DETAILED ACTION
This action is responsive to the Applicant’s response filed 5/08/26.
As indicated in Applicant’s response, claims 1-2, 6-9, 14-17 have been amended, and claims 4-5, and 12-13 cancelled. Claims 1-3, 6-11, 14-17 remain pending in the next office action.
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.
Claim 1 rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial
exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly
more. Claim(s) 1 is/are directed to Abstract Idea. The claim(s) does/do not include additional
elements that are sufficient to amount to significantly more than the judicial exception because of
the following 2 step Analysis.
Step 1:
Claim 1 is directed to a method category of subject matter.
Step2A
Prong 1:
The steps recited as "determining an operation intended by a statement in NL", "determining one or more parameters", "identifying a template", "populating the template with the operation and the parameters", “generating synthetic data … defining new text statements written in NL” without further more concrete details can be construed as activities that can be performed by a human process (e.g. determining an operation, generating synthetic data defining new text in NL) and methods of organizing human activity (e.g. populating a "template"), whereas translating natural language into an operation/intent so to come up new sentences (synthetic text) are processes that can be performed in the human mind or via pen/paper. Per MPEP 2106.04(a)(3) filling out predefined forms/templates or rule-based translation from one language format to another format falls under generic data organization, rule following under this MPEP section. Per MPEP 2106.04(a)(1), machine learning models (including neural network, matrix operations) are treated as mathematical calculations or algorithms. Simply specifying that the process (“determining”) is performed “by a NN” does not strip it of its mathematical character. That is, the claim recite a Judicial Exception and is directed to an Abstract Idea type. See MPEP §2106.04(a)(2)
Prong 2:
The claim does not improve the operational efficiency of the neural network or computer system (memory management, GPU compute, NW architecture); it merely uses standard NN training technique to solve a generic translation/text generation problem. MPEP § 2106.04(d)(1)
The elements recited as “template”, “synthetic data”, ”natural language”, “source code”, “domain specific language” are described functionally in rather basic/stand-alone level of generality; the claim simply applies general data-manipulation concepts on conventional hardware without having restriction on how the hardware operates at a more concrete, structural level. MPEP 2106.04(d)(2)
The claim is directed to an abstract idea because it does not integrate the judicial exception into a practical application.
Step 2b:
The additional elements, construed individually include:
"Determining by a NN..." which merely amounts to conventional use of machine learning for classification/intent detection (MPEP § 2106.05(d)).
"Populating a template..." which merely amounts to generic software operation for code generation (MPEP § 2106.05(d)).
"Training the NN on real-world NL statements..." which merely amounts to conventional, necessary step for supervised/unsupervised machine learning (MPEP § 2106.05(d)).
"Generating synthetic data..." which merely amounts to well-known data augmentation technique in AI training models (MPEP § 2106.05(d)).
In a ordered combination, the above elements are viewed as steps that simply follow a standard pipeline: Collect data [Wingdings font/0xE0]Train neural network [Wingdings font/0xE0] Predict intent [Wingdings font/0xE0]Fill template [Wingdings font/0xE0] Output code [Wingdings font/0xE0] Generate synthetic data
This represents a standard computer workflow operating as intended. It does not alter computer functionality or solve a specific hardware problem
The claim lacks an inventive concept (MPEP § 2106.05). The elements, taken individually and as an ordered combination, amount to nothing more than generic computer processing steps (data collection/augmentation, model training, intent classification, and template population) performing their expected functions. Thus, the claim fails to recite "significantly more" than the abstract idea itself.
Claim 1 is not eligible under the 35 USC § 101 statute.
Claim 9 rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial
exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claim(s) 9 is/are directed to Abstract Idea. The claim(s) does/do not include additional
elements that are sufficient to amount to significantly more than the judicial exception because of
the following 2 step Analysis.
