DETAILED ACTION
This examination is in response to the communication filed on 06/03/2026. Claims 1-20 are currently pending, where claims 1, 8 and 17 are independent.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment/arguments
Applicant’s arguments with respect to the rejection of amended claims 1-16 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Applicant's arguments with respect to the rejection of claims 17-20 have been fully considered but they are not persuasive.
Applicant argues that “Golany describes training/evaluating agent models against reference answers/attributions, not extracting feature results from a platform about the synthetic data and not diagnosing deficiencies by comparing to pre-specified role parameters” and “Wang’s doctor/patient roleplay does not apply the missing role-parameter comparison or targeted deficiency loop” (Response, p. 12) The Examiner respectfully disagrees.
Golany teaches evaluating the agent models, i.e., the prompted LLMs along two dimensions. One being the quality of the generated responses. One skilled in the art would readily appreciate that the quality of the generated response includes whether the model performed as prompted. This in combination with Wang’s teaching of prompting with role characteristics teaches that the quality of the generated response would include whether the agent models response in the prompted role. Therefore, under a broadest reasonable interpretation the fine-tuning of persona-prompted LLMs using pre-defined standards reads on the comparing and refining limitations of claim 17.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 112(d):
(d) REFERENCE IN DEPENDENT FORMS.—Subject to subsection (e), a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
The following is a quotation of pre-AIA 35 U.S.C. 112, fourth paragraph:
Subject to the following paragraph [i.e., the fifth paragraph of pre-AIA 35 U.S.C. 112], a claim in dependent form shall contain a reference to a claim previously set forth and then specify a further limitation of the subject matter claimed. A claim in dependent form shall be construed to incorporate by reference all the limitations of the claim to which it refers.
Claims 15 and 16 are rejected under 35 U.S.C. 112(d) or pre-AIA 35 U.S.C. 112, 4th paragraph, as being of improper dependent form for failing to further limit the subject matter of the claim upon which it depends, or for failing to include all the limitations of the claim upon which it depends. Claim 15 recites that the same language incorporated into independent claim 8 from which it depends and claim 16 improper broaden the limitation incorporated into independent claim 8 regarding the characteristics. It appears that claim 15, like claim 6 should have been canceled, and claim 16, like claim 7 should have been amended to recite “at least three” rather than “at least one”.
Applicant may cancel the claim(s), amend the claim(s) to place the claim(s) in proper dependent form, rewrite the claim(s) in independent form, or present a sufficient showing that the dependent claim(s) complies with the statutory requirements.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1-4, 7-11 and 15-16 are rejected under 35 U.S.C. 103 as being unpatentable over Golany et al. “Efficient Data Generation for Source-grounded Information-seeking Dialogs: A Use Case for Meeting Transcripts” Google Research arXiv:2405.01121v1 [cs.CL:] May 2, 2024 (herein “Golany”) in view of Gersoa et al. “Can AI Serve as a Substitute for Human Subjects in Software Engineering Research?” arXiv:2311.1081v1 [csSE] 18 Nov 2023 pp. 1-25 (herein “Gersoa”).
