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 .
Claims 1-2, 4-8, 10-16, and 21-26 are pending. This Office Action is responsive to the arguments filed on 03/10/2026.
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 1-2, 4-8, 10-16, and 21-26 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Independent Claims
Step 1 – Claim 1 is drawn to a system, claim 11 is drawn to a method, and claim 16 is drawn to a computer program product comprising a computer-readable storage medium having program code embodied therewith. Therefore, each of these claims fall under one of the four categories of statutory subject matter (process/method, machine/product/apparatus, manufacture or composition of matter).
Step 2A Prong 1 – Claims 1, 11 and 16 are directed to a judicially recognized exception of an abstract idea without significantly more. Claims 1, 11 and 16 recite:
split unlabeled data into a plurality of groups corresponding to different perspectives – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0020] of the applicant’s specification, it states “as one specific example in a multi-perspective dialog between customers and support agents, the unlabeled data perspectives 112A and 112B may include a perspective from a customer 112A including utterances from the customer and a perspective from an agent 112B including utterances from the agent.” BRI in light of the specification would support that “split[ting] unlabeled data into a plurality of groups” would encompass a mental process with or without the aid of pen and paper of grouping utterances in a dialogue by the identity of the speaker.
define a respective associated heuristic for each of the plurality of groups – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0017] of the applicant’s specification, it states “In various examples, a heuristic may be separately defined for each of the unlabeled data perspectives 112A, 112B, and 112C... For example, the heuristics 114A, 114B, and 114C may include a lead heuristic and a long heuristic. A lead heuristic, as used herein, refers to a heuristic in which the first utterance of a perspective that contains at least five tokens is taken as a summary of a given perspective. For example, a token may be a word. For example, the tokens may be words. A long heuristic, as used herein, refers to a heuristic in which the longest utterance of a given perspective is selected as the respective summary. For example, the length of the utterance may be measured in tokens. Both of these heuristics may be used to efficiently extract weak summaries for each of the perspectives 112A, 112B, and 112C. In various examples, any other suitable heuristics may be used.” BRI in light of the specification would support that “defin[ing] a respective associated heuristic” would encompass a mental process with or without the aid of pen and paper of establishing a rule.
generate weakly labeled data for each of the plurality of groups using a respective associated heuristic – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0020] of the applicant’s specification, it states “The heuristic-based labeler 104 may thus use heuristics 114A and 114B to generate weakly labeled data 116A including a question and weakly labeled data 116B including a longest sentence, for each of the conversations in unlabeled data 110 from the customer perspective 112A and agent perspective 112B, respectively.” BRI in light of the specification would support that “generat[ing] weakly labeled data” would encompass a mental process with or without the aid of pen and paper of generating a label (e.g. the longest sentence) of an utterance according to a heuristic.
Step 2A Prong 2 – The following additional limitations are recited:
inter-train a pre-trained model for each perspective based on respective weakly labeled data to generate an inter-trained weak label-based model for each perspective – This limitation merely recites the idea of inter-training a model and fails to recite details of how the inter-training is accomplished. Reciting the idea of a solution or outcome without detailing how the result is accomplished is equivalent to saying to "apply" weakly labeled data to the pre-trained model (see MPEP § 2106.05(f)) and thus, fails to integrate the exception into a practical application.
fine-tune each inter-trained weak label-based model for each perspective based on few-shot training data for each different perspective to generate a final model for each different perspective – This limitation merely recites the idea of fine-tuning a model and fails to recite details of how the fine-tuning is accomplished. Reciting the idea of a solution or outcome without detailing how the result is accomplished is equivalent to saying to "apply" few-shot trained data to each model (see MPEP § 2106.05(f)) and thus, fails to integrate the exception into a practical application.
Claims 11 and 16:
receiving, via a processor, unlabeled data, few-shot training data, and a pre-trained model – This limitation recites an insignificant extra-solution activity of mere data gathering (see MPEP § 2106.05(g)) and thus, fails to integrate the exception into a practical application.
Step 2B – The additional elements in Step 2A Prong 2, view individually or wholistically, do not provide an inventive concept or otherwise amount to significantly more than the abstract idea itself.
inter-train a pre-trained model for each perspective based on respective weakly labeled data to generate an inter-trained weak label-based model for each perspective – This limitation merely recites the idea of fine-tuning a model and fails to recite details of how the fine-tuning is accomplished. Reciting the idea of a solution or outcome without detailing how the result is accomplished is equivalent to saying to "apply" few-shot trained data to each model (see MPEP § 2106.05(f)) and thus, fails to provide significantly more to the judicial exception.
fine-tune each inter-trained weak label-based model for each perspective based on few-shot training data for each different perspective to generate a final model for each different perspective – This limitation merely recites the idea of fine-tuning a model and fails to recite details of how the fine-tuning is accomplished. Reciting the idea of a solution or outcome without detailing how the result is accomplished is equivalent to saying to "apply" few-shot trained data to each model (see MPEP § 2106.05(f)) and thus, fails to provide significantly more to the judicial exception.
Claims 11 and 16:
receiving, via a processor, unlabeled data, few-shot training data, and a pre-trained model – This limitation recites an insignificant extra-solution activity of mere data gathering (see MPEP § 2106.05(g)) which is well-understood, routine, and conventional activity similar to cases reviewed by the courts involving receiving or transmitting data over a network (see MPEP § 2106.05(d)(II)) and thus, fails to provide significantly more to the judicial exception.
As such, claims 1, 11 and 16 are not patent eligible.
Dependent Claims
Claims 2, 4-8, 10, 12-15 and 21-26 merely narrow the previously cited abstract idea limitations. For the reasons described above with respect to independent claims 1, 11 and 16 these judicial exceptions are not meaningfully integrated into a practical application, nor amount to significantly more than the abstract ideas. The claims disclose similar limitations described for the independent claims above and do not provide anything more than the mental processes that are practically capable of being performed in the human mind with the assistance of pen and paper. Therefore, claims 2, 4-8, 10, 12-15, 17-19 and 21-23 also recite abstract ideas that do not integrate into a practical application or amount to significantly more than the judicial exception, and are rejected under U.S.C. § 101.
Step 1 – Claims 2, 4-8, 10, and 21-26 are drawn to a system and claims 12-15 are drawn to a method. Therefore, each of these claims fall under one of the four categories of statutory subject matter (process/method, machine/product/apparatus, manufacture or composition of matter).
Step 2A Prong 1 – These claims are directed to a judicially recognized exception of an abstract idea without significantly more.
