DETAILED ACTION
This examination is in response to the communication filed on 01/21/2026. Claims 1-20 are currently pending, where claims 1, 9 and 17 are independent.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Applicant's arguments filed 01/21/2026 have been fully considered but they are not persuasive.
Applicant argues on page 11 that “in contrast to the Applicant’s claimed invention, in Pedersen the training space is used to classify only English language content because only English langue content is embedded in the training space…Wick does not teach or suggest using a trained embedding space to classify English and non-English content. Instead Wick only teaches using a multilingual embedding space for natural language processing such as translating content” Applicant further asserts that “any combination of Pedersen and Wick at most would teach a method and apparatus for classification of text (as taught by Pedersen) in which non-English language is first translated using an embedding space (as taught in Wick) and then classified and moderated (as taught in Pedersen)” Therefore, Applicant concludes the combination of Pedersen and Wick fails to disclose or suggest the claimed “classifying” limitation because the combination would allegedly teach translating before classifying and remediating. The Examiner respectfully disagrees.
First, even if, arguendo, the combination of Pedersen and Wick would have only suggested to one skilled in the art to translate first then classify as asserted by Applicant, such a system would still read the currently pending claims. Independent claim 1 recites “classifying, using a first machine learning system, the content data by projecting the content data into a trained embedding space to determine at least one English-language classification for the content data, wherein the embedding space is trained such that embedded English-language content data and embedded non-English-language data that are similar occur closer in the embedding space…”. Under a broadest reasonable interpretation the language “a trained embedding space to determine” does not preclude the use of multiple embedding spaces, only that at an embedding space is trained as suggested. Accordingly, projecting the non-English language in a first trained embedding space as taught by Wick and then utilizing another embedding space as taught by Pedersen reads on the current claim language.
Second, contrary to Applicant’s assertion on skilled in the art would have appreciated that the multilingual embedding space taught by Wick could have been utilized by the classification system of Pedersen to achieve the predictable result of classifying multilingual content.
Accordingly, the previous rejections are maintained.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1, 2, 5, 6, 8-10, 13, 14, 16-18, 19 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Pedersen (US 2020/0142999 A1; herein “Pedersen”) in view of Wick et al. (US 2016/0350288 A1; “Wick”).
Regarding claims 1, 9 and 17, Pederson teaches a method, an apparatus comprising a processor (Fig. 1, processor(s) 112) and a memory ( Fig. 1, memory 114) accessible to the processor, the memory having stored therein at least one of programs or instructions (Fig. 1, components 122, 124, 128, and 130) executable by the processor to configure the apparatus, and a non-transitory compute readable storage medium ( Fig. 1, memory 114) having stored thereon instructions, for moderating multilingual content data, comprising:
receiving or pulling content data that can include content (Fig. 5 step 502 “Receive unclassified Text (e.g., comments that are to be presented in an online discussion forum”; See also Fig. 6, step 606 and ¶[0060]);
classifying, using a first machine learning system, the content data by projecting the content data into a trained embedding space to determine at least one English-language classification for the content data (¶[0027]-[0028] teaches “The word embedding component 122 may be configured to provide the word embedding vectors 134 as input to a trained machine learning model(s) 136, and to determine, based on output from the trained machine learning model(s) 136, clusters of associated text…the trained machine learning model(s) 136 is trained to make accurate predictions as to text (e.g., words) that can be grouped together into the clusters 138.”; and Fig. 5, steps 504 and 506 and ¶¶[0054]-[0055] teaches “At step 504, the text classifier 128 of the remote system 106 may provide the unclassified text as input to the (second) trained machine learning model(s) 148….At 506, the text classifier 128 may generate, as output from the trained machine learning model(s) 148, a classification of the unclassified text…as one of multiple class labels to obtain classified text (e.g., one or more classified comments 120”; See also Fig. 6, steps 608 and 610 and ¶¶[0061]-[0062] ); and
determining, using a second machine learning system, if the content data violates at least one predetermined moderation rule, wherein the second machine learning system is trained to determine from English-language classifications determined by the first machine learning system if the content data violates moderation rules (Fig. 5, steps 508 and 510 and ¶[0056] teaches “At step 508, the text presenter 130…may cause presentation of the classified text (e.g., the classified comment(s) 120 containing text) on a display(s) of a client machine(s) 104. As shown by sub-block 510, the text moderation component 150…may moderate text of any classified comment(s) 120 classified as a particular class label that is to be moderated…”; see also Fig. 6, step 612 and 614 and ¶[0063]).