Per Step I, claim 9 is directed to an apparatus/system category
Per step 2a, prong 1: claim 9 recites elements such as "determine an operation intended by a statement in NL", "determine one or more parameters", "identify a template", "populate the template with the operation and the parameters", “generating synthetic data … defining new text statements written in NL” and as set forth in prong step1a, prong one of claim 1 analysis, the elements are standard activities of a mental process or standard use of mathematical methods, all grouped under the Abstract Idea subgroup of a Judicial Exception.
Per step 2a, prong 2:
The elements recited here again as “template”, “synthetic data”, ”natural language”, “source code”, “domain specific language” are described functionally in rather basic/stand-alone level of generality; the claim simply applies general data-manipulation concepts on conventional hardware without having restriction on how the hardware operates at a more concrete, structural level. MPEP 2106.04(d)(2)
The claim, based on the level of generality of its elements, does not improve the operational efficiency of the neural network or computer system (memory management, GPU compute, NW architecture); it merely uses standard NN training technique to solve a generic translation/text generation problem. MPEP § 2106.04(d)(1)
The claim thus fails to integrate the judicial exception into a practical application
Per step 2b.
This claim recites the same additional elements recited in claim 1; as set forth per step2B analysis of claim 1; these additional elements fail to add significantly more to the Abstract Idea of prong 1; i.e. the elements, taken individually and as an ordered combination, amount to nothing more than generic computer processing steps (data collection/augmentation, model training, intent classification, and template population) performing their expected functions
The claim lacks an inventive concept (MPEP § 2106.05) per step2B; and is deemed non-eligible under 35 USC § 101 statute.
Step 2B analysis of the dependent claims.
Claim 2 recites domain specific language as set of operations, and operation determined by the NN as one of the set of operations expressed in a very high level of generality fail to add substantial technical improvement to the mental process identified per step 2A.
Claim 3 recites the neural network for determining respective probabilities of classes corresponding to a respective operation; but the probabilities of classes being a standalone characterization not tied with the identification of a particular operation or template, cannot be viewed as a way to improve upon the step of determining an operation in the direction of populating a template, as part of the Judicial Exception of step 2A.
Claim 6 recites replacing words with new words SO to define synonyms; but this finding of new words fails to particularly make the template identifying and/or populating a significant computer improvement to the NL processing and/or the source code generating; nor does it render the mental process from step IIA SO to make it amount to significantly more than the Judicial Exception.
Claim 7 recites rearranging order of words and defining new order of words; but as a whole, this rearranging fails to render the mental process of claim 1 so to make it significantly more than
the well-understood organization of information associated with a mental process set forth by the Judicial Exception
Claim 8 recites that neural network is also trained on the synthetic data, but this well-known and very generic limitation recited disjoint from the identifying and determining steps of claim 1, make it impossible to see if this training would bring a novelty or improvement to field of template identification and determination observed in claim 1.
Claims 10-11, 14-16 are repeat of claims 2-3, 6-8, hence are deemed insufficient to make claim 9 for it to amount much more than a Judicial Exception.
Claim 17 recites a medium version of claim 1, hence incorporates by default the Judicial Exception of this claim.
In all, claims 1-3, 6-11, 14-17 are non-eligible under the 35 USC § 101 statute.
