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/Status of Claims
Claims 1, 2, 11, 12, and 20 were amended.
Claims 1-20 are pending and examined herein.
Claims 1-20 are rejected under 35 U.S.C. 101.
Claims 1-20 are rejected under 35 U.S.C. 103.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on January 28, 2026 was filed. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Arguments
Applicant's arguments filed 10/16/2025 regarding the 35 U.S.C. 101 rejection of claims 1-20 have been fully considered but they are not persuasive.
Applicant argues, see pages 8-10, that, similar to McRO v Bandai, claim 1 recites a process that implements specific rules to automate a task that would otherwise require manual human intervention, and that the claim is not carried out in the same way as would have been performed by a human. Thus, applicant argues, claim 1 is integrated into a practical application.
Examiner respectfully disagrees. MPEP 2106.05(a) states "For example, in McRO, the court relied on the specification’s explanation of how the particular rules recited in the claim enabled the automation of specific animation tasks that previously could only be performed subjectively by humans, when determining that the claims were directed to improvements in computer animation instead of an abstract idea."
MPEP 2106.05(a)(I), list 2, ex. iii, in reference to examples that courts have indicated may not be sufficient to show an improvement in computer-functionality states "i. Mere automation of manual processes, such as using a generic computer to process an application for financing a purchase, Credit Acceptance Corp. v. Westlake Services, 859 F.3d 1044, 1055, 123 USPQ2d 1100, 1108-09 (Fed. Cir. 2017) or speeding up a loan-application process by enabling borrowers to avoid physically going to or calling each lender and filling out a loan application, LendingTree, LLC v. Zillow, Inc., 656 Fed. App'x 991, 996-97 (Fed. Cir. 2016) (non-precedential);"
Thus, the claims in McRO were directed to an improvement to a specific technology, namely computer animation. The automation of a manual process itself does not show an improvement to technology. Therefore, claim 1 is not integrated into a practical application.
Applicant further argues, "Second, under Step 2A, Prong Two, even if considered as reciting an abstract idea (which Applicant does not concede), claim 1 as a whole integrates the alleged abstract idea into a practical application of automatic generation of a visual layout, and thus is eligible under Step 2A, Prong Two. Similar to the claim in McRO (Fed. Cir., 2016), the "claimed process uses a combined order of specific rules that renders information into a specific format that is then used and applied to create desired results." (Id., at 1315). The Specification describes a number of technical benefits including "allow[ing] for high accuracy in reconstruction of input layouts, ensuring that the generated model is attending to the inputs." (Specification, [0020])."
Examiner respectfully disagrees. The cited portion of the specification is not actually present in the specification. The specification does not mention automatic generation of a visual layout. Additionally, the claims do not include automatic generation of a visual layout. Applicant additionally presents case law for improvements to machine learning systems. However, the specification and claims do not appear to reflect improvements to machine learning systems. Applicant also asserts that the pending claims reflect an improvement to fintech systems. However, the specification nor the claims appear to reflect improvements to fintech systems. Therefore, the claims do not integrate the abstract into a practical application nor provides significantly more.
Applicant’s arguments, see pages 10-11, filed 10/16/2025, with respect to the rejection(s) of claim(s) 1-20 under 35 U.S.C. 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Liu et al. (K. Liu, J. Zhao, S. He and Y. Zhang, "Question Answering over Knowledge Bases," in IEEE Intelligent Systems, vol. 30, no. 5, pp. 26-35, Sept.-Oct. 2015, doi: 10.1109/MIS.2015.70.), Allen et al. (US 20160048514 A1), Nallapati et al. (US 11604794 B1), and Gu et al. (“Beyond I.I.D.: Three Levels of Generalization for Question Answering on Knowledge Bases,” April 19, 2021, Proceedings of the Web Conference 2021 (WWW ’21)) [provided by Applicant via IDS].
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim(s) recite(s) mental processes and mathematical concepts. This judicial exception is not integrated into a practical application because the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception, as explained below.
Step 1 for all Claims:
Claims 1-10 are directed to a process. Claims 11-19 are directed to a machine. Claim 20 is directed to a manufacture. Therefore, claims 1-20 are directed to one of the statutory categories of invention, i.e., process, machine, manufacture, or composition of matter.