Regarding claims 1 and 8, Golany teaches a method for training a neural network (Page 4, Section 5 teaches “This test set was created in order to assess whether training models on the MISeD data leads to better performance…”) and a system for generating synthetic data for training and/or testing of neural network (Page 4, Section 5 teaches “This test set was created in order to assess whether training models on the MISeD data leads to better performance…”) comprising: at least one Large Language Model (LLM) module, a data storing unit, and a bot builder service module, implemented by at least one processor (page 1, Figure 1 “LLM” and page 2, 1st column teaches “In summary, our main contributions are as follows: (1) presenting an LLM-based data generation methodology for information-seeking dialogs; (2) creating the MISeD dataset” and Fig. 1, full generated dialog transcript requires a storage unit), configured to perform the method comprising:
generating synthetic data by defining roles for a plurality of speakers (Page 1, Abstract teaches “Instead of the labor-intensive Wizard-of-Oz (WOZ) method, where two annotators generate a dialog from scratch, role-playing agent and user, we use LLM generation to simulate the two roles.” The agent and user roles are interpreted as a plurality of speakers),
inputting the roles to at least one Large Language Model (LLM) implemented by at least one first processor (Page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” See also, Fig. 1, “Query Prompt”),
requesting the at least one LLM to generate a first statement based on the role of a first speaker of the plurality of speakers (Page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” See also, Fig. 1, “Query Prompt”),
instructing the at least one LLM to generate a second statement based on the role of a second speaker of the plurality of speakers that is responsive to the first statement (Page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” See also, Fig. 1, “Response Prompt”),
storing a dialog between the first speaker and the second speaker comprising the first and second statements (Fig. 1, transcript of full generated dialog),
iterating (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”) the requesting, instructing and storing such that
the first statement is responsive to the second statement of a preceding iteration of the requesting (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…” ),
the second statement is responsive to the first statement of a current iteration of the requesting instructing and storing (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…” ),
the storing comprises adding the first and second statements of a current iteration to the dialog such that the dialog comprises the first and second statements of each previous iteration of the requesting, instructing and storing ( Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”), and
each instance of the requesting and instructing comprises providing the at least one LLM with the dialog of a preceding iteration of the storing (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”),
ceasing said iterating in response to a termination condition to obtain the stored dialog in a final iteration of the iterating, wherein the stored dialog in the final iteration is the synthetic data (Page 1, 2nd column Figure 1 description teaches “… Iterating this automatic process yields a full dialog, which is then validated by annotators, who further augment it with response attributions.” The iteration process necessarily requires a termination condition in order to trigger the validation process, therefore, the iteration process is inherently ceased upon detection of the termination condition); and
training a neural network, implemented by at least one second processor, based on the synthetic data (Page 7, Section 7.2 teaches “We train the finetuned agent models using the MISeD training set…” ),
wherein at least one of the roles of the first speaker or the second speaker comprises characteristics of the first speaker or the second speaker (page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” the user or agent role is interpreted as a characteristic of the speaker, i.e., either user or agent.), and
wherein each instance of the requesting and instructing further comprises providing the at least one LLM with the characteristics (page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” the user or agent role is interpreted as a characteristic of the speaker, i.e., either user or agent. In addition, Fig. 1 illustrates that at each turn, the LLM is fed the query prompt or response prompt. These in combination with the fact that the query prompt and response prompt include role/characteristics of speaker suggests that each instance includes providing the LLM with the characteristics).
Golany fails to disclose that the characteristics comprise at least two of: name, gender, age, address or occupation.
Gerosa teaches a method of providing AI-generated synthetic text as an alternative source of qualitative data in the form of persona-based prompting for interviews, multi-persona dialogue for focus groups and mega-persona responses for surveys. Section 2.1 of Gerosa teaches “By constructing detailed persona and corresponding prompts, researchers can guide the AI to produce text the provides insights into how different groups might think, feel, or react to various stimuli or situations…can engineer the prompt by including extensive details about the persona…may specify the persona’s age, gender, geographical location, occupation, preferred programming language, and additional contextual elements” In addition, Fig. 1 illustrates an example prompt including the characteristics of Name (“John”), age (“45 years old”), gender (“man”), and address (“Arizona, United States”). Therefore, Gerosa teaches the prompted characteristics comprise at least two of: name, gender, age, address or occupation.
Golany differs from the claimed invention, as defined by claims 1 and 8, in the Golany fails to disclose that the role characteristics for the LLM comprise at least two of: name, gender, age, address or occupation. Persona-based prompting for LLMs to generate synthetic data which includes multiple characteristics including two or more of name, gender, age, address or occupation are known in the art as evidenced by Gerosa. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified the system of Golany to include the persona-based prompting for the LLMs to roleplay as taught by Gerosa as it merely constitutes the combination of known processes to achieve the predictable result of generating synthetic data which replicates human responses and behaviors in order to “offer scalable and efficient means of data generation” (see Gerosa, Abstract).
Regarding claim 2, the combination of Golany and Gerosa teaches all of the elements of claim 1 (see detailed mapping above). In addition, Golany further teaches the training comprises performing a first learning by the neural network based on other data and performing a second learning by the neural network based on the synthetic data to refine the neural network (Page 1, Abstract teaches “Models finetuned with MISeD demonstrate superior performance…” and page 7, section 7.1 teaches “Finetuned Encoder-Decoder…We finetuned the open-source LongT5 XL…on the MISeD training set, using a context length of 16 thousand tokens” by definition “finetuned” is performed on a pre-trained model).
