Prosecution Insights
Last updated: November 21, 2026
Application No. 19/146,521

METHODS, APPARATUSES AND COMPUTER-READABLE MEDIUMS WITH PROGRAM CODE FOR GENERATING CONVERSATIONAL TEXT, CLASSIFYING CONVERSATIONAL TEXT, AND TRAINING A MACHINE-LEARNING MODEL

Non-Final OA §103
Filed
Jul 09, 2025
Priority
Feb 24, 2023 — EU 23158554.8 +1 more
Examiner
WON, MICHAEL YOUNG
Art Unit
2443
Tech Center
2400 — Computer Networks
Assignee
Sony Group Corporation
OA Round
1 (Non-Final)
80%
Grant Probability
Favorable
1-2
OA Rounds
1y 7m
Est. Remaining
99%
With Interview

Examiner Intelligence

Grants 80% — above average
80%
Career Allowance Rate
680 granted / 850 resolved
+22.0% vs TC avg
Strong +28% interview lift
Without
With
+28.4%
Interview Lift
resolved cases with interview
Typical timeline
2y 11m
Avg Prosecution
41 currently pending
Career history
898
Total Applications
across all art units

Statute-Specific Performance

§101
8.4%
-31.6% vs TC avg
§103
47.8%
+7.8% vs TC avg
§102
31.3%
-8.7% vs TC avg
§112
8.7%
-31.3% vs TC avg
Black line = Tech Center average estimate • Based on career data from 850 resolved cases