Pedersen fails to disclose that the content data include multilingual text, or that the embedding space is trained such that embedded English-language content data and embedded non-English-language content data that are similar occur closer in the embedding space than embedded English-language content data and embedded non-English- language content data that are not similar.
Wick teaches using a multilingual embedding space for natural language processing, more specifically, Wick teaches the embedding space is trained such that embedded English-language content data and embedded non-English-language content data that are similar occur closer in the embedding space than embedded English-language content data and embedded non-English- language content data that are not similar (Fig. 5 and ¶[0021] teaches “…system 10 is configured to train an NLP model in one language and apply that NLP model to a different language. System 10 can use large collections of unlabeled multilingual data to find a common representation in which structure is shared across languages…”; ¶[0039] teaches “an embedding model…is learned that maps each word type to a k-dimensional vector such that the vectors capture syntactic and semantic relations between words in a way that generalizes across the languages”; and ¶[0053] teaches “Constraint updates discussed above at 408 of Fig. 4 are illustrated as a dotted line 502, representing the constrains pulling the vectors for “rouge” and “red” closer together. As these two words are pulled together the context words also move closer”).
Pedersen differs from the claimed invention, as defined in claims 1, 9 and 17, in that Pedersen fails to explicitly disclose utilizing a multilingual embedding space. Multilingual embedding spaces for clustering cross lingual context are known in the art as evidenced by Wick. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified to the embedding space as taught by Pedersen to have been a multilingual embedding space as taught by Pedersen in order detect obfuscation of offensive words using different languages.
Regarding claims 2, 10 and 18, the combination of Pedersen and Wick teaches all of the elements of claims 1, 9 and 17 (see detailed element mapping above). In addition, Pedersen further teaches prohibiting a presentation of the content data related to the at least one English-language classification determined to violate the at least one predetermined moderation rule (¶[0039] teaches “For example, the text 304 of the comment 120(2) may include toxic language, and, as a result, the text classifier 128 will have classified the text 304 of the comment 120(2) as a particular class label: toxic. The text moderation component 150 moderates this text 304 based on its classification so that the user 102 of the client machine 104 is not subjected to the toxic language in the comment 120(2).”).
Regarding claims 5 and 13, the combination of Pedersen and Wick teaches all of the elements of claim 1 (see detailed element mapping above). In addition, Pedersen further teaches the content data comprises English-language content data and wherein at least one English-language classification is determined for the English-language content data using the embedding of the English-language content data (Fig. 2 and ¶[0027] teaches “The word embedding component 122 may be further configured to provide the word embedding vectors 134…and to determine, based on output from the trained machine learning model(s)136, clusters of associated text 138(1)-(p)…” and ¶[0037] teaches “…the computing system 106 can label the corpus of comments 120 appropriately to create labeled comments…”).
Regarding claims 6 and 14, the combination of Pedersen and Wick teaches all of the elements of claims 1 and 9 (see detailed element mapping above). In addition, Wick further teaches the content data comprises non-English- language content data and similar English-language content data is determined for the non-English-language content data by projecting the non-English-language content data into the embedding space, and wherein at least one English-language classification is determined for the non-English-language content data using the determined similar English language content data (Fig. 5 and ¶[0021] teaches “…system 10 is configured to train an NLP model in one language and apply that NLP model to a different language. System 10 can use large collections of unlabeled multilingual data to find a common representation in which structure is shared across languages…”; ¶[0039] teaches “an embedding model…is learned that maps each word type to a k-dimensional vector such that the vectors capture syntactic and semantic relations between words in a way that generalizes across the languages”; and ¶[0053] teaches “Constraint updates discussed above at 408 of Fig. 4 are illustrated as a dotted line 502, representing the constrains pulling the vectors for “rouge” and “red” closer together. As these two words are pulled together the context words also move closer”).
Pedersen differs from the claimed invention, as defined in claims 6 and 14, in that Pedersen fails to explicitly disclose utilizing a multilingual embedding space. Multilingual embedding spaces for clustering cross lingual context are known in the art as evidenced by Wick. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified to the embedding space as taught by Pedersen to have been a multilingual embedding space as taught by Pedersen in order detect obfuscation of offensive words using different languages.