Regarding claim 3, the combination of Golany and Gerosa teaches all of the elements of claim 2 (see detailed mapping above). In addition, Golany further teaches the other data is real data based on at least one real dialog (page 7, section 7.1 teaches “Finetuned Encoder-Decoder…We finetuned the open-source LongT5 XL…on the MISeD training set, using a context length of 16 thousand tokens” As evidenced by Guo et al. “LongT5:Efficient Text-To-Text Transformer for Long Sequences” Findings of the Association for computational Linguistics: NAACL 2022, pages 724-736 July 10-15, 2022, the LongT5 XL was trained on the MediaSum dataset which is large-scale media interview, i.e., dialog, dataset).
Regarding claims 4 and 10, the combination of Golany and Gerosa teaches all of the elements of claims 1 and 9 (see detailed mapping above). In addition, Golany further teaches the synthetic data is text data (Page 3, 1st column description of Figure 2 teaches “The agent receives the source text (meeting transcript), dialog history…” Thus, the meeting transcript which corresponds to LLM generated dialog transcript is text).
Regarding claim 7, the combination of Golany and Gerosa teaches all of the elements of claim 1 (see detailed element mapping above). In addition, Gerosa further teaches wherein the characteristics comprise at least three of: name, gender, age, address or occupation (Fig. 1 illustrates an example prompt including the characteristics of Name (“John”), age (“45 years old”), gender (“man”), and address (“Arizona, United States”).
Golany differs from the claimed invention, as defined by claim 7, in the Golany fails to disclose that the role characteristics for the LLM comprise at least three of: name, gender, age, address or occupation. Persona-based prompting for LLMs to generate synthetic data which includes multiple characteristics including two or more of name, gender, age, address or occupation are known in the art as evidenced by Gerosa. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified the system of Golany to include the persona-based prompting for the LLMs to roleplay as taught by Gerosa as it merely constitutes the combination of known processes to achieve the predictable result of generating synthetic data which replicates human responses and behaviors in order to “offer scalable and efficient means of data generation” (see Gerosa, Abstract).
Regarding claim 9, the combination of Golany and Gerosa teaches all of the elements of claim 8 (see detailed mapping above). In addition, Golany further teaches the at least one LLM module provides each instance of the first and second statement as text data (Page 1, Figure 1, query 1 and response 1 and Page 3, 1st column description of Figure 2 teaches “The agent receives the source text (meeting transcript), dialog history…” Thus, the meeting transcript which corresponds to LLM generated dialog transcript is text).
Regarding claim 11, the combination of Golany and Gerosa teaches all of the elements of claim 9 (see detailed mapping above). In addition, Golany further teaches the dialog is modeled for implementation on a dialog channel that is a text-based platform (Page 3, 1st column description of Figure 2 teaches “The agent receives the source text (meeting transcript), dialog history…” Thus, the meeting transcript which corresponds to LLM generated dialog transcript is text).
Regarding claim 15, the combination of Golany and Gerosa teaches all of the elements of claim 8 (see detailed element mapping above). In addition, Golany further teaches at least one of the roles of the first speaker or the second speaker comprises characteristics of the first speaker or the second speaker (page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” the user or agent role is interpreted as a characteristic of the speaker, i.e., either user or agent.).
Regarding claim 16, the combination of Golany and Gerosa teaches all of the elements of claim 15 (see detailed element mapping above). In addition, Gerosa further teaches wherein the characteristics comprise at least one of: name, gender, age, address or occupation (Fig. 1 illustrates an example prompt including the characteristics of Name (“John”), age (“45 years old”), gender (“man”), and address (“Arizona, United States”).
Golany differs from the claimed invention, as defined by claim 7, in the Golany fails to disclose that the role characteristics for the LLM comprise at least three of: name, gender, age, address or occupation. Persona-based prompting for LLMs to generate synthetic data which includes multiple characteristics including one or more of name, gender, age, address or occupation are known in the art as evidenced by Gerosa. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified the system of Golany to include the persona-based prompting for the LLMs to roleplay as taught by Gerosa as it merely constitutes the combination of known processes to achieve the predictable result of generating synthetic data which replicates human responses and behaviors in order to “offer scalable and efficient means of data generation” (see Gerosa, Abstract).