Office Action

§103
Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . DETAILED ACTION 2. This action is in response to the application filed July 9, 2025. 3. Claims 1-20 have been examined and are pending with this action. 4. The Information Disclosure Statement filed July 9, 2025, has been considered. 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. 5. Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Luzhnica et al. (US 11,516,158 B1) in view of Tian et al. (US 2021/0142110 A1). As per claim 1, Luzhnica teaches a method for generating conversational text, the method comprising: training a plurality of machine-learning-based agents to generate conversational text (see Luzhnica, col.6, lines 42-57: “the training data includes training pairs, each training pair including a semantic representation and a corresponding reference sequence in a natural language; receiving human-generated utterances in the form of speech or text; predicting a current dialog state of a natural language dialog between a virtual agent and a user, based on the utterances; generating a semantic representation of a next utterance, based on the current dialog state, the semantic representation including a sequence of characters; and generating a target sequence in a natural language from the semantic representation”; and col.27, lines 2-5: “(3) training the first neural network or at least one neural network in the first group of neural networks with an initial training set, the initial training set comprising a plurality of training set message records”), by: in a first training phase, training the plurality of machine-learning-based agents to output conversational text based on a first objective, the first objective being to develop mutually different linguistic writing styles (see Luzhnica, col.3, lines 32-47: “… receiving input text and an indication of the target author for which the input text is to be stylized; determining, by the at least one computing device, the writing style of the target author; rewriting, by the at least one computing device, the input text in the writing style of the target author using a language model trained with a corpus of text written in a language of the input text and with a corpus of text associated with the target author, the language model being trained by: separating the corpus of text written in the language of the input text into individual sentences and masking a subset of words in each of the individual sentences; and causing the language model to predict, for each word in the masked subset of words, an original value of the word based on at least one preceding word in the corresponding one of the individual sentences; and outputting, by the at least one computing device, the rewritten input text”; and col.122, lines 26-39: “such as selection of initial training set data for the neural network(s) or selection engine to draw from, as noted in FIG. 5), a “friendly” tone setting, #614, using a more informal communication style, and a #616 “formal” tone setting, mimicking classic business-style communications. Readers will understand that this represents an exemplary list, which could alternatively reflect communication styles of different cultures (e.g., Latin American style vs. Asian style vs. American style vs. European style) or additional types of “tones” (e.g., inquisitive style, directive style, driving (results-oriented/commanding) style, supportive style (for audience members struggling with challenges), humorous style, and the like”), in a second training phase, training the plurality of machine-learning-based agents to output conversational text based on a second objective being different from the first objective (see Luzhnica, col.116, lines 40-44: “… e.g., a large amount of TSD may be used to training the system (e.g., in several cycles of training with large data sets) prior to input/upload of prompt data”). Luzhnica does not explicitly teach that the second training phase is based on a second objective being different from the first objective of the first training phase; and wherein the first and second training phase are alternated repeatedly during training of the plurality of machine-learning-based agents. Tian teaches the second training phase is based on a second objective being different from the first objective (see Tian, [0076]: “In some embodiments, the process 300 comprises three distinct phases for developing (e.g., training) the image processing model: (1) a data reduction phase 301, (2) a first training phase 303 for training a deep learning model 310 for local feature extraction, and (3) a second training phase 305 for training support vector machines 314 for object detection and classification”); and wherein the first and second training phase are alternated repeatedly during training of the plurality of machine-learning-based agents (see Tian, [0082]: “Next, during the second training phase 305, the computing system uses the integrated feature vectors for each image to train one or more support vector machines (SVMs) 314. An SVM is a supervised learning model that performs bi-classification of data using regression analysis. For example, each of the support vector machines 314 can be used to predict whether an image includes a particular data type given the feature vector (e.g., the integrated feature vector) extracted from the image (e.g., by the deep learning model 310)”; [0086]: “The process 500 comprises two distinct phases: (1) a feature extraction phase 503 and (2) an object classification phase 505”; and [0090]: “The computing system then sends the integrated feature vector 508 to a plurality of binary classifiers such as SVMs 510a-c to predict classification results 512. In some embodiments, each SVM has previously been trained (e.g., by the second training phase 305 of FIG. 3) to receive a feature vector and to predict whether a particular type of object (e.g., type of clothes) is present on the”). It would have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to modify the system of Luzhnica in view of Tian so that the second training phase is based on a second objective being different from the first objective of the first training phase; and wherein the first and second training phase are alternated repeatedly during training of the plurality of machine-learning-based agents. One would be motivated to do so because it is well-known, routine, and conventional to training ML models in various phases such as in sequential training or staged optimization enables first phase to establish a broader useful set of learned behaviors and the second phase to adapt those behaviors to a targeted task, while reducing inference between the competing objectives. Furthermore, Luzhnica teaches in column 98, lines 18-23, “In aspects, initial draft messages are subjected to further editing, enhancement, or used to generate additional messages, by, e.g., subjecting the messages to a message variation neural network (MVNN), which is trained to review message(s) and to generate variable messages based on such input”, emphasis added. As per claim 2, which depends on claim 1, Luzhnica further teaches wherein the plurality of machine-learning-based agents are trained by the machine-learning-based agents providing generated conversational text to at least one of a) one or more other machine-learning-based agents of the plurality of machine-learning-based agents and b) one or more human agents, and by the machine-learning-based agents receiving feedback from the one or more other machine-learning-based agents or one or more human agents that is based on the generated conversational text and the objective of the respective training phase, with the training being based on the received feedback (see Luzhnica, col.1, lines 45-50: “With the development of artificial intelligence methods and tools, such as neural networks, there has been some progress in the development of fully automated or nearly fully automated messaging. However, successful commercial embodiments of such systems are primarily limited to very short messages, such as chat bots”; and col.19, lines 40-50: “Neural networks also may undergo some amount of training by human system administrators or by operation of system components as described herein. In aspects where a system is specifically trained to “onboard” a new user, adapt to a new type of recipient, deal with new triggers, etc., methods can specifically include performing several rounds of collecting training data before considering a system fully functional or ready for intended/regular use, and the training data is intentionally intended to increase during such steps of the method”). As per claim 3, which depends on claim 2, Luzhnica further teaches wherein the plurality of machine-learning-based agents are trained