DETAILED ACTION
This communication is in response to the Arguments and Remarks filed on 2/27/2026. Claims 1, 4, 6-14, and 16-24 are pending and have been examined. Hence, this Action has been made FINAL.
Any previous objection/rejection not mentioned in this Office Action has been withdrawn by the examiner.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 13, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
With regards to the rejections under 35 U.S.C. 112(b) and 112(d), the applicant has amended the claims to no longer lack antecedent basis. Furthermore, claim 5 has been cancelled thus the issue regarding than improper dependent form is no longer relevant.
With regards to the rejections under 35 U.S.C. 101, the applicant asserts that the claimed invention involves the interaction of pre-determined language processing rules and a GenAI model. In one step, a natural language description is processed in order to identify one or more tables to be queried. In another step, the natural language description and identified tables are used by a rules-based translator to generate a first database query. In yet another step, the first database query is evaluated to see if the first database query satisfies one or more language processing rules. In a further step, when the first database query is not rules complaint, a second database query is received from a GenAI model by providing the model with the natural language description or the first database query. These features provide a specific technical improvement over the prior art as described in Applicant's specification.
Examiner respectfully disagrees, the improvement presented is directed to a mental process rather than a technical one. Natural language understanding, translation to another form, and following translation rules are all processes the human mind is capable of performing. The technical aspect within claims revolves around the GenAI model which is treated as an additional component. The GenAI model’s only purpose within the claim is to translate a natural language input to a database query which is something the human mind is capable of doing. The GenAI model is merely applying the mental process and is given no specific form, implementation, or training which would suggest a technical improvement.
The applicant further asserts that the claimed invention avoids invoking the GenAI model for each input natural language description. Instead, a database query from the less computationally intense rules-based translator is used when it satisfies the one or more language processing rules.
Examiner respectfully disagrees, an improvement to computational intensity is not clearly presented within the claim language as the process is capable of being performed by the human mind and the computational components are only used to apply the process via a computer.
With regards to the rejections under 35 U.S.C. 102, the applicant asserts that the claims have been amended to obviate the present rejection under § 102, and that Tan does not disclose each and every element in the arrangement as claimed
Examiner respectfully disagrees, Tan et al. is still relevant prior art and teaches each of the limitations of amended independent claims. A portion of the amendments consist of implementing claims 2 and 3 within the independent claim 1 which both were previously rejected using the same reference. Furthermore, the inclusion of “components” such as a table determiner, rules-based translator, and rules-based evaluator do not narrow the claim further as there is no described structure to any of them and thus, they can be equated to the model components of Tan et al. Further details can be found in the updated prior art rejections found below rejection below.
With regards to the rejections under 35 U.S.C. 103, the applicant asserts that the rejections should be withdrawn as claims 6-7, 9-11, 16-17, and 19 each depend from claims that are not anticipated under 35 U.S.C. § 102 as described above.
Examiner respectfully disagrees for the same reasons presented above in response to the 35 U.S.C. 102 rejections.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 6, 7, 16, and 17 rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
The limitation “verifying the second database query” does not have support within the specification. Prior to amendment the limitation read “verifying a result of the large language model” which is a broader interpretation of the limitation. The closest supporting portion in the specification is “A result of the large language model is verified based on one or more guardrails, wherein the one or more guardrails comprise syntactic rules. A result of the large language model is verified based on one or more guardrails” in paragraphs 30 and 33. Furthermore, paragraphs 53-58 and Figs. 3A-3C show the verification happening but is only seen being done on general model output.
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, 4, 6-14, and 16-24 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claims 1, 14, and 20 recite receiving a natural language description; processing, using a table determiner, the natural language description, wherein the table determiner identifies one or more tables within a database to be queried based on the natural language description; generating, using a rules-based translator, a first database query based on the natural language description and the one or more tables; determining, by a rules-based evaluator, that the first database query does not satisfy one or more language processing rules; providing, to a [generative artificial intelligence (GenAI)] model, a portion of the first database query or a portion of the natural language description; receiving, from the GenAI model, a second database query based on the portion of the first database query or the portion of the natural language description; in response to receiving the second database query, causing the second database query to be executed at the database; and receiving data from the database in response to the second database query.