Claims 2 and 12:
split the conversation into a second plurality of groups corresponding to the different perspectives – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0024] of the applicant’s specification, it states “the perspective splitter 102 can split a received conversation 202 into a number of groups corresponding to different perspectives. For example, the groups may each include a list of sentences associated with a particular perspective. As one examples, the groups may include a first group 204A associated with an agent perspective and a second group 204B associated with a customer perspective.” BRI in light of the specification would support that “split[ting] the conversation into a plurality of groups” would encompass a mental process with or without the aid of pen and paper of grouping utterances in a dialogue by the identity of the speaker.
Claims 4 and 13:
wherein the processor is to add a prefix of indirect speech clause to a generated conversation summary in response to detecting that a generated conversation summary does not begin with any prefix of indirect speech clause – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0026] of the applicant’s specification, it states “As one example, the prefix of indirect speech clause for the example of customer and agent described in Fig. 1, would be "The customer asks:" for customer perspective summaries and "The agent answers:" for agent perspective summaries. In this example, such prefix of indirect speech clause may be added to customer perspective summaries and agent perspective summaries, respectively, in response to detecting that the summaries do not start with the prefix "[The] Customer/Agent".” BRI in light of the specification would support that “add[ing] a prefix to a generated conversation summary” would encompass a mental process with or without the aid of pen and paper of adding a label indicating the speaker to a piece of text.
Claim 5:
wherein the weakly labeled data comprises a summary automatically generated using the respective associated heuristic – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0020] of the applicant’s specification, it states “The heuristic-based labeler 104 may thus use heuristics 114A and 114B to generate weakly labeled data 116A including a question and weakly labeled data 116B including a longest sentence, for each of the conversations in unlabeled data 110 from the customer perspective 112A and agent perspective 112B, respectively.” BRI in light of the specification would support that “generat[ing] weakly labeled data” would encompass a mental process with or without the aid of pen and paper of generating a summary (e.g. the longest sentence) of an utterance according to a heuristic.
Claim 7:
wherein a respective associated heuristic for one of the plurality of groups is different from a respective associated heuristic for another of the plurality of groups – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C) describing evaluating each group of utterances by a different heuristic.
Claim 8:
wherein the plurality of groups each comprise a list of sentences associated with a particular perspective of the plurality of different perspectives – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0024] of the applicant’s specification, it states “the perspective splitter 102 can split a received conversation 202 into a number of groups corresponding to different perspectives. For example, the groups may each include a list of sentences associated with a particular perspective. As one examples, the groups may include a first group 204A associated with an agent perspective and a second group 204B associated with a customer perspective”. BRI in light of the specification would support that “where the plurality of groups each comprise a list of sentences” would encompass a mental process with or without the aid of pen and paper of grouping sentences by the identity of the speaker.
Claim 10:
wherein the weakly labeled data comprises dialog-summary pairs – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C) describing associating a line of dialogue with a summary of that line of dialogue.
Claim 14:
concatenating, via the processor, the generated conversation summaries to generate a final multi-perspective summary – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0023] of the applicant’s specification, it states “the concatenator 208 can receive the conversation perspective summaries 206A, 206B, 206C and output a final summary 210. The final summary 210 may thus be a concatenation of the conversation perspective summaries 206A, 206B, 206C.” BRI in light of the specification and the plain meaning of the term “concatenate” would support that “concatenate[ing] the generated conversation summaries” would encompass a mental process with or without the aid of pen and paper of joining a plurality of text pieces into one.
Claim 15:
wherein inter-training the pre-trained model comprises masking a target utterance – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). BRI of the term “masking” in light of its plain meaning in the context of training a model would support that “mask[ing] a target utterance” would encompass a mental process with or without the aid of pen and paper of obscuring or hiding a piece of data.
Claim 21:
identify an important part of an input with the fine-tuned inter-trained model for each perspective – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0018] of the applicant’s specification, it states “In some examples, the target utterance may be masked during inter-training. For example, the target utterance may be the longest utterance or lead utterance in the example of the use of long and lead heuristics. In this manner, the model inter-trainer 106 can train the model 118 to locate the most important part of an input dialog for each perspective and output it as a summary.” BRI in light of the specification would support that “identify[ing] an important part of an input” would encompass a mental process with or without the aid of pen and paper of evaluating the significance of a portion of dialogue.
Claim 22:
generate an abstractive third-person summary with the fine-tuned inter-trained model for each perspective – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0019] of the applicant’s specification, it states “The fine-tuning may enable the resulting final models 124A, 124B, and 124C to learn to generate an abstractive third-person summary.” BRI in light of the specification would support that “generat[ing] an abstractive third-person summary” would encompass a mental process with or without the aid of pen and paper of summarizing a dialogue from a third-person perspective.
Claim 24:
generating a conversation perspective summary for each perspective with the final model for each perspective – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0033] of the specification, it states “At block 310, the processor fine-tunes each inter-trained model based on the few-shot training data for each different perspective to generate a final model for each different perspective. For example, the final models may then be used to summarize conversations”. BRI in light of the specification would support that “generating a conversation perspective summary for each perspective” would encompass a mental process with or without the aid of pen and paper of summarizing a dialogue from each perspective.
Claim 26:
post-processing the conversation perspective summary for each perspective – This limitation recites the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgement or opinion (see MPEP § 2106.04(a)(2), subsection III, C). In Paragraph [0026] of the specification, it states “In some examples, the generated conversation perspective summaries 206A, 206B, and 206C may be post-processed before concatenation by the concatenator 208. For example, in response to detecting that a conversation perspective summary does not begin with a prefix of indirect speech clause, a post-processing unit (not shown) may add a prefix of indirect speech clause to the conversation perspective summary. As one example, the prefix of indirect speech clause for the example of customer and agent described in Fig. 1, would be "The customer asks:" for customer perspective summaries and "The agent answers:" for agent perspective summaries.” BRI in light of the specification would support that “post-processing the conversation perspective summary for each perspective” would encompass a mental process with or without the aid of pen and paper of adding a label indicating the speaker to a piece of text.
Step 2A Prong 2 – These limitations do not recite any additional elements which integrate the abstract idea into a practical application.
Claims 2 and 12:
wherein the processor is to further: receive a conversation to summarize – This limitation recites an insignificant extra-solution activity of mere data gathering (see MPEP § 2106.05(g)) and thus, fails to integrate the exception into a practical application.
input each of the second plurality of groups into a respective final model for each different perspective and receive a generated conversation summary for each of the second plurality of groups from each respective final model – This limitation amounts to no more than generally linking the use of the judicial exception to a particular technological environment or field of use (see MPEP § 2106.05(h)). It merely recites a model to perform the abstract idea of generating a conversation summary and thus, fails to integrate the exception into a practical application.