to carry out a conversation with the one or more other machine-learning-based agents or one or more human agents (see Luzhnica, col.3, lines 48-56: “Neural networks also may undergo some amount of training by human system administrators or by operation of system components as described herein. In aspects where a system is specifically trained to “onboard” a new user, adapt to a new type of recipient, deal with new triggers, etc., methods can specifically include performing several rounds of collecting training data before considering a system fully functional or ready for intended/regular use, and the training data is intentionally intended to increase during such steps of the method”). As per claim 4, which depends on claim 3, Luzhnica further teaches wherein a conversation is carried out between one of a) a single machine-learning-based agent and a single human agent, b) two or more machine-learning-based agents selected in a round-robin manner, and c) two or more machine-learning-based agents and one or more human agents (see Luzhnica, col.3, lines 48-56: “Neural networks also may undergo some amount of training by human system administrators or by operation of system components as described herein. In aspects where a system is specifically trained to “onboard” a new user, adapt to a new type of recipient, deal with new triggers, etc., methods can specifically include performing several rounds of collecting training data before considering a system fully functional or ready for intended/regular use, and the training data is intentionally intended to increase during such steps of the method”). As per claim 5, which depends on claim 2, Luzhnica further teaches wherein a training phase is terminated when a termination condition is met, the termination condition being one of a) a target number of participants in a conversation being carried out between the machine-learning-based agents being reached and b) the feedback reaching a consensus (see Luzhnica, col.96, lines 29-46: “The system typically performs such additions until either a system engine determines that a message should be terminated (e.g., based on size of the message, repetition of message statements/elements, or both), until the EPDNN sends a stop signal to the selection engine, or both. In aspects, a detection in redundancy of statements in a draft message is a trigger that promotes or causes termination of message drafting. Whenever the message drafting is terminated a draft message is ready for further analysis and possibly for presentation to a user, dependent on whether such a message will be modified/enhanced, filtered, etc. In aspects, the system appends a small amount of message content after termination of drafting. For example, the system can determine to cease drafting of paragraphs but append a closing/regards statement, a post scription (p.s.) statement, or the like, based on system message generation settings (editor/engine operation), NN training, etc.”). As per claim 6, which depends on claim 1, Luzhnica further teaches wherein the second objective comprises at least one of a) to provide sufficiently informative conversational text, b) to carry out coherent conversation and c) to mimic a linguistic writing style of another agent (see Luzhnica, col.107, lines 26-36: “Other output that can be generated by a system, whether through transformation or as direct output of the system, include, e.g., emails, long text/SMS messages, social media posts, short blog posts, advertising copy and short advertisements, advertorials, and the like. In aspects, output is in the form of an email. In aspects, output is in the form of a sales-related email or hiring related email. In aspects, TSD, prompts, or both, reflected the intended output of the system (e.g., in aspects, TSD, prompts, or both, mostly, generally, or essentially entirely relate to sales emails)”). As per claim 7, which depends on claim 1, Luzhnica further teaches wherein each of the plurality of machine-learning-based agents comprises at least one machine-learning model being trained to generate the conversational text (see Luzhnica, col.20, lines 5-13: “a first neural network/EPDNN will typically be significantly trained on one or more natural language models, which can be significantly larger than training sets used in other methods/systems of the invention (e.g., in aspects a system could utilize a powerful language modeling neural network platform, such as GPT-3, which was trained on Common Crawl (commoncrawl.org), believed to be one of the largest corpus ever used to train a neural network system”). As per claim 8, which depends on claim 7, Luzhnica further teaches wherein training the machine-learning-based agents comprises training the at least one machine-learning model, using reinforcement learning, to generate the conversational text, with a reward of the reinforcement learning being based on feedback of one or more human agents (see Luzhnica, col.101, lines 43-56: “In aspects, human reviewers input instructions to the system to provide reward/penalty feedback to the EDPDNN, e.g., based on a rating of 1-10, 1-5, 0-1, or other scale/grade system (A-D and F), or other form of feedback, etc. submitted by reviewers, based on a comparative review of initial/earlier prompts and messages generated, thereby causing a change in the performance of the EPDNN, and thereby further changing message generation outcomes in a manner that typically, mostly, generally only, or significantly enhances truthfulness, message efficacy, etc., either instantly (in the next iteration/application), over time, or both”). As per claim 9, which depends on claim 8, Luzhnica further teaches wherein training the machine-learning-based agents comprises continuing training the at least one machine-learning model, using supervised learning, to generate the conversational text, based on a plurality of utterances of the one or more human agents (see Luzhnica, col.89, lines 48-63: “Accordingly, in aspects, systems can, e.g., comprise an API where an administrator can interact with/modify and obtain information from or concerning system components and their operation. Administrators also can be trainers, individuals that provide supervised learning to NNs, such as in pre-training steps, adjustment steps, aspects where humans review or input prompts based on PIMs, and the like. In aspects, systems/methods comprise automatically facilitating PIM review and prompt data collection from human trainers, at least in some iterations of the method/uses of the system (e.g., in the first 10, 20, 50, 100 iterations, etc., or until some indicator of system maturity/convergence has been determined)”; and col.98, lines 18-23: “In aspects, initial draft messages are subjected to further editing, enhancement, or used to generate additional messages, by, e.g., subjecting the messages to a message variation neural network (MVNN), which is trained to review message(s) and to generate variable messages based on such input”). As per claim 10, which depends on claim 8, Luzhnica further teaches wherein training the machine-learning-based agents comprises training a reward determination machine-learning model to output a reward based on feedback of one or more human agents and continuing training the at least one machine-learning model, using reinforcement learning, to generate the conversational text, with a reward of the reinforcement learning being based on the reward output by the reward determination machine-learning model (see Luzhnica, col.101, lines 43-56: “In aspects, human reviewers input instructions to the system to provide reward/penalty feedback to the EDPDNN, e.g., based on a rating of 1-10, 1-5, 0-1, or other scale/grade system (A-D and F), or other form of feedback, etc. submitted by reviewers, based on a comparative review of initial/earlier prompts and messages generated, thereby causing a change in the performance of the EPDNN, and thereby further changing message generation outcomes in a manner that typically, mostly, generally only, or significantly enhances truthfulness, message efficacy, etc., either instantly (in the next iteration/application), over time, or both”). As per claim 11, which depends on claim 7, Luzhnica further teaches wherein the at least one machine-learning model is trained to generate the conversational text based on a conversational topic and based on prior conversational text provided by a machine-learning-based agent or a human agent (see Luzhnica, col.98, lines 18-23: “In aspects, initial draft messages are subjected to further editing, enhancement, or used to generate additional messages, by, e.g., subjecting the messages to a message variation neural network (MVNN), which is trained to review message(s) and to generate variable messages based on such input”). As