Regarding Claim 1:
Step 2A, Prong 1:
A method of knowledge base question answering, the method comprising:
generating a set of candidate logical forms by starting from each entity of the set of entities and searching for paths reachable within two hops of each entity in a knowledge base; (Generating a set of candidate logical forms by searching a knowledge base can be practically performed in the human mind. This is a mental process.)
ranking… the set of candidate logical forms based on similarity scores between the question and the set of candidate logical forms, respectively; (Ranking the candidate logical forms can be practically performed in the human mind. This is a mental process.)
generating, by a generation model, a target logical form conditioned on the question and a subset of the ranked set of candidate logical forms; (Generating a logical form can be practically performed in the human mind. This is a mental process.)
and generating an answer to the question by applying the target logical form on the knowledge base (Generating an answer using the logical form on the knowledge base can be practically performed in the human mind. This is a mental process. MPEP 2106.04(a)(2)(III) Examples of mental processes include observations, evaluations, judgments, and opinions. Judgements, particularly “collecting information, analyzing it…” are mental processes);
Step 2A, Prong 2:
receiving, via a communication interface, a question that mentions a set of entities (MPEP 2106.05(g) mere data gathering (or where storing data in memory) is considered insignificant extra-solution activity);
…by a ranking model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application).
Step 2B:
receiving, via a communication interface, a question that mentions a set of entities (MPEP 2106.05(d) well-understood, routine, conventional activities previously known to the industry, which is recited at a high level of generality, are not eligible, e.g. “receiving or transmitting data over a network”);
…by a ranking model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application). The claim does not recite any additional elements, alone or in combination, that integrate the judicial exception into a practical application.
Regarding Claim 2:
Step 2A, Prong 1:
The method of claim 1, wherein the set of candidate logical forms is generated by:
and converting relation labels along the paths to the set of candidate logical forms (Converting relation labels to logical forms can be practically performed in the human mind. This is a mental process. MPEP 2106.04(a)(2)(III) Examples of mental processes include observations, evaluations, judgments, and opinions. Judgements, particularly “collecting information, analyzing it…” are mental processes);
Step 2A, Prong 2 and Step 2B:
The claim does not recite any further additional elements.
Regarding Claim 3:
Step 2A, Prong 1:
Refer to claim 1.
Step 2A, Prong 2:
The method of claim 1, wherein the ranking model comprises a language model based bi- encoder and a linear projection layer (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component).
Step 2B:
The method of claim 1, wherein the ranking model comprises a language model based bi- encoder and a linear projection layer (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component). The claim does not recite any additional elements, alone or in combination, that integrate the judicial exception into a practical application.
Regarding Claim 4:
Step 2A, Prong 1:
The method of claim 1, wherein the ranking… the set of candidate logical forms further comprises:
forming an input for the ranking model by concatenating the question and a first candidate logical form from the set of candidate logical forms (MPEP 2106.04(a)(2)(III) Examples of mental processes include observations, evaluations, judgments, and opinions. Judgements, particularly “collecting information, analyzing it…” are mental processes);
and generating… a first logit representing a similarity score between the question and the first candidate logical form (MPEP 2106.04(a)(2)(I) A claim that recites a mathematical calculation will be considered as falling within the "mathematical concepts" grouping).
Step 2A, Prong 2:
…by the ranking model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application).
Step 2B:
…by a ranking model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application). The claim does not recite any additional elements, alone or in combination, that integrate the judicial exception into a practical application.
Regarding Claim 5:
Step 2A, Prong 1:
Refer to Claim 1.
Step 2A, Prong 2:
The method of claim 1, wherein the generation model is a transformer-based sequence- to-sequence model (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component).
Step 2B:
The method of claim 1, wherein the generation model is a transformer-based sequence- to-sequence model (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component). The claim does not recite any additional elements, alone or in combination, that integrate the judicial exception into a practical application.)
Regarding Claim 6:
Step 2A, Prong 1:
The method of claim 1, wherein the generating… the target logical form further comprises:
constructing an input to the generation model by concatenating the question and the subset of the ranked set of candidate logical forms; (Constructing an input by concatenation can be practically performed in the human mind. This is a mental process.)
and generating… the target logical form based on the constructed input (Generating the target logical form can be practically performed in the human mind. This is a mental process. MPEP 2106.04(a)(2)(III) Examples of mental processes include observations, evaluations, judgments, and opinions. Judgements, particularly “collecting information, analyzing it…” are mental processes).