Claim 6:
wherein the pre-trained model comprises a pre-trained generative model – This limitation amounts to no more than generally linking the use of the judicial exception to a particular technological environment or field of use (see MPEP § 2106.05(h)). It merely limits the use of the abstract idea to a generative model and thus, fails to integrate the exception into a practical application.
Claim 14:
output the final multi-perspective summary – This limitation recites an insignificant extra-solution activity of mere data output (see MPEP § 2106.05(g)) and thus, fails to integrate the exception into a practical application.
Claim 21:
output the important part of the input as a summary – This limitation recites an insignificant extra-solution activity of mere data output (see MPEP § 2106.05(g)) and thus, fails to integrate the exception into a practical application.
Claim 23:
wherein the few-shot training data includes conversation summaries from a plurality of perspectives – This limitation recites an insignificant extra-solution activity of selecting a data source or type to manipulate (see MPEP § 2106.05(g)) and thus, fails to integrate the exception into a practical application.
Claim 24:
using the conversation perspective summary for each perspective to output a final summary – This limitation merely recites the idea of outputting a final summary and fails to recite details of how the conversation perspective summary is used to do so. Reciting the idea of a solution or outcome without detailing how the result is accomplished is equivalent to saying to "apply" the conversation perspective summary (see MPEP § 2106.05(f)) and thus, fails to integrate the exception into a practical application.
Claim 25:
wherein a concatenator receives the conversation perspective summaries and outputs a final summary – This limitation recites an insignificant extra-solution activity of mere data gathering and output (see MPEP § 2106.05(g)) and thus, fails to integrate the exception into a practical application.
Step 2B – These limitations, as a whole, do not amount to significantly more than the judicial exception.
Claims 2 and 12:
wherein the processor is to further: receive a conversation to summarize – This limitation recites an insignificant extra-solution activity of mere data gathering (see MPEP § 2106.05(g)) which is well-understood, routine, and conventional activity similar to case reviewed by the courts involving receiving or transmitting data over a network (see MPEP § 2106.05(d)(II)) and thus, fails to provide significantly more to the judicial exception.
input each of the second plurality of groups into a respective final model for each different perspective and receive a generated conversation summary for each of the second plurality of groups from each respective final model – This limitation amounts to no more than generally linking the use of the judicial exception to a particular technological environment or field of use (see MPEP § 2106.05(h)). It merely recites a model to perform the abstract idea of generating a conversation summary and thus, fails to provide significantly more to the judicial exception.
Claim 6:
wherein the pre-trained model comprises a pre-trained generative model – This limitation amounts to no more than generally linking the use of the judicial exception to a particular technological environment or field of use (see MPEP § 2106.05(h)). It merely limits the use of the abstract idea to a generative model and thus, fails to provide significantly more to the judicial exception.
Claim 14:
output the final multi-perspective summary – This limitation recites an insignificant extra-solution activity of mere data output (see MPEP § 2106.05(g)) which is well-understood, routine, and conventional activity similar to case reviewed by the courts involving receiving or transmitting data over a network (see MPEP § 2106.05(d)(II)) and thus, fails to provide significantly more to the judicial exception.
Claim 21:
output the important part of the input as a summary – This limitation recites an insignificant extra-solution activity of mere data output (see MPEP § 2106.05(g)) which is well-understood, routine, and conventional activity similar to case reviewed by the courts involving receiving or transmitting data over a network (see MPEP § 2106.05(d)(II)) and thus, fails to provide significantly more to the judicial exception.
Claim 23:
wherein the few-shot training data includes conversation summaries from a plurality of perspectives – This limitation amounts to no more than generally linking the use of the judicial exception to a particular technological environment or field of use (see MPEP § 2106.05(h)). It merely limits the use of the abstract idea to multi-perspective conversations and thus, fails to provide significantly more to the judicial exception.
Claim 24:
using the conversation perspective summary for each perspective to output a final summary – This limitation merely recites the idea of outputting a final summary and fails to recite details of how the conversation perspective summary is used to do so. Reciting the idea of a solution or outcome without detailing how the result is accomplished is equivalent to saying to "apply" the conversation perspective summary (see MPEP § 2106.05(f)) and thus, fails to provide significantly more to the judicial exception.
Claim 25:
wherein a concatenator receives the conversation perspective summaries and outputs a final summary – This limitation recites an insignificant extra-solution activity of mere data gathering and output (see MPEP § 2106.05(g)) which is well-understood, routine, and conventional activity similar to case reviewed by the courts involving receiving or transmitting data over a network (see MPEP § 2106.05(d)(II)) and thus, fails to provide significantly more to the judicial exception.
As such, claims 2, 4-8, 10, 12-15, and 21-26 are not patent eligible.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-2, 5-8, 10-12, 14, 16, 21 and 23-25 are rejected under 35 U.S.C. 103 as being unpatentable over Liang et al. (“HERALD: An Annotation Efficient Method to Detect User Disengagement in Social Conversations”, published 06/02/2021), hereinafter Liang; in view of Wu et al. (US 20220108086 A1, provisional filed 10/02/2020), hereinafter Wu; in further view of Zhang et al. (“Unsupervised Abstractive Dialogue Summarization for Tete-a-Tetes”, published 09/15/2020), hereinafter X. Zhang. Liang, Wu, and X. Zhang were cited in previous Office Actions.
Regarding Claim 1, Liang discloses method steps to:
define a respective associated heuristic for each of the plurality of groups (Liang: “Since labeling large-scale training data is time consuming, we propose heuristic labeling functions to label training data automatically. The heuristic functions focus on detecting disengagement from user responses, as it directly indicates poor user experience. To build the heuristics functions, we first summarize the heuristic rules shared among users. We investigate the disengaged dialog turns from the four datasets mentioned above and identify four groups of user disengagement patterns” [Section 5.1 Stage 1: Auto-label Training Data with Heuristic Function]);
generate weakly labeled data for each of the plurality of groups using a respective associated heuristic (Liang: “instead of manually labeling training samples, we first use a set of labeling heuristics to label training samples automatically” [Abstract]);
inter-train a pre-trained model based on respective weakly labeled data to generate an inter-trained weak label-based model (Liang: “We load and fine-tune pre-trained BERT as the feature extractor φ” [Appendix A.1 Implementation Details of HERALD], “as we received clean training data, we use them to fine-tune a BERT based model and obtain the final user disengagement detection model.” [Section 5. Method]; BRI of “fine-tuning” is that it means to further train a pre-trained model); and
fine-tune the inter-trained weak label-based model based on few-shot training data to generate a final model (Liang: “After fine-tuning BERT on the weakly labeled training data (BERT(Auto), 80.55%, 78.76%), having an additional fine-tuning step using the development set slightly improves the model’s performance (BERT(Auto+dev), 80.73%, 80.46%)” [Section 6. Experiments - Results], “The test set Dtest = {(xi, yi)} Ntest1 contains the ground-truth label yi. The development set Ddev has a similar structure as the test set Dtest but the development set can be much smaller than a train set (i.e., Ndev << Ntrain), making it economical to obtain” [Section 3. Problem Foundation]; BRI of “few-shot” learning is that it means teaching a model when there is a small amount of data).