per claim 12, which depends on claim 7, Luzhnica further teaches wherein the at least one machine-learning model is trained to provide feedback to one or more other machine-learning-based agents (see Luzhnica, col.30, line 60-col.31, line 2: “Also or alternatively, methods can include providing other feedback to the system, e.g., providing the neural network(s) of the system with feedback, such as a performance grade, which impacts the behavior of the system. Such feedback can come from human supervision, application of a different type of neural network (e.g., a continent grading neural network), or both. The invention also provides systems that can carry out such methods, e.g., by comprising at least one or two neural networks that perform such functions”). As per claim 13, which depends on claim 12, Luzhnica further teaches wherein the at least one machine-learning model is trained to provide the feedback based on conversational text provided by the respective other machine-learning-based agent and based on the objective of the respective training phase (see Luzhnica, col.30, line 60-col.31, line 2: “Also or alternatively, methods can include providing other feedback to the system, e.g., providing the neural network(s) of the system with feedback, such as a performance grade, which impacts the behavior of the system. Such feedback can come from human supervision, application of a different type of neural network (e.g., a continent grading neural network), or both. The invention also provides systems that can carry out such methods, e.g., by comprising at least one or two neural networks that perform such functions”). As per claim 14, which depends on claim 1, Luzhnica teaches further comprising providing the generated conversational text as training data for training a machine-learning model, the training data representing a plurality of different linguistic writing styles, or providing the generated conversational text as entertainment content (see Luzhnica, col.85, lines 34-45: “they impart a level of communication that is not capable with, e.g., a chat or typical SMS/text or instant message format communication, while also not being of such an extensive size that they impart too many different contexts, statements, and the like for systems to be able to efficiently process, effectively process, or both. E.g., in aspects, messages generated herein are not in the form in, and typically lack the characteristics of, e.g., a published technical article, a longer journalistic work, a mid-to-long blog posting, a book, a long poetic piece, a short story, an essay, or other types of similar works of such length, complexity, or both”). As per claim 15, which depends on claim 1, Luzhnica further teaches wherein each agent comprises a machine-learning model being trained to provide feedback to one or more other machine-learning-based agents, the machine-learning model being trained as classifier for classifying text according to the respective second objective, the method comprising providing the machine-learning model (see Luzhnica, col.62, line 62-col.63, line 3: “Neural networks typically execute numerous and interrelated algorithms at different levels (e.g., a neuron/node can execute an algorithm; a NN layer can perform a function/algorithm, such as classification, clustering, and the like; and interconnected nodes/neurons, such as interconnected neurons in a dense NN or a densely connected NN, which systems can include, can be considered to collectively perform complex algorithms/functions including sub-algorithms/functions)”). As per claim 16, Luzhnica further teaches non-transitory, computer-readable medium comprising training data for training a machine-learning model, the training data being generated according to the method of claim 15 (see Luzhnica, col.111, lines 10-17: “Different programming techniques can be employed e.g., procedural or object-oriented approaches in system(s)/method(s). Any particular routine can execute on a single processor, multiple processors, or even in multiple devices, as suitable. Data can be stored in a single storage medium or distributed through multiple storage mediums and may reside in a single database or multiple databases (or other data storage techniques)”). As per claim 17, Luzhnica further teaches method for classifying conversational text, the method comprising: inputting the conversational text into one or more machine-learning models that are trained according to the method of claim 16 (see Luzhnica, col.98, lines 18-23: “In aspects, initial draft messages are subjected to further editing, enhancement, or used to generate additional messages, by, e.g., subjecting the messages to a message variation neural network (MVNN), which is trained to review message(s) and to generate variable messages based on such input”); and determining a classification of the conversational text based on an output of the one or more machine-learning models (see Luzhnica, col.62, line 62-col.63, line 3: “Neural networks typically execute numerous and interrelated algorithms at different levels (e.g., a neuron/node can execute an algorithm; a NN layer can perform a function/algorithm, such as classification, clustering, and the like; and interconnected nodes/neurons, such as interconnected neurons in a dense NN or a densely connected NN, which systems can include, can be considered to collectively perform complex algorithms/functions including sub-algorithms/functions)”). As per claim 18, Luzhnica further teaches the method for training a machine-learning model, the method comprising: obtaining training data, the training data comprising a plurality of samples of generated conversational text representing a plurality of different linguistic writing styles ,the training data being generated according to the method of claim 15 (see Luzhnica, col.127, lines 9-19: “Feeding prompt inputs to the model and using a sampling strategy, the completion engine, #1620, completes message(s) and returns the message or set of messages back, #1622, to the content generator service, #1614. The completion engine, #1620, contains, as an example, standard sampling settings such as temperature, top k, presence/frequency penalty, maximum tokens to sample, beam search size, etc., such as are known in connection with sampling engine/neural network natural language processing system controls”); and using the training data to train a machine-learning model (see Luzhnica, col.104, lines 37-38: “Additional training of NN(s) also occurs in systems/methods by provision of additional training data”). As per claim 19, Luzhnica further teaches a non-transitory, computer-readable medium comprising a program code that, when the program code is executed on a processor, a computer, or a programmable hardware component, causes the processor, computer, or programmable hardware component to perform the method of claim 1, or the method of claim 17, or the method of claim 18 (see Luzhnica, col.15, line 60-col.16, line 1: “Uncontradicted, a “system” of the invention comprises an electronic device/collection of devices that is composed of interrelated/interacting components/devices (including, e.g., separate networked device(s)/component(s)), including (1) memory component(s) comprising computer-readable media including preprogrammed instructions for carrying out Function(s) (“engines”) and (2) a computer/processing unit capable of reading/executing such instructions causing Function(s) to be performed.”). As per claim 20, Luzhnica further teaches an apparatus comprising memory circuitry, machine-readable instructions, and processing circuitry to execute the machine-readable instructions to perform the method of claim 1, or the method of claim 17, or the method of claim 18 (see Luzhnica, col.15, line 60-col.16, line 1: “Uncontradicted, a “system” of the invention comprises an electronic device/collection of devices that is composed of interrelated/interacting components/devices (including, e.g., separate networked device(s)/component(s)), including (1) memory component(s) comprising computer-readable media including preprogrammed instructions for carrying out Function(s) (“engines”) and (2) a computer/processing unit capable of reading/executing such instructions causing Function(s) to be performed.”). Conclusion 6. For the reasons above, claims 1-20 have been rejected and remain pending. 7. Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHAEL Y WON whose telephone number is (571)272-3993. The examiner can normally be reached on Wk.1: M-F: 8-5 PST & Wk.2: M-Th: 8-7 PST. 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, Nicholas R Taylor can be reached on 571-272-3889. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /Michael Won/Primary Examiner, Art Unit 2443
Read full office action

