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 .
This non-final rejection is in response to the RCE filed on: 06/02/2026.
The following rejections are withdrawn in view of applicant’s amendments:
Claims 1-20 rejected under 35 U.S.C. 101.
Claim(s) 1, 3--8 , 10-15, and 17-20 rejected under 35 U.S.C. 103 as being unpatentable over Sahayaraj et al (US Application: US 2022/0180068, published: Jun. 9, 2022, filed: Dec. 7, 2020) in view of Ahmed et al (US Application: US 20240126995, published: Apr. 18, 2024, filed: Oct. 12, 2022).
Claim(s) 2, 9 and 16 rejected under 35 U.S.C. 103 as being unpatentable over Sahayaraj et al (US Application: US 2022/0180068, published: Jun. 9, 2022, filed: Dec. 7, 2020) in view of Ahmed et al (US Application: US 20240126995, published: Apr. 18, 2024, filed: Oct. 12, 2022) in view of Gaur et al (US Application: US 20180341637, published: Nov. 29, 2018, filed: May 24, 2017).
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 06/02/2026 has been entered.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 3--8 , 10-15, and 17-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sahayaraj et al (US Application: US 2022/0180068, published: Jun. 9, 2022, filed: Dec. 7, 2020) in view of Ahmed et al (US Application: US 20240126995, published: Apr. 18, 2024, filed: Oct. 12, 2022) in view of Friedman et al (US Application: US 2023/0316003, published: Oct. 5, 2023, filed: Mar. 7 , 2023).
With regards to claim 1, Sahayaraj et al a computer-implemented method comprising:
identifying, by one or more processors and within a syntactic debiased document (Fig. 2, Fig. 5: processor and memory is implemented to perform bias processing and correction steps and the document could have undergone at least one round of debiasing of portion(s) of document text and can undergo subsequent iterations of debiasing), one or more document segments that each comprise a sequence of terms from a … document (paragraph 0033: document word segments are generated in the form of vectors);
identifying, by the one or more processors, candidate semantic bias terms from a document segment of the one or more document segments based on a semantic bias corpus (paragraph 0033 – 0036, 0041-43: a ‘first word embedding’ contains data about how a word (‘first word’) relates to other words in the document and the embedding is compared to secondary embedding(s) to identify whether the ‘first word’ is considered a candidate bias term via a model and determined bias metric );
in response to the identifying of the candidate semantic bias , generating, by the one or more processors and using a classification model, a bias classification for the document segment (paragraph 0041, 0043: the segment can be classified/labeled as being above a bias threshold and as requiring ‘debiasing’ while accounting for context of other terms within the document and the segment(s)/subset(s) of text is/are highlighted ); and
in response to the bias classification indicating a positive bias classification, providing, by the one or more processors and using a semantic debiasing model, one or more replacement tokens for the one or more candidate semantic bias terms … automatically replacing, by the one or more processors, the candidate semantic bias term with the replacement token within the syntactic debiased document … (paragraph 0029, 0034, 0039, 0041, 0043, 0044, 0049: a positive bias indicating a degree/magnitude of bias includes a number and a ‘X’. A candidate term/text can be suggested to the user as highlighted and option(s) for substitution of presented identified /highlighted subset of text are provided (based upon syntactic and semantic processing)) … identify the replacement token for the candidate semantic bias term based on a position of a masked token corresponding to the candidate semantic bias term within a tokenized subset of the syntactic debiased document (Fig 2: masked text (position of characters associated with token-text) can be updated with a replacement token for the candidate bias term).