Step 2A, Prong 2:
…by the generation model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application).
Step 2B:
…by the generation model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application). The claim does not recite any additional elements, alone or in combination, that integrate the judicial exception into a practical application.
Regarding Claim 7:
Step 2A, Prong 1:
Refer to Claims 6 and 1.
Step 2A, Prong 2:
The method of claim 6, further comprising: decoding the subset of candidate logical forms using beam search; and querying the knowledge base using each candidate logical form from the subset until a valid answer is returned (MPEP 2106.05(g) mere data gathering (or storing data in memory) is considered insignificant extra-solution activity);
Step 2B:
The method of claim 6, further comprising: decoding the subset of candidate logical forms using beam search; and querying the knowledge base using each candidate logical form from the subset until a valid answer is returned (MPEP 2106.05(d) well-understood, routine, conventional activities previously known to the industry, which is recited at a high level of generality, are not eligible, e.g. iv. Storing and retrieving information in memory);
Regarding Claim 8:
Step 2A, Prong 1:
The method of claim 7, further comprising: in response to determining that no valid answer is returned after exhausting the subset of candidate logical forms, determining that a top-ranked candidate logical form in the subset is the target logical form (These determinations can be practically performed in the human mind. This is a mental process. MPEP 2106.04(a)(2)(III) Examples of mental processes include observations, evaluations, judgments, and opinions. Judgements, particularly “collecting information, analyzing it…” are mental processes);
Step 2A, Prong 2:
Refer to Claims 7, 6, and 1.
Step 2B:
Refer to Claims 7, 6, and 1.
Regarding Claim 9:
Step 2A, Prong 1:
The method of claim 1, further comprising: determining, for a first entity mentioned in the question, a first set of candidate entities in the knowledge base that match the first entity;
and determining linking relations between a second entity mentioned in the question and the first set of candidate entities (These determinations can be practically performed in the human mind. This is a mental process. MPEP 2106.04(a)(2)(III) Examples of mental processes include observations, evaluations, judgments, and opinions. Judgements, particularly “collecting information, analyzing it…” are mental processes).
Step 2A, Prong 2:
Refer to Claim 1.
Step 2B:
Refer to Claim 1.
Regarding Claim 10:
Step 2A, Prong 1:
The method of claim 9, further comprising: concatenating, for a first candidate entity from the first set of candidate entities, the question with a corresponding linking relations to form a first input to the ranking model; (Concatenation can be practically performed in the human mind. This is a mental process.)
generating, …, a first similarity score between the question and the first candidate entity based on the first input; (Generating a similarity score can be practically performed in the human mind. This is a mental process.)
ranking the first set of candidate entities based on generated similarity scores; (Ranking a set of candidate entities can be practically performed in the human mind. This is a mental process.)
and selecting a top-ranked candidate entity from the first set as a matching entity for the first entity mentioned in the question (Selecting a top-ranking entity can be practically performed in the human mind. This is a mental process. MPEP 2106.04(a)(2)(III) Examples of mental processes include observations, evaluations, judgments, and opinions. Judgements, particularly “collecting information, analyzing it…” are mental processes);
Step 2A, Prong 2:
…by ranking model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application).
Step 2B:
…by ranking model… (MPEP 2106.05(f) mere instructions to apply an abstract idea on a computer is not enough to integrate the claim into a practical application). The claim does not recite any additional elements, alone or in combination, that integrate the judicial exception into a practical application.
Regarding Claim 11:
Step 2A, Prong 1:
Claim 11, aside from the below additional elements, recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis. Refer to Step 2A, Prong 1 for claim 1.
Step 2A, Prong 2:
A system for knowledge base question answering, the system comprising: (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
a communication interface receiving a question that mentions a set of entities; (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
a memory storing a plurality of processor-executable instructions; (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
and a processor reading and executing the instructions from the memory to perform operations comprising (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
Claim 11, aside from the above additional elements, recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis. Refer to Step 2A, Prong 2 for claim 1.
Step 2B:
A system for knowledge base question answering, the system comprising: (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
a communication interface receiving a question that mentions a set of entities; (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
a memory storing a plurality of processor-executable instructions; (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
and a processor reading and executing the instructions from the memory to perform operations comprising (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component):
Claim 11, aside from the above additional elements, recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis. Refer to Step 2B for claim 1.
