Notice of Pre-AIA or AIA Status
1. 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 Arguments
2. Applicant's arguments filed 12/15/2025 have been fully considered but they are not persuasive.
First, the amendments do not add subject matter beyond what is already taught by Sabapathy’s schema/template based processing. Sabapathy discloses selecting a template/schema for the received transcript object based on type/classification and then identifying and populating the dynamic fields of that selected template via rule-based mappings to external data sources. In other words, once the extraction schema instance (template) is selected, the system determines which fields are to be extracted/filled and performs extraction to obtain corresponding values from the object in accordance with the selected schema instance. Therefore, the added language “based on the extraction schema instance, determine to extract a first value of the first extraction field from the object using a first model”, “and, based on the extraction schema instance, determine to extract a second value of the second extraction field from the object using a default model; input the object to the default model to output the second value of the second extraction field;” are taught by Sabapathy’s disclosure that the selected template/schema specifies the fields to be populated and drives the extraction/filling of those fields for the object.
Second, Pasko expressly discloses generating confidence scores for model outputs, comparing those confidence scores to one or more threshold and when a confidence threshold is not satisfied, proceeding to evaluate the input using an additional model associated with the same semantic task. This constitutes determining a fallback model and obtaining an alternative output in response to a confidence based determination. The claimed default model and fallback model merely recite a known prioritization and sequencing of models based on confidence thresholds, which is explicitly taught by Pasko’s round-by-round evaluation and multi-threshold framework. Incorporating this confidence based fallback behavior into Sabapathy’s schema-driven extraction system would have been an obvious to one of ordinary skill in the art.
Claim Rejections - 35 USC § 103
3. In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
4. Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Sabapathy (US 2023/0385557) in view of Pasko (US 11,132,509).
Regarding Claim 1:
A system comprising: a memory storing processor-executable program code (Sabapathy: p[0005] memory with code stored);
and a processing unit to execute the processor-executable program (Sabapathy: p[0005] processor to run the instructions) to cause the system to: code to cause the system to:
receive an object on which to perform entity extraction (Sabapathy: Fig. 5 step 501 discloses a multi-party communication transcript data object to perform extraction);
from a plurality of extraction schema instances, identify an extraction schema instance associated with the object (Sabapathy: p[0080] and Fig. 5 step 502 discloses selecting a summarization template from a plurality of templates based on hybrid classification);
determine a first extraction field (Sabapathy: p[0022], Figs. 7 and 8 discloses identifying dynamic data fields in the selected template, these are extraction fields, associates with a rule-based mapping to data sources like APIs) (1)
(2)
(3)
(4)
determine a second extraction field (Sabapathy: p[0025] discloses identifying standard sentences within each template, meaning each template includes multiple call-specific fields that need to be populated after the first field is processed (e.g., the first caller’s name) is processed, the system continues identifying additional fields such as account number, reason for call etc.) (5)
(6)
(7) ;
determine a confidence threshold associated with the second extraction field by the extraction schema instance;
determine that the confidence score is not greater than the confidence threshold in response to the determination that the confidence score is not greater than the threshold;
determine a fallback model associated with the second extraction field by the extraction schema instance ();
generate a second input payload according to an input format of the fallback model;
transmit the second input payload to the fallback model;
and receive a third value of the second extraction field output by the fallback model;
and return the first value and the third value.
Sabapathy does not disclose the crossed out limitations above, however Pasko discloses a (1) first model associated with the first extraction field based on the extraction schema instance (Pasko: Col 15:8-22 within the selected domain (schema instance), the system determines fields (referred to as slots) and uses the domain specific NER model associated with those slots which are interpreted as a first extraction field and a first model associated with the first extraction field based on the schema);
(2) generate an input payload according to an input format of the first model (Pasko: Col 23:3-9 discloses exactly what the model expects as input, the NER models (interpreted as the first model) take in ASR text (or audio), the IC models take ASR text + NER result. Preparing and routing that specific structure is generating an input payload according to the model’s input format);
(3) transmit the input payload to the first model (Pasko: Col 26:37-50 discloses inputting this data to a domain specific NER model meaning the data was transmit);
(4) receive a first value of the first extraction field output by the first model (Pasko: Col 26:37-50 discloses the first model (NER) outputs slot data, this is a value of an extraction field);
(5) and, based on the extraction schema instance, determine to extract a second value of the second extraction field from the object using a default model (Pasko: Col 16:42-57 discloses a second model, domain specific IC models associated with another field for figuring out intent of the NER);
(6) input the object to the default model to output the second value of the second extraction field (Pasko: Col 18:50-54 discloses once the NER model identifies entities, the same input object is fed into the IC model, this intent data that is output is the second value of the second extraction field);
(7) receive the second value and a second confidence score from the default model (Pasko: Col 18:55-58 discloses the system receives the IC model’s output which is the second value).