However although Sahayaraj et al teaches identifying, … a document segment; … generating … using a classification model, … wherein the classification model comprises a first machine learning model that is trained: (i) using a training dataset that includes a training document segment assigned with a semantic context label corresponding to one of a plurality of semantic contexts based on a context corresponding to context of use of at least one term within the training document segment, and (ii) to classify the document segment into one of the plurality of semantic contexts to generate the bias classification based on which of the plurality of semantic contexts the document segment is classified into; … wherein the semantic debiasing model comprises a second machine learning model configured to identify the replacement token for the candidate semantic bias term within a tokenized subset of the syntactic debiased document; … wherein the replacement token preserves context of the document segment
Yet Ahmed et al teaches identifying … a document segment; … generating … using a classification model, … wherein the classification model comprises a first machine learning model (Ahmed et al, paragraphs 0011, 0046, 0047, 0058 and 0059: classification can be implemented using a BERT learning model which can take in text tokenized and represented in vector format ) that is trained using a training dataset that includes a training document segment assigned with a semantic context label based on a semantic context corresponding to context of use of at least one term within the training document segment (Ahmed et al, paragraphs 0042 and 0045, training data (document(s)) for the classification model includes for particular terms in the training data/document are associated with semantic labels of positive bias or non-bias (negative for bias identification) based on how term(s) are used within context of an area/domain of corporate communication document(s). The examiner notes interpretation of negative bias in the manner of the explanation/citation above is consistent with the instant application’s explanation of negative bias in paragraph 0121 of the instant application); … wherein the semantic debiasing model comprises a second machine learning model configured to identify the replacement token for the candidate semantic bias term within a tokenized subset of the syntactic debiased document (Ahmed et al, paragraph 0039 and 0042: a second machine learning engine can be implemented to identify replacement token for a term of a plurality of terms from the document to replace a bias term with a non-bias term (based on score/weighting of the replacement terms amongst other terms)).
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to have modified Sahayaraj et al’s ability to generate segments from a document to perform document processing/correction to debias (via highlight of candidate terms and selectable alternative/replacement terms) terms in the document (using semantic and syntactic processing/analysis), such that the document could have undergone initial iterative document processing debiasing and then further implementation of first and second machine learning models for classification and replacement, as taught by Ahmed et al. The combination would have allowed Sahayaraj et al to have implemented an effective and efficient way to recognize and control unconscious bias (Ahmed et al, paragraph 0002).
However the combination does not expressly teach … a semantic context label corresponding to one of a plurality of semantic contexts …, and (ii) to classify the document segment into one of the plurality of semantic contexts to generate the bias classification based on which of the plurality of semantic contexts the document segment is classified into … ; … wherein the replacement token preserves context of the document segment.
Yet Friedman et al teaches … a semantic context label corresponding to one of a plurality of semantic contexts …, and (ii) to classify the document segment into one of the plurality of semantic contexts to generate the bias classification based on which of the plurality of semantic contexts the document segment is classified into … ; … wherein the replacement token preserves context of the document segment (paragraphs 0091, 0092, 0093, 0110, 0117, 0121, 0129, 0163, Fig. 8, 9: different semantic labels relating to entity and attributes are used in training to classify document segments/spans of text , and subsequently predict bias (bias classification/identification is made based upon the labels). It is noted that due to the training and analysis taking into account a span of text in order to make a judgement/classification of bias, then the replacement performed accounting for context as an act of ‘preserve’ consideration of context).
It would have been obvious to one or ordinary skill in the art before the effective filing of the invention to have modified Sahayaraj and Ahmet et al’s ability to perform iterative debiasing of terms in a document using semantic and syntactic processing/analysis via a trained bias identification model, such that a context label of a plurality of semantic contexts can be included as part of the training to implement a replacement, as taught by Friedman et al. The combination would have implemented an efficient way to recognize bias present within text and providing an opportunity to address the recognized bias.
With regards to claim 3 , which depends on claim 1, the combination of Sahayaraj, Ahmed et al and Friedman et al teaches wherein the first machine learning model is previously trained based on semantic bias criteria defining the plurality of semantic contexts for a prediction domain, as explained in the rejection of claim 1 and is repeated here for ease of reference: “Ahmed, paragraphs 0042 and 0045, training data (document(s)) for the classification model includes for particular terms in the training data/document are associated with semantic labels of positive bias or non-bias (negative for bias identification) based on how term(s) are used within context of an area/domain of corporate communication document(s). The examiner notes interpretation of negative bias in the manner of the explanation/citation above is consistent with the instant application’s explanation of negative bias in paragraph 0121 of the instant application)”).
With regards to claim 4, which depends on claim 3, the combination of Sahayaraj, Ahmed et al and Friedman et al teaches wherein the semantic bias criteria defines the positive bias classification and a negative bias classification, as similarly explained in the rejection of claim 3 above, and is rejected under similar rationale.