Claims 5, 12 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Golany and Gerosa as applied to claim 1 above, and further in view of Lee et al. “Exploring The Viability of Synthetic Audio Data for Audio-Based Dialogue Tracking” arXiv:2312.01841v1 [cs.SD] December 4, 2023 (herein “Lee”).
Regarding claims 5, the combination of Golany and Gerosa teaches all of the elements of claim 1 (see detailed mapping above). However, the combination of Golany and Gerosa fails to disclose that the synthetic data is audio data.
Lee teaches a method and system for extending text-based dialogue state tracking (DST) to the audio domain. More specifically, Lee teaches by leveraging text-to-speech (TTS) models it is possible to generate a diverse and customizable synthetic audio dialogue dataset that covers a wide range of scenarios and contexts (see Lee page 1, 2nd column, 2nd to last paragraph). More specifically, Lee teaches in Section 4.1 that “SynthWOZ is a comprehensive multi-turn synthetic audio dataset…utilizing the text transcripts required for audio generations, wherein were obtained from MultiWOZ 2.1…a widely acknowledged benchmark resource for text-based DST.” Accordingly, Lee teaches generation synthetic audio datasets for DST from text-based transcripts.
The combination of Golany and Gerosa differs from the claimed invention, as defined in claim 5, in that the combination fails to disclose the synthetic data is in the audio domain. Generation synthetic audio dialogue data from text-based dialogue is known in the art as evidenced by Lee. Therefore, it would have been obvious to one having ordinary skill in the art to have modified the synthetic dialog generation method/system taught by the combination of Golany and Gerosa to include utilizing text-to-speech (TTS) models to generate a diverse and customizable synthetic audio dialogue dataset that covers a wide range of scenarios and contexts (Lee, page 1, 2nd column).
Regarding claim 12, the combination of Golany and Gerosa teaches all of the elements of claim 9 (see detailed mapping above). However, the combination of Golany and Gerosa fails to disclose a Text-to-Speech (TTS) Service module, implemented by the at least one processor, wherein the TTS Service module is configured to convert the text data to audio data; and a voice cloning service module, implemented by the at least one processor, wherein the voice cloning service module is configured to clone at least one voice and convert the audio data into cloned audio data in the at least one voice such that the synthetic data is stored as the cloned audio data.
Lee teaches a Text-to-Speech (TTS) Service module, implemented by the at least one processor, wherein the TTS Service module is configured to convert the text data to audio data (page 1, 2nd column teaches “By leveraging text-to-speech (TTS) models…to generate a diverse and customizable synthetic audio dialogue dataset”); and
a voice cloning service module, implemented by the at least one processor, wherein the voice cloning service module is configured to clone at least one voice and convert the audio data into cloned audio data in the at least one voice such that the synthetic data is stored as the cloned audio data (page 3, 1st column teaches “…we employed multiple speaker voices during the synthesis process by passing speaker labels to a fixed lookup table ).
The combination of Golany and Gerosa differs from the claimed invention, as defined in claim 5, in that the combination fails to disclose the synthetic data is in the audio domain. Generation synthetic audio dialogue data from text-based dialogue is known in the art as evidenced by Lee. Therefore, it would have been obvious to one having ordinary skill in the art to have modified the synthetic dialog generation method/system taught by the combination of Golany and Gerosa to include utilizing text-to-speech (TTS) models to generate a diverse and customizable synthetic audio dialogue dataset that covers a wide range of scenarios and contexts (Lee, page 1, 2nd column).
Regarding claim 13, the combination of Golany, Gerosa and Lee teaches all of the elements of claim 12 (see detailed mapping above). In addition, Lee further teaches the dialog is modeled for implementation on a dialog channel that is a voice-based platform (page 1, section 1 teaches “Audio-based DST offers several advantages in enhancing user experience…” ).
The combination of Golany and Gerosa differs from the claimed invention, as defined in claim 13, in that the combination fails to disclose the synthetic data is modeled for a voice-based platform. Generating synthetic audio dialogue data for voice-based platforms is known in the art as evidenced by Lee. Therefore, it would have been obvious to one having ordinary skill in the art to have modified the synthetic dialog generation method/system taught by the combination of Golany and Gerosa to include utilizing text-to-speech (TTS) models to generate a diverse and customizable synthetic audio dialogue dataset that covers a wide range of scenarios and contexts (Lee, page 1, 2nd column).