Claims 12-18 recites substantially similar subject matter to claim 2-8 and is rejected with the same rationale, mutatis mutandis.
Claim 19 recites substantially similar subject matter to claims 9 and 10 and is rejected with the same rationale, mutatis mutandis.
Regarding Claim 20:
Step 2A, Prong 1:
Claim 20, aside from the below additional elements, recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis. Refer to Step 2A, Prong 1 for claim 1.
Step 2A, Prong 2:
A processor-readable non-transitory storage medium storing a plurality of processor- executable instructions for knowledge base question answering, the instructions being executed by one or more processors to perform operations comprising: (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component);
Claim 20, aside from the above additional elements, recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis. Refer to Step 2A, Prong 2 for claim 1.
Step 2B:
A processor-readable non-transitory storage medium storing a plurality of processor- executable instructions for knowledge base question answering, the instructions being executed by one or more processors to perform operations comprising: (MPEP 2106.05(f) computer implementation which are recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component);
Claim 20, aside from the above additional elements, recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis. Refer to Step 2B for claim 1.
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 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Liu et al. (K. Liu, J. Zhao, S. He and Y. Zhang, "Question Answering over Knowledge Bases," in IEEE Intelligent Systems, vol. 30, no. 5, pp. 26-35, Sept.-Oct. 2015, doi: 10.1109/MIS.2015.70.), Allen et al. (US 20160048514 A1), Nallapati et al. (US 11604794 B1), and Gu et al. (“Beyond I.I.D.: Three Levels of Generalization for Question Answering on Knowledge Bases,” April 19, 2021, Proceedings of the Web Conference 2021
(WWW ’21)) [provided by Applicant via IDS], hereinafter referred to as Liu, Allen, Nallapati, and Gu respectively.
Regarding Claim 1:
Liu teaches: A method of knowledge base question answering, the method comprising (Pg. 26 Col 1 Par 1 “Deep Web search is on the cusp of a profound change, from simple document retrieval to natural language question answering (QA). Ultimately, search needs to precisely understand the meanings of users’ natural language questions, extract useful facts from all information on the Web, and select appropriate answers.” “The Web” is implied to be a form of knowledge base.):
ranking, by a ranking model, the [answers] based on similarity scores between the question and the set of candidate logical forms, respectively (Pg. 32 Col 1 Par 3, Col 2 Par 1 “PowerAqua is an ontology-based question-answering system that contains four major components. The linguistic component aims to identify syntactic relations among terms and output several triple-based representations… The merging and ranking component handles questions that require the association of different KBs, using a merging process that involves identifying semantically equivalent or overlapping information. A set of ranking criteria (such as mapping confidence to the facts in KBs, disambiguation, and the merging process) is then applied to select the correct answer.” Pg. 31 Col 1 Par 3 “Consequently, the task of question answering can be converted to a problem of similarity computation between the embeddings of the questions and answers in this space. Such methods are suitable for large-scale usage because the embeddings are free from domain restriction and require weak supervision.”);
and generating an answer to the question by applying the target logical form on the knowledge base (Pg. 26 Col 2 Par 2 and Col 1 Par 1, 2 “The dominant methods usually convert a natural language question into a complete and formal meaning representation (FMR) first, such as logical form. Based on FMR, the structured query is then smoothly generated.” “Semantic items in the text, including entities, classes, and their semantic relations, can be extracted from the raw data—answers corresponding to users’ questions can be grasped through direct matching in the KB.” “…structured query languages (such
as SPARQL) have been designed and provided for visiting these structured data…” “KB” and “structured data” both imply knowledge base.).
Liu does not appear to distinctly disclose: receiving, via a communication interface, a question that mentions a set of entities.