(8) determine a confidence threshold associated with the second extraction field by the extraction schema instance (Pasko: Col 18:40-67 discloses comparing the confidence score for a domain specific NLU result to a threshold confidence score, tis threshold is the criterion used for the domain/model output, i.e., the confidence threshold applied to the model output associated with the field/model);
(9) determine that the confidence score is not greater than the confidence threshold in response to the determination that the confidence score is not greater than the threshold (Pasko: Col 18:67 – Col 19:5, “the threshold confidence score is not satisfied…” expressly teaches the non meeting threshold condition and continuing evaluation);
(10) determine a fallback model associated with the second extraction field by the extraction schema instance (Pasko: Col 18:67 – Col 19:5 “for the next highest coring candidate domain “discloses selecting another model when the current one fails the threshold, i.e., fallback domain/model for the same input; Col 19:1-15 discloses a next round of evaluation runs a domain specific NER model and domain specific IC model for the next candidate domain);
(11) generate a second input payload according to an input format of the fallback model (Pasko: Col 19:1-15: teaches providing the same input data to the next domain’s model (the fallback models) are provided the required inputs in the required form (ASR text, ASR text + NER result);
(12) transmit the second input payload to the fallback model (Pasko: Col 19:1-15: teaches providing the same input data to the next domains models);
(13) and receive a third value of the second extraction field output by the fallback model (Pasko: Col 18:67 – Col 19:15 discloses teaching that in the subsequent round the domain specific NER model outputs a NER result and the domain specific IC models outputs an IC result which are combined into a domain specific NLU result);
(14) and return the first value and the third value (Pasko: Col 19:1-15 explains that when a later round produces an NLU result meeting threshold, the evaluation is stopped and that NLU result may be selected).
Sabapathy and Pasko are combinable because they are from the same field of endeavor of information extraction using machine learning models. Each discloses receiving an input object, selecting among multiple processing options, invoking those options with model-specific inputs and returning structured values. It would have been obvious to a person of ordinary skill in the art before the effective filing date to implement Sabapathy’s schema-based extraction with a second intent-oriented model and the domain-based routing taught by Pasko to improve accuracy and latency when extracting multiple fields from the same object. The motivation for doing so is “optimizing the utilization of local computing resources (e.g., processing resources, etc.) of the speech interface device 102 can a reduce latency so that the user experience with the speech interface device 102 is not negatively impacted by local processing tasks taking too long.” As disclosed by Pasko in Column 4 lines 46-50.
Regarding Claim 2:
The combination of Sabapathy and Pasko further disclose a system according to Claim 1, wherein the first model is a large language model, and the input payload includes a prompt and the object (Pasko: Col 15:10-16 discloses the NER model is trained on natural language text for entity recognition), and the input payload includes a prompt and the object (Sabapathy: teaches generating an input payload for invoking a selected ruleset).
It would have been obvious to one of ordinary skill in the art before the effective filing date to use a large language model for performing entity extraction because LLMs are a well-known advanced machine learning model capable of handling natural language inputs. Pasko teaches the use of machine learning models such as NER and IC models for similar purposes, making substitution of an LLM a predictable variation to possibly improve performance and accuracy.
Regarding Claim 3:
A system according to Claim 2, wherein the default model is a pre-trained model (Pasko: Col 15:12-20 the NER and IC models are trained and the downloaded to the device for execution).
