DETAILED ACTION
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 with respect to 35 U.S.C. 101 Abstract Idea in regards to claims 1-20 have been considered but are moot due to new grounds of rejection necessitated by amendments. See detailed reason below.
Applicant's arguments with respect to 35 U.S.C. 102 in regards to claims 1, 2, 5, 10-11, 14 and 19-20 have been considered but are moot due to new grounds of rejection necessitated by amendments. See detailed reason for rejection below.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1, 3-5, 7-10, 12-14, 16-19 and 21-22 are rejected under 35 U.S.C. 101 Abstract Idea.
Claim 1, 10 and 19 are directed to an abstract idea. The claims mainly cover a high-level process of receiving a question, extracting a query, looking up information in a database, and generating a response from that information. The claims that can practically be characterized as collecting information, analyzing it, and presenting a result, or as steps that can be performed as a mental process, are treated as abstract ideas.
The claim recites a “processor,” a “knowledge retriever,” a “dialogue model,” and different “prompt information,” but it does not explain any specific technical way those components improve computer performance or solve a technical problem in computer operation. The claim states the result at a functional level: one prompt is used to get a query, another prompt is used to get a response.
The claims also lack an “inventive concept.” The added elements are conventional components doing ordinary jobs: the processor receives text, the retriever queries a database, and the dialogue model produces outputs based on prompts. The claim does not recite a particular model architecture, a concrete data structure, a specialized retrieval mechanism, or any technical implementation detail that would transform the abstract idea into patent-eligible subject matter.
The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the claims are (i) mere instructions to implement the idea on a computer, and/or (ii) recitation of generic computer structure that serves to perform generic computer functions that are well-understood, routine, and conventional activities previously known to the pertinent industry. Viewed as a whole, these additional claim element(s) do not provide meaningful limitation(s) to transform the abstract idea into a patent eligible application of the abstract idea such that the claim(s) amounts to significantly more than the abstract idea itself. Therefore, the claim(s) are rejected under 35 U.S.C. 101 as being directed to non-statutory subject matter. There is further no improvement to the computing device.
Dependent claims 3-5, 7-9, 12-14, 16-18 and 21-22 further recite an abstract idea performable by a human and do not amount to significantly more than the abstract idea as they do not provide steps other than what is conventionally known.
Claims 3 and 12, this just adds that the system gets the query by feeding the dialogue text and a prompt into a generic model, which is still just abstract information processing done on a computer.
Claims 4 and 13, this just adds that the system gets the response by feeding the retrieved information, dialogue text, and a prompt into a generic model, which is still the abstract idea of using information to produce an answer.
Claims 5 and 14, this only splits the same abstract query-and-answer idea into a query network and a response network, without claiming a concrete technological improvement in computer functionality.
Claims 7 and 16, this adds searching records, ranking them by correlation, and picking the top results, which is classic data collection, analysis, and selection—an abstract idea unless tied to a specific technical improvement.
Claims 8 and 17, this just adds training the model with example dialogue/query and dialogue/knowledge/response data, which is still generic model training for the same abstract information-processing idea, not a specific improvement to how the computer operates.
Claims 9 and 18, this only says the query side and response side are trained separately, which is still just organizing generic AI training steps rather than claiming a particular technological improvement.
Claims 21 and 22, this adds classifying the query into a field and routing it to the matching database, which is organizing and analyzing information, not improving the computer itself.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 3-5, 8-9, 10, 12-14 and 17-19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Campbell et al. (US 2019/0278792) in view of Shuster et al. (“Large Models that Seek for Knowledge: Modular Search & Generation for Dialogue and Prompt Completion”; Mar. 29, 2022; Facebook AI Research).