Claim 14 is rejected under 35 U.S.C. 103 as being unpatentable over the combination of Golany, Gerosa and Lee as applied to claim 12 above, and further in view of Lee et al (US 2024/0143916 A1 herein “Lee ‘916”).
Regarding claim 14, the combination of Golany, Gerosa and Lee teaches all of the elements of claim 12 (see detailed element mapping above). However, the combination fails to explicitly disclose the dialog is modeled for implementation on a dialog channel that is both a textual-based platform and a voice-based platform.
Lee ‘916 teaches an electronic apparatus that performs a voice assistant functions using a language model, wherein the voice assistant provides a response to user interactions through text or voice. Thus, Lee teaches an LLM system which includes both a textual-based and voice-based platform (see ¶[0003] of Lee ‘916).
The combination of Golany, Gerosa and Lee differs from the claimed invention, as defined in claim 14, in that combination fails to disclose the synthetic data is modeled for a system that is both a textual-based platform and a voice-based platform. LLM based assistants which are implemented on both textual-based and voice-based platforms are known in the art as evidenced by Lee ‘916. Therefore, it would have been obvious to one having ordinary skill in the art to have utilized the synthetic audio and textual dialog generated by the system taught by the combination of Golany, Gerosa and Lee to train an LLM provided in a textual and voice based platform as taught by Lee ‘916 as it merely constitutes the combination of known systems to achieve the predictable result of providing synthetic dialogue data for both text-based and audio-based systems.
Claims 17-20 are rejected under 35 U.S.C. 103 as being unpatentable over Golany and further in view of Wang et al. “NoteChat: A Dataset of Synthetic Patient-Physician Conversations Conditioned on Clinical Notes” arXiv:2310.15959v1 [cs.CL] Oct 24 2023 (herein “Wang”).
Regarding claim 17, Golany teaches a method for refining a conversation analytics platform (under a BRI conversation analytic platform is interpreted as any platform which process conversational data to generate an output or response; page 3, section 3 teaches “our goal is to generate datasets for agent models in source-grounded information-seeking dialogs” the source-grounded information-seeking dialogs system is interpreted as a conversation analytics platform in as much as it is designed to process/analyze conversational data to provide a response/output ) comprising:
generating synthetic data by defining roles for a plurality of speakers (Page 1, Abstract teaches “Instead of the labor-intensive Wizard-of-Oz (WOZ) method, where two annotators generate a dialog from scratch, role-playing agent and user, we use LLM generation to simulate the two roles.” The agent and user roles are interpreted as a plurality of speakers),
inputting the roles to at least one Large Language Model (LLM), implemented by at least one first processor (Page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” See also, Fig. 1, “Query Prompt”),
requesting the at least one LLM to generate a first statement based on the role of a first speaker of the plurality of speakers (Page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” See also, Fig. 1, “Query Prompt”),
instructing the at least one LLM to generate a second statement based on the role of a second speaker of the plurality of speakers that is responsive to the first statement (Page 1, 2nd column teaches “We utilize separate prompts to guide the LLM’s generation of both the user queries and the agent responses” See also, Fig. 1, “Response Prompt”),
storing a dialog between the first speaker and the second speaker comprising the first and second statements (Fig. 1, transcript of full generated dialog),
iterating (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”) the requesting, instructing and storing such that
the first statement is responsive to the second statement of a preceding iteration of the requesting (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”),
the second statement is responsive to the first statement of a current iteration of the requesting instructing and storing (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”),
the storing comprises adding the first and second statements of a current iteration to the dialog such that the dialog comprises the first and second statements of each previous iteration of the requesting, instructing and storing (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”), and
each instance of the requesting and instructing comprises providing the at least one LLM with the dialog of a preceding iteration of the storing, ceasing said iterating in response to a termination condition to obtain the stored dialog in a final iteration of the iterating, wherein the stored dialog in the final iteration is the synthetic data (Page 1, 2nd column Figure 1 description teaches “Iterative dialog generation flow. In each turn, a query prompt guides the LLM to generate a user query given the transcript, the accumulated dialog history and a query template. Then, a response prompt, accompanied by the full context so far, generates the agent response. Iterating this automatic process yields a full dialog…”);
inputting the synthetic data to the conversation analytics platform, which is implemented by at least one second processor (page 7, section 7.2 teaches “We train the finetuned agent models using the MISeD training set…” as discussed above and noted in Section 3 of Golany, the agent model is part of the source-grounded information-seeking dialog system, i.e., the conversation analytics platform );
receiving feature results characterizing the synthetic data from the conversation analytics platform (page 7, section 7.2 teaches “We train the finetuned agent models using the MISeD training set…” training the finetuned agent models );
comparing the feature results to initial parameters (under a broadest reasonable interpretation this limitation is interpreted as an automated validation of the conversation analytics platform based on the synthetic training dataset; page 6, section 6 teaches “We evaluate the agent models along two dimensions; the quality of the generated responses (§6.1), and the accuracy of the provided attributions (§6.2), through both automatic and human evaluations” response quality is interpreted as including adherence to prompted task);
refining the at least one model portion of the conversation analytics platform in response to determining that the at least one model portion of the conversation analytics platform is deficient (page 7, sections 7.1 teaches “we finetuned the open-source5 LongT5 XL (3 billion parameters) on the MISeD training set…we finetuned the Gemini Pro Model7 on the MISeD training set…”).