Claim Objections
Claims 8 and 17 are objected to under 37 CFR 1.75(c) as being recited in an improper form because a multiple dependent claim; that is, claims 8 and 17 should refer to other claims in a alternative dependency only and cannot depend from any other multiple dependent claims – e.g. claim 8: dependent on claims 1, 6, and 7; and claim 17: dependent to claims 1-3, 6 and 7. See MPEP § 608.01(n). Accordingly, claims 8 and 17 will not be fully treated based on the dependency merits as proffered in the claim language.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-2, 9-10, 17 is/are rejected under § 35 U.S.C. 103 as being unpatentable over Gutta et al, USPubN: 2023/0038529 (herein Gutta) in view of Sellam et al, USPubN: 2022/0067309 (herein Sallam) and Jalaluddin et al, USPubN: 2021/0304733 (herein Jalaluddin)
As per claim 1, Gutta discloses a computer-implemented method of generating source code in a target domain-specific language, the method comprising:
receiving a statement written in natural language text (text documents, text section of the natural language document – para 0005; natural language document - para 0039; natural language
document - para 0018);
determining, by a neural network (set of neural networks - para 0005; AI learning models, machine learning models, employs a neural network with multiple layers - para 0022; a set of deep
neural network models - para 0017), an operation (indicating an entity, a protocol operation ("action") type associated with the code template - para 0017; a smart contract may include execution or "calling" - para 0025 – Note1: processing text statement using set of neural networks to identify values or entity such as a "action" - protocol operation: step 416 - Fig. 4 - as well as set of numeric values to populate a corresponding template targeted to generate program code for smart contract "calling" within a blockchain network reads on determining by a NN an operation and associated set of parameters with which to generate a program language expressed as bytecode specific to the smart-contract interactions and blockchain NW domain - see para 0035) intended by the statement (see natural language text section from above);
based on the operation (a protocol operation ("action") - para 0017), determining one or more parameters that correspond (code template may be selected and populated based on numeric values for example an entity identifier, a value of a conditional statement, a first date, a second date a first set of numeric values - para 0005) to the operation (protocol operation ("action") type associated with the code template - para 0017; step 416 - Fig. 4);
based on the operation, identifying a template (associated with the code template – para 0017; a code template may be selected - para 0005) in the target domain-specific language (program code that aggregates transactions between smart contract hosted on a blockchain network - para 0066; bytecode version of program code to a blockchain network or distributed ledger - para 0067; bytecode version - para 0035; generated bytecode on a blockchain network - para 0017);
populating the template with the operation and the one or more parameters (code templates may be selected and populated based on numeric values for example an entity identifier, a value of a conditional statement, a first date, a second date a first set of numeric values - para 0005), so as to generate the source code (generating associated program code based on n-grams and first set of numeric values based on the document - para 0005; populate fields of the template to generate program code to generate program code compile program code into bytecode - para 0017) in the target domain-specific language (see above);
training the neural network on training data associated with the target domain-specific
language (bytecode specific to the smart-contract interactions and blockchain NW domain – see para 0035), the training data comprising real-world text statements written in natural language (natural language document - para 0039; natural language document - para 0018; para 0005)
Gutta does not explicitly disclose
generating synthetic data from the real-world text statements written in natural language, the synthetic data defining new text statements written in natural language.
As far as synthetizing a text document into another form, Gutta discloses use of TF-IDF vectorization as part of classifying element corpus of the document into categories or features from unstructured text into fixed length numerical representation, the TF-IDF being a form of synthetization that transforms a large corpus into most relevant grouping (fixed-length vector), based on weighing on frequency of occurrence of words (term frequency or TF), semantic contribution of each term, and/or its corpus-wide rarity (IDF), filtering of noise coupled with consideration of n-grams or phrase-level context.
Performing synthetization of natural language data is shown in Sallam method of training a neural network where a pair of synthetic version (original and modified) of text passage can form the input to the training (para 0004), where a plurality of sentence pairs to the neural network enable fine-tuning a grade allocated for each pair (para 0005), where creation of the synthetic sentence pair is done by effecting random substitution in terms of replacing words in passage A with replacement word resulting in a second passage B (para 0028) or random omission of text (para 0029); hence, synthetic transformation to the input text in terms of omission or substitution of the original text passage for generating input to the NN learning entails replacement or omission to form new version of the original text.
Similar to grouping of terms of contextual weight by a TF-IDF from a large unstructured corpus, replacement of the original words with equivalent terms as part of a synthetization of text or utterance data is shown in Jalaluddin's natural language processing and configuration of training models, where training includes pre-labeling of text (phrase or sentence) in accordance to an intent, sing data augmentation or irrelevant text addition (para 0116-0119) by way of randomly augmenting the original text with synonym insertion or replacement, position swapping or deletion, substitution of word or n-gram of the original stream with a vector and numerical score (TF-IDF score - para 0124) to render the synthetized training set of for the training more agnostic (para 0123) or render the training more robust (para 0117). Hence, modifying a statement with synonym, position swapping or word deletion entails text re-arrangement and generating of new statement resulting from the modification.