With regards to claim 5. The computer-implemented method of claim 4, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein providing, using the semantic debiasing model, the replacement token for the candidate semantic bias term comprises: identifying a subset of document segments within the syntactic debiased document; and generating, using a tokenizer model, semantic debiasing model, the one or more replacement tokens for the one or more candidate semantic bias terms based on the subset of document segments, (as similarly explained in the rejection of claim 1, the teachings of Sahayaraj et al, Ahmed et al and Friedman et al were combined to address these limitations, which allow highlighting of candidate terms and selection of one or more alternative /replacement terms for the candidate terms), and is rejected under similar rationale.
With regards to claim 6. The computer-implemented method of claim 1, Sahayaraj teaches wherein the replacement token is selected from one or more candidate replacement tokens based on comparing the one or more candidate replacement tokens with the semantic bias corpus (paragraph 0041, 0043, 0044, 0049: a positive bias indicating a degree/magnitude of bias includes a number and a ‘X’. A candidate term/text can be suggested to the user as highlighted and option(s) for substitution of presented identified /highlighted subset of text are provided. It is noted as explained in paragraphs 0039 and 0043, bias graphs are compared to determine bias scores/degrees and a second bias graph (semantic bias corpus data) is further referenced to produce better alternative text suggestions).
With regards to claim 7. The computer-implemented method of claim 6, Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein generating the one or more candidate replacement tokens comprises: assigning a relevancy score to a candidate replacement token from the one or more candidate replacement tokens for the one or more candidate semantic bias terms (as similarly explained in the rejection of claim 1, paragraph 0039 of Ahmed was explained to show that that candidate is selected based upon weightings amongst candidates, and is rejected under similar rationale).
With regards to claim 8, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches a computing system comprising: one or more processors; and one or more memories storing processor-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising: identifying within a syntactic debiased document, a document segment that comprises a sequence of terms; identifying candidate semantic bias term from the document segment based on a semantic bias corpus; in response to the identifying of the one or more candidate semantic bias term, generating, using a classification model, a bias classification for the document segment, wherein the classification model comprises a first machine learning model that is trained: (i) using a training dataset that includes a training document segment assigned with a semantic context label corresponding to one of a plurality of semantic contexts based on context of use of at least one term within the training document segment, and (ii) to classify the document segment into one of the plurality of semantic contexts to generate the bias classification based on which of the plurality of semantic contexts the document segment is classified into; and in response to the bias classification indicating a positive bias classification providing, using a semantic debiasing model, a replacement token for the candidate semantic bias term, wherein the semantic debiasing model comprises a second machine learning model configured to identify the replacement token for the candidate semantic bias term based on a position of a masked token corresponding to the candidate semantic bias term within a tokenized subset of the syntactic debiased document; and automatically replacing the candidate semantic bias term with the replacement token within the syntactic debiased document to generate a debiased document, wherein the replacement token preserves context of the document segment, as similarly explained in the rejection of claim 1, and is rejected under similar rationale.
With regards to claim 10 , which depends on claim 8, the combination of Sahayaraj, Ahmed et al and Friedman et al teaches wherein the first machine learning model is previously trained based on semantic bias criteria defining the plurality of semantic contexts for prediction domain, as explained in the rejection of claim 3, and is rejected under similar rationale.
With regards to claim 11, which depends on claim 10, the combination of Sahayaraj, Ahmed et al and Friedman et al teaches wherein the semantic bias criteria defines the positive bias classification and a negative bias classification, as similarly explained in the rejection of claim 4 above, and is rejected under similar rationale.
With regards to claim 12. The computer-implemented method of claim 11, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein providing, using the semantic debiasing model, the replacement token for the candidate semantic bias term comprises: identifying a subset of document segments within the syntactic debiased document; and generating, using a tokenizer model, semantic debiasing model, the one or more replacement tokens for the one or more candidate semantic bias terms based on the subset of document segments, as similarly explained in the rejection of claim 5, and is rejected under similar rationale.
With regards to claim 13. The computing system of claim 8, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein the replacement token is selected from one or more candidate replacement tokens based on comparing the one or more candidate replacement tokens with the semantic bias corpus, as similarly explained in the rejection of claim 6, and is rejected under similar rationale.
With regards to claim 14. The computing system of claim 13, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein generating the one or more candidate replacement tokens comprises : assigning a relevancy score to a candidate replacement token from the one or more candidate replacement tokens, as similarly explained in the rejection of claim 7, and is rejected under similar rationale.
