DETAILED ACTION
This communication is in response to the Amendments and Arguments filed on 30 June 2026. Claims 1-15 are pending and have been examined. The Applicants’ amendment and remarks have been carefully considered, but are moot in view of new grounds for rejection. Hence, this Action has been made FINAL.
All previous objections and rejections directed to the Applicant’s disclosure and claims not discussed in this Office Action have been withdrawn by the Examiner.
Terminal Disclaimer
The terminal disclaimer filed on 30 June 2026 disclaiming the terminal portion of any patent granted on this application which would extend beyond the expiration date of 12271696 has been reviewed and is accepted. The terminal disclaimer has been recorded.
Response to Amendments and Arguments
The 101 rejections have been reconsidered and removed.
The Applicant’s arguments have been considered but are moot because the new ground of rejection (Mao et al.) does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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, 4, 6, 9, 11, and 14 and is/are rejected under 35 U.S.C. 103 as being unpatentable over US 20240419246, hereinafter referred to as Ullrich et al., in view of CN 117556014, hereinafter referred to as Guo et al., and further in view of US 20250131919, hereinafter referred to as Mao et al.
Regarding claim 1 (currently amended), Ullrich et al. discloses a computer system comprising:
a hardware memory (Ullrich et al., fig. 38(3820).), wherein the computer system is configured to execute software instructions stored on nontransitory machine-readable storage media (Ullrich et al., para [1063].) that:
tokenize a set of training data into a plurality of training tokens (“FIG. 8 illustrates an exemplary tokenizer 800 in accordance with one embodiment. Input data 802 may be provided to the exemplary tokenizer 800 in the form of a text string typed by a user. In one embodiment, the text string may be generated by performing voice-to-text conversion on an audio stream. The exemplary tokenizer 800 may detect tokenizable elements 804 within the input data 802,” Ullrich et al., para [0132]. And, “In the above embodiments, it may be understood by one skilled in the art that a record of inputs and responses may be used to retrain and enhance the performance of any of the system components. For example, a record of natural language outputs from the GenAI 118 may be scored based on some external measure and this data may then be used to retrain or fine-tune the GenAI 118,” Ullrich et al., para [0125]. See also fig. 8. Thus, the tokenized elements may be used to train the model.);
create a codeword dictionary by assigning unique codewords to each of the plurality of training tokens (“Each tokenizable element 804 may be converted into a token 806. The set of tokens 806 created from the tokenizable elements 804 of the input data 802 may be sent from the exemplary tokenizer 800 as tokenized output 808. The set of tokens 806 may be such as are used to create the prompt 144 of FIG. 1,” Ullrich et al., para [0134].);
convert all training tokens into a plurality of training codewords using the codeword dictionary (“In some alternate embodiments, multimodal data may be quantized and converted into tokens 806 using a codebook,” Ullrich et al., para [0134]. The codebook is interpreted as a codeword dictionary. And, “In the above embodiments, it may be understood by one skilled in the art that a record of inputs and responses may be used to retrain and enhance the performance of any of the system components. For example, a record of natural language outputs from the GenAI 118 may be scored based on some external measure and this data may then be used to retrain or fine-tune the GenAI 118,” Ullrich et al., para [0125].);
train a large language model using the plurality of training codewords (Ullrich et al., para [0125].);
receive a text prompt from a user (Ullrich et al., fig. 8(802).);
tokenize the prompt into a plurality of prompt tokens (“FIG. 8 illustrates an exemplary tokenizer 800 in accordance with one embodiment. Input data 802 may be provided to the exemplary tokenizer 800 in the form of a text string typed by a user. In one embodiment, the text string may be generated by performing voice-to-text conversion on an audio stream. The exemplary tokenizer 800 may detect tokenizable elements 804 within the input data 802. Each tokenizable element 804 may be converted into a token 806. The set of tokens 806 created from the tokenizable elements 804 of the input data 802 may be sent from the exemplary tokenizer 800 as tokenized output 808. The set of tokens 806 may be such as are used to create the prompt 144 of FIG. 1,” Ullrich et al., para [0132]. See also fig. 8.); and
convert the plurality of prompt tokens into a plurality of prompt codewords using the codeword dictionary (Ullrich et al., fig. 8. The input data 802 is a prompt.).