Claims 1, 10 and 19,
Campbell teaches a method for processing a dialogue, performed by a device comprising a processor ([0066] dialog handler component 402 includes a processor), a knowledge retriever, and a Query-driven Task-oriented Dialogue System (Q-TOD) ([0003] task-oriented dialog systems provide a computer-based interface for explaining information in a repository (e.g. database) to a user via a dialog),
obtaining, by the processor, a dialogue text of the dialogue, wherein the dialogue text comprises a current question text, or the dialogue text comprises a current question text and a historical dialogue text ([0077] a context is provided that encodes the current user utterance and the history of user-system utterances);
obtaining, by the knowledge retriever, a knowledge query result for the current query text by querying a knowledge database based on the current query text ([0067] [0070] [0085] parsing the user’s query and searching database 410 efficiently for information pertinent to the user’s intent; the query action queries the database; returning information from the database 410 that answers the user’s query).
The difference between the prior art and the claimed invention is that Campbell does not explicitly teach a query generator and a response generator, the method comprising: extracting, by the query generator, a current query text from the dialogue text; determining, by the response generator, a response text for the current question text based on the knowledge query result and the dialogue text; wherein the method further comprises: using a same dialogue model by the query generator and by the response generator to obtain the current query text and to determine the response text respectively based on different prompt information, wherein in a case where an input of the dialogue model does not comprise the knowledge query result, the prompt information is first prompt information configured to prompt the dialogue model to extract the current query text, and in a case where an input of the dialogue model comprises the knowledge query result, the prompt information is second prompt information configured to prompt the dialogue model to generate the response text.
Shuster teaches a query generator and a response generator ([3.] Search Module and Response Module), the method comprising:
extracting, by the query generator, a current query text from the dialogue text ([3.] Search Module; given the encoded input context, a search query is generated);
determining, by the response generator, a response text for the current question text based on the knowledge query result and the dialogue text ([3.] Response Module; given the encoded input context concatenated with the knowledge response, the final response is generated);
wherein the method further comprises: using a same dialogue model by the query generator and by the response generator to obtain the current query text and to determine the response text respectively based on different prompt information ([3.] this same encoder-decoder model is used in a modular way multiple times; see Fig. 1; a single transformer architecture is called successively to invoke three different modules; for each module, special tokens are used to indicate which module is being invoked),
wherein in a case where an input of the dialogue model does not comprise the knowledge query result, the prompt information is first prompt information configured to prompt the dialogue model to extract the current query text ([3.2] [App. C.2] we used control tokens appended to the context; for search tasks, this was “__generate-query__” token), and
in a case where an input of the dialogue model comprises the knowledge query result, the prompt information is second prompt information configured to prompt the dialogue model to generate the response text ([3.2] [App. C.2] we used control tokens appended to the context; for dialogue, we surrounded the concatenated knowledge with “__knowledge__ and __endknowledge__ tokens”).
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the teachings of Campbell with teachings of Shuster by modifying the dialog agent for conducting task-oriented computer-based communications as taught by Campbell to include a query generator and a response generator, the method comprising: extracting, by the query generator, a current query text from the dialogue text; determining, by the response generator, a response text for the current question text based on the knowledge query result and the dialogue text; wherein the method further comprises: using a same dialogue model by the query generator and by the response generator to obtain the current query text and to determine the response text respectively based on different prompt information, wherein in a case where an input of the dialogue model does not comprise the knowledge query result, the prompt information is first prompt information configured to prompt the dialogue model to extract the current query text, and in a case where an input of the dialogue model comprises the knowledge query result, the prompt information is second prompt information configured to prompt the dialogue model to generate the response text for the benefit of generating more factual responses (Shuster [Abstract]).
Claims 3 and 12,
Shuster further teaches the method of claim 1, wherein extracting, by the query generator, the current query text from the dialogue text comprises: obtaining the current query text outputted by the dialogue model by inputting the dialogue text and the first prompt information into the dialogue model ([3.] [3.2] we append special tokens to the input context to indicate that the transformer is performing the search task, via predicting a relevant search query; Search Module – given the encoded input context, a search query is generated).
Claims 4 and 13,
Shuster further teaches the method of claim 1, wherein determining, by the response generator, the response text for the current question text based on the knowledge query result and the dialogue text ([3.] Response Module; given the encoded input context concatenated with the knowledge response, the final response is generated) comprises:
determining second prompt information, wherein the second prompt information is configured to prompt the dialogue model to generate the response text ([3.2] [App. C.2] we used control tokens appended to the context; for dialogue, we surrounded the concatenated knowledge with “__knowledge__ and __endknowledge__ tokens”); and
obtaining the response text outputted by the dialogue model by inputting the knowledge query result, the dialogue text and the second prompt information into the dialogue model (3.2] the context contains the usual dialogue, concatenated to the gold knowledge response … surrounded by special tokens; the new target is the standard dialogue response).