With regards to claim 15 the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches one or more non-transitory computer-readable media storing processor-executable instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising: identifying within a syntactic debiased document, a document segment that comprises a sequence of terms; identifying a candidate semantic bias term from the document segment based on a semantic bias corpus; in response to the identifying of the candidate semantic bias term, generating, using a classification model, a bias classification for the document segment, wherein the classification model comprises a first machine learning model that is trained: (i) using a training dataset includes a training document segment assigned with a semantic context label corresponding to one of a plurality of semantic contexts based on context of use of at least one term within the training document segment, and (ii) to classify the document segment into one of the plurality of semantic contexts to generate the bias classification based on which of the plurality of semantic contexts the document segment is classified into; in response to the bias classification indicating a positive bias classification, providing, using a semantic debiasing model, a replacement token for the candidate semantic bias term, wherein the semantic debiasing model comprises a second machine learning model configured to identify the replacement token for the candidate semantic bias term based on a position of a masked token corresponding to the candidate semantic bias term within a tokenized subset of the syntactic debiased document; and automatically replacing the candidate semantic bias term with the replacement token within the syntactic debiased document to generate a debiased document, wherein the replacement token preserves context of the document segment as similarly explained in the rejection of claim 1, and is rejected under similar rationale.
With regards to claim 17 , which depends on claim 15, the combination of Sahayaraj, Ahmed et al and Friedman et al teaches wherein the first machine learning model is previously trained based on semantic bias criteria defining the plurality of semantic contexts for a prediction domain, as explained in the rejection of claim 3, and is rejected under similar rationale.
With regards to claim 18, which depends on claim 17, the combination of Sahayaraj, Ahmed et al and Friedman et al teaches wherein the semantic bias criteria defines the positive bias classification and a negative bias classification, as similarly explained in the rejection of claim 4 above, and is rejected under similar rationale.
With regards to claim 19. The computer-implemented method of claim 18, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein providing, using the semantic debiasing model, the replacement token for the candidate semantic bias term comprises: identifying a subset of document segments within the syntactic debiased document; and generating, using a tokenizer model, semantic debiasing model, the one or more replacement tokens for the one or more candidate semantic bias terms based on the subset of document segments, as similarly explained in the rejection of claim 5, and is rejected under similar rationale.
With regards to claim 20. The computing system of claim 19, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein the replacement token is selected from one or more candidate replacement tokens based on comparing the one or more candidate replacement tokens with the semantic bias corpus, as similarly explained in the rejection of claim 6, and is rejected under similar rationale.
Claim(s) 2, 9 and 16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Sahayaraj et al (US Application: US 2022/0180068, published: Jun. 9, 2022, filed: Dec. 7, 2020) in view of Ahmed et al (US Application: US 20240126995, published: Apr. 18, 2024, filed: Oct. 12, 2022) in view of Friedman et al (US Application: US 2023/0316003, published: Oct. 5, 2023, filed: Mar. 7 , 2023) in view of Gaur et al (US Application: US 20180341637, published: Nov. 29, 2018, filed: May 24, 2017).
With regards to claim 2. The computer-implemented method of claim 1, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches wherein the syntactic debiased document is previously generated using syntactic debiasing criteria by: … ; generating a corresponding non-bias term for the syntactic bias term based on the syntactic debiasing criteria; and generating the syntactic debiased document by replacing the syntactic bias term with the corresponding non-bias term (as similarly explained in the rejection of claim 1, the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches a non-bias term is generated for replacement of a bias term via semantic and syntactic processing/analysis, and is rejected under similar rationale)..
However the combination of Sahayaraj et al, Ahmed et al and Friedman et al teaches identifying a syntactic bias term in a grammar corrected document; … generating the syntactic debiased document by replacing the syntactic bias term with the corresponding non-bias term within the grammar corrected document.
Yet Gaur et al teaches identifying a syntactic bias term in a grammar corrected document; … generating the syntactic debiased document by replacing the syntactic bias term with the corresponding non-bias term within the grammar corrected document (Fig 1B, Fig. 2, paragraphs 0014 and 0015 of Gaur et al teaches a pronoun term ‘he’ is identified in a document (that could have undergone grammar processing/correction first) based on selection of ‘Uncon Bias’ as a follow-up proofread option/action. Non-bias alternative terms are provided such as ‘he/she’, ‘they’ and ‘the selected candidate’ and the document can be updated/debiased when the user selects one of the alternate terms).