Ullrich et al., though, does not specifically disclose processing the sequence of prompt codewords through the large language model to generate a codeword response; and convert the codeword response into a text response.
Guo et al. is cited to disclose processing the sequence of prompt codewords through the large language model to generate a codeword response (“Step 105 is a process of processing the above-mentioned complement user problem by using a large language model, and specifically includes: inputting the complement user problem into a trained large language model to obtain an output result of the large language model; the output result comprises: coding a target answer; and searching in the corresponding relation between the answer codes and the answer texts according to the target answer codes to obtain the answer texts corresponding to the target answer codes. For example, in the case where the target answer code is "a0001", the correspondence shown in table 4 may be queried to obtain an answer text corresponding to the target answer code,” Guo et al., Highlight 1. Here, the coded target answer corresponds to a codeword response.); and
convert the codeword response into a text response (Guo et al., Highlight 1. And, “The conversion module 417 is configured to search in a correspondence between the answer codes and the answer text according to the target answer codes, so as to obtain the answer text corresponding to the target answer codes. The conversion module 417 may also output answer text to an answer output device. The answer output device may be an electronic device where the client is located,” Guo et al., Highlight 2.). Guo et al. benefits Ullrich et al. by providing data compression methods for minimizing calculation resources and storage space as the scale of the LLM increases (Guo, Background). Therefore, it would be obvious for one skilled in the art to combine the teachings of Ullrich et al. with those of Guo et al. to improve the human augmentation platform efficiency of Ullrich et al.
Neither Ullrich et al. nor Guo et al., though, disclose wherein the codeword dictionary defines an operating vocabulary of a large language model such that the large language model processes codewords and not natural language tokens; train a large language model using the plurality of training codewords such that the large language model learns statistical relationships between codewords rather than between natural language tokens; and wherein the large language model generates the codeword response as a sequence of codewords drawn from the codeword dictionary.
Mao et al. is cited to disclose wherein the codeword dictionary defines an operating vocabulary of a large language model such that the large language model processes codewords and not natural language tokens (“FIG. 7 illustrates a system 700 that includes an audio encoder 710 that outputs compact speech vectors 720 that are provided to an LLM embedding translator process 730. The LLM embedding translator process 730 converts or translates the acoustic embeddings (a sequence of compact speech vectors 720 and the codebook indices that are derived from the speech vectors) to LLM embeddings. The extracted speech representations preserve rich speech semantics and are separated from background noise. These acoustic embeddings are generated during a fixed temporal window (e.g., every 10 ms),” Mao et al., para [0077].);
train a large language model using the plurality of training codewords such that the large language model learns statistical relationships between codewords rather than between natural language tokens (“At the bottom of FIG. 8 is a pre-trained audio encoder 820. The audio encoder 820 has already been trained using the techniques described above in connection with FIGS. 1 and 2. The audio encoder 820 receives speech audio input 822 and generates compact speech vectors (codeword indices) 824. The translator 830 (which is the same as the LLM tokenizer process 730 shown in FIG. 7) maps codeword indices 824 to language embeddings also known as LLM embeddings. The translator 830 is trained to perform this mapping,” Mao et al., para [0088].); and
wherein the large language model generates the codeword response as a sequence of codewords drawn from the codeword dictionary (“The receive side 104 obtains receive (RX) packets from the network 106. At the receive side 104, there are a jitter buffer 120, vector de-quantizer 122, codebook 124 and an audio decoder 126. The jitter buffer 120 keeps track of the incoming packets, putting them in order and deciding when to process and play a packet. The vector de-quantizer 122 de-quantizes received codeword indices and using the codebook 124, outputs recovered embedding vectors. The audio decoder 126 decodes the embedding vectors to produce an output audio stream. Again, in some implementations, the audio decoder 126 may generate the directly perform vector de-quantization without the need for a separate vector de-quantizer 122,” Mao et al., para [0021].). Mao et al. benefits Ullrich et al. by processing codewords instead of natural language tokens, thereby preserving the acoustic nuances of speech. Therefore, it would be obvious for one skilled in the art to combine the teachings of Ullrich et al. with those of Mao et al. to improve the speech processing functionality of Ullrich et al.
