DETAILED ACTION
Introduction
This office action is in response to Applicant’s amendment filed on January 28, 2026.
Claims 1, 3, 6, 9, 12, 14, 16, 24-25 and 27 have been amended. Claims 2, 5, 7, 10, 15, 21 and 22 have been previously cancelled. Claim 27 has been previously added. Claims 1, 3-4, 6, 8-9, 11-14, 16-20 and 23-27 are now pending in the application. As such, claims 1, 3-4, 6, 8-9, 11-14, 16-20 and 23-27 have been examined.
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 .
Drawings
The drawings were received on August 16, 2022. These drawings have been accepted and considered by the Examiner.
Response to Amendments and Arguments
In view of the amendments to claims, the amendments to claims have been acknowledged and entered.
In view of the amendments to claims, the objections claim 6 are withdrawn.
In view of the amendments to claims, the rejections to claims 1, 3-4, 6, 8-9, 11-20 and 23-27 under 35 U.S.C. 103 have been withdrawn.
In light of the amendments to the claims, new grounds for rejection for claims 1, 3-4, 6, 8-9, 11-14, 16-20 and 23-27 under 35 U.S.C. 103 are provided in the response below. New grounds for rejection is based at least upon the following new elements:
generating, by one or more machine learning models and based at least on processing first input data representing text, first output data representing one or more first vectors associated with one or more prompts, the one or more prompts associated with converting the text into a textual structure that one or more dialogue managers are able to use to interpret at least a novel intent that was not included in training data for the one or more dialogue managers;
generating, by one or more language models and based at least on processing the first input data and the first output data, second output data representing one or more second vectors;
determining, based at least on the second output data, a canonical form that includes the textual structure by representing one or more first words from the text that describe the novel intent [[for]] that was not included in the training data for the one or more dialogue managers and one or more second words from the text that are associated with one or more second intents that the one or more dialogue managers are trained to interpret; and
causing, based at least on the one or more dialogue managers processing second input data to interpret the novel intent using the canonical form represented by the second input data, at least one action to occur with respect to the novel intent from the text
Applicant’s arguments regarding the prior art rejections under 35 U.S.C 103, received on January 28, 2026, have been fully considered.
Applicant reiterates the previous argument on page 5 that Van Durme does not teach or suggest that the “prompt generator” includes a "machine learning model," and on page 6 that Van Durme does not teach or suggest that the “prompt generator” includes a component that generates the multi-dimensional vector representations. Applicant previously refered to Fig. 1 of Van Durme, provided below for reference.
PNG
media_image1.png
662
896
media_image1.png
Greyscale
Examiner maintains the obviousness rejection and provides further clarification as follows: the three separate boxes as presented in the diagram, can also be interpreted as being within a single larger box, and the larger box can be considered to be the “prompt generator.” This is how the examiner has interpreted the prior art, and therefore, the “prompt generator” does includes a "machine learning model," and does generate the multi-dimensional vector representations. Please see modified Fig. 1 as annotated by examiner for clarification and explanation below.
PNG
media_image2.png
662
896
media_image2.png
Greyscale
Furthermore, Examiner respectfully reminds the applicant that the rejection employed in the previous Office Action was based on the “Obviousness statutory” and not “Anticipation statutory” and, as such, exact teaching is not necessary. See Appeal Br. 14. The test for obviousness is not whether the claimed invention is expressly suggested in any one or all of the references, but whether the claimed subject matter would have been obvious to those of ordinary skill in the art in light of the combined teachings of those references. See In re Keller, 642 F.2d41 3, 425 (CCPA 1981).
Additionally, Donnelly (US Patent Pub. No. 20190047584 A1) in [0050] teaches prompts being generated by a MLM. This is provided as further evidence that it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have used a MLM to generate prompts.
Applicant’s remaining arguments with respect to claims 1, 3-4, 6, 8-9, 11-14, 16-20 and 23-27 have been considered, are directed to the newly amended matter in the claims, are not considered to be persuasive, and are addressed accordingly in the updated rejection rationale below.
Claim Objections
Claims 11, 16-20 and 25-27 are objected to.
Claim 11 line 2 recites “one or more prompts,” examiner believes this to be a clerical error and it should read “one or more first prompts”.
Claim 16 line 17-18 recites “based at least on third output data,” examiner believes this to be a clerical error and it should read “based at least on the third output data”. Claims 17-20 and 25-27 depend from claim 16 and therefore inherit this objection.
Appropriate correction is required.
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.