However, Allen teaches: receiving, via a communication interface, a question that mentions a set of entities (Pg. 9 Claim 1 “A computer-implemented method for handling a plurality of input questions, the method comprising: receiving, using a computer, the plurality of input questions…”; [0021] “In some embodiments, remote devices 102, 112 may enable users to submit questions (e.g., search requests or other user queries) to host devices 122 to retrieve search results. For example, the remote devices 102, 112 may include a query module 110, 120 (e.g., in the form of a web browser or any other suitable software module) and present a graphical user interface or other interface (e.g., command line prompts, menu screens, etc.) to solicit queries from users for submission to one or more host devices 122 and to display answers/results obtained from the host devices 122 in relation to such user queries.” [0032] “Consistent with various embodiments, semantic relationship identifier 220 may be a computer module that can identify semantic relationships of recognized entities (e.g., words, phrases, etc.) in questions posed by users. In some embodiments, semantic relationship identifier 220 may determine functional dependencies between entities and other semantic relationships.” “Remote devices” and “graphical user interface or other interface” imply a communication interface.);
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the method of knowledge base question-answering of Liu with the ability to receive of questions via communication interface in order to allow the question-answering system to communicate with users over large distances and organized into desired configurations ([0018] In some embodiments, the computing environment 100 may include one or more remote devices 102, 112 and one or more host devices 122. Remote devices 102, 112 and host device 122 may be distant from each other and communicate over a network 150 in which the host device 122 comprises a central hub from which remote devices 102, 112 can establish a communication connection. Alternatively, the host device and remote devices may be configured in any other suitable relationship (e.g., in a peer-to-peer or other relationship)).
Liu and Allen do not distinctly disclose: generating, by a generation model, a target logical form conditioned on the question and a subset of the ranked set of candidate logical forms.
However, Nallapati teaches: generating, by a generation model, a target logical form conditioned on the question and a subset of the ranked set of candidate logical forms (Col 19 Lines 47-51 “Intermediate representation generation model 530 responsible for predicting the intermediate representation given the natural language query and upstream predictions from entity recognition model 410, entity linkage model 512 and data set selection model 520. Col 21 Lines 24-27 “In various embodiments, query restatement generation 610 may be implemented. Query restatement generation 610 may generate from intermediate representation a query restatement 612”; “Intermediate representation” implies a logical form; “natural language query” implies the question; “query restatement” implies a generated target logical form);
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the knowledge base QA system of Liu and the remote capability of Allen with the generation of a target logical form of Nallapati in order to offer natural language query performance benefits (Col 21 Lines 28-33 “Query restatement 612 may offer natural language query performance benefits. For example, a developer can easily check result to see if the IR is correct or not. The restatement 612 may provide better interaction with user. A user can correct query based on the interpretable result, and increase user confidence in the system and erase frictions.”)
Liu, Allen, and Nallapati do not appear to explicitly teach
generating a set of candidate logical forms by starting from each entity of the set of entities and searching for paths reachable within two hops of each entity in a knowledge base, wherein each logical form of the set of logical forms includes one or more functions for operating on set-based semantics over the knowledge base
However, Gu—directed to analogous art—teaches
generating a set of candidate logical forms by starting from each entity of the set of entities and searching for paths reachable within two hops of each entity in a knowledge base, wherein each logical form of the set of logical forms includes one or more functions for operating on set-based semantics over the knowledge base (Page 3483 states "For Ranking, instead of using Seq2Seq as a generator we use it as a ranker to score each candidate logical form and return the top-ranked candidate. We employ a simple yet effective strategy to generate candidate logical forms: We enumerate all logical forms, optionally with a count function, within 2 hops starting from each entity identified in the question" The count function is the function for operating on set-based semantics over the knowledge base, as it counts a number of a set of entities in the knowledge base.)
ranking, by a ranking model, the set of candidate logical forms (Page 3483 states "For Ranking, instead of using Seq2Seq as a generator we use it as a ranker to score each candidate logical form and return the top-ranked candidate.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine the knowledge base QA system of Liu, the remote capability of Allen, and the generation of a target logical form of Nallapati with the logical form generation of Gu because, as Gu states on page 3483, "We employ a simple yet effective strategy to generate candidate logical forms: We enumerate all logical forms, optionally with a count function, within 2 hops starting from each entity identified in the question.7 The recall is 80% on GrailQA."
Regarding Claim 2:
Liu further teaches: The method of claim 1, wherein the set of candidate logical forms is generated by: converting relation labels along the paths to the set of candidate logical forms (Pg. 32 Col 1 Par 3 “The triple mapping component determines the most likely interpretation of a query, taking information from the previous two components into account.” Col 2 Par 2 “Shekarpour’s work addresses this complicated question via two cascade procedures: resource disambiguation and query construction.