Regarding claims 8, 16 and 20, the combination of Pedersen and Wick teaches all of the elements of claims 1, 9 and 17 (see detailed element mapping above). In addition, Pedersen further teaches soliciting information from at least one source of the received or pulled content data to assist in determining an accuracy of the at least one English-language classification determined for the received or pulled content data (Fig. 3B, element 312 and ¶[0042] teaches “For example, the feedback element 312 may ask the user 102 if the comment 120(2) with the temporarily revealed text is in fact toxic (from the user’s perspective)…”).
Regarding claim 19, the combination of Pedersen and Wick teaches all of the elements of claim 17 (see detailed element mapping above). In addition, Pedersen further teaches if the content data comprises English-language content data, at least one English- language classification is determined for the English-language content data using the embedding of the English-language content data (¶[0027]-[0028] teaches “The word embedding component 122 may be configured to provide the word embedding vectors 134 as input to a trained machine learning model(s) 136, and to determine, based on output from the trained machine learning model(s) 136, clusters of associated text…the trained machine learning model(s) 136 is trained to make accurate predictions as to text (e.g., words) that can be grouped together into the clusters 138.”; and Fig. 5, steps 504 and 506 and ¶¶[0054]-[0055] teaches “At step 504, the text classifier 128 of the remote system 106 may provide the unclassified text as input to the (second) trained machine learning model(s) 148….At 506, the text classifier 128 may generate, as output from the trained machine learning model(s) 148, a classification of the unclassified text…as one of multiple class labels to obtain classified text (e.g., one or more classified comments 120”; See also Fig. 6, steps 608 and 610 and ¶¶[0061]-[0062]) and
Wick further teaches if the content data comprises non-English-language content data, similar English-language content data is determined for the non-English-language content data by projecting the non- English-language content data into the embedding space, and at least one English- language classification is determined for the non-English-language content data using the determined similar English language content data (Fig. 5 and ¶[0021] teaches “…system 10 is configured to train an NLP model in one language and apply that NLP model to a different language. System 10 can use large collections of unlabeled multilingual data to find a common representation in which structure is shared across languages…”; ¶[0039] teaches “an embedding model…is learned that maps each word type to a k-dimensional vector such that the vectors capture syntactic and semantic relations between words in a way that generalizes across the languages”; and ¶[0053] teaches “Constraint updates discussed above at 408 of Fig. 4 are illustrated as a dotted line 502, representing the constrains pulling the vectors for “rouge” and “red” closer together. As these two words are pulled together the context words also move closer”).
Pedersen differs from the claimed invention, as defined in claims 1, 9 and 17, in that Pedersen fails to explicitly disclose utilizing a multilingual embedding space. Multilingual embedding spaces for clustering cross lingual context are known in the art as evidenced by Wick. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified to the embedding space as taught by Pedersen to have been a multilingual embedding space as taught by Pedersen in order detect obfuscation of offensive words using different languages.
Claims 3, 4, 11 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Pedersen and Wick as applied to claims 1, 9 and 17 above, and further in view of Huffman et al. (US 2023/0396457 A1; herein “Huffman”).
Regarding claims 3 and 11, the combination of Pedersen and Wick teaches all of the elements of claims 1 and 9 (see detailed element mapping above). In addition, Pedersen further teaches the content data is presented during a communication session (¶[0038] teaches “an online discussion forum 110 may be presented within the GUI 300, thereby enabling the exchange of comments 120 between the user 102 of the client machine 104 and other users 102 who may be playing the same video game at the same time.”).
In addition, Pedersen teaches providing a user interface which allows the user whose interface is being moderated to temporally view blocked messages and provide feedback. However, the combination of Pedersen and Wick fails to explicitly disclose presenting semantic parameters related to the determined at least one English-language classification to at least one participant of the communication session.
Huffman teaches a method and system for content moderation that includes, inter alia, presenting semantic parameters related to the determined at least one English-language classification to at least one participant of the communication session (¶[0191] teaches “the system provides toxic speech content with an associated toxicity score to the moderator 106. As discussed above, the toxic speech content is scored based on the platform content policy provided in step 304”).