Claims 1, 3 and 4 are rejected under 35 U.S.C. 103 as being unpatentable over Van Durme et al. (US Patent Pub. No. 20220327288 A1), hereinafter Van Durme, in view of Johnson et al. (US Patent Pub. No. 20210064828 A1), hereinafter Johnson, in view of Chen et al. (US Patent Pub. No. 20220101834 A1), hereinafter Chen.
Regarding claim 1, Van Durme teaches a method (Van Durme in [0020] teaches methods for automatically generating instructions based on a received natural language utterance using semantic parsing)
comprising:
generate, one or more machine learning models and based at least on processing first input data representing text, first output data representing one or more first vectors associated with one or more prompts (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts (first output), and using a natural language model (second model), and using the received natural language utterance (first input), and in [0021] teaches the input words are converted to text, and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations (first vectors), and in [0018] teaches the semantic parser may be a machine learnt semantic parser, which uses a trained machine learning model to generate the canonical utterance),
the one or more prompts associated with converting the text into a textual structure that [one or more dialogue managers] are able to use to interpret [at least a novel intent that was not included in training data for the one or more dialogue managers] (Van Durme in [0018] teaches the prompt has a textual structure, and in [0057, Fig. 4] teaches using a prompt generator to create input into a natural language machine where the prompt includes examples of input/output pairs);
generating, one or more language models and based at least on processing the first input data and the first output data, second output data representing one or more second vectors (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts, and using a natural language model (second model), and using the received natural language utterance, and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations (one or more second vectors));
determining, based at least on the second output data, a canonical form that includes the textual structure [by representing one or more first words from the text that describe the novel intent that was not included in the training data for the one or more dialogue managers and one or more second words from the text that are associated with one or more second intents that the one or more dialogue managers are trained to interpret] (Van Durme in [0020] teaches generating a canonical utterance, using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences).
Van Durme teaches prompts, a textual structure, second output data, and the canonical form.
Van Durme does not teach, however Johnson teaches
[the one or more prompts associated with a textual structure that] one or more dialogue managers are able to use to interpret at least a novel intent [that was not included in training data for the one or more dialogue managers] (Johnson in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager), and in [0006] teaches determine an intent (novel intent) based on the canonical translation)
[determining, based at least on the second output data, a canonical form that includes the textual structure and] represents an intent associated with the text (Johnson in [0006] teaches determine an intent based on the canonical translation)
[determining, based at least on the second output data, a canonical form that includes the textual structure] by representing one or more first words from the text that describe the novel intent [that was not included in the training data] for the one or more dialogue managers and one or more second words from the text that are associated with one or more second intents that the one or more dialogue managers are trained to interpret (Johnson in [0011] teaches a machine translation model is trained on training data (second intents) in order to be able to handle any input (novel intent) while in use after being trained, and in [0006] teaches determine an intent based on the canonical translation [determining an intent can be repeated to determine a second intent], and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots for date, time, and text, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager))
causing, based at least on the one or more the dialogue managers processing second input data to interpret the novel intent using the canonical form, represented by the second input data, at least one action to occur with respect to the text, (Johnson in [0006] teaches determine an intent (second input data) based on the canonical translation, and in [0004] teaches determining an intent of the speaker by the automated assistant, so that one or more responsive actions can be performed based on the speaker's intent, and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots for date, time, and text, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager)).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme further in view of Johnson to allow for determining intent based on canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, does not teach, however Chen teaches
a novel intent that was not included in training data for the one or more dialogue managers (Chen in [0042] teaches determining a new intent which results in updating the training which indicates it was not originally in the training data).
Chen is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Chen to allow for determining a new intent. Motivation to do so would allow for using a dialogue manager which may track one more responses from a user cancelling execution of a process initiated by a previously provided speech command and providing a new speech command that is executed (Chen [0035]).
Regarding claim 3, Van Durme, as modified above, teaches the method of claim 1.
Van Durme further teaches
wherein the determining the canonical form
[determining], based at least on the second output data, [a final vector] based at least in part on the one or more second vectors (Van Durme in [0034] teaches the encoder encodes a set of words (second output data) or tokens in the prompt into multi-dimensional vector representations (one or more second vectors))
Van Durme, as modified above, does not teach, however Johnson teaches
determining, [based at least on the second output data], a final vector [based at least in part on the one or more second vectors] (Johnson in [0076] teaches generating embeddings for various versions of the input [here these embeddings map to final vectors]);
and
determining that the final vector is associated with the canonical form (Johnson in [0014] teaches mapping multiple outputs of an encoder portion—e.g., one or more semantic embeddings representing a plurality of semantically-related phrases in a first language—to the same canonical translation [here any one embedding may be the final vector]).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for determining a vector associated with a canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Regarding claim 4, Van Durme, as modified above, teaches the method of claim 3.