Prosecution Timeline

Jul 09, 2025
Application Filed
Sep 09, 2026
Non-Final Rejection mailed — §103
Nov 02, 2026
Interview Requested
Nov 12, 2026
Applicant Interview (Telephonic)
Nov 12, 2026
Examiner Interview Summary

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12786053
SYSTEMS AND METHODS FOR CREATING VIRTUAL DEVICES
2y 1m to grant Granted Nov 03, 2026
Patent 12775733
ANTI-SCRAPING METHOD AND SYSTEM FOR PROTECTION OF TEXT-BASED DATA
2y 7m to grant Granted Oct 27, 2026
Patent 12775722
DATA COMMUNICATION METHOD, MEMORY POOLING SWITCH DEVICE, CLOUD COMPUTING SYSTEM, AND STORAGE MEDIUM
2y 2m to grant Granted Oct 27, 2026
Patent 12764109
Trusted Identification of Enrolling Users Based on Images and Unique Identifiers Associated With Sponsoring Users
2y 0m to grant Granted Oct 13, 2026
Patent 12755016
RANSOMWARE DETECTION IN MEMORY OF A DATA PROCESSING UNIT USING MACHINE LEARNING DETECTION MODELS
2y 1m to grant Granted Oct 06, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

1-2
Expected OA Rounds
80%
Grant Probability
99%
With Interview (+28.4%)
2y 11m (~1y 7m remaining)
Median Time to Grant
Low
PTA Risk
Based on 850 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month