It would have been obvious to one of ordinary skill in the art before the effective filing date to disclose a pre-trained model for entity extraction would have been obvious as Pasko discloses machine learning models trained on prior data, such as domain-specific NER and IC models. Pre-trained models are a standard design choice which is used by Sabapathy for the classification model and Pasko for the extraction models. Incorporating Pasko’s pre-trained models into Sabapathy’s (who also discloses pre-trained models for classification purposes) would have been a simple additional
Regarding Claim 5:
A system according to Claim 1,the processing unit to execute the processor-executable program code to cause the system to:
receive a second object on which to perform entity extraction (Sabapathy: p[0019] discloses one of the purposes is for storing multiple data objects with more efficiency, Examiner Interpretation {Sabapathy places no limit on the number of objects that analyzed});
from a-the plurality of extraction schema instances, identify the extraction schema instance as associated with the second object (Sabapathy: p[0020] selects a hybrid class from a classification space based on the type of input object, this matches the claim step of identifying which schema to use for the second object);
determine the first extraction field (Sabapathy: p[0022], Figs. 7 and 8 discloses identifying dynamic data fields in the selected template, these are extraction fields, associates with a rule-based mapping to data sources like APIs) and, based on the extraction schema instance, determine to extract a fourth value of the first extraction field from the second object using the first mode (Pasko: Col 15:8-22 within the selected domain (schema instance), the system determines fields (referred to as slots) and uses the domain specific NER model associated with those slots which as previously reasoned in other claims can be done any number of times and are interpreted as a first model associated with the first extraction schema instance);
generate a third input payload according to the input format of the first model (Pasko: Col 23:3-9 discloses exactly what the model expects as input, the NER models take in ASR text (or audio), preparing and routing that specific structure is generating an input payload according to the model’s input format, in this case the third input payload would be the input to the NER model in another iteration);
transmit the third input payload to the first model (Pasko: Col 23:3-9 generates input for the models; Col 21:7-Col 22 line 17 discloses there can be any number of domain classifications, and that pruning steps may occur or not occur on how iterations these models will go through i.e., transmitting this third input payload to a first model (NER model));
receive a fourth value of the first extraction field output by the first model (Pasko: Col 26:37-50 discloses the first model (NER) outputs slot data, this is a value of an extraction field and if a second iteration were to occur could reasonably by interpreted as a fourth value);
determine the second extraction field and, based on the extraction schema instance, determine to extract a fifth value of the second extraction field from the second object using the default model (Pasko: Col 16:42-57 discloses a second model, domain specific IC models associated with another field for figuring out intent of the NER);
input the second object to the default model to output the fifth value of the second extraction field (Pasko: Col 18:50-54 discloses once the NER model identifies entities, the same input object is fed into the IC model, this intent data that is output is the second value of the second extraction field);
receive the fifth value and a second confidence score from the default model (Pasko: Col 18:45 – Col 19:15 discloses that a domain specific intent classification model outputs an IC result and a “second confidence score”, together. The IC result constitutes the received value and the associated confidence score is expressly generated and received from the same model);
determine that the second confidence score is greater than the confidence threshold (Pasko: Col 17: 43-45 and Col 18:40- Col 19:5 compares the confidence scores to the threshold to determine if it satisfies); and
based on the determination that the second confidence score is greater than the confidence threshold, return the fourth value and the fifth value (Pasko: Col 23:31-42 the system determines when processing is no longer needed based on the scores of the thresholds).
Sabapathy and Pasko are combinable because they are from the same field of endeavor of information extraction using machine learning models. Each discloses receiving an input object, selecting among multiple model option, invoking those models with model-specific inputs and returning structured values. It would have been obvious to a persona of ordinary skill in the art before the effective filing date to implement Sabapathy’s schema-based extraction with a second intent-oriented model and the domain-based routing taught by Pasko to improve accuracy and latency when extracting multiple fields from the same object. The motivation for doing so is “Optimizing the utilization of local computing resources (e.g., processing resources, etc.) of the speech interface device 102 can a reduce latency so that the user experience with the speech interface device 102 is not negatively impacted by local processing tasks taking too long.” As disclosed by Pasko in Column 4 lines 46-50.