However, Liang fails to expressly disclose a system, comprising a processor to: split unlabeled data into a plurality of groups corresponding to different perspectives; and generate a model for each different perspective.
In the same field of endeavor, Wu teaches a system, comprising a processor (Wu: “As shown in FIG. 1, computing device 100 includes a processor 110 coupled to memory 120. Operation of computing device 100 is controlled by processor 110” [0017]) to:
split unlabeled data into a plurality of groups corresponding to different perspectives (Wu: “Each dialogue turn corresponds to an utterance made by one speaker before an utterance is made by another speaker. In some embodiments, dialogue conversational history 140 may be defined as D={X.sub.1, X.sub.2, . . . , X.sub.N} where each X.sub.i is a sequence of words in a dialogue turn and N is a total number of dialogue turns. In some instances, dialogue conversation history 140 may include more than two speakers, each speaker speaking during a corresponding dialogue turn.” [0020]);
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated a system, comprising a processor to: split unlabeled data into a plurality of groups corresponding to different perspectives, as taught by Wu, to the system of Liang because both of these systems are directed towards training and applying a machine learning model to analyze conversational data. In making this combination and splitting the unlabeled data into groups as taught by Wu, it allows the system of Liang to summarize conversations and dialogue, which “often involves multiple speakers that may have different points of view” and is also “different from a standard writing style” in that it “has more abbreviations and typos” and “the important information may be scattered” (Wu: [0004]).
Liang and Wu still fail to expressly disclose to generate a model for each different perspective.
In the same field of endeavor, X. Zhang teaches to generate a model for each different perspective (X. Zhang: “In a tete-a-tete, such as a customer-agent conversation, SuTaT aims to summarize for each speaker by modeling the customer utterances and the agent utterances separately while retaining their correlations.” [Abstract]; “To make the unsupervised summarization baseline models adapt to the two-speaker scenario in tete-a-tetes, we train two models for each baseline with either customer utterances or agent utterances. During testing, the customer summaries and agent summaries are generated by the two trained models of each baseline, which are used either separately for automatic and human evaluation or concatenated together for the classification experiment.” [Section 4.2 Baselines]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to have incorporated to generate a model for each different perspective, as taught by X. Zhang, to the system of Liang and Wu because both of these systems are directed towards generating summaries for segments of multi-speaker dialogue. In making this combination and training a model for each different perspective, it would allow the system of Liang and Wu to accommodate the “different semantic contents and choices of vocabularies” that occur in dialogue due to each speaker having “different roles, goals, and language styles”, while allowing the system to “capture highly dependent conversation histories and produce coherent discourses” (X. Zhang: [Section 1. Introduction]).
Regarding Claim 2, Liang, Wu and X. Zhang teach the system of Claim 1, wherein the processor is to further: receive a conversation to summarize (Wu: “CorDial model 130 may receive input, such as a dialogue conversational history 140” [0020]);
split the conversation into a second plurality of groups corresponding to the different perspectives (Wu: “Each dialogue turn corresponds to an utterance made by one speaker before an utterance is made by another speaker. In some embodiments, dialogue conversational history 140 may be defined as D={X.sub.1, X.sub.2, . . . , X.sub.N} where each X.sub.i is a sequence of words in a dialogue turn and N is a total number of dialogue turns. In some instances, dialogue conversation history 140 may include more than two speakers, each speaker speaking during a corresponding dialogue turn.” [0020]); and
input each of the second plurality of groups into a respective final model for each different perspective (X. Zhang: “Let X = {x1, · · · , xn} denote a set of customer utterances and Y = {y1 , · · · , yn} denote a set of agent utterances in the same dialogue. Our aim is to generate a customer summary and an agent summary for the utterances in X and Y.” [Section 2. Methodology]; “Summary representations sX and sY are sampled from the latent spaces taking the weighted combined utterance representations e˜X and e˜Y as inputs. To limit the amount of novelty in the generated summary, we set the variances of the latent spaces close to zero so that sX ≈ µx and sY ≈ µy ∙ sX and sY containing key information from the dialogue are decoded into a customer summary and an agent summary using the same decoders from the conditional generative module, which makes the generated summaries similar to the utterances in pronouns and language styles.” [Section 2.2 Unsupervised Summarization Module]) and
receive a generated conversation summary for each of the second plurality of groups from each respective final model (X. Zhang: “Summary representations sX and sY are sampled from the latent spaces taking the weighted combined utterance representations e˜X and e˜Y as inputs. To limit the amount of novelty in the generated summary, we set the variances of the latent spaces close to zero so that sX ≈ µx and sY ≈ µy ∙ sX and sY containing key information from the dialogue are decoded into a customer summary and an agent summary using the same decoders from the conditional generative module, which makes the generated summaries similar to the utterances in pronouns and language styles.” [Section 2.2 Unsupervised Summarization Module]).
Regarding Claims 11-12 and 16, they are method and computer program product claims corresponding to Claims 1-2. Therefore, they are rejected for the same reasons as Claims 1-2 above.
Regarding Claim 5, Liang, Wu and X. Zhang teach the system of Claim 1, wherein the weakly labeled data comprises a summary automatically generated using the respective associated heuristic (Liang: “Specifically, instead of manually labeling each training datum, we automatically label the training samples with a set of labeling heuristics.” [Section 1. Introduction]).
Regarding Claim 6, Liang, Wu and X. Zhang teach the system of Claim 1, wherein the pre-trained model comprises a pre-trained generative model (X. Zhang: “To accommodate the two-speaker scenario, SuTaT processes the utterances of a customer and an agent separately in a conditional generative module.” [Section 1. Introduction]; Wu: “CorDial model 130 may be a neural network that includes one or more networks or modules and/or pre-trained language models that perform natural language processing tasks. CorDial model 130 may receive input, such as a dialogue conversational history 140 and generate output which may be a dialogue summary 150 of dialogue conversational history 140” [0020]; “A dialogue summary is generated using the generative language model trained using the summary draft.” [0021]).