Resource disambiguation maps natural language questions to KB resources…” Col 3 Par 1 “Once the resources are determined, the query construction procedure uses a graph-generating method to group the mapped resources together, with the resources of entities represented as vertices and the resources of relations corresponding to edges. The connections of these elements are defined by a series of rules induced by each KB structure.” The map structures are implied to represent logical forms according to Fig 2.
PNG
media_image1.png
252
267
media_image1.png
Greyscale
).
Regarding Claim 3:
Nallapati further teaches: The method of claim 1, wherein the ranking model comprises a language model based bi- encoder and a linear projection layer (Pg. 4 Col 1 Par 2, Col 2 Par 4 In this work, we introduce the Bidirectional Gated Graph Neural Network (BiGGNN) which extends GGSNN by learning node embeddings from both incoming and outgoing directions in an interleaved fashion when processing a directed graph. A similar bidirectional approach has been exploited in [43], [51] to extend other GNN variants. While their methods simply learn the node embeddings of each direction independently and concatenate them at last step, BiGGNN fuses the intermediate node embeddings from both directions at every iteration… To compute the graph-level embedding, we first apply a linear projection to the node embeddings, and then apply maxpooling over all node embeddings to get a d-dim vector hG).
Regarding Claim 4:
Liu further teaches: The method of claim 1, wherein the ranking, by the ranking model, the set of candidate logical forms further comprises: forming an input for the ranking model by concatenating the question and a first candidate logical form from the set of candidate logical forms (Pg. 29 Col 1 Par 3-4, Col 2 Par 1 “Automatically learning the lexicon indicates a mapping from natural language texts to logical forms. The previously described methods usually investigate such learning on sentences paired with their labeled logical forms, < sentence (S) -logical form (L) >:
Sentence: Utah borders Idaho
Logic Form: borders (utah, Idaho)”
The “automatic learning” is referring to the machine learning model that has a ranking component.);
Liu does not distinctly disclose: and generating, by the ranking model, a first logit representing a similarity score between the question and the first candidate logical form.
Nallapati further teaches: and generating, by the ranking model, a first logit representing a similarity score between the question and the first candidate logical form (Col 18 Lines 1-4 “Entity linkage model 512 may be implemented as a deep learning model, utilizing a neural network trained to identify and rank entity linkages in a given query text string and other input data, in various embodiments.” Lines 16-20 “A linear layer may then be applied on the classification token [CLS] to produce a logit score (BERT score). During training, cross-entropy loss may be calculated on all the linking candidates, including one positive candidate and at most Y (e.g., 63) negative candidates.” The “given query text string” implies question. The “linking candidates” refer to the candidate logical form, including its entities.).
Regarding Claim 5:
Nallapati further teaches: The method of claim 1, wherein the generation model is a transformer-based sequence- to-sequence model (Col 19 Lines 12-24 “For each candidate dataset, data set selection model 520 may then use a denoising autoencoder for pretraining sequence-to-sequence models (e.g., a fine-tuned BART encoder) to encode the (1) NLQ (natural language query) and (2) Column names.” BART stands for “Bidirectional and Auto-Regressive Transformer”.)).
Regarding Claim 6:
Liu further teaches: The method of claim 1, wherein the generating, by the generation model, the target logical form further comprises: constructing an input to the generation model by concatenating the question and the subset of the ranked set of candidate logical forms (Pg. 29 Col 1 Par 3-4, Col 2 Par 1 “Automatically learning the lexicon indicates a mapping from natural language texts to logical forms. The previously described methods usually investigate such learning on sentences paired with their labeled logical forms, < sentence (S) -logical form (L) >:
Sentence: Utah borders Idaho
Logic Form: borders (utah, idaho)
…To resolve this problem, Zettlemoyer and Collins constructed a probabilistic model by first designing 10 templates to produce the initialized entries in the aimed lexicon and then using these entries to parse all sentences in the training data, resulting in one or more high-scoring parsers; extra entries are extracted from the results with higher scores. This procedure is repeated until no more entries are extracted.” The “probabilistic model” is a learning model that fulfills the role of the “generation model”. The “entries” produced by the 10 templates is analogous to the set of candidate logical forms as this is used as input to result in parsers for the learning model);
and generating by the generation model the target logical form based on the constructed input (Pg. 29 Col 2 Par 3, Col 3 Par 1 “We can write the learning model for converting a question to a structured logic form as
PNG
media_image2.png
57
327
media_image2.png
Greyscale
where S is the question, L is the final logical form expression for S (such as borders(utah, idaho) in Figure 3), and T indicates the progress of deriving L. Normally, we can denote T as a tree structure, as in Figure 3’s sequence of steps. One L can derive from many pairing trees T: the probability of logical forms L is marginalized out by summing over all the parses that produce L; q ∈ Rd represents the probabilistic model’s parameters.”).