Therefore, as training of real-world text statement in Gutta system is based from a initial data synthesis or decomposition into significant context grouping, it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to implement training and synthetizing a text source into another form as set forth in Gutta’s vectorization of training data so that the preprocessing of text data into the training would include
generating synthetic data from the real-world text statements written in natural language, the synthetic data defining new text statements written in natural language - as shown in the replacement by Sallem and Jalaluddin; because
input in natural language most often comes in unstructured format that makes it difficult to
parse out terms or grouping or phrase of application specific significance, functional weight or
contextual merits, and by restricting the data into more organized or structure form – via synthetization techniques, the structured or re-arranged format would enable particular group to be extracted and assigned with a value, substitution word or a label, via deletion or replacement, as set forth above - in that a large unstructured corpus can be restructured into organized subset of synthetized data which in turn can be easily processed by mathematical or numerical means (vectorization) associated with the classification aspect of a NN or machine learning model configured to train the textual input,
a synthesis stage to pre-process the initial unstructured text data as set forth above in order to separate or distinguish portion thereof in terms of intentional text augmentation or modification thereof - as set forth above, via synonym replacement, word omission, or word swapping, or word order altering or creation of new statement – would enable this restructuring to reinforce the agnostic aspect of data synthetizing prior to submitting it to a training model, thus stripping off any biased context of the input, minimizing likelihood of mis-evaluation and scaling up possibilities for re-evaluating result/inference by the training in the course of filtering out unacceptable classification outcome and for better finetuning of a hyperparameter targeted by the machine learning
As per claim 2, Gutta discloses computer-implemented method as recited claim 1, wherein
the target domain-specific language (refer to Notel from above) defines a set of operations (smart
contract may include execution or "calling" or protocol action - para 0025, 0017), and the operation
determined by the neural network is one of the operations (a protocol operation ("action") type
associated with the code template - para 0017; step 416 - Fig. 4) in the set of operations (refer to
calling instances in a smart-contract and blockchain network from Note1).
As per claim 9, Gutta discloses a computing system configured to generate source code in a plurality of domain-specific languages, the computing system comprising: one or more processors; and a memory storing instructions that, when executed by the one or more processors, cause the computing system to:
receive a statement written in natural language text;
determine an operation intended by the statement;
based on the operation, determine one or more parameters that correspond to the operation;
based on the operation, identify a template in a target domain-specific language of the plurality of domain-specific languages;
populate the template with the operation and the one or more parameters, so as to generate the source code in the target domain-specific language;
train a neural network on training data associated with the target domain-specific language, the training data comprising real-world text statements written in natural language; and
generate synthetic data from the real-world text statements written in natural language, the synthetic data defining new text statements written in natural language.
( All of which having been addressed in claim 1)
As per claim 10, refer to rejection of claim 2.
As per claim 17, Gutta discloses a non-transitory computer-readable storage medium
including instructions that, when processed by a computing system, configure the computing
system to perform the method (refer to claim 1) according to any one of claims 1-3, 6, and 7 (refer to Claim Objection).
Claims 3, 11 is/are rejected under § 35 U.S.C. 103 as being unpatentable over Gutta et al, USPubN: 2023/0038529 (herein Gutta) in view of Sellam et al, USPubN: 2022/0067309 (herein Sallam) and Jalaluddin et al, USPubN: 2021/0304733 (herein Jalaluddin)in view of Xu et al, USPubN: 2022/0171947 (herein Xu)
As per claim 3, Gutta does not explicitly disclose computer-implemented method as recited in claim 2, wherein determining the operation intended by the statement further comprises:
determining, by the neural network, respective probabilities associated with a plurality of
classes, each class in the plurality of classes corresponding to a respective operation in the set of
operations.