Golang fails to explicitly disclose that the comparison parameters utilized for the validation of the conversation analytics platform includes the roles for the plurality of speakers.
Wang teaches a system and method for generating a synthetic dialog dataset using LLM chatbots that includes, inter alia, a Roleplay module where two ChatGPT agents take on the roles of doctor and patient, respectively. See Wang Figure 1 and page 2, 2nd column. In addition, Wang further teaches that the extrinsic evaluation parameters/training labels includes the role, i.e., physician/doctor, of the speakers. (See section 3.2 of Wang).
Golany differs from the claimed invention, as defined by claim 17, in the Golany fails to explicitly disclose that validation features includes the roles for the plurality of speakers. Utilizing speaker role as a training label/validation features is known as evidenced by Wang. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified the system of Golany to include validation of the speaker role features as taught by Wang as it merely constitutes the combination of known processes to achieve the predictable result of generating synthetic medical datasets without violating privacy regulations associated with real Docter-patient dialog data.
Regarding claim 18, the combination of Golany and Wang teaches all of the elements of claim 17 (see detailed element mapping above). In addition, Golany further teaches [wherein] the synthetic data is first synthetic data (page 3, section 4 teaches “Its first stage automatically generates a dialog using LLMs (§4.1), simulating the typical WOZ process…”) and
the method further comprises: generating second synthetic data (page 4, section 4.2 teaches “following the automatic generation of the dialog, we present it to trained annotators who asses the generated query and response, and identify corresponding attributions within the source text” the addition of the attributions is interpreted as second synthetic data),
wherein the refining comprises refining the at least one model portion of the conversation analytics platform with the second synthetic data (page 7, sections 7.1 teaches “we finetuned the open-source5 LongT5 XL (3 billion parameters) on the MISeD training set…we finetuned the Gemini Pro Model7 on the MISeD training set…”)).
Regarding claim 19, the combination of Golany and Wang teaches all of the elements of claim 18 (see detailed element mapping above). In addition, Golany further teaches the second synthetic data is provided in a model training dataset and wherein the refining comprises training the at least one model portion with the model training dataset (page 3, Figure 3 teaches the dialog history includes the Attribution information; and page 7, sections 7.1 teaches “we finetuned the open-source5 LongT5 XL (3 billion parameters) on the MISeD training set…we finetuned the Gemini Pro Model7 on the MISeD training set…”).
Regarding claim 20, the combination of Golany and Wang teaches all of the elements of claim 19 (see detailed element mapping above). In addition, Golany further teaches the model training dataset comprises the first synthetic data (page. 3, Figure 3 teaches that that Dialog history include the User and Agent responses, i.e., the first synthetic data).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PENNY L CAUDLE whose telephone number is (703)756-1432. The examiner can normally be reached M-Th 8:00 am to 5:00 pm eastern.
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, Daniel Washburn can be reached at 571-272-5551. 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.
/PENNY L CAUDLE/Examiner, Art Unit 2657
/HAI PHAN/Supervisory Patent Examiner, Art Unit 2654
Prosecution Insights