Regarding Claim 7, Liang, Wu and X. Zhang teach the system of Claim 1, wherein a respective associated heuristic for one of the plurality of groups is different from a respective associated heuristic for another of the plurality of groups (Liang: “We perform ablation studies to analyze the importance of each of the four heuristics groups in Table 1… each heuristics group has different importance in different datasets” [Section 6. Experiments – Heuristic Group Analysis]).
Regarding Claim 8, Liang, Wu and X. Zhang teach the system of Claim 1, wherein the plurality of groups each comprise a list of sentences associated with a particular perspective of the plurality of different perspectives (X. Zhang: “Summary representations sX and sY are sampled from the latent spaces taking the weighted combined utterance representations e˜X and e˜Y as inputs. To limit the amount of novelty in the generated summary, we set the variances of the latent spaces close to zero so that sX ≈ µx and sY ≈ µy ∙ sX and sY containing key information from the dialogue are decoded into a customer summary and an agent summary using the same decoders from the conditional generative module, which makes the generated summaries similar to the utterances in pronouns and language styles.” [Section 2.2 Unsupervised Summarization Module]; Wu: “Each dialogue turn corresponds to an utterance made by one speaker before an utterance is made by another speaker. In some embodiments, dialogue conversational history 140 may be defined as D={X.sub.1, X.sub.2, . . . , X.sub.N} where each X.sub.i is a sequence of words in a dialogue turn and N is a total number of dialogue turns. In some instances, dialogue conversation history 140 may include more than two speakers, each speaker speaking during a corresponding dialogue turn” [0020]).
Regarding Claim 10, Liang, Wu and X. Zhang teach the system of Claim 1, wherein the weakly labeled data comprises dialog-summary pairs (Liang: See [Figure 1], where weakly labeled training data includes original data from dialog corpora and their corresponding labels).
Regarding Claim 14, Liang, Wu and X. Zhang teach the method of Claim 12, further comprising:
concatenating, via the processor, the generated conversation summaries to generate a final multi-perspective summary (Wu: “Concatenation module 220 may receive segment summaries 204 that decoder 215 generates from multiple dialogue segments 202 and concatenates multiple segment summaries 204 into dialogue summary 150” [0022]; X. Zhang: “During testing, the customer summaries and agent summaries are generated by the two trained models of each baseline, which are used either separately for automatic and human evaluation or concatenated together for the classification experiment.” [Section 4.2 Baselines]); and
outputting, via the processor, the final multi-perspective summary (Wu: “A dialogue summary is generated using the generative language model trained using the summary draft.” [Abstract]).
Regarding Claim 21, Liang, Wu and X. Zhang teach the system of Claim 1, wherein the processor is to further:
identify an important part of an input with the fine-tuned inter-trained model for each perspective (X. Zhang: “SuTaT employs a setence-level self-attention mechanism to highlight more significant utterances and neglect uninformative ones.” [Section 1. Introduction]); and
output the important part of the input as a summary (X. Zhang: “We employ a sentence-level self-attention mechanism on the utterances embeddings to highlight the more informative ones and combine the weighted embeddings. A summary representation is drawn from the low-variance latent space using the combined utterance embedding, which is then decoded into a summary with the same decoder and a partial copy mechanism.” [Section 2. Methodology]).
Regarding Claim 23, Liang, Wu and X. Zhang teach the system of Claim 1, wherein the few-shot training data includes conversation summaries from a plurality of perspectives (X. Zhang: “We fine-tune the parameters of SuTaT on the validation set.” [Section 4.3 Settings]; “MultiWOZ consists of 10438 goal-oriented human-human written dialogues between customers and agents, spanning over 7 domains such as booking hotels, booking taxis, etc. 3406 of them are single label and 7302 of them are multi-label. In the experiment, we split the dataset into 8438, 1000, and 1000 dialogues for training, testing, and validation.” [Section 4.1 Dataset]; Liang: “The development set Ddev has a similar structure as the test set Dtest but the development set can be much smaller than a train set (i.e., Ndev << Ntrain), making it economical to obtain.” [Section 3. Problem Formulation]).
Regarding Claim 24, Liang, Wu and X. Zhang teach the system of Claim 1, further comprising:
generating a conversation perspective summary for each perspective with the final model for each perspective (X. Zhang: “During testing, the customer summaries and agent summaries are generated by the two trained models of each baseline, which are used either separately for automatic and human evaluation or concatenated together for the classification experiment.” [Section 4.2 Baselines]); and
using the conversation perspective summary for each perspective to output a final summary (X. Zhang: “During testing, the customer summaries and agent summaries are generated by the two trained models of each baseline, which are used either separately for automatic and human evaluation or concatenated together for the classification experiment.” [Section 4.2 Baselines]).
Regarding Claim 25, Liang, Wu and X. Zhang teach the system of Claim 24, wherein a concatenator receives the conversation perspective summaries and outputs a final summary (X. Zhang: “During testing, the customer summaries and agent summaries are generated by the two trained models of each baseline, which are used either separately for automatic and human evaluation or concatenated together for the classification experiment.” [Section 4.2 Baselines]; Wu: “Concatenation module 220 may receive segment summaries 204 that decoder 215 generates from multiple dialogue segments 202 and concatenates multiple segment summaries 204 into dialogue summary 150.” [0022]).
Claims 4, 13, 22 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over Liang in view of Wu and X. Zhang, as applied to Claims 1, 2, 12 and 24 above, in further view of Zhu et al. (“A Hierarchical Network for Abstractive Meeting Summarization with Cross-Domain Pretraining”, published 09/20/2020), hereinafter Zhu. Zhu was cited in a previous Office Action.
Regarding Claim 4, Liang, Wu and X. Zhang teach the system of Claim 2. However, they fail to expressly disclose wherein the processor is to add a prefix of indirect speech clause to a generated conversation summary in response to detecting that a generated conversation summary does not begin with any prefix of indirect speech clause.