Xu discloses prediction models (para 0167) such as DNN, CNN for natural language processing, the models to train sentences from utterances or text input, then representing the word level or n-grams as topic vector of the respective text/NLP features (para 0139-0141), the training output including probability of class whose values are evaluated each time in determining whether a given utterance portion corresponds to a respective class among the set of classes yielded as intermediate outputs from a given classifier instance(para 0168) using a logit function to calculate a value for each utterance instance so to fit a distribution of probability values of classes (para 0169) with that utterance, yielding a logarithm of odds which can be weighted by a centroid effect (para 0182) in accordance with effect to correspond set of binary classifiers with a logit function (cross- entropy function) that underlies validity of each utterance target (para 0170), such that by adapting modification of this function the classification system would govern distance between the intermediate output and the centroid representation enabling finetuning the target (hyperparameter) of the training (para 0184); hence use of activation a cross-entropy function (para 0191) and managing it (Fig. 7-8) in conjunction with computed probability values of classes to correlate correspondence of an item of the utterance input with a set of classes generated as intermediate outputs by the NN entails improving prediction accuracy of the model in minimizing difference between class probability values output with a centroid representing a item (utterance) of the input stream.
That is, if the neural network as in Gutta is to weigh significance of a lexical item among the input text against probability values of intermediate classes generated by a stage of NN predictive model so to finetune accuracy of the lexical item target under evaluation by the training to identify the most relevant lexical item within a cross-entropy space in that probability values of respective intermediate classes output from a given classifier run are used with a loss function to attain the lowest distance separating each probability with the lexical item targeted by the training.
Therefore, it would have been obvious for one of ordinary skill in the art before the effective filing date of the invention to implement the neural network in determining a operation as part of the natural language text or document into Gutta NLP system so that confirming validity or intent of a feature among the words or lexical element of the natural language would include use of the Neural network to calculate for each class (or probabilities associated with classes) generated as intermediate output by a respective classifier run, one or more probability values so to correlate distance from the probability values (of the class) versus a centroid values under establishment of loss function as set forth in Xu, whereby the loss function and adjustment thereof enables minimizing the distance from the probability values respective to the centroid representation, for the NN to consolidate weight and significance of a given lexical element, or a word indicative of a function or operation among a set of thereof underlying "calling" activities within the business domain of smart-contract or blockchain network in Gutta; because
use of prediction model with layered execution or outputs of a neural network model, so that classes of intermediate output generated from topic vectors formed from a natural language stream of lexical elements from a text document or utterance statements can be evaluated against a minimizing function whereby probability values of the intermediate class of outputs can be correlated to the centroid representation of a lexical item targeted via hyperparametric setting of the Neural network model run would enable iterative improvement of the entropy space thereby minimizing the distance from probability of intermediate output respective to central point set by this minimizing function, thereby enable the NN model to determine the best and optimal set of output classes that best consolidate validity of the lexical item established as a hyperparameter to tune as part of the multi-layered classification model; e.g. whereby a validated lexical item as well as its contextual metadata can be set forth by the NN as an element/action of functional significance along with pertinent parameters to form a specific package by which domain-specific code generation can be initiated to programmatically suit the code calling of business activities that belong to the smart contract and blockchain domain in Gutta.
As per claim 11, refer to rationale of claim 4.
Claims 6-8, 14-16 is/are rejected under § 35 U.S.C. 103 as being unpatentable over Gutta et
al, USPubN: 2023/0038529 (herein Gutta) in view of Sellam et al, USPubN: 2022/0067309 (herein Sallam) and Jalaluddin et al, USPubN: 2021/0304733 (herein Jalaluddin) and further in view of Stabler et al, USPubN: 2021/0224486 (herein Stabler) and Nakao et al, USPubN: 2019/0179908 (herein Nakao)
As per claims 6-8, Gutta does not explicitly disclose computer-implemented method as recited in claim 1, wherein generating the synthetic data further comprises
(i) replacing one or more words of the real-world text statements with one or more synonyms of the one or more words, so as to define the new text statements written in natural language that include the one or more synonyms;
(ii) rearranging an original order of one or more words of the real- world text statements, so as to define the new text statements written in natural language that include words in a different order as compared to the original order;
(iii) wherein the training data further comprises the synthetic data such that
the neural network is also trained on the synthetic data.