Regarding Claim 6:
A system according to Claim 1, the processing unit to execute the processor-executable program code to cause the system to:
receive a second object on which to perform entity extraction (Sabapathy: p[0019] discloses one of the purposes is for storing multiple data objects with more efficiency, Examiner Interpretation {Sabapathy places no limit on the number of objects that analyzed});
from the plurality of extraction schema instances, identify a second extraction schema instance associated with the second object (Sabapathy: p[0020] selects a hybrid class from a classification space based on the type of input object, this matches the claim step of identifying which schema to use for the second object);
determine the first extraction field (Sabapathy: discloses identifying dynamic data fields in the selected template, these are extraction fields, associates with a rule-based mapping to data sources like APIs) and the first model associated with the first extraction field based on the second extraction schema instance (Pasko: Col 15:8-22 within the selected domain (schema instance), the system determines fields (referred to as slots) and uses the domain specific NER model associated with those slots which as previously reasoned in other claims can be done any number of times and are interpreted as a first model associated with the first extraction schema instance);
generate a third input payload according to the input format of the first model (Pasko: Col 23:3-9 discloses exactly what the model expects as input, the NER models take in ASR text (or audio), the IC models take ASR text + NER result. Preparing and routing that specific structure is generating an input payload according to the model’s input format, in this case the third input payload would be the input to the NER model if deemed necessary by the threshold fallback system);
transmit the third input payload to the first model (Pasko: Col 23:3-9 generates input for the models; Col 21:7-Col 22 line 17 discloses there can be any number of domain classifications, and that pruning steps may occur or not occur on how iterations these models will go through i.e., transmitting this third input payload to a first model (NER model));
receive a fourth value of the first extraction field output by the first model (Pasko: Col 26:37-50 discloses the first model (NER) outputs slot data, this is a value of an extraction field and if a second iteration were to occur could reasonably by interpreted as a fourth value);
determine a third extraction field and a third model associated with the third extraction field based on the second extraction schema instance (Pasko: Col 16:42-57 discloses a at least a third model, domain specific IC models associated with another field for figuring out intent of the NER that it can fallback or switch to which are interpreted as third models);
generate a fourth input payload according to the input format of the third model (Pasko: Col 23:3-9 discloses, the IC model (which in this multiple iteration scenario is interpreted as a third model as a new domain has been choses) takes ASR text + NER result. Preparing and routing that specific structure is generating an input payload according to the model’s input format);
transmit the fourth input payload to the third model (Pasko: Col 26:37-50 discloses transmitting this data to a domain specific NER model meaning the data was transmit); and
receive a fifth value of the third extraction field output by the third model (Pasko: Col 26:37-50 discloses the third model (IC model) outputs intent data and confidence results and in this specific iteration is interpreted as a fifth extraction value received from the third extraction field by the third model which is an IC model).
Sabapathy and Pasko are combinable because they are from the same field of endeavor of information extraction using machine learning models. Each discloses receiving an input object, selecting among multiple model option, invoking those models with model-specific inputs and returning structured values. It would have been obvious to a persona of ordinary skill in the art before the effective filing date to implement Sabapathy’s schema-based extraction with a second intent-oriented model and the domain-based routing taught by Pasko to improve accuracy and latency when extracting multiple fields from the same object. The motivation for doing so is “Optimizing the utilization of local computing resources (e.g., processing resources, etc.) of the speech interface device 102 can a reduce latency so that the user experience with the speech interface device 102 is not negatively impacted by local processing tasks taking too long.” As disclosed by Pasko in Column 4 lines 46-50.