Claims 5 and 14,
Shuster further teaches the method of claim 1, wherein the dialogue model includes a query generation network for obtaining the current query text and a response generation network for determining the response text ([Fig. 1] [3.] a single transformer architecture is called successively to invoke three different modules; search and generate final response; this same encoder-decoder is used in a modular way multiple times);
wherein extracting the current query text from the dialogue text comprises: obtaining the current query text outputted by the query generation network by inputting the dialogue text into the query generation network ([3.] [3.2] Search Module; given the encoded input context, a search query is generated; we append special tokens to the input context via predicting a relevant search query); and
wherein determining the response text for the current question text based on the knowledge query result and the dialogue text comprises: obtaining the response text outputted by the response generation network by inputting the knowledge query result and the dialogue text into the response generation network ([3.] [3.2] Response Module; given the encoded input context concatenated with the knowledge response, the final response is generated; the input context contains the usual dialogue, concatenated to the gold knowledge response; the new target is the standard dialogue response).
Claims 8 and 17,
Shuster further teaches the method of claim 1, wherein the dialogue model is obtained by training an initial dialogue model using a first training sample and the first prompt information and training the initial dialogue model using a second training sample and the second prompt information ([3.1] [App. B.3/B.4] we consider the GPT2 transformer as a base model, and fine-tune it to become a SeeKeR 2.7B R2C2 model was fine-tuned on all of the search, knowledge and dialogue response tasks simultaneously),
wherein the first training sample comprises a sample dialogue text and a sample query text ([3.2] 42,306 human-authored relevant search queries given the dialogue context; we can use the search query data as targets to directly train the search module),
the second training sample comprises a sample dialogue text, a sample knowledge query result and a sample response text ([3.2] the input context contains the usual dialogue, concatenated to the gold knowledge response; the new target is the standard dialogue response), and
the sample knowledge query result is a knowledge query result for the sample query text ([3.] [3.2] Search Module; a search is generated and returns results; Knowledge Module; a knowledge response is generated; response training uses the gold knowledge response (the target in the previous task)).
Claims 9 and 18,
Shuster further teaches the method of claim 8, wherein the initial dialogue model comprises a query generation network and a response generation network ([3.] this same encoder-decoder model is used in a modular way multiple times; Search Module; Response Module); and
wherein a trained query generation network is obtained by training the query generation network in the dialogue model using the first training sample ([3.2] the search query data as targets to directly train the search module); and
a trained response generation network is obtained by training the response generation network in the dialogue model using the second training sample ([3.2] the input context contains the usual dialogue, concatenated to the gold knowledge response.. the new target is the standard dialogue response).
Claim(s) 21-22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Campbell et al. (US 2019/0278792) in view of Shuster et al. (“Large Models that Seek for Knowledge: Modular Search & Generation for Dialogue and Prompt Completion”; Mar. 29, 2022; Facebook AI Research) and further in view of Feng et al. (US 11,288,322).
Claims 21 and 22,
Campbell teaches the method of claim 1, wherein there are more than one knowledge database ([0070] database 410 includes multiple databases, in which a first database is a database of golf tournaments, the second database is a database of golf players and a third database is a database of tournament statistics).
The difference between the prior art and the claimed invention is that Campbell nor Shuster explicitly teach the knowledge databases respectively correspond to different fields; and wherein obtaining, by the knowledge retriever, the knowledge query result for the current query text by querying the knowledge database based on the current query text comprises: determining a field to which the current query text belongs; and obtaining the knowledge query result for the current query text by querying, based on the current query text, the knowledge database corresponding to the field to which the current query text belongs.