As for (i)
Text re-arrangement/swapping, replacing, deleting and modifying the original NL text is shown in Sellam and Jalaluddin technique using re-arrangement, modification of unstructured input corpus text into a particular synthetized format (new text) destined for a machine learning – refer to rationale A in claim 1.
Converting the original text into new data is also shown in the synthetization by Nakkao, in terms of voice recognition or sentence evaluation by way of using neural network (para
0114) for training information provided as n-gram models (para 0056, 0125; claim 5, pg. 11) via
use of a controller that performs synthesis on the voice data (para 0127; voice recognition text -
para 0079; S18 - Fig. 10), evaluate utterance value by correlating previous and present occurrence
of the text utterance (para 0130), respective to an appearance probability (para 0080; Fig. 8) or
predetermined threshold (Fig. 7) thereby to determine a new data (0089-0091; claim 1, pg. 11) from
the voice text recognition process. Hence, performing synthesis of input text from utterance stream
using a neural network set on n-grams model to derive a new data from the utterance text is
recognized.
As for (ii),
Similar to creating new words in Jalaluddin, Stabler discloses adversarial training for NL using substitution to the original text in terms of replacing of words/phrases with equivalent words/phrases, by removing words or phrases, swapping characters, switching order of words or phrases in association with preserving composition of the original linguistic set and maintaining a semblance of symmetries (para 0036; permutation symmetries - para 0139), e.g. to simulate higher likelihood of misclassification (para 0006-0008) by the machine learning as part of adversarial training (Fig. 5-7), whereby model accuracy can be more effective (para 0124) over simple strategy of data augmentation.
As for (iii)
Sellam discloses training over the set of synthetized data in form of modified and original
sentence pair (para 0004-0005)
Stabler discloses application of irrelevant data augmentation in a synthetic agnostic manner
to natural language input into a training of original utterances (para 0030, 0111, 0121)
Thus, as restructuring incoming text stream via a synthetizing technique enables grouping
and identifying phrases of significant weight and context – e.g. TF-IDF as in Gutta - by which input into a training can be represented by a numerical entity/structure identifiable by a value that in turn would facilitate classification and unbiased evaluation thereof by stages of a neural network, the synthetizing of input to render a training of natural language stream or data set more robust is recognized.
Therefore, based on use of TF-IDF technique to synthetize text input in Gutta, it would have been obvious at the time of the invention for one skill in the art to implement the training of real-world text statement in Gutta system so that the training would be configured to run with result from a initial data synthesis or decomposition into significant context grouping, including
1) replacing one or more words of the real-world text statements with one or more synonyms of the one or more words -- as shown in Jalaluddin; so as to define the new text statements written in natural language that include the one or more synonyms - as set forth in Jalaluddin or word replacement by Sallem;
2) rearranging an original order of one or more words of the real- world text statements, - as set forth in position swapping by Jalaluddin and words order switch by Stabler - so as to define the new text statements written in natural language that include words in a different order as compared to the original order; wherein
3) the training data further comprises the synthetic data such that the neural network is also trained on the synthetic data - as set forth in Jalaluddin or word replacement by Sallem; because
re-arranging text order, generating new text, replacing, removing and augmenting some of the original text data would facilitate formation of agnostic data into a vectorization stage (into a machine learning model as set forth in Gutta) so that this representation becomes invariant to arbitrary or superficial syntax/artifacts, thereby forcing the machine learning to strictly extract meaning regardless of casual and influence of surface irregularities, diverting a machine learning from heavily relying on key trigger words, making the feature space more independent from impact by too particular word choices; that is, rendering judgement by the machine learning more domain-neutral, invariant to real-world structural irregularities, superficial artifacts, minor typos or omitted words, and more focused to a deeper context in which transient noises or inconsistencies have been altered or replaced in that only solid feature of relevancy between the words can be retained in the ML results thus ensuring that the trained (NN) model would invariably generalize seamless results across domains, writing styles or user-specific demographics.