As to claim 6, method claim 6 and system claim 1 are related as system and method of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 6 is similarly rejected under the same rationale as applied above with respect to system claim.
As to claim 11, CRM claim 11 and system claim 1 are related as system and CRM of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 11 is similarly rejected under the same rationale as applied above with respect to system claim. Also, Ullrich et al., para [0449]-[0450], teaches CRM. And, Ullrich et al., para [00435], teaches processor(s).
Regarding claim 4 (original), Ullrich et al., as modified by Guo et al. and Mao et al., discloses the system of claim 1, wherein the large language model uses a transformer architecture (It is well-known in the art that LLMs rely on a transformer architecture, such as GPT. See also Ullrich et al., para [0121].).
As to claim 9, method claim 9 and system claim 4 are related as system and method of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 9 is similarly rejected under the same rationale as applied above with respect to system claim.
As to claim 14, CRM claim 10 and system claim 4 are related as system and CRM of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 14 is similarly rejected under the same rationale as applied above with respect to system claim. Also, Ullrich et al., para [0449]-[0450], teaches CRM. And, Ullrich et al., para [00435], teaches processor(s).
Claim(s) 2-3, 7-8, and 12-13 and is/are rejected under 35 U.S.C. 103 as being unpatentable over US 20240419246hereinafter referred to as Ullrich et al., in view of CN 117556014, hereinafter referred to as Guo et al., further in view of US 20250131919, hereinafter referred to as Mao et al., and further in view of US 20240296031, hereinafter referred to as Shaffer et al.
Regarding claim 2 (currently amended), Ullrich et al., as modified by Guo et al. and Mao et al., discloses the system of claim 1, but not wherein the text prompt is received, tokenized, and converted from tokens to codewords and from codewords back to tokens on an edge device using a locally stored copy of the codeword dictionary resident on the edge device. Schaffer et al. is cited to disclose wherein the text prompt is received, tokenized, and converted from tokens to codewords and from codewords back to tokens on an edge device (“FIG. 3 illustrates an environment in which the executable application 110 and codebook 200 may be deployed. A client system 300, such as an edge computing device which may have limited network bandwidth and computational resources, runs the executable application 110, that comprises the replacement codes 202, to produce compressed output statements 302. Strings in the compressed output statements are replaced with replacement codes 202 in a codebook 200 generated from the source code 102 for the executable application 110 to produce uncompressed output statements. The client system 300 transmits the compressed output statements 302 over a network 304 to a system 306, such as a server or central network hub receiving output statements from different client systems 300 running the executable application 110, or other executable applications whose output code is replaced with replacement codes from different codebooks 200 for the applications 110,” Schaffer et al., para [0020].). Schaffer et al. benefits Ullrich et al. by providing improved computer technology and techniques for reducing the size of generated log files and messages, and other data, without burdening processing and network bandwidth resources (Schaffer et al., Background). Therefore, it would be obvious for one skilled in the art to combine the teachings of Ullrich et al. with those of Schaffer et al. to improve the human augmentation platform efficiency of Ullrich et al.
As to claim 7, method claim 6 and system claim 2 are related as system and method of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 7 is similarly rejected under the same rationale as applied above with respect to system claim.