The combination of Pedersen and Wick differs from the claimed invention, as defined in claims 3 and 11, in that the combination fails to disclose displaying semantic parameters related to the moderated content. Displaying semantic information, such as a toxicity score, in association with the respective content being scored is known in the art as evidenced by Huffman. Therefore, it would have been obvious to one having ordinary skilled in the art, before the effective filing date of the invention, to have modified the user interface taught by the combination of Pedersen and Wick to include displaying a toxicity score to the participant as taught by Huffman as it merely constitutes the combination of known processes to achieve the predictable result to providing the participant with the additional information regarding the toxicity level in order to allow the user to determine whether or not to display the moderated content.
Regarding claims 4 and 12, the combination of Pedersen, Wick and Huffman teaches all of the elements of claims 3 and 11 (see detailed element mapping above). In addition, Huffman further teaches the semantic parameters related to the determined at least one English-language classification include at least one of an intent of the received or pulled content data (the “or” makes this element optional), an emotion of the received or pulled content data (the “or” makes this element optional), an offensiveness of the received or pulled content data (the “or” makes this element optional), an abuse level of the received or pulled content data ((¶[0191] teaches “the system provides toxic speech content with an associated toxicity score to the moderator 106. As discussed above, the toxic speech content is scored based on the platform content policy provided in step 304” the toxicity score is interpreted as an abuse level), or a topic of the received or pulled content data (the “or” makes this element optional),.
The combination of Pedersen and Wick differs from the claimed invention, as defined in claims 3 and 11, in that the combination fails to disclose displaying semantic parameters related to the moderated content. Displaying semantic information, such as a toxicity score, in association with the respective content being scored is known in the art as evidenced by Huffman. Therefore, it would have been obvious to one having ordinary skilled in the art, before the effective filing date of the invention, to have modified the user interface taught by the combination of Pedersen and Wick to include displaying a toxicity score to the participant as taught by Huffman as it merely constitutes the combination of known processes to achieve the predictable result to providing the participant with the additional information regarding the toxicity level in order to allow the user to determine whether or not to display the moderated content.
Claims 7 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Pedersen and Wick as applied to claims 1, 9 and 17 above, and further in view of Cecchi et al. (US 9985912 B2; herein “Cecchi”).
Regarding claims 7 and 15, the combination of Pedersen and Wick teaches all of the elements of claims 1 and 9 (see detailed element mapping above). In addition, Pedersen further teaches the content data is presented during a communication session (¶[0038] teaches “an online discussion forum 110 may be presented within the GUI 300, thereby enabling the exchange of comments 120 between the user 102 of the client machine 104 and other users 102 who may be playing the same video game at the same time.”).
However, the combination of Pedersen and Wick fails to explicitly disclose clustering participants of the communication session based on semantic characteristics of respective content data posted to the communication session by each of the participants.
Cecchi teaches a method and system for moderating an online discussion that includes, inter alia, clustering participants of the communication session based on semantic characteristics of respective content data posted to the communication session by each of the participants (Col. 4, lines 44-60 teaches “The clusters repository 110 stored different clusters of graphs that represent different behaviors of participants that previously led to one or more different problems in discussions. Examples of problems in the discussions include dominance, unproductiveness, inappropriateness, insulting, bullying, threatening dynamics… The graphical text analyzing module 106 compares the graph for all participants or the graph for each participant with the clusters of previously generated graphs stored in the clusters repository 110 in order to determine a behavior of the participant”).
The combination of Pedersen and Wick differs from the claimed invention, as defined by claims 7 and 15, in that the combination fails to explicitly disclose clustering participants of the discussion based on their respective comments. Clustering participated based on their comments is known in the art as evidenced by Cecchi. Therefore, it would have been obvious to one having ordinary skill in the art, before the effective filing date of the invention, to have modified the moderation system taught by the combination of Pedersen and Wick to include clustering participants as taught by Cecchi in order to allow moderating actions to change the course of discussion or alleviate potential problems if a participant’s comments representing a determined behavior exceeds a threshold likelihood (Cecchi, col. 5, lines 25-30).
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 PENNY L CAUDLE whose telephone number is (703)756-1432. The examiner can normally be reached M-Th 8:00 am to 5:00 pm eastern.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Daniel Washburn can be reached at 571-272-5551. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/PENNY L CAUDLE/ Examiner, Art Unit 2657
/DANIEL C WASHBURN/ Supervisory Patent Examiner, Art Unit 2657