Feng teaches the knowledge databases respectively correspond to different fields ([col. 6 lines 28-52] a plurality of libraries of knowledge representations with each library defined by a particular product or service domain); and
wherein obtaining, by the knowledge retriever, the knowledge query result for the current query text by querying the knowledge database based on the current query text comprises: determining a field to which the current query text belongs ([col. 5 lines 42-67] the dialog simulator (140) identifies potential virtual locations, e.g. web sites, based on the subject matter of the received input); and
obtaining the knowledge query result for the current query text by querying, based on the current query text, the knowledge database corresponding to the field to which the current query text belongs ([col. 7 line 56 to col. 8 line 17] identify relevant data from a corresponding knowledge base).
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the teachings of Campbell with teachings of Feng by modifying the dialog agent for conducting task-oriented computer-based communications as taught by Campbell to include the knowledge databases respectively correspond to different fields; and wherein obtaining, by the knowledge retriever, the knowledge query result for the current query text by querying the knowledge database based on the current query text comprises: determining a field to which the current query text belongs; and obtaining the knowledge query result for the current query text by querying, based on the current query text, the knowledge database corresponding to the field to which the current query text belongs as taught by Feng for the benefit of developing the corresponding conversational agent for goal oriented information retrieving tasks over structured knowledge (Feng [Background]).
Claim(s) 7 and 16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Campbell et al. (US 2019/0278792) in view of Shuster et al. (“Large Models that Seek for Knowledge: Modular Search & Generation for Dialogue and Prompt Completion”; Mar. 29, 2022; Facebook AI Research) in view of Feng et al. (US 11,288,322) and further in view of Thulke et al. (“Efficient Retrieval Augmented Generation from Unstructured Knowledge for Task-Oriented Dialog”; Feb. 9, 2021).
Claims 7 and 16,
Feng teaches the method of claim 21, wherein obtaining the knowledge query result for the current query text by querying, based on the current query text, the knowledge database corresponding to the field to which the current query text belongs ([col. 11 lines 27-55] the query or API call is processed, and a domain, e.g. virtual location or website, to satisfy the query or API calls is identified (612); a corresponding domain schema is identified) comprises:
obtaining a search result based on the current query text by querying, based on the current query text, the knowledge database corresponding to the field to which the current query text belongs ([col. 11 lines 27-55] a corresponding domain schema is identified (614) and data from the schema is selectively associated).
The difference between the prior art and the claimed invention is that Campbell, Shuster nor Feng explicitly teach obtaining a sorting result by sorting, based on respective correlations between a plurality of knowledge records contained in the search result and the current query text, the plurality of knowledge records in a descending order; and determining a preset number of knowledge records that are ranked first in the sorting result as the knowledge query result for the current query text.
Thulke teaches obtaining a sorting result by sorting, based on respective correlations between a plurality of knowledge records contained in the search result and the current query text, the plurality of knowledge records in a descending order ([4.1] [4.3] the method returns the knowledge snippet with the highest relevance score; we propose to use a siamese network structure made of a dialog context encoder and a knowledge snippets encoder so that the distance between their representation builds a suitable ranking function; we use the dot product as a similarity measure); and
determining a preset number of knowledge records that are ranked first in the sorting result as the knowledge query result for the current query text ([4.4] top n snippets, we typically use n=5).
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the teachings of Feng with teachings of Thulke by modifying the conversational agents over domain structured knowledge as taught by Feng to include obtaining a sorting result by sorting, based on respective correlations between a plurality of knowledge records contained in the search result and the current query text, the plurality of knowledge records in a descending order; and determining a preset number of knowledge records that are ranked first in the sorting result as the knowledge query result for the current query text as taught by Thulke for the benefit of generate responses based on multiple selected snippets and we show how the method can be used to fine-tune trained embeddings (Thulke [Abstract]).
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.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SHREYANS A PATEL whose telephone number is (571)270-0689. The examiner can normally be reached Monday-Friday 8am-5pm PST.
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, Pierre Desir can be reached at 571-272-7799. 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.
SHREYANS A. PATEL
Primary Examiner
Art Unit 2653
/SHREYANS A PATEL/ Examiner, Art Unit 2659
Prosecution Insights