The limitations in these claims, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. The human mind is capable of understanding natural language and creating a query. For example, somebody could be asked, in English, to find something in a database. They could use their prior knowledge of said database to immediately identify relevant tables in the database. They could then follow a rule-based approach to creating the query by following a SQL guidebook and using the tables they identified as potentially being relevant. If the guidebook wasn’t working for this question, they could use their own prior knowledge of SQL to create a query or to modify the original question to work with the guidebook. Finally, the human mind is capable of “executing” a query by manually gathering the information that the initial question was asking for. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. In particular, the claims recite the additional component of a generative AI model. The generative AI model is detailed in paragraph 74 specification with a generic example model provided. Claim 14 specifically lists additional components of a processor and a memory. The processor is detailed in paragraph 77 specification with a generic description of the component. The memory is detailed in paragraph 78 specification with a generic description of the component. Claim 20 specifically lists additional component of a non-transitory computer readable storage medium. The non-transitory computer readable storage medium is detailed in paragraph 82 specification with a generic description of the component. Accordingly, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claims are not patent eligible.
Claim 4 recites wherein the one or more language processing rules are based at least in part on Backus-Naur Form (BNF).
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable of following rules in Backus-Naur form as it is merely a format to follow when making the conversion and a guidebook/prior knowledge could be used to follow this format. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claims 6 and 16 recite verifying the second database query based on one or more guardrails, wherein the one or more guardrails comprise syntactic rules.
The limitations in these claims, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human can verify that the query they created is corrected by using syntactic guardrails. This would be equivalent to double checking to make sure they used the correct syntax when creating their query. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The additional element of a large language model is being interpreted as the same element as the generative AI (GenAI) model from the independent claims. Otherwise, the claims do not recite any additional components that were not present in the independent claims. Accordingly, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claims are not patent eligible.
Claims 7 and 17 recite verifying the second database query based on one or more guardrails, wherein the one or more guardrails comprise semantic rules, wherein the semantic rules comprise semantic rules corresponding to one or more of the following: column types, choice values, numbers, dates, or time.
The limitations in these claims, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human can verify that the query they created is corrected by using semantic guardrails. This would be equivalent to double checking to make sure they used the correct names for all column types, choice values, numbers, dates, and times when creating their query. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claims do not recite any additional components that were not present in the independent claims. Accordingly, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claims are not patent eligible.
Claims 8 and 18 recite collecting training data; and pre-training or fine-tuning the GenAI model based on the collected training data.
The limitations in these claims, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable of collecting training data, for example, by writing down/gathering a large amount of natural language questions and database query pairs that they know correspond together. Then they are capable of making a design decision like using that gathered data to train a language model. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claims do not recite any additional components that were not present in the independent claims. Accordingly, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claims are not patent eligible.
Claims 9 and 19 recite wherein collecting the training data comprises collecting training data based on one or more of the following: crowdsourced data, data augmentation, or paraphrasing.
The limitations in these claims, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human can gather crowd sourced data by finding a textbook with examples of natural language to database query translations. A human can augment data by using prior knowledge to change words that they know will still have the same meaning to the database. A human can paraphrase data using their prior knowledge of the language. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claims do not recite any additional components that were not present in the independent claims. Accordingly, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claims are not patent eligible.
Claim 10 recites wherein the data augmentation comprises generating variations in one or more of the following: dates, years, or numbers.
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable of identifying dates, years, and numbers in a dataset and swapping them out for different values. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 11 recites wherein the data augmentation comprises generating variations in one or more of the following: questions or multi-conditions with choice values.
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable of identifying choice values in a query language and changing the choices to different values. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 12 recites wherein the data augmentation comprises generating variations in spelling mistakes.
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable of manually creating spelling mistakes using their prior knowledge of the language. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 13 recites adding tokens to a tokenizer for the GenAI model, wherein the tokens include one or more of the following: operators, table names, or column names.
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable of tokenizing operators, table names, and column names. They could do this by converting them to their ASCII equivalent or some other form of embedding them. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 21 recites further comprising: modifying, by a post-processor, the second database query, wherein the post-processor modifies the second database query by correcting one or more errors in the second database query.
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable correcting errors found in a query using their prior knowledge of the database query language. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 22 and 24 recites further comprising: receiving a further natural language description; processing, using the table determiner, the further natural language description, wherein the table determiner identifies one or more further tables within the database to be queried based on the further natural language description; generating, using the rules-based translator, a further database query based on the further natural language description and the one or more further tables; determining, by the rules-based evaluator, that the further database query satisfies the one or more language processing rules; causing the further database query to be executed at the database; and receiving data from the database in response to the further database query.
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. The human mind is capable of understanding further natural language questions and creating a query. For example, somebody could be asked, in English, to find something in a database. They could use their prior knowledge of said database to immediately identify relevant tables in the database. They could then follow a rule-based approach to creating the query by following a SQL guidebook and using the tables they identified as potentially being relevant. If the guidebook wasn’t working for this question, they could use their own prior knowledge of SQL to create a query or to modify the original question to work with the guidebook. Finally, the human mind is capable of “executing” a query by manually gathering the information that the initial question was asking for. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
Claim 23 recites further comprising: prior to causing the further database query to be executed at the database, verifying that the further database query is an output of the rules-based translator.
The limitations in this claim, as drafted, are a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. A human is capable of creating a rules-based version of a query translation as well as verifying if a query came from that approach. For example, using prior knowledge of the rules/guide used to create a query to recognize how it would have instructed someone to create the query. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. The claim does not recite any additional components that were not present in the independent claim. Accordingly, the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. The claim is not patent eligible.
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.
Claims 1, 8, 13, 14, 18, and 20-24 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by US Patent Application Publication US 20250217351 A1 (Tan et al.).
Regarding Claims 1, 14, and 20, Tan et al. teaches A method comprising:
(An aspect of the disclosure provides a computer-implemented method) (Paragraph 4).
Alternatively, claim 14 states a processor configured to:
(The methods 500, 530, 560, or each of the aforementioned methods' individual functions, routines, subroutines, or operations can be performed by a processing device, having one or more processing units (CPU)) (Paragraph 75).
Alternatively, claim 20 states A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:
(The data storage device 608 may include a non-transitory computer-readable storage medium 624 on which is stored the sets of instructions 626 of the system architecture 100) (Paragraph 88).
receiving a natural language description;
(the client device 130 may send a prompt 202 to the prompt generator 112. The prompt 202 may include text data that includes a natural language input that requests the MLM 122 to generate a database query.) (Paragraph 51).
A prompt is received from a user device in the form of natural language input.
processing, using a table determiner, the natural language description, wherein the table determiner identifies one or more tables within a database to be queried based on the natural language description;
(The request 302 may include text data describing the client device's 130 request. As noted above, the request 302 may be based on the prompt 202 provided by the client device 130. The modified prompt 300 may include one or more table definitions 304. The one or more table definitions 304 may include text data that describe a table of the database, including the table name (e.g., “STUDENTS”), the one or more columns in the table (e.g., “name,” “studentID,” “GPA,” etc.), and a data type of a column (e.g., “TEXT,” “NUMBER,” “DATE,” etc.). The modified prompt 300 may include one or more table references 306. A table reference may indicate that a column from one table references a column in another table (e.g., “STUDENTS(studentID)->ENROLLMENT(studentID)” may indicate that the “studentID” column in the “STUDENTS” table references the “studentID” column in the “ENROLLMENT” table. A table reference 306 may indicate use of a foreign key or other referencing data in the database. The table definitions 304 or table references 306 may be based on the context information 203.) (Paragraph 54).
Context information from the DBMS is used to associate the natural language with tables from the database. The table determiner is represented by the context information 203 shown in Fig. 2A.
generating, using a rules-based translator, a first database query based on the natural language description and the one or more tables;
(The DBMS 116 may include a metadata catalog, which may store data about the database, such as table or column names, column data types, a database schema, data indicating relationships between tables, a knowledge graph that indicates relationships between database objects for generating database queries, etc.) (Paragraph 30).
(the prompt generator 112 may receive the prompt 202 from the client device 130. In some embodiments, the prompt generator 112 may modify the prompt 202 to generate a modified prompt 204. In some embodiments, modifying the prompt 202 may include rewording the text of the prompt.) (Paragraph 52).
(modifying the prompt 202 may include adding context information 203 to the prompt 202. Adding context information 203 may include adding text data that describes a database schema or tables of a database.) (Paragraph 53).
(The MLM 122 may accept the modified prompt 204 as input and may execute the modified prompt 204. The MLM 122 may generate a first query 206 in response to executing the modified prompt 204.) (Paragraph 55).
The natural language description (prompt) is modified by the prompt generator using techniques that can be considered language processing rules. The modified prompt is then used to generate a query via an MLM. The table names are incorporated into the prompts which are then used by the MLM to generate database queries. This can be visualized in Fig. 2A where the rules-based generator is represented by the Prompt Generator 112.
determining, by a rules-based evaluator, that the first database query does not satisfy one or more language processing rules;
(At operation 506, responsive to processing logic determining that the first query includes an uncorrectable error, processing logic generates a prompt element. The prompt element may describe the uncorrectable error. The prompt element may be structured for inclusion in a prompt requesting the MLM to generate a modified first query that corrects the uncorrectable error. In one embodiment, the prompt element may include the prompt element 216. The prompt requesting the MLM to generate the modified first prompt may include the second prompt 218. The modified first query may include the modified first query 222.) (Paragraph 78).
If errors are detected in the query (language processing rules are not satisfied) then a new query is generated using the originally modified prompt and a second prompt (second portion of natural language description). The rules-based evaluator is represented by the AST 208 in Fig. 2A.
providing, to a generative artificial intelligence (GenAI) model, a portion of the first database query or a portion of the natural language description;
(In some embodiments, the second prompt 218 may include the second database query 214 and the information describing the one or more uncorrectable errors. Similar to the modified prompt 204, as discussed above, the second prompt 218 may include context information that may help the MLM 122 generate a database query, in some embodiments.) (Paragraph 69).
This is generated using an MLM which can be considered a form of GenAI model.
receiving, from the GenAI model, a second database query based on the portion of the first database query or the portion of the natural language description;
(responsive to receiving the second prompt 218 as depicted in FIG. 2A, the MLM 122 may output a modified first query 222. In some embodiments, the modified first query 222 may include a database query that corrects the uncorrectable error(s) included in the second database query generated by the SQL generator 230.) (Paragraph 71).
A modified query is generated by the MLM based on the error information and second prompt.
in response to receiving the second database query, causing the second database query to be executed at the database; and
(FIG. 2C continues the example flow of data depicted in FIG. 2B. In one embodiment, responsive to the AST 224 not including any errors, the SQL generator 230 may submit the modified first query 222 to an external DBMS 240 for execution. The external DBMS 240 may include a database that includes data on which the modified first query 222 may execute.) (Paragraph 72).
Queries generated from this method are executed on a database to retrieve data.
receiving data from the database in response to the second database query.
(The database response 226 may include data indicating whether the query was successful (e.g., for an INSERT query, data indicating whether the data in the INSERT query was successfully added to the database). The external DBMS 240 may provide the database response 226 to the prompt generator 112. The prompt generator 112 may generate a response 228 and provide the response 228 to the client device 130.) (Paragraph 73).
Response is received from the execution of the query and provided back to the user.
Additionally, claim 14 states a memory coupled to the processor and configured to provide the processor with instructions.
(The methods 500, 530, 560, or each of the aforementioned methods' individual functions, routines, subroutines, or operations can be performed by a processing device, having one or more processing units (CPU) and memory devices communicatively coupled to the CPU(s).) (Paragraph 75).
Regarding Claims 8 and 18, Tan et al. teaches the method of claims 1 and 14, further comprising: collecting training data;
(the MLM 122 can be a model that is first pre-trained on a corpus of data to create a foundational model, … The foundational model can first be pre-trained using a corpus of data that can include data in the public domain, licensed content, and/or proprietary content. … In some embodiments, this first, foundational model can be trained using self-supervision, or unsupervised training on such datasets.) (Paragraph 39).
The MLM is trained using data from multiple sources. Thus, training data is collected from public domain, licensed content, and proprietary content sources.
and pre-training or fine-tuning GenAI model based on the collected training data.
(the MLM 122 can be a model that is first pre-trained on a corpus of data to create a foundational model, and afterwards is fine-tuned on more data pertaining to a particular set of tasks to create a more task-specific, or targeted, model. The foundational model can first be pre-trained using a corpus of data that can include data in the public domain, licensed content, and/or proprietary content. Such a pre-training can be used by the MLM 122 to learn broad elements including, image or speech recognition, general sentence structure, common phrases, vocabulary, natural language structure, computer code structure (including SQL queries), and other elements. In some embodiments, this first, foundational model can be trained using self-supervision, or unsupervised training on such datasets.) (Paragraph 39).
Pre-training and fine-tuning are done on the MLM using the data gathered from the prior mentioned sources.
Regarding Claim 13, Tan et al. teaches the method of claim 1, Adding tokens to a tokenizer for the GenAI model,
(In one embodiment, the query compiler 114 may provide one or more prompt elements 216 to the prompt generator 112. In some embodiments, a prompt element 216 may include information that describes the one or more uncorrectable errors in the second database query 214. In some embodiments, a prompt element 216 may include information that is different from, or in addition to, a conventional error message (if any) from a conventional database query compiler. For example, in some embodiments, a prompt element 216 can be formatted in a format that is acceptable to the MLM 122 (e.g., valid input).) (Paragraph 68).
The prompts are converted into a format that can be accepted by the MLM. This process is considered tokenization.
wherein the tokens include one or more of the following: operators, table names, or column names.
(The DBMS 116 may include a metadata catalog, which may store data about the database, such as table or column names, column data types, a database schema, data indicating relationships between tables, a knowledge graph that indicates relationships between database objects for generating database queries, etc.) (Paragraph 30).
Prompts are created using context information from the DBMS. The DBMS includes table and column names. Fig. 2A shows the flow from the DBMS to the prompt generator to the MLM.
Regarding Claim 21, Tan et al. teaches the method of claim 1, modifying, by a post-processor, the second database query, wherein the post-processor modifies the second database query by correcting one or more errors in the second database query.
(In some embodiments, the second prompt 218 may include the second database query 214 and the information describing the one or more uncorrectable errors. Similar to the modified prompt 204, as discussed above, the second prompt 218 may include context information that may help the MLM 122 generate a database query, in some embodiments.) (Paragraph 69).
(responsive to receiving the second prompt 218 as depicted in FIG. 2A, the MLM 122 may output a modified first query 222. In some embodiments, the modified first query 222 may include a database query that corrects the uncorrectable error(s) included in the second database query generated by the SQL generator 230.) (Paragraph 71).
(At operation 506, responsive to processing logic determining that the first query includes an uncorrectable error, processing logic generates a prompt element. The prompt element may describe the uncorrectable error. The prompt element may be structured for inclusion in a prompt requesting the MLM to generate a modified first query that corrects the uncorrectable error. In one embodiment, the prompt element may include the prompt element 216. The prompt requesting the MLM to generate the modified first prompt may include the second prompt 218. The modified first query may include the modified first query 222.) (Paragraph 78).
A second database query is modified using the identified uncorrectable errors. The query has those errors corrected and then a new SQL query is generated from it.
Regarding Claim 22 and 24, Tan et al. teaches the method of claim 1 and 14, receiving a further natural language description;
In Fig. 2A it can be seen that the process starts with the client device and Figs. 2B-2C show the process ending with client device. This shows a round-trip process in which further queries participate in the same process.
processing, using the table determiner, the further natural language description, wherein the table determiner identifies one or more further tables within the database to be queried based on the further natural language description;
(The request 302 may include text data describing the client device's 130 request. As noted above, the request 302 may be based on the prompt 202 provided by the client device 130. The modified prompt 300 may include one or more table definitions 304. The one or more table definitions 304 may include text data that describe a table of the database, including the table name (e.g., “STUDENTS”), the one or more columns in the table (e.g., “name,” “studentID,” “GPA,” etc.), and a data type of a column (e.g., “TEXT,” “NUMBER,” “DATE,” etc.). The modified prompt 300 may include one or more table references 306. A table reference may indicate that a column from one table references a column in another table (e.g., “STUDENTS(studentID)->ENROLLMENT(studentID)” may indicate that the “studentID” column in the “STUDENTS” table references the “studentID” column in the “ENROLLMENT” table. A table reference 306 may indicate use of a foreign key or other referencing data in the database. The table definitions 304 or table references 306 may be based on the context information 203.) (Paragraph 54).
Context information from the DBMS is used to associate the natural language with tables from the database.
generating, using the rules-based translator, a further database query based on the further natural language description and the one or more further tables;
(The DBMS 116 may include a metadata catalog, which may store data about the database, such as table or column names, column data types, a database schema, data indicating relationships between tables, a knowledge graph that indicates relationships between database objects for generating database queries, etc.) (Paragraph 30).
(the prompt generator 112 may receive the prompt 202 from the client device 130. In some embodiments, the prompt generator 112 may modify the prompt 202 to generate a modified prompt 204. In some embodiments, modifying the prompt 202 may include rewording the text of the prompt.) (Paragraph 52).
(modifying the prompt 202 may include adding context information 203 to the prompt 202. Adding context information 203 may include adding text data that describes a database schema or tables of a database.) (Paragraph 53).
(The MLM 122 may accept the modified prompt 204 as input and may execute the modified prompt 204. The MLM 122 may generate a first query 206 in response to executing the modified prompt 204.) (Paragraph 55).
The natural language description (prompt) is modified by the prompt generator using techniques that can be considered language processing rules. The modified prompt is then used to generate a query via an MLM. The table names are incorporated into the prompts which are then used by the MLM to generate database queries. This can be visualized in Fig. 2A.
determining, by the rules-based evaluator, that the further database query satisfies the one or more language processing rules;
(At operation 506, responsive to processing logic determining that the first query includes an uncorrectable error, processing logic generates a prompt element. The prompt element may describe the uncorrectable error. The prompt element may be structured for inclusion in a prompt requesting the MLM to generate a modified first query that corrects the uncorrectable error. In one embodiment, the prompt element may include the prompt element 216. The prompt requesting the MLM to generate the modified first prompt may include the second prompt 218. The modified first query may include the modified first query 222.) (Paragraph 78).
If errors are detected in the query (language processing rules are not satisfied) then a new query is generated using the originally modified prompt and a second prompt (second portion of natural language description).
causing the further database query to be executed at the database;
(FIG. 2C continues the example flow of data depicted in FIG. 2B. In one embodiment, responsive to the AST 224 not including any errors, the SQL generator 230 may submit the modified first query 222 to an external DBMS 240 for execution. The external DBMS 240 may include a database that includes data on which the modified first query 222 may execute.) (Paragraph 72).
Queries generated from this method are executed on a database to retrieve data.
and receiving data from the database in response to the further database query.
(The database response 226 may include data indicating whether the query was successful (e.g., for an INSERT query, data indicating whether the data in the INSERT query was successfully added to the database). The external DBMS 240 may provide the database response 226 to the prompt generator 112. The prompt generator 112 may generate a response 228 and provide the response 228 to the client device 130.) (Paragraph 73).
Response is received from the execution of the query and provided back to the user.
Regarding Claim 23, Tan et al. teaches the method of claim 22, prior to causing the further database query to be executed at the database, verifying that the further database query is an output of the rules-based translator.
(At operation 506, responsive to processing logic determining that the first query includes an uncorrectable error, processing logic generates a prompt element. The prompt element may describe the uncorrectable error. The prompt element may be structured for inclusion in a prompt requesting the MLM to generate a modified first query that corrects the uncorrectable error. In one embodiment, the prompt element may include the prompt element 216. The prompt requesting the MLM to generate the modified first prompt may include the second prompt 218. The modified first query may include the modified first query 222.) (Paragraph 78).
(FIG. 2C continues the example flow of data depicted in FIG. 2B. In one embodiment, responsive to the AST 224 not including any errors, the SQL generator 230 may submit the modified first query 222 to an external DBMS 240 for execution. The external DBMS 240 may include a database that includes data on which the modified first query 222 may execute.) (Paragraph 72).
Queries generated from this method are executed on a database to retrieve data. If errors are detected in the query (language processing rules are not satisfied) then a new query is generated using the originally modified prompt and a second prompt (second portion of natural language description).
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over US Patent Application Publication US 20250217351 A1 (Tan et al.) in view of “A Generic Model for Natural Language Interface to Database” (Sujatha et al.).
Regarding Claim 4, Tan et al. teaches the method of claim 1.
Tan et al. does not explicitly teach: wherein the one or more language processing rules are based at least in part on Backus-Naur Form (BNF).
However, Sujatha et al. teaches wherein the one or more language processing rules are based at least in part on Backus-Naur Form (BNF).
(To convert natural language query to first order logic conversion uses the notation to define the Backus-Naur Form BNF. There are many other techniques available to describe the language syntax such as Extended Backus-Naur Form (EBNF), Augmented Backus-Naur Form (ABNF) etc. BNF is a formal language to encrypt grammar for human utilization. Each rule in BNF satisfies the format as Term: : =expansion. The basic formula to convert a natural language statement into First Order Logic is given by:) (Section 3.1, Paragraph 5).
Sujatha et al. describes a system of converting a natural language statement to a database query by using a rule-based approach that takes a Backus-Naur Form.
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify the natural language to database query method as taught by Tan et al. to use an initial rules-based approach in Backus-Naur Form as taught by Sujatha et al. This would have been an obvious substitution to this step in the method as the BNF is one of many ways the language syntax can be represented when forming a query. (Sujatha et al. Section 3.1, Paragraph 5).
Claim 6-7 and 16-17 are rejected under 35 U.S.C. 103 as being unpatentable over US Patent Application Publication US 20250217351 A1 (Tan et al.) in view of “Robust Text-to-SQL Generation with Execution-Guided Decoding” (Wang et al.).
Regarding Claims 6 and 16, Tan et al. teaches the method of claim 1 and 14.
Tan et al. does not explicitly teach: verifying the second database query based on one or more guardrails, wherein the one or more guardrails comprise syntactic rules.
However, Wang et al. teaches verifying a result of the second database query based on one or more guardrails,
(We introduce a new mechanism, execution guidance, to leverage the semantics of SQL. It detects and excludes faulty programs during the decoding procedure by conditioning on the execution of partially generated program. The mechanism can be used with any autoregressive generative model … We demonstrate that execution guidance universally improves model performance on various text-to-SQL datasets with different scales and query opponent = UEFA opponent > Haugar opponent = Haugar competition = UEFA Partial Programs complexity:)) (Abstract).
Wang et al. teaches a method of verifying the outputs of natural language to database query systems. This method employs a form of guardrail by executing partially generated queries to detect faulty code produced by generative models such as an LLM.
wherein the one or more guardrails comprise syntactic rules.
(A program p causes a parsing error if it is syntactically incorrect. This kind of error is more common for complex queries (as appearing in the GeoQuery and ATIS datasets). Autoregressive models are more prone to such errors than template-based and slot-filling-based models.) (Section “Execution-Guided Decoding”, Subsection “Execution Errors”, Paragraph 1).
The guardrails comprise checks for syntactical errors.
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify the natural language to database query method as taught by Tan et al. to use guardrails for the output of the LLM as taught by Wang et al. This would have been an obvious improvement as it would allow detection of errors in the results of the LLM prior to it being executed at a database and provided to the user. (Wang et al. Section “Execution Guided Decoding”, Paragraph 1).
Regarding Claims 7 and 17, Tan et al. teaches the method of claim 1 and 14.
Furthermore, Wang et al. teaches verifying the second database query based on one or more guardrails,
(We introduce a new mechanism, execution guidance, to leverage the semantics of SQL. It detects and excludes faulty programs during the decoding procedure by conditioning on the execution of partially generated program. The mechanism can be used with any autoregressive generative model … We demonstrate that execution guidance universally improves model performance on various text-to-SQL datasets with different scales and query opponent = UEFA opponent > Haugar opponent = Haugar competition = UEFA Partial Programs complexity:)) (Abstract).
Wang et al. employs a form of guardrail by executing partially generated queries to detect faulty code produced by generative models such as an LLM.
wherein the one or more guardrails comprise semantic rules,
(A program p throws a run-time error if it has a component whose operator type mismatches its operands types. Such an error could be caused by a mis-match between an aggregation function and its target column (e.g., sum over a column with string type) or a mis-match between a condition operator and its operands (e.g., applying > to a column of float type and a constant of string type).) (Section “Execution-Guided Decoding”, Subsection “Execution Errors”, Paragraph 2).
The guardrails comprise checks for semantic errors such as incorrect column names or operators used.
and wherein the semantic rules correspond to one or more of the following: column types, choice values, numbers, dates, or time.
(Some examples where execution guidance (EG) for Coarse2Fine leads to correct prediction. In the first example, the table column is corrected by execution guidance due to an empty output) (Fig. 5 Description).
(An execution-guided decoder evaluates partially generated queries at appropriate timesteps and then excludes those candidates that cannot be completed to a correct SQL query (red background). Here, “opponent > Haugar” would yield a runtime error, whereas “opponent = UEFA” would yield an empty result.) (Fig. 1 Description).
Fig. 5 shows an example where incorrect column types are detected by the “guardrails”. Fig. 1 shows an example where incorrect choice values (Hauger vs. UEFA) are detected by the “guardrails”.
Claim 9-11 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over US Patent Application Publication US 20250217351 A1 (Tan et al.) in view of US 12399892 B2 (Cao et al.)
Regarding Claims 9 and 19, Tan et al. teaches the method of claim 8 and 18.
Tan et al. does not explicitly teach: Wherein collecting the training data comprises collecting training data based on one or more of the following: crowdsourced data, data augmentation, or paraphrasing.
However, Cao et al. teaches Wherein collecting the training data comprises collecting training data based on one or more of the following: crowdsourced data, data augmentation, or paraphrasing.
(As the starting point, an ontology 412 of the domain of interest is built, which includes database schema, schema descriptions and other necessary meta-data. The direct labelling 414 approach refers to experts labelling SQL queries given the questions and the database schema. The indirect labelling 416 approach refers to crowd-source workers rewriting machine generated canonical utterances for SQL queries sampled from a grammar.) (Col. 6, Lines 8-19)
(FIG. 10 illustrates an example 900 of data augmentation, in accordance with some embodiments. After data cleaning 418, the resulting parallel corpus between questions and SQL could still be too small, therefore, data augmentation 420 may be performed.) (Col. 23, Lines 10-21).
(The two types of augmentation are context-free swaps of column and values names as well as automatic paraphrasing of input-questions.) (Col. 24, Lines 22-34).
Cao et al. teaches a natural language to database query system that builds its training data from crowdsourced examples, data augmentation, and paraphrasing of inputs.
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify the natural language to database query method as taught by Tan et al. to use gather training data from various sources as taught by Cao et al. This would have been an obvious improvement to expand the training of the MLM and create new data sets to be built on augmented data which can allow the system to expand to new domains if wanted (Cao et al. Col. 5, Lines 60-67)
Regarding Claim 10, Tan et al. teaches the method of claim 9.
Furthermore, Cao et al. teaches Wherein the data augmentation comprises generating variations in one or more of the following: dates, years, or numbers.
(The two types of augmentation are context-free swaps of column and values names as well as automatic paraphrasing of input-questions. The context-free swaps modify both the natural language questions as well as the corresponding SQL queries, and is only performed if the name of the column or values is a contiguous text-span.) (Col. 24, Lines 22-34).
It can be seen in Fig. 10 that numbers and time frames are augmented in the dataset examples. In the example, “YTD return” is changed to “3 month RoR” and “10%” was changed to be “5%”.
Regarding Claim 11, Tan et al. teaches the method of claim 9.
Furthermore, Cao et al. teaches Wherein the data augmentation comprises generating variations in one or more of the following: questions or multi-conditions with choice values.
(The two types of augmentation are context-free swaps of column and values names as well as automatic paraphrasing of input-questions. The context-free swaps modify both the natural language questions as well as the corresponding SQL queries) (Col. 24, Lines 22-34).
Choice values are represented by column and value names which is something Cao et al. states augmenting.
Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over US Patent Application Publication US 20250217351 A1 (Tan et al.) in view of US 12399892 B2 (Cao et al.) and further in view of“Data augmentation in natural language processing: a novel text generation approach for long and short text classifiers“ (Bayer et al.).
Regarding Claims 12, Tan et al. teaches the method of claim 9.
Tan et al. in view of Cao et al. does not explicitly teach: Wherein the data augmentation comprises generating variations in spelling mistakes.
However, Bayer et al. teaches Wherein the data augmentation comprises generating variations in spelling mistakes.
(In NLP, data augmentation is considered a difficult task [19] since textual transformations that preserve the label are difficult to define [24, 59]. Thus, many methods have been tried out in research so far. Among them are methods for swapping [59], deleting [16, 38], inducing spelling mistakes [6, 10], paraphrasing [28], and replacing of synonyms [25, 61, 66], close embeddings [2, 58] and words predicted by a language model [11, 18, 24] on word-level.) (Section 2.1, Paragraph 2).
Bayer et al. teaches a data augmentation method in which spelling mistakes are one of the types of augmentation.
It would have been obvious to a person having ordinary skill in the art before the time of the effective filing date of the claimed invention of the instant application to modify the natural language to database query method as taught by Tan et al. in view of Cao et al. to have the additional data augmentation technique as taught by Bayer et al. This would have been an obvious addition as it is a form of data augmentation that can be done without changing the label corresponding to those words. (Bayer et al. Section 2.1, Paragraph 2).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NICHOLAS DANIEL LOWEN whose telephone number is (571)272-5828. The examiner can normally be reached Mon-Fri 8:00am - 4:00pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Paras D Shah can be reached at (571) 270-1650. 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.
/NICHOLAS D LOWEN/Examiner, Art Unit 2653
/Paras D Shah/Supervisory Patent Examiner, Art Unit 2653
07/08/2026
Prosecution Insights