Regarding Claim 7:
A system according to Claim 1, the processing unit to execute the processor-executable program code to cause the system to:
receive a second object on which to perform entity extraction (Sabapathy: p[0019] discloses one of the purposes is for storing multiple data objects with more efficiency, Examiner Interpretation {Sabapathy places no limit on the number of objects that analyzed});
from the plurality of extraction schema instances, identify a second extraction schema instance associated with the second object (Sabapathy: p[0020] selects a hybrid class from a classification space based on the type of input object, this matches the claim step of identifying which schema to use for the second object);
determine the first extraction field (Sabapathy: discloses identifying dynamic data fields in the selected template, these are extraction fields, associates with a rule-based mapping to data sources like APIs) and the first model associated with the first extraction field based on the second extraction schema instance (Pasko: Col 15:8-22 within the selected domain (schema instance), the system determines fields (referred to as slots) and uses the domain specific NER model associated with those slots which as previously reasoned in other claims can be done any number of times and are interpreted as a first model associated with the first extraction schema instance);
generate a third input payload according to the input format of the first model (Pasko: Col 23:3-9 discloses exactly what the model expects as input, the NER models take in ASR text (or audio), the IC models take ASR text + NER result. Preparing and routing that specific structure is generating an input payload according to the model’s input format, in this case the third input payload would be the input to the NER model if deemed necessary by the threshold fallback system);
transmit the third input payload to the first model (Pasko: Col 23:3-9 generates input for the models; Col 21:7-Col 22 line 17 discloses there can be any number of domain classifications, and that pruning steps may occur or not occur on how iterations these models will go through i.e., transmitting this third input payload to a first model (NER model));
receive a fourth value of the first extraction field output by the first model (Pasko: Col 26:37-50 discloses the first model (NER) outputs slot data, this is a value of an extraction field and if a second iteration were to occur could reasonably by interpreted as a fourth value);
determine the second extraction field and a third model associated with the second extraction field based on the second extraction schema instance (Pasko: Col 16:42-57 discloses a at least a third model, domain specific IC models associated with another field for figuring out intent of the NER that it can fallback or switch to which are interpreted as third models);
generate a fourth input payload according to the input format of the third model; transmit the fourth input payload to the third model (Pasko: Col 23:3-9 discloses, the IC model (which in this multiple iteration scenario is interpreted as a third model as a new domain has been choses) takes ASR text + NER result. Preparing and routing that specific structure is generating an input payload according to the model’s input format); and
receive a fifth value of the second extraction field output by the third model (Pasko: Col 26:37-50 discloses the third model (IC model) outputs intent data and confidence results and in this specific iteration is interpreted as a fifth extraction value received from the second extraction field by the third model which is an IC model).
Sabapathy and Pasko are combinable because they are from the same field of endeavor of information extraction using machine learning models. Each discloses receiving an input object, selecting among multiple model option, invoking those models with model-specific inputs and returning structured values. It would have been obvious to a persona of ordinary skill in the art before the effective filing date to implement Sabapathy’s schema-based extraction with a second intent-oriented model and the domain-based routing taught by Pasko to improve accuracy and latency when extracting multiple fields from the same object. The motivation for doing so is “Optimizing the utilization of local computing resources (e.g., processing resources, etc.) of the speech interface device 102 can a reduce latency so that the user experience with the speech interface device 102 is not negatively impacted by local processing tasks taking too long.” As disclosed by Pasko in Column 4 lines 46-50.
Regarding Claim 8:
Claim 8 has been analyzed with regard to claims 1 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 9:
Claim 9 has been analyzed with regard to claims 2 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 10:
Claim 10 has been analyzed with regard to claims 3 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 12:
Claim 12 has been analyzed with regard to claims 5 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 13:
Claim 13 has been analyzed with regard to claims 6 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 14:
Claim 14 has been analyzed with regard to claims 7 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 15:
Claim 15 has been analyzed with regard to claims 1 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 16:
Claim 16 has been analyzed with regard to claims 2 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 18:
Claim 18 has been analyzed with regard to claims 5 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 19:
Claim 19 has been analyzed with regard to claims 6 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Regarding Claim 20:
Claim 20 has been analyzed with regard to claims 7 (see rejection above) and is rejected for the same reasons of obviousness as used above.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to IAN SCOTT MCLEAN whose telephone number is (703)756-4599. The examiner can normally be reached "Monday - Friday 8:00-5:00 EST, off Every 2nd Friday".
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, Hai Phan can be reached at (571) 272-6338. 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.
/IAN SCOTT MCLEAN/Examiner, Art Unit 2654
/HAI PHAN/Supervisory Patent Examiner, Art Unit 2654