In the same field of endeavor, Zhu teaches wherein the processor is to add a prefix of indirect speech clause to a generated conversation summary in response to detecting that a generated conversation summary does not begin with any prefix of indirect speech clause (Zhu: “To incorporate the participants’ information, we integrate the speaker role component. In the experiments, each meeting participant has a distinct role, e.g., program manager, industrial designer. For each role, we train a vector to represent it as a fixed-length vector rp, 1 ≤ p ≤ P, where P is the number of roles... This vector is appended to the embedding of the speaker's turn” [Section 3.1.1 Role Vector], See [Table 1 - Summary from our model (23 sentences)] where each line of summary includes the role with a term of indirect speech).
It would have been obvious to one of ordinary skill in the art to have incorporated wherein the processor is to add a prefix of indirect speech clause to a generated conversation summary in response to detecting that a generated conversation summary does not begin with any prefix of indirect speech clause, as taught by Zhu to the system of Liang, Wu and X. Zhang as both of these systems are directed towards training a model to summarize transcripts with multiple participants. In making this combination and including a prefix before the summary of each segment of dialogue, it would allow the person reviewing the summary to know who is speaking by “incorporat[ing] the role of each speaker to encode different semantic styles and standpoints among participants” [Section 1. Introduction].
Regarding Claim 13, it is a method that corresponds to Claim 4 above. Therefore, it is rejected for the same reason as Claim 4 above.
Regarding Claim 22, Liang, Wu and X. Zhang teach the system of Claim 1. However, they fail to expressly disclose wherein the processor is to further: generate an abstractive third-person summary with the fine-tuned inter-trained model for each perspective.
In the same field of endeavor, Zhu teaches wherein the processor is to further: generate an abstractive third-person summary with the fine-tuned inter-trained model for each perspective (Zhu: See [Table 8] – [Table 10]).
It would have been obvious to one of ordinary skill in the art to have incorporated wherein the processor is to further: generate an abstractive third-person summary with the fine-tuned inter-trained model for each perspective, as taught by Zhu to the system of Liang, Wu and X. Zhang as both of these systems are directed towards training a model to summarize transcripts with multiple participants. In making this combination and generating a third-person summary for each perspective, it would allow the system of Liang, Wu and X. Zhang to “incorporate[[s]] the role of each speaker to encode different semantic styles and standpoints among participants” (Zhu: [Section 1. Introduction]).
Regarding Claim 26, Wu and X. Zhang teach the system of Claim 24. However, they fail to expressly disclose post-processing the conversation perspective summary for each perspective.
In the same field of endeavor, Zhu teaches post-processing the conversation perspective summary for each perspective (Zhu: “To incorporate the participants’ information, we integrate the speaker role component. In the experiments, each meeting participant has a distinct role, e.g., program manager, industrial designer. For each role, we train a vector to represent it as a fixed-length vector rp, 1 ≤ p ≤ P, where P is the number of roles... This vector is appended to the embedding of the speaker's turn” [Section 3.1.1 Role Vector], See [Table 1 - Summary from our model (23 sentences)] where each line of summary includes the role with a term of indirect speech; In light of paragraph [0026] of the specification, which states “the generated conversation perspective summaries 206A, 206B, and 206C may be post-processed before concatenation by the concatenator 208. For example, in response to detecting that a conversation perspective summary does not begin with a prefix of indirect speech clause, a post-processing unit (not shown) may add a prefix of indirect speech clause to the conversation perspective summary”, BRI would support that “post-processing” encompasses appending a prefix of indirect speech clause to a summary).
It would have been obvious to one of ordinary skill in the art to have incorporated post-processing the conversation perspective summary for each perspective, as taught by Zhu to the system of Liang, Wu and X. Zhang as both of these systems are directed towards training a model to summarize transcripts with multiple participants. In making this combination and post-processing the conversation summaries for each perspective, it would allow the person reviewing the summary to know who is speaking by “incorporat[ing] the role of each speaker to encode different semantic styles and standpoints among participants” [Section 1. Introduction].
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Liang in view of Wu and X. Zhang, as applied to Claim 11 above, in further view of Zhang et al. (“PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization”, published 06/02/2020), hereinafter J. Zhang. J. Zhang was cited in a previous Office Action.
Regarding Claim 15, Liang, Wu and X. Zhang teach the method of Claim 11. However, they fail to expressly disclose wherein inter-training the pre-trained model comprises masking a target utterance.
In the same field of endeavor, J. Zhang teaches wherein inter-training the pre-trained model comprises masking a target utterance (J. Zhang: “In PEGASUS, important sentences are removed/masked from an input document and are generated together as one output sequence from the remaining sentences, similar to an extractive summary.” [Abstract]).
It would have been obvious to one of ordinary skill in the art to have incorporated wherein inter-training the pre-trained model comprises masking a target utterance, as taught by J. Zhang to the system of Liang, Wu and X. Zhang because both of these methods are directed towards pre-training a model for text summarization. In making this combination and masking target utterances while training the model as taught by J. Zhang, it would allow the system of Liang, Wu and X. Zhang to better handle downstream summarization tasks, as “it closely resembles the downstream task, encouraging whole-document understanding and summary-like generation” [Section 1. Introduction].
Response to Arguments
Applicant's arguments, filed 03/10/2026, traversing the rejection of Claims 1, 2, 4-8, 10-16, and 21-26 under 35 U.S.C. § 103 have been fully considered but are not persuasive.
Applicant alleges, on pages 10-22 of the Remarks, that the combination of Liang, Wu and X. Zhang fails to teach each and every element of the claims, specifically:
Liang fails to teach or suggest “inter-train a pre-trained model for each perspective based on respective weakly labeled data” as it fails to acknowledge the “for each perspective” part of the claim limitation requiring per perspective model training. The addition of Wu does not rectify this deficiency because Wu also fails to teach or suggest the limitation. The addition of X. Zhang fails to rectify this deficiency because it fails to teach or suggest the limitation because it does not offer inter-training of any models as a training step in between a pre-training step and a fine-tuning step, fails to use “weakly-labeled data” to inter-train any models, and uses entirely different mechanisms to train models than that required by the claims which are not combinable with the other cited art. Additionally X. Zhang uses “unsupervised abstractive dialogue summarization” on “a conditional generative module and two unsupervised summarization modules”, which is distinct from the “respective associated heuristic” mechanism used by the claims to “generate weakly labeled data” and thereby “inter-train a pre-trained model for each perspective”, and therefore not combinable with Liang.
Liang fails to teach or suggest “define a respective associated heuristic for each of the plurality of groups” as required by the pending claims. Specifically, Liang assigns heuristic labeling functions to groups of dialogue segments based on “user disengagement patterns”, not based on perspective.