As per claims 14-16, refer to rationale of claims 6, 7, 8 from above.
Response to Arguments
Applicant's arguments filed 5/8/26 have been fully considered but they are not persuasive. Following are the Examiner’s observations in regard thereto.
(A) The Applicant has submitted that the limitations in claims 1 and 9 are not isolated instances of observation, judgement or opinion nor instance of formula, equation or named mathematical calculation; but rather as computer-implemented operations involving neural network, real-world text in NL and synthetic data therefrom, templates and source code generation for domain-specific language, describing much more than characterization of a mental process as alleged in step2A prong 2 of the 101 rejection; when the recitation in claims 1 and 9 include training data associated with target domain-specific language, synthetic data and real-world data, all underlying computer-system functionality directed to training data associated with the above languages or data, all related to the claimed source code generation technology, which cannot be insignificant data gathering or post-solution activity. (Applicant's Remarks pg. 8-10)
The steps of “determining” (using a neural network) , for instance, has been treated as a process of deriving information via use of a computer tool such as that of a numerical technique operative with mathematical core functionality; and this falls into the use of a generic computer by a human in identifying and determining information using mathematical means to going about that data recognizing step, the result of which can be retained inside the human mind, as opposed to having to be carried out via a concrete use of machine and deployed as a real-world product. Applying a mental process via tools in a field of use cannot strip away the abstract idea concept of the applying – see MPEP 2106.05 (b)(f)(h) – notably when the “determining” or “identifying” as recited fail to set forth a transformation to the computer core, or depict a improvement to the deep functionality of this computer field. Further, as recited, the acts of populating a template with data, training with statements in NL, and generating real-world statements as new statements can all be done by a user with use of a computer and mathematical support of tool, especially when “training” as generically recited amounts to no more than applying a mathematical method (via a generic computer) to the Abstract Idea sequence of determining, identifying, populating and generating of synthetized statements in natural language (i.e. something a human can perform via pen/paper). Therefore, the analysis under step2A, prong 2 will stand, as no transformation of significance is evident from the steps as recited, notably when “training” or “neural network” as recited lacks teachings specifying how in details, this operation is implemented so to render it non-conventional and distinctive over analogous concepts; whereas “source code” generating has been interpreted as mere result from “populating a template”, which is a well-understood technique of restructuring data typical to an information recognition context.
(B) The Applicant has submitted that the order in which the “additional elements” have been cited in the construction of claims 1, 9 has not been established by the step 2B of the Office action; nor does the Office action clearly show how this ordered combination is merely well-understood and conventional, when, considered as a whole, the independent claim language now recites a workflow for source code generation in a target domain-specific language via neural network and synthetic data techniques (Applicant's Remarks pg. 11-12)
The step 2b has come to the conclusion that the additional elements, taken individually and as an ordered combination, amount to nothing more than generic computer processing steps (data collection/augmentation, model training, intent classification, and template population) performing their expected functions; That is, the claimed scenario reflecting simply follow a standard pipeline: Collect data [Wingdings font/0xE0]Train neural network [Wingdings font/0xE0] Predict intent [Wingdings font/0xE0]Fill template [Wingdings font/0xE0] Output code [Wingdings font/0xE0] Generate synthetic data
… where construction of the components of this workflow is established as following:
"Determining by a NN..." which merely amounts to conventional use of machine learning for classification/intent detection (MPEP § 2106.05(d)).
"Populating a template..." which merely amounts to generic software operation for code generation (MPEP § 2106.05(d)).
"Training the NN on real-world NL statements..." which merely amounts to conventional, necessary step for supervised/unsupervised machine learning (MPEP § 2106.05(d)).
"Generating synthetic data..." which merely amounts to well-known data augmentation technique in AI training models (MPEP § 2106.05(d)).