Van Durme, as modified above, does not teach, however Johnson teaches
wherein the determining that the final vector is associated with the canonical form comprises:
comparing the final vector to a set of one or more vectors (Johnson in [0076] teaches comparing the embeddings to the canonical translation),
wherein at least one individual vector of the set of one or more vectors is associated with a respective canonical form associated with the text (Johnson in [0014] teaches mapping multiple outputs of an encoder portion—e.g., one or more semantic embeddings representing a plurality of semantically-related phrases in a first language—to the same canonical translation [here any one embedding may be the final vector]);
determining, based at least on the comparing, that the final vector is similar to a vector from the set of one or more vectors (Johnson in [0076] teaches comparing the embeddings to the canonical translation to determine errors, using techniques such as gradient descent and/or back propagation);
and
determining that the vector is associated with the canonical form (Johnson in [0014] teaches mapping multiple outputs of an encoder portion—e.g., one or more semantic embeddings representing a plurality of semantically-related phrases in a first language—to the same canonical translation [here any one embedding may be the final vector]).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for determining a vector associated with a canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Chen, in view of Nguyen et al. (US Patent No. 10546001 B1), hereinafter Nguyen.
Regarding claim 6, Van Durme, as modified above, teaches the method of claim 1.
Van Durme further teaches
further comprising:
generating, based at least on the one or more machine learning models processing third input data representing second text, third output data representing one or more third vectors associated with the one or more prompts (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts, and using a natural language model (second model), and using the received natural language utterance (process may be repeated converting a third input to text), and in [0021] teaches the input words are converted to text (third text), and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations (third vectors), and in [0018] teaches the semantic parser may be a machine learnt semantic parser, which uses a trained machine learning model to generate the canonical utterance),
second text (Van Durme in [0094] teaches the pre-trained natural language model includes an auto-regressive natural language model based on a transformer for paraphrasing a first sequence of tokens as a second sequence of tokens, and wherein the first sequence of tokens and the second sequence of tokens are distinct)
wherein the text is different from the second text and the one or more third vectors are also associated with the textual structure (Van Durme in [0094] teaches the pre-trained natural language model includes an auto-regressive natural language model based on a transformer for paraphrasing a first sequence of tokens as a second sequence of tokens, and wherein the first sequence of tokens and the second sequence of tokens are distinct, and in [0018] teaches the prompt has a textual structure, and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations (second vectors));
generating, based at least on the one or more language models processing the third input data and the fourth output), and using a natural language model (second model), and using the received natural language utterance (third input), and in [0021] teaches the input words are converted to text (third output), and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations (fourth vectors));
and
determining, based at least on the fourth output data, the canonical form that includes the textual structure and represents [a third intent] associated with the second text (Van Durme in [0020] teaches generating a canonical utterance, using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences, in [0049] teaches a prompt generator (first model) which generates prompts (fourth output), and in [0018] teaches the prompt has a textual structure).
Van Durme, as modified above, does not teach, however Johnson teaches
a third intent (Johnson in [0006] teaches determine an intent based on the canonical translation (this can be repeated to obtain a third intent)).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for determining intent based on canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, does not teach, however Nguyen teaches
the third intent being different than the novel intent (Nguyen in [claim9] teaches the second identified intent being different from the first identified intent).
Nguyen is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Nguyen to allow for determining intent which are different. Motivation to do so would allow for a data analysis system which provides suggestions of terms to the user for adding to the prefix to improve accuracy and efficiency (Nguyen [Abstract]).
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Chen, in view of Mundie et al. (US Patent Pub. No. 20230412731 A1), hereinafter Mundie.
Regarding claim 8, Van Durme, as modified above, teaches the method of claim 1.