Liang fails to teach or suggest “fine-tune the inter-trained weak label-based model based on few-shot training for each different perspective” as it fails to acknowledge the “for each perspective” part of the claim limitation requiring per perspective model training. The addition of X. Zhang still cannot teach or suggest “fine-tune the inter-trained weak label-based model based on few-shot training for each different perspective” as required by the claims as X. Zhang does not teach or suggest few-shot training data nor a weak label-based model. Further, X. Zhang is not combinable with Liang as they offer entirely different mechanisms to achieve wholly different goals, as they employ distinct training mechanisms that cannot be substituted with one another and have entirely different objectives to be fulfilled.
Examiner respectfully disagrees.
Regarding points (a) and (c), it is acknowledged that none of the references teach all of the required limitations on their own, however, this is not required under U.S.C. § 103. As such, the point of contention appears to be over the ability to combine the teachings of Liang with those of X. Zhang. Applicant tries to assert that the references differ in ways that make their teachings incompatible with one another such that there is no motivation to combine. However, the differences in these references do not discredit nor discourage the teachings of the other such that the cited elements cannot be combined to teach the limitations of the independent claim. On a fundamental level, the two references are addressing the same bottleneck: training a dialogue model when labeled conversation data is scarce or expensive. The “weakly-supervised” approach of Liang and the “unsupervised” approach of X. Zhang are similarly attempting to reduce the cost of manually labeling data in a way that, while not explicitly interchangeable, represent desirable alternatives that a POSITA could reasonably consider and are both well-known strategies in the art of machine learning. Similarly, while the ultimate objectives of the two references are different (i.e., user disengagement detection vs. dialogue summarization), these objectives fall with the same relevant field of endeavor, in that they both fit within the domain of conversational AI, dialogue representation learning, dialogue segmentation, textual feature extraction to a latent space, and resource efficient training for dialogue related tasks. It is also worth pointing out that the training pipeline employed by Liang of utilizing a pre-trained model, generating weak labels for dialogue segments based on heuristics, and performing multiple rounds and types of training are not specific to user disengagement, nor to a particular model architecture, and can be applied broadly to other tasks in the dialogue representation learning domain or even NLP in general. A POSITA could reasonably consider applying this training pipeline to fulfill a related task, especially given that both of these tasks involve multi-turn conversational text, defined speaker roles (i.e., chat bot and user vs. agent and customer), and limited labeled data. Similarly, a POSITA could reasonably consider apply the teachings of X. Zhang to the training pipeline of Liang as that they both involve a conversation between distinct speakers, and X. Zhang provides a clear motivation why one may desire to train speaker-specific models in order to capture the correlations between the utterances of each speaker. As such, it would not be unreasonable to combine the cited teachings of Liang and X. Zhang to arrive at the independent claim’s limitations of “inter-train a pre-trained model for each perspective based on respective weakly labeled data” and “fine-tune each inter-trained weak label-based model for each perspective based on few-shot training data for each different perspective”.
Regarding point (b), the independent claim does not provide details of what heuristics are being applied, nor how they are determined to be associated with a given group of data, merely that weakly labeled data is generated for each of the plurality of data groups using a respective associated heuristic, for which the groups of data correspond with the different perspectives. While Liang does not teach dividing data into groups corresponding to different perspectives, it does define heuristic groups based on semantic differences in the conversation data and applies the heuristic functions to the data to map them to these heuristic groups. X. Zhang expressly splits the utterances of into groups based on different speaker roles, as it processes them separately to train the separate models, in order to reflect the differences between the speakers’ dialogue utterances (e.g., different roles, goals, language styles, semantic contents, choice of vocabularies, etc.). Once the dialogue data is partitioned into separate groups, a POSITA would have found it obvious to apply the concept taught in Liang, that heuristics can be used to generate weakly labeled data for groups of dialogue segments, to define heuristics that model the differences in the already-grouped speaker utterances and label the respective speaker data groups of X. Zhang using those heuristics, as this represents the application of a known technique to a known method ready for improvement to yield predictable results. While the heuristics needed to properly represent the data of X. Zhang would probably differ from the heuristic functions in Liang, the application of the general technique would be well within the capabilities of a POSITA at the time of the invention to arrive at defining a respective associated heuristic for each of a plurality of groups corresponding to different perspectives and generating weakly labeled data for each of the plurality of groups using the respective associated heuristic, as recited in the independent claim.
As such, Examiner asserts that the rejection establishes a prima facie case of obviousness under 35 U.S.C. § 103 for the independent claims. The dependent claims are similarly ineligible for their dependency on an ineligible independent claim as well as for their own deficiencies outlined in the 35 U.S.C. § 103 rejection above.
Applicant's arguments, filed 03/10/2026, traversing the rejection of Claims 1, 2, 4-8, 10-16, and 21-26 under 35 U.S.C. § 101 have been fully considered but are not persuasive.
Applicant alleges, on Pages 23-28 of the Remarks, that no claims as a whole is directed to an abstract idea because any such concept is integrated into a particular, technical use that improves the technical field of “training machine learning networks with unlabeled data” while also improving computer operation given the specific training regimen and clear objective. In accordance with the present disclosure, models can automatically generate multi-perspective summaries of conversations using a very small amount of annotated data during training, thereby averting problems caused by the available training data including very few labeled instances or incorrectly labeled instances of training data, specifically models requiring large amounts of labeled data including summaries for training as well as manually generating such summaries which may be time consuming and sometimes inaccurate. These functions are not practically performed in the human mind nor do they merely require generic use of computer components. In contrast to generically applying a generic machine learning model as discussed in Recentive Analytics, the claims offer a specific model training regimen to achieve a specific objective to overcome a technology-specific problem. The claims do not offer a sort of result-oriented scheduling and/or commercial optimization as seen in ineligible subject matter but rather, the claims offer a field-specific technological advancement that enables continued compute during a manager outage. Based on the guidance presented in Ex Parte Desjardins and Memo integrating the decision into the MPEP, the pending claims improve a technical field and/or the operation of the machine learning system by improving the technical field of few-shot fine-tuning while training machine learning networks with unlabeled data and the specific operation of “training of models to automatically generate multi-perspective summaries of conversations using a very small amount of annotated data during training. The claims therefore map closely to the types of improvements the Appeals Review Panel credited in the Desjardins decision. The Memo stresses the improvement must be apparent to a person having ordinary skill in the art in the specification and reflected in the claims but that the claim itself does not need to explicitly recite the improvement described in the specification as improvements tantamount to how the machine learning model itself would function in operation are not subsumed in the identified abstract idea. The specification explains the problem solved by the pending claims, including few-shot fine-tuning machine learning network training with unlabeled data. Applying the guidance of Desjardins and the Memo to the present claims, which include, inter alia, a specific model training regimen for a specific technological problem faced when training machine learning networks with unlabeled data, the claims are not directed to an abstract idea at least at Step 2A, Prong Two.