Under the constraints governing the ruling under step 2B analysis, the elements forming the ordered combination of the claim have to demonstrate that a transformation of inventive nature has taken place to bring forth a particular improvement to an otherwise conventional functional machine or field of use. Instead, as learned from the ordered combination, the flow pipeline: Collect data [Wingdings font/0xE0]Train neural network [Wingdings font/0xE0] Predict intent [Wingdings font/0xE0]Fill template [Wingdings font/0xE0] Output code [Wingdings font/0xE0] Generate synthetic data clearly describes a conventional flow associated with collecting, training, distinguishing intent, fill template and output synthetic data. The analysis for lack of significant inventiveness per step 2B will stand
(C ) The Applicant has submitted that (Applicant's Remarks pg. 13-14) rejection of claims 3 and 11 (per a § 103 rejection) is traversed because claim 3 (for instance) relies on the limitations now incorporated in claim 1, such that the support by Xu as relied upon by the Office action fails to teach “training the neural network…” and “generating synthetic data …”, which also recited in claim 9.
The argument is deemed largely misplaced as the added limitations in claims 1 and 9 have been met with an adjusted ground of rejection that does not rely on Xu as a stand-alone reference.
(D) The Applicant has submitted that for obviousness of claim 5, Sellam disclosure (para 0003, 0028-0029), on synthetic sentence pairs from sample document text cannot be same (Applicant's Remarks pg. 15) as “synthetic data” generated from “the real-world text statements” written in NL, “the synthetic data defining new statements” written in NL, as recited. Synthetic data as recited in a very high level of generality is deemed falling under the definition of data synthesis, which Sellam has provided in form of synthesis pairs derived from real-world text documents. Applicant allegation to the contrary is deemed largely non-persuasive.
( E ) The Applicant has submitted that, based on individual teaching by Sellam, by Stabler, Jalaluddin, and Nakao compared to Gutta’s operation-to-template-to-populate source code pipeline, the alleged obviousness set forth by the § 103 rejection on basis of what appears to be a combination of non-analogous teachings (Applicant's Remarks pg. 16-17) so to over-reach an obviousness assertion by the office action is deemed a conclusory statement that fails to articulate burden to provide an articulate reason and relevant underpinning for the proposed modification (MPEP § 2143), particularly when the claim itself includes generating “synthetic data” from “the real-world text statements” written in NL, “the synthetic data defining new statements” written in NL and none of the references has been shown as generating new statements in NL from real-world text statements (Applicant's Remarks pg. 18)
The added limitations to claims 1 and 9 have been addressed with adjusted grounds of rejection showing what constitutes “new statement” “synthetic data” and “real-world statements” written in NL, and what synthetizing of unstructured document is all about along with obviousness prongs using a particular set of references. Accordingly, the above argument does not appear to make actual reference to the very prongs of a specific 103 rationale now set forth to meet this newly added limitation.
The allegation that conclusory aspect of 103 rejection (combining Sellam, Stabler, Jalaluddin, and Nakao to Gutta) is basically overreaching and lacking proper rationale is thereby deemed inconclusive, and request by applicant to withdraw all rejections of the claims related thereto cannot be honored.
In all, the claims as submitted with this amendment, will stand rejected.
Conclusion
THIS ACTION IS MADE FINAL. The 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 extension fee 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 Tuan A Vu whose telephone number is (571) 272-3735. The examiner can normally be reached on 8AM-4:30PM/Mon-Fri.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Chat Do can be reached on (571)272-3721.
The fax phone number for the organization where this application or proceeding is assigned is (571) 273-3735 ( for non-official correspondence - please consult Examiner before using) or 571-273-8300 ( for official correspondence) or redirected to customer service at 571-272-3609.
Any inquiry of a general nature or relating to the status of this application should be directed to the TC 2100 Group receptionist: 571-272-2100.
/Tuan A Vu/
Primary Examiner, Art Unit 2193
July 28, 2026
Prosecution Insights