Van Durme further teaches
further comprising:
determining, using the one or more machine learning models and based at least on third input data representing [second] text, third output data representing one or more second prompts (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts (second prompts), and using a natural language model (second model), and using the received natural language utterance (third input), and in [0021] teaches the input words are converted to text (third output), and in [0018] teaches the semantic parser may be a machine learnt semantic parser, which uses a trained machine learning model to generate the canonical utterance);
second text (Van Durme in [0094] teaches the pre-trained natural language model includes an auto-regressive natural language model based on a transformer for paraphrasing a first sequence of tokens as a second sequence of tokens (second text), and wherein the first sequence of tokens and the second sequence of tokens are distinct)
determining, using the one or more language models and based at least on the third input data and the third output data, a second canonical form associated with the [second] text (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts, and using a natural language model (second model), and using the received natural language utterance (third input), and in [0021] teaches the input words are converted to text (third output), and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations, and in [0020] teaches generating a canonical utterance (second canonical form), using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences));
updating one or more parameters associated with the one or more machine learning models (Van Durme in [0043] teaches the training the natural language model using triplet data (a natural language utterance, a canonical utterance, and an instruction)).
Van Durme, as modified above, does not teach, however Mundie teaches
based at least on [the second canonical form]
and ground truth data associated with the [second] text (Mundie in [0036] teaches using a ground-truth label that describes that the corresponding entry is associated with the IVR interaction),
the ground truth data indicating that the [second] text is associated with the [canonical form] (Mundie in [0036] teaches using a ground-truth label that describes that the corresponding entry is associated with the IVR interaction).
Mundie is considered to be analogous to the claimed invention because it is in the same field of using prompt models. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Mundie to allow for using a ground-truth label. Motivation to do so would allow for facilitating efficient and effective automated interactions with IVR systems (Mundie [Abstract]).
Claims 9 and 11-14 are rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Baidsy et al. (US Patent Pub. No. 20230395061 A1), hereinafter Baidsy, in view of Lester et al. (US Patent Pub. No. 20230325725 A1), hereinafter Lester.
Regarding claim 9, Van Durme teaches one or more processors (Van Durme in [0028] teaches an instruction processor which processes or executes the instructions)
comprising:
processing circuitry to (Van Durme in [0028] teaches an instruction processor which processes or executes the instructions):
the first canonical form including a textual structure representing at least [an intent] [that one or more dialogue managers are trained to interpret] (Van Durme in [0020] teaches generating a canonical utterance, using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences, and in [0057, Fig. 4] teaches using a prompt generator to create input into a natural language machine where the prompt includes examples of input/output pairs)
generate, based at least on the one or more machine learning models processing text data representing text, prompt data representing one or more first prompts [associated with converting the text into the textual structure representing at least the intent that the one or more dialogue managers are trained to interpret] (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts, and using a natural language model (second model), and using the received natural language utterance, and in [0021] teaches the input words are converted to text, and in [0018] teaches the semantic parser may be a machine learnt semantic parser, which uses a trained machine learning model to generate the canonical utterance);
generate, based at least on the one or more language models processing the text data and the prompt data, a second canonical form that predicts the textual structure and is associated with the text (Van Durme in [0020] teaches generating a canonical utterance, using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences, and in [0057, Fig. 4] teaches using a prompt generator to create input into a natural language machine where the prompt includes examples of input/output pairs),
such that the one or more language models are able to generate one or more third canonical forms that include the textual structure and represent one or more [novel intents that the one or more dialogue managers are able to interpret based on the textual structure] (Van Durme in [0020] teaches generating a canonical utterance [this can be repeated to create the third canonical form], using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences, and in [0057, Fig. 4] teaches using a prompt generator to create input into a natural language machine where the prompt includes examples of input/output pairs).
Van Durme teaches the prompt model, the language model, the canonical form and the textual structure.
Van Durme, as modified above, does not teach, however Johnson teaches
the [first canonical form including] a textual structure representing at least an intent that one or more dialogue managers are trained to interpret (Johnson in [0011] teaches a machine translation model is trained on training data (second intents) in order to be able to handle any input (novel intent) while in use after being trained, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager), and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots for date, time, and text)
one or more first prompts associated with converting the text into the textual structure representing at least the intent that the one or more dialogue managers are trained to interpret (Johnson in [0011] teaches a machine translation model is trained on training data (second intents) in order to be able to handle any input (novel intent) while in use after being trained, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager), and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots for date, time, and text).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme further in view of Johnson to allow for using slots and slot values known as parameters. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, teaches the prompt model, the language model, the first canonical form, the textual structure, and the dialog manager.