Examiner respectfully disagrees.
When reviewing the claim limitations and the specification, it becomes clear that the concept of grouping unlabeled data by perspective and generating weakly trained data based on a respective associated heuristic recites mental processes. First, splitting conversation data based on perspective is simple sorting, which is routinely performed by humans. Second, the example of a lead heuristic and long heuristic, which respectively defines a summary as the first utterance containing at least five tokens/words and the longest utterance. Applying these heuristics to conversational data is not only well within the capabilities of the human mind to perform, but also is presumably done manually by humans to label the limited amount of labeled instances in the training data. At least with these limitations, the independent claim recites an abstract idea. In Step 2A Prong Two, as recited in the Desjardin guidance, “the specification should be evaluated to determine if the disclosure provides sufficient details such that one of ordinary skill in the art would recognize the claimed invention as providing an improvement in the functioning of a computer, or an improvement to other technology or a technical field.” The recited problem presented in Paragraph [0013] is that “very few labeled instances or incorrectly labeled instances of training data” and that “existing methods may require large amounts of labeled data including summaries for training and manually generating such summaries may be time consuming and sometimes inaccurate”. The proposed solution as presented in Paragraph [0014] is a method to “enable training of models to automatically generate multi-perspective summaries of conversations using a very small amount of annotated data during training”. If the problem is a lack of annotated training data and the solution is a method to enable training using a small amount of labeled data, then the training itself is not the solution, but what enables the training. For example, if one was trying to teach another person to recognize a song from sheet music with a lot of sheet music but very few professional recordings, so they record themselves playing the music, have the other person learn by listen to the amateur recordings with the sheet music, and ensure their recognition is correct by listening to the professional recordings, the solution to enabling learning to recognize songs is not the learning while listening to the recordings but making the recordings. In the present disclosure, the solution of “enabl[ing] the training of models to automatically generate multi-perspective summaries of conversations using a very small amount of annotated data during training” is provided exclusively by the generating weakly labeled data using the heuristics and the model training steps are the conclusory result, as training does not “enable training”. The other additional results also fail to contribute to the purported improvement, as training a model for each perspective is not described in the specification as contributing to the solution but rather seems to be a design choice, and the fine-tuning merely involves labeled training data which is barely relevant to the solution as the problem is not the annotated data but the lack thereof. Otherwise, if the solution is to “automatically generate multi-perspective summaries of conversations using a very small amount of annotated data” then the model training is merely linking the abstract idea of generating summaries of conversations to the technological field of machine learning and is irrelevant to the stated problem of unlabeled data for model training. Or, if the solution is a method for the training to be more efficient, the models to be more efficient, and/or the models to be more accurate than prior art with unlabeled data, those are not provided in enough detail by the disclosure such that one would be able to recognize the proposed solution as fulfilling that goal. As such, when evaluating the claim as a whole, the improvement is either provided by the abstract idea alone, the claim does not reflect the improvement, or the specification does not detail the improvement. Regardless, the independent claim does not integrate the abstract idea into a practical application.
Applicant alleges, on Pages 28-30 of the Remarks, that the present claims contribute an inventive concept under an eligibility analysis under Step 2B. The claims recite an “inventive concept” or “an element or combination of elements that is sufficient to ensure that the patent in practice amounts to significantly more than a patent upon the abstract idea itself. The present claims’ contribution of “an inventive concept” is evidenced by the conclusion that the claims are inventive over the prior art. Like the claims at issue in DDR Holdings, the present claims do not recite “mathematical algorithms” or “fundamental economic or longstanding commercial practices” but instead are “necessarily rooted in computer technology in order to overcome a problem specifically arising in the realm of computers” and provide a technology-specific solution to a technology-specific problem that amounts to more than an abstract idea. The claims also introduce an advancement in the technology of training machine learning networks with unlabeled data that is patent eligible. The improvement to machine learning network training with unlabeled data to “enable training of models to automatically generate multi-perspective summaries of conversations using a very small amount of annotated data during training” is a “critical aspect” of the present application and therefore amounts to an inventive concept.
Examiner respectfully disagrees.
Regardless of the state of the prior art, a judicial exception alone cannot provide an inventive concept. As noted above, the purported improvement is either provided by the abstract idea alone, the claim does not reflect the improvement, or the specification does not detail the improvement. Further, the additional elements presented in the independent claims merely recite well-understood, routine, conventional activity in the field of machine learning. The recited “inter-training” describes a training step after pre-training in which the model is trained on the weakly labeled data generated by the abstract idea of applying a heuristic to create a summary of dialogue. Weakly supervised learning is a well-known concept in the art and is a form of semi-supervised learning that is widely used in any field in which labeled data is scarce, such as in the medical field or with brand new technologies. The recited “fine-tuning” describes a training step after weakly supervised training in which the few-shot labeled training data that is provided by a client is used to correct the model from tending towards extractive summarization. Fine-tuning and few-shot learning are also well-known concepts in machine learning and are employed widely in semi-supervised learning. Additionally, the present disclosure is not comparable to DDR Holdings or Enfish as the abstract idea of dialogue summarization is not exclusive rooted in computer technology, as it can be performed in the human mind, exclusively provides the recited improvement in enabling the training of a summarization model with limited annotated data, and does not amount to more than the judicial exception when considered independently or as a whole with the well-understood, routine, and conventional training techniques recited in the claim and thus, the Examiner asserts that the 35 U.S.C. § 101 rejection set forth above is proper and establishes a prima facie case of patent ineligibility.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wu et al. (“Controllable Abstractive Dialogue Summarization with Sketch Supervision”) discusses improving abstractive dialogue summarization quality through a two-stage generation strategy in which a weakly supervised signal in the form of pseudo-labeled interrogative pronoun categories and key phrases are provided by user defined heuristic labeling functions.
Ratner et al. (“Snorkel: Rapid Training Data Creation with Weak Supervision”) discusses allowing users to write labeling functions expressing arbitrary heuristics to allow the training of models while bypassing the bottleneck of limited labeled training data.
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MEGAN E HWANG whose telephone number is (703)756-1377. The examiner can normally be reached Monday-Friday 10:00-7:30 ET.
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, Jennifer Welch can be reached at (571) 272-7212. 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.
/M.E.H./Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143
Prosecution Insights