As to claim 12, CRM claim 12 and system claim 2 are related as system and CRM of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 11 is similarly rejected under the same rationale as applied above with respect to system claim. Also, Ullrich et al., para [0449]-[0450], teaches CRM. And, Ullrich et al., para [00435], teaches processor(s).
Regarding claim 3 (original), Ullrich et al., as modified by Guo et al., Shaffer et al., and Mao et al., discloses the system of claim 2, wherein the codeword dictionary is a local codeword dictionary lookup on the edge device (Shaffer et al., para [0020].) The codebook is a codeword dictionary lookup.).
As to claim 8, method claim 8 and system claim 3 are related as system and method of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 8 is similarly rejected under the same rationale as applied above with respect to system claim.
As to claim 13, CRM claim 13 and system claim 3 are related as system and CRM of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 13 is similarly rejected under the same rationale as applied above with respect to system claim. Also, Ullrich et al., para [0449]-[0450], teaches CRM. And, Ullrich et al., para [00435], teaches processor(s).
Claim(s) 5 is/are rejected under 35 U.S.C. 103 as being unpatentable over US 20240419246hereinafter referred to as Ullrich et al., in view of CN 117556014, hereinafter referred to as Guo et al., further in view of US 20250131919, hereinafter referred to as Mao et al., and further in view of US 20220366898, hereinafter referred to as Qian et al.
Regarding claim 5 (currently amended), Ullrich et al., as modified by Guo et al. and Mao et al., discloses the system of claim 1, but not wherein the large language model uses a latent transformer architecture, wherein the latent transformer architecture applies attention mechanisms over sequences of codewords from the codeword dictionary. Qian et al. is cited to disclose wherein the large language model uses a latent transformer architecture (Qian et al., fig. 1. The bytes are codewords.). Qian et al. benefits Ullrich et al. by training a machine learning model to learn phonetically aware speech representations (Qui et al., para [0005]). Therefore, it would be obvious for one skilled in the art to combine the teachings of Ullrich et al. with those of Qui et al. to improve the speech recognition capabilities of Ullrich et al.
Claim(s) 10 and 15 and is/are rejected under 35 U.S.C. 103 as being unpatentable over US 20240419246 hereinafter referred to as Ullrich et al., in view of CN 117556014, hereinafter referred to as Guo et al., further in view of US 20250131919, hereinafter referred to as Mao et al., and further in view of US 20250356120, hereinafter referred to as Mohajer et al.
Regarding claim 10 (original), Ullrich et al., as modified by Guo et al. and Mao et al., discloses the system of claim 1, but not wherein the large language model uses a variational autoencoder based architecture. Mohajer et al. is cited to disclose wherein the large language model uses a variational autoencoder based architecture (“Model Architecture Selection: The LLM engine 120 may be configured in different model architectures, including for example a transformer architecture, generative adversarial network (GAN), a variational autoencoder (VAE), an autoregressive model, or other types of models designed for generative tasks,” Mohajer et al., para [0023].). Mohajer et al. benefits Ullrich et al. by using parallel processing of individual token groups, which reduces the time to analyze a query and return the results compared to that needed for a conventional serial search of a query. Therefore, it would be obvious for one skilled in the art to combine the teachings of Ullrich et al. with those of Mohajer et al. to improve the human augmentation platform efficiency of Ullrich et al.
As to claim 15, CRM claim 10 and system claim 10 are related as system and CRM of using same, with each claimed element’s function corresponding to the system step. Accordingly claim 15 is similarly rejected under the same rationale as applied above with respect to system claim. Also, Ullrich et al., para [0449]-[0450], teaches CRM. And, Ullrich et al., para [00435], teaches processor(s).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
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 ANNE L THOMAS-HOMESCU whose telephone number is (571)272-0899. The examiner can normally be reached on Mon-Fri 8-6.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Bhavesh Mehta can be reached on 5712727453. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/ANNE L THOMAS-HOMESCU/Primary Examiner, Art Unit 2656
Prosecution Insights