Regarding Claim 7:
Liu further teaches: The method of claim 6, further comprising: decoding the subset of candidate logical forms using beam search (Pg. 30 Col 1 Par 2 “Beam-search and dynamic programming are popular technologies used for improving efficiency in semantic parsing.”).
Liu does not explicitly teach: and querying the knowledge base using each candidate logical form from the subset until a valid answer is returned.
However, Allen further teaches: and querying the knowledge base using each candidate logical form from the subset until a valid answer is returned ([0041] “Next, the candidate generation module 306 may formulate queries from the output of the question analysis module 304 and then pass these queries on to search module 308 which may consult various resources such as the internet or one or more knowledge resources, e.g., databases or corpora, to retrieve documents that are relevant to answering the user question. The candidate generation module 306 may extract, from the search results obtained by search module 308, potential (candidate) answers to the question, which it may then score (e.g., with confidence scores) and rank.”).
Regarding Claim 8:
Allen further teaches: The method of claim 7, further comprising: in response to determining that no valid answer is returned after exhausting the subset of candidate logical forms, determining that a top-ranked candidate logical form in the subset is the target logical form (Pg. 10 Claim 7 “…receiving, using the computer, feedback from a user, the feedback including an indication that none of a set of candidate answers to a first input question of the plurality of input questions is accurate, and the feedback further including a correct answer to the first input question; and in response to the receiving the feedback, identifying, using the computer, the correct answer to the first input question using the updated information source.”).
Regarding Claim 9:
Liu further teaches: The method of claim 1, further comprising: determining, for a first entity mentioned in the question, a first set of candidate entities in the knowledge base that match the first entity (Pg. 31 Col 2 Par 3 “Yih and colleagues presented a method that takes advantage of a convolutional neural networks (CNNs)-based semantic model (CNNSM) to build two parsers, one for entity mapping and the other for relation mapping. The core idea is that the relation pattern expressed in natural language and the relation in the KB can be projected to the same low-dimensional space through CNNs. The label of entity in the KB is the same as the phrase of entity in the question. CNNSM scores relational triples in the KB by using CNN-provided measures for each question and selects the top scoring relational triple as the final answer.”);
and determining linking relations between a second entity mentioned in the question and the first set of candidate entities (Pg. 30 Col 2 Par 3 “Berant and colleagues proposed a lexicon construction method to map textual relations to more than 19,000 predicates in a KB, adopting two strategies for this challenge. First, the method aligns a large text corpus to Freebase and, considering the neighboring predicates, a bridge operation is made to produce more difficult alignments. For example, in, “What government does Chile have?”, the predicate is expressed with the light verb have. Suppose the phrases Chile and government are parsed as Chile and Type.FormOfGovernment, respectively. Using the bridging predicate GovernmentTypeOf, the two parsed phrases are connected together: Type.FormOfGovernment ┌┐ GovernmentTypeOf.Chile”).