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to have modified Sahayaraj et al, Ahmed et al and Friedman et al’s ability to generate segments from a document to perform document processing/correction to debias (via highlight of candidate terms and selectable alternative/replacement terms) terms in the document (using semantic and syntactic processing/analysis), such that the document could have undergone initial grammar correction prior to replacement of a bias term, as taught by Gaur et al. The combination would have allowed Sahayaraj et al, Ahmed et al and Friedman et al to have detected instances of unconscious bias and brought [them] to the user's attention and presented alternative suggestions to the user to avoid the unconscious bias (Gaur et al, paragraph 0007).
With regards to claim 9. The computing system of claim 8, the combination of Sahayaraj et al, Ahmed et al, Friedman et al and Gaur et al teaches wherein the syntactic debiased document is previously generated using syntactic debiasing criteria by: identifying a syntactic bias term in a grammar corrected document; generating a corresponding non-bias term for the syntactic bias term based on the syntactic debiasing criteria; and generating the syntactic debiased document by replacing the syntactic bias term with the corresponding non-bias term within the grammar corrected document, as similarly explained in the rejection of claim 2, and is rejected under similar rationale.
With regards to claim 16. The one or more non-transitory computer-readable media of claim 15, the combination of Sahayaraj et al, Ahmed et al, Friedman et al and Gaur et al teaches wherein the syntactic debiased document is previously generated using syntactic debiasing criteria by: identifying a syntactic bias term in a grammar corrected document; generating a corresponding non-bias term for the syntactic bias term based on the syntactic debiasing criteria; and generating the syntactic debiased document by replacing the syntactic bias term with the corresponding non-bias term within the grammar corrected document, as similarly explained in the rejection of claim 2, and is rejected under similar rationale.
Response to Arguments
Applicant's arguments filed 06/02/2026 have been fully considered but they are not persuasive.
With regards to claim 1, the applicant argues ‘Ahmed does not disclose the corresponding label is a semantic context label corresponding to one of a plurality of semantic contexts based on context of use of at least one term within the phrases’. However the examiner respectfully points out that the applicant appears to be requiring an aspect of ‘context of use’ that is not present in the claim language. More specifically, the applicant appears to be arguing that ‘context of use’ is somehow assessed in a certain manner (maybe through an assessment of an adjacent or surrounding term). Yet the claim language only require one term (interpreted as any one term) in a phrase, which can be interpret to encompass a single term that is associated with a ‘biased’ or ‘not biased’ semantic label, and Ahmed was disclosed to teach this in paragraphs 0042 and 0045, where training data (document(s)) for the classification model includes for particular terms in the training data/document are associated with semantic labels of positive bias or non-bias (negative for bias identification) based on how term(s) are used within context of an area/domain of corporate communication document(s). With regards to the amended ‘plurality of semantic contexts’, the examiner has applied a new reference (Friedman et al) which recognizes a plurality of semantic labels corresponding to a plurality of entity and attribute contexts. The examiner respectfully directs the applicant to the rejection of claim 1 above for how the newly amended claim is now rejected under this new combination (Sahayaraj, Ahmet et al and Friedman et al).
The applicant argues that Sahayaraj does not disclosed ‘masked token corresponding to the candidate semantic bias term’ because Sahayaraj discloses masking text such that the biased text is visible. This argument is not persuasive because masking can be interpreted to encompass identifying a position/range of the biased text. This interpretation is consistent with applicant’s own specification which explains that masking involves identifying the position of the token (see paragraph 0123). Since the claim does not require that masking involves anything about visibility, then the applicant’s argument is not persuasive, and the claim’s limitation (about ‘masked’) is maintained to be rejected using at least Sahayaraj (for identifying a range/position of the text).
The applicant argues claims 8 and 15 are allowable for reasons presented by the applicant for claim 1. However this argument is not persuasive since claim 1 has been shown/explained to be rejected above.
The applicant argues claims that depend directly or indirectly upon independent claims 1, 8 or 15 are allowable, however this argument is not persuasive since claims 1, 8 and 15 have been shown/explained to be rejected above.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to WILSON W TSUI whose telephone number is (571)272-7596. The examiner can normally be reached Monday - Friday 9 am -6 pm.
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, Adam Queler can be reached at (571) 272-4140. 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.
/WILSON W TSUI/Primary Examiner, Art Unit 2172
Prosecution Insights