAG configured LLM. However, while Gurari discloses dividing source information into processable segments, Gurari relies on broad disclosures of utilizing natural language processing techniques for proper segmentation of the received documents. It is well understood that LLMs have a limited maximum token size for an input and limited context windows for maintaining attention. Musoles describes an n-gram based contextual unit segmentation procedure that partitions text into defined chunk sizes. It would have been obvious to a person having ordinary skill in the art at the time of filing to implement the segmentation engine of Gurari using the n-gram based segmentation procedure described in Musoles, as this would mathematically ensure that the resulting segments never exceed the strict token limits of the downstream language model, thereby preventing context window overflow while efficiently preparing large documents for processing, as recognized in light of Musoles. (Musoles, ¶ [0068]-[0070]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Boué (US 2026/0004081 A1) discloses systems and methods for processing a plurality of query-answer pairs associated with a generative artificial intelligence (AI) model. A first set of query variations are generated from the plurality of query-answer pairs using a genetic algorithm and a plurality of content portions associated with the first set of query variations are identified using a Retrieval Augmentation Generation (RAG) system.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Sean E. Serraguard whose telephone number is (313)446-6627. The examiner can normally be reached 07:00-17:00 M-F.
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, Daniel C. Washburn can be reached at (571) 272-5551. 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 E Serraguard/Primary Examiner, Art Unit 2657
Prosecution Insights