Regarding Claim 10:
Liu teaches: The method of claim 9, further comprising: concatenating, for a first candidate entity from the first set of candidate entities, the question with a corresponding linking relations to form a first input to the ranking model (Pg. 31 Col 1 Par 4 “Bordes and colleagues used an approach that jointly learns representations of words, entities, and semantic items in KBs (WordNet43). This work focuses on utterances that can be represented by a single relation form, that is, a relation (subject, direct object). An utterance is first analyzed by a semantic role labeler…” Col 2 Par 2 “This approach adopts semantic
matching energy ∈ to address the problem, which uses triples (lhs, rel, rhs) in the KB as supervision data… The training object is to minimize the energy of the observed triples in the KB.” Col 2 Par 3 “Yih and colleagues presented a method that takes advantage of a convolutional neural networks (CNNs)-based semantic model (CNNSM) to build two parsers, one for entity mapping and the other for relation mapping. The core idea is that the relation pattern expressed in natural language and the relation in the KB can be projected to the same low-dimensional space through CNNs. The label of entity in the KB is the same as the phrase of entity in the question. CNNSM scores relational triples in the KB by using CNN-provided measures for each question and selects the top scoring relational triple as the final answer.” The “relation form” or “relational triples” imply candidate entities and their linking relations with other entities which become the input for the energy computations done in the ranking model.);
generating, by the ranking model, a first similarity score between the question and the first candidate entity based on the first input (Pg. 31 Col 3 Par 1 “Bordes and colleagues proposed a more straightforward technique: instead of mapping entity mentions and relation patterns to corresponding entities and predicates in the KB, their work directly maps the question to a triple tuple in the KB. The natural language questions and the triples are represented by low-dimensional embeddings, and the similarities between them are easy and efficient to compute. A paraphrase corpus is also used for multitask training, with a goal of ensuring that the similarities between similar utterances are high.”);
ranking the first set of candidate entities based on generated similarity scores (Pg. 32 Col 1 Par 3, Col 2 Par 1 “The element mapping component (PowerMap) identifies the possible relevant semantic resources for each term; word sense disambiguation (WSD) techniques then perform semantic validation to determine a possible mapping of the term to semantic items in the KBs… A set of ranking criteria (such as mapping confidence to the facts in KBs, disambiguation, and the merging process) is then applied to select the correct answer.” Also refer to the pervious limitations’ reasonings.);
and selecting a top-ranked candidate entity from the first set as a matching entity for the first entity mentioned in the question (Pg. 31 Col 2 Par 3 Line 4 “CNNSM (convolutional neural network semantic model) scores relational triples in the KB (knowledge base) by using CNN-provided measures for
each question and selects the top scoring relational triple as the final answer.”).
Regarding Claim 11:
Allen further teaches: a memory storing a plurality of processor-executable instructions ([0070] “The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device.”);
and a processor reading and executing the instructions from the memory to perform operations comprising ([0069] “The present invention may be a system, a method, and/or a computer program product. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.”):
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the QA system of Liu with the innovations of Allen and Nallapati (as described in Claim 1) along with the additional features of an instruction-storing memory and processor taught by Allen in order to carry out the aspects of the invention in a tangible way ([0069] “The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.”)
The remainder of claim 11 recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis.
Regarding Claim 12:
The claim’s rejection reasoning is identical to that of Claim 2.
Regarding Claim 13:
The claim’s rejection reasoning is identical to that of Claim 3.
Regarding Claim 14:
The claim’s rejection reasoning is identical to that of Claim 4.
Regarding Claim 15:
The claim’s rejection reasoning is identical to that of Claim 5.
Regarding Claim 16:
The claim’s rejection reasoning is identical to that of Claim 6.
Regarding Claim 17:
This claim’s rejection reasoning is identical to that of Claim 7.
Regarding Claim 18:
This claim’s rejection reasoning is identical to that of Claim 8.
Regarding Claim 19:
This claim’s rejection reasoning is identical to those of Claims 9 and 10 combined.
Regarding Claim 20:
Allen teaches: A processor-readable non-transitory storage medium storing a plurality of processor- executable instructions for knowledge base question answering, the instructions being executed by one or more processors to perform operations comprising ([0069] “The present invention may be a system, a method, and/or a computer program product. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.”):
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the knowledge base QA system and all of its functions from Liu with the innovations of Allen and Nallapati (as described in Claim 1) along with the additional feature of a processor-readable non-transitory storage medium that stores executable instructions taught by Allen in order to carry out the aspects of the invention in a tangible way, and the instructions can be shared through a network ([0069] “The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.” [0070] “The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device.” [0071] “Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet…”).
The remainder of claim 20 recites substantially similar subject matter to claim 1 and is rejected with the same rationale, mutatis mutandis.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 JESSICA THUY PHAM whose telephone number is (571)272-2605. The examiner can normally be reached Monday - Friday, 9 A.M. - 5:00 P.M.
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, Li Zhen can be reached at (571) 272-3768. 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.
/J.T.P./Examiner, Art Unit 2121
/Li B. Zhen/Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights