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 .
DETAILED ACTION
This office action is in response to Applicant’s submission filed on 2/18/2025. Claims 10-16 are pending. As such, claims 10-16 have been examined.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 10-16 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 10 recites the limitation “the language” in line 23. There is insufficient antecedent basis for this limitation in the claim. For sake of compact prosecution, it would be interpreted it as “general natural language”.
Claim 13 recites the limitation “the external databases” in line 1-2. There is insufficient antecedent basis for this limitation in the claim. For sake of compact prosecution, it would be interpreted it as if the claim depends from claim 12 which recites “external databases”.
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 10 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Applicant supplied reference, Lin et al: "Pretrained Transformers for Text Ranking: BERT and Beyond", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 13. Oktober 2020 (2020-10-13), XP081785445, in view of Staub (US 20210216576).
Regarding Claim 10, Lin discloses:10. (New) A computer-implemented method for determining relevant answer passages, from an answer corpus containing a plurality of answer passages indexed in a vector-free manner, concerning a query relating to a legal matter, the computer-implemented method comprising the following steps: (pages 40, sect 3 “Multi-stage Ranking Architecture”, pages 104-105: “3.7.2 Legal Domain”, also see pg. 36, “BM25 ranking” which reads on vector-free manner);
using a ranking function to assess the relevance of the plurality of answer passages of the answer corpus to the query and to determine a list containing the relevant answer passages, wherein the relevant answer passages are a fixed number of the plurality of answer passages of the answer corpus that are possible candidates for answering the query based on the ranking function, (page 36: “BM25 ranking”; See also page 45, middle paragraph: “… This retrieval stage is called candidate generation, initial retrieval, or first-stage retrieval, the output of which is a ranked list of texts, typically ordered by a scoring function based on exact term matches such as BM25 (see Section 1.2).”)
sorting best candidates of the relevant answer passages within the list with descending relevance using at least one machine learning semantic language model, which embeds text passages context-sensitively in high-dimensional vector spaces in order to determine semantic similarities of the query and the relevant answer passages from the answer corpus, the query and the relevant answer passages determined using the ranking function being vectorized in real time by the at least one machine learning semantic language model only during the step of sorting after the relevant answer passages are determined using the ranking function, (page 36, penultimate paragraph: “most neural ranking models today (all of those discussed in Section 3) act as rerankers over a list of candidates,"; page 45, figure 5, and page 36, figure 6; pages 45: “3.2 Simple Relevance Classification: monoBERT”: page 45, third-from-last paragraph: “BERT inference is then applied to rerank these candidates to generate & score si for each text di in the candidates list.”; figure 6 on page 46 describes a re-ranking using “monoBERT”, wherein the monoBERT model receives both the query and the “candidate text” together and simultaneously as input and directly outputs a “revance score s”. This is only possible as soon as the query is available.);
the at least one semantic language model being trained in a two-stage process before the step of sorting by: a first pre-training step, the at least one machine learning semantic language model is trained on the basis of general texts (pages 98-99; a BERT model is a pre-trained transformer model and is always trained in a two-stage process comprising a pretraining step and a fine-tuning step. The use of specialist legal texts, or legal matters, in these steps is not disclosed explicitly in D1, however)
and a second fine-tuning step, the at least one machine learning semantic language model (pages 98-99; a BERT model is a pre-trained transformer model and is always trained in a two-stage process comprising a pretraining step and a fine-tuning step. The use of specialist legal texts, or legal matters, in these steps is not disclosed explicitly in D1, however)
and outputting the best candidates of the relevant answer passages in the form of a list with the descending relevance. (page 45: figure 5: “Ranked List”).
Lin does not explicitly disclose the following features.
Staub in the related art discloses: wherein the at least one machine learning semantic model is Sentence-BERT, which uses Siamese and triplet network structures to derive semantically meaningful sentence embeddings that can be compared using cosine similarities, ([0028] Sentence-BERT using the BERT architecture provides sentence embedding that is fine-tuned for sentence similarity using a Siamese network. Other search systems may improve retrieval with the co-learning of the retriever and the reader from question-passage pairs, or pre-training the language representation on a retrieval task.) para 0063 and 0085 discuss cosine similarity.
the at least one semantic language model being trained in a two-stage process before the step of sorting by: a first pre-training step, the at least one machine learning semantic language model is trained on the basis of general texts and specialist legal texts in a general understanding of the language and taking into account specialist legal language, ([0040]the embeddings service may employ a pretrained BERT language model that is trained on a legal corpus. This Legal BERT is trained on hundreds of millions of legal examples, such as case law headnotes and RFCs and secondary sources, such as treatises. In this example, the Legal BERT used by the offline system 150 comprises highly optimized, dense vector indices generated from an entire corpus for serving the purpose of performing similarity searches with respect to queries of the online process. The data analytics engine 152 is configured to convert the legal content into embeddings using the Legal BERT embedding service 151.)
and a second fine-tuning step, the at least one machine learning semantic language model uses training data containing questions and answers concerning legal matters to learn to calculate the relevance of possible answers concerning the questions in order to find relevant answer passages in the answer corpus, ([0040]the embeddings service may employ a pretrained BERT language model that is trained on a legal corpus. This Legal BERT is trained on hundreds of millions of legal examples, such as case law headnotes and RFCs and secondary sources, such as treatises. In this example, the Legal BERT used by the offline system 150 comprises highly optimized, dense vector indices generated from an entire corpus for serving the purpose of performing similarity searches with respect to queries of the online process. The data analytics engine 152 is configured to convert the legal content into embeddings using the Legal BERT embedding service 151. [0072] The answer finder module is fine-tuned on Legal-BERT. An additional SoftMax layer is added to the classifier to provide the probability of the input question-passage membership to each class, P=softmax(Z). The Softmax function normalizes the output to a probability distribution over predicted output classes (“invalid” and “valid” question and answer pairs). The main output of Answer Finder is P1=softma(Z)[1] which is the answer finder module's prediction on how likely a question-passage pair is a good question-passage pair.) Also see para 0096.
Staub also discloses: and outputting the best candidates of the relevant answer passages in the form of a list with the descending relevance. ([0048] the orchestrator module 141 is the primary controller of the open domain questioning system 100. It coordinates questions recognition, coordinates the query engagement decision, preprocesses the question, initiates search and retrieve workflows, aggregates search results from disparate sources, initiates the answer finder (i.e., the re-ranker) workflow, and post-processes the results.) [0095] Curve 301 is the answer finder run on BM25+L S BERT, curve 302 is the answer finder run on bm25, curve 303 is the answer finder run on L S BERT, curve 304 is the answer finder run on Legal GloVe, curve 305 is bm25 (more like this), curve 306 is Legal Siamese BERT, and curve 307 is Legal GloVe. Using the answer finder module, the answers were sorted based on answer probability. Thus, answers with higher probability were promoted to the top of the list, and answers in the top of the list are more probable to be good answers to the user's question.)
Lin and Staub are considered analogous art. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lin to combine the teaching of Staub, because Sentence-BERT enables fast vector comparison involving sentences (Staub, [0028]).
Regarding Claim 14, Lin in view of Staub discloses: Although Lin reference would require use of a computer and computer readable medium to use Sentence-BERT, Staub explicitly disclose the following: 14. (New) A non-transitory computer-readable medium storing a computer program product comprising computer-executable commands that, when the program is executed by a computer, cause said computer to carry out the method steps as claimed in claim 10. ([0010] In another embodiment, a system for providing answers to a question includes one or more processors, and a non-transitory computer-readable medium storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to retrieve a plurality of passages relevant to a search query, and to generate a plurality of question-passage pairs)
As for the rest of the claim, they recite the method of claim 10, therefore the rationale and grounds used in rejection of claim 10 is also applicable.
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Applicant supplied reference, Lin in view of Staub, and further in view of Gandhi (US 20080126335).
Regarding Claim 11, Lin and Staub is silent regarding wherein, in order to provide the answer corpus, legal documents are broken down into partial documents that each contain an answer passage of the plurality of answer passages, and the plurality of answer passages are indexed in a vector-free manner.
Gandhi in the related art discloses: wherein, in order to provide the answer corpus, legal documents are broken down into partial documents that each contain an answer passage of the plurality of answer passages, and the plurality of answer passages are indexed in a vector-free manner. ([0014] A query has a set of terms. A query term may have a set of grams. Candidate documents that may be relevant to a query may be identified by searching an inverted index using query term grams. Selected (e.g., less than all) candidate documents can then be partially reconstructed using the matching grams and offset information available in the inverted index. Documents may be selected for partial reconstruction based on an initial similarity measure (e.g., number of matched grams, percentage of grams matched, number of matched grams divided by number of grams in a document, rank).) [inverted index reads on indexed in a vector free manner]
Lin/Staub/Gandhi are considered analogous art. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lin and Staub to combine the teaching of Gandhi, because the technique provided enable faster and more efficient retrieval process (Gandhi, [0005]).
Claims 12 is rejected under 35 U.S.C. 103 as being unpatentable over Applicant supplied reference, Lin in view of Staub, further in view of Gandhi (US 20080126335), and furthermore in view of Deibler (US 20210191981).
Lin/Staub/Gandhi are silent regarding the following feature,
Deibler in the related art discloses: wherein the legal documents come from external databases. ([0038] Regulatory documents may include but may not be limited to legal documents including regulations, and laws, medical documents including research papers, and case studies, and documents related to any other field of study or research. The regulatory documents may be retrieved and accessed from external connected databases.)
Lin/Staub/Gandhi/Deibler are considered analogous art. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lin, Staub and Gandhi to combine the teaching of Deibler, because the having access to external database may enable user to access more resource which may lead to more relevant text (Deibler, [0038]).
Claims 13 is rejected under 35 U.S.C. 103 as being unpatentable over Applicant supplied reference, Lin in view of Staub, further in view of Gandhi (US 20080126335), and furthermore in view of Deibler (US 20210191981) and Kalinski (US 20220374797)
Regarding claim 13, the combination of Lin/Staub/Gandhi/Deibler do not disclose the following feature,
Kalinski in the related art discloses: wherein the external databases comprise software systems of the user of the method ([0109] external databases …can be access by user utilizing client software.)
and also databases of third parties containing anonymized question-answer pairs. ([0211] third parties databases …is anonymized)
Lin/Staub/Gandhi/Deibler/Kalinski are considered analogous art. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of teachings to combine with the teaching of Kalinksi, because the having access to external database may enable user find information that they need (Kalinksi, [0109]).
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Applicant supplied reference, Lin in view of Staub, further in view of Londhe (US 20190260694).
Regarding claim 15, Staub further discloses: wherein the computer program product is part of a software application that a user ([0032] user application that user can receive answer from system regarding legal matters) Also see para 0097, software.
Lin and Staub are silent regarding user ability to provide answer to query.
Londhe in the related art discloses: part of a software application that a user uses to answer queries ([0054] user providing answer to other user)
Lin/Staub/Londhe are considered analogous art. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lin, and Staub to combine the teaching of Londhe, because user may be able to provide accurate answer to other user (Londhe, [0054]).
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Applicant supplied reference, Lin in view of Staub, further in view of Londhe (US 20190260694), and furthermore in view of Breuer (US 20010012007).
Regarding Claim 16, the combination of Lin/Staub/Londhe is silent regarding an office application.
Breuer in the related art discloses: wherein the computer program product is part of an office application. ([0088] computer program includes office application …)
Lin/Staub/Londhe/Breuer are considered analogous art. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Lin/ Staub/Londhe to combine the teaching of Breuer, because office application enable user to perform more computer related task (Breuer, [0088]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure: Shalmashi US 20240241902 – discloses multistage ranking process using various BERT based models. See Abstract, para 0074-0076, and figs 11, and 13-14 for additional details.
Nogueira, R., & Cho, K. (2019). Passage Re-ranking with BERT. arXiv preprint arXiv:1901.04085.- using a two stage retrieve and rerank where the first stage can be BM25 for retrieval and BERT in the second stage for reranking. See Abstract and sections 1-3 for additional details.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Philip H Lam whose telephone number is (571)272-1721. The examiner can normally be reached 9 AM-3 PM Pacific time.
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, Bhavesh Mehta can be reached on 571-272-7453. 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.
/PHILIP H LAM/ Examiner, Art Unit 2656