Van Durme, as modified above, does not teach, however Baidsy teaches
obtain ground truth data [for updating one or more corresponding ground-truth canonical speech sample of the training output audio data to generate the loss),
the ground truth data representative of at least a first canonical form [to be output by one or more language models] (Baidsy in [0052] teaches comparing the output to the corresponding ground-truth canonical speech sample of the training output audio data to generate the loss),
determine one or more losses based at least on comparing the [second canonical form] to the [first canonical form] represented by the ground truth data (Baidsy in [0052] teaches comparing the output [here output maps to first canonical form] to the corresponding ground-truth canonical speech sample of the training output audio data to generate the loss),
based at least on the one or more losses (Baidsy in [0052] teaches comparing the output to the corresponding ground-truth canonical speech sample of the training output audio data to generate the loss).
Baidsy is considered to be analogous to the claimed invention because it is in the same field of using ground truth canonical speech. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Baidsy to allow for using a ground-truth canonical speech sample to generate a loss. Motivation to do so would allow for automatically determining the breakpoints, to provide for a more natural user-experience as the turn-by-turn nature of the system mimics a normal conversation (Baidsy [0027]).
Van Durme, as modified above, teaches the prompt model, the language model, the one or more machine learning models, the second canonical form, the textual structure, the ground truth data, one or more losses, and the dialog manager.
Van Durme, as modified above, does not teach, however Lester teaches
[obtain ground truth data for] updating one or more
[the ground truth data representative of at least] a first canonical form to be output by one or more language models (Lester in [0105] teaches using canonical forms with a machine learning model, and in [0007] teaches the MLM can output text data (which may be canonical)),
the first canonical form [including a textual structure representing at least an intent that one or more dialogue managers are trained to interpret] (Lester in [0105] teaches using canonical forms with a machine learning model)
update, [based at least on the one or more losses], the one or more
Lester is considered to be analogous to the claimed invention because it is in the same field of using machine learning models for prompt tuning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Lester to allow for updating parameters of a machine learning model while the language model parameters are frozen. Motivation to do so would be that prompt tuning can allow for the circumvention of retraining the full pre-trained machine-learned model (Lester [0062]).
Regarding claim 11, Van Durme, as modified above, teaches the one or more processors of claim 9.
Van Durme further teaches
wherein the data representing one or more prompts comprises data representing one or more vectors associated with the one or more prompts (Van Durme in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations).
Regarding claim 12, Van Durme, as modified above, teaches the one or more processors of claim 9.
Van Durme further teaches
wherein the determination of the second canonical form associated with the text comprises:
determining,
based at least on the one or more language models (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts, and using a natural language model (second model), and using the received natural language utterance, and in [0021] teaches the input words are converted to text)
processing the text data (Van Durme in [0021] teaches the input words are converted to text)
and
the prompt data, one or more vectors associated with one or more words (Van Durme in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations).
Van Durme, as modified above, does not teach, however Johnson teaches
determining a final vector based at least on the one or more vectors (Johnson in [0076] teaches generating embeddings for various versions of the input [here these embeddings map to final vectors]);
and
determining that the final vector is associated with the second canonical form (Johnson in [0014] teaches mapping multiple outputs of an encoder portion—e.g., one or more semantic embeddings representing a plurality of semantically-related phrases in a first language—to the same canonical translation [here any one embedding may be the final vector]).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for determining a vector associated with a canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Regarding claim 13, Van Durme, as modified above, teaches the one or more processors of claim 12.
Van Durme, as modified above, does not teach, however Johnson teaches
wherein the determination that the final vector is associated with the second canonical form comprises:
comparing the final vector to a set of one or more vectors (Johnson in [0076] teaches comparing the embeddings to the canonical translation),
wherein at least one individual vector of the set of one or more vectors is associated with a respective canonical form associated with the text (Johnson in [0014] teaches mapping multiple outputs of an encoder portion—e.g., one or more semantic embeddings representing a plurality of semantically-related phrases in a first language—to the same canonical translation [here any one embedding may be the final vector]);
determining, based at least on the comparing, that the final vector is similar to a vector from the set of one or more vectors (Johnson in [0076] teaches comparing the embeddings to the canonical translation to determine errors, using techniques such as gradient descent and/or back propagation);
and
determining that the vector is associated with the second canonical form (Johnson in [0014] teaches mapping multiple outputs of an encoder portion—e.g., one or more semantic embeddings representing a plurality of semantically-related phrases in a first language—to the same canonical translation [here any one embedding may be the final vector]).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for determining a vector associated with a canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Regarding claim 14, Van Durme, as modified above, teaches the one or more processors of claim 9.
Van Durme further teaches
wherein the processing circuitry is further to:
second text data (Van Durme in [0094] teaches the pre-trained natural language model includes an auto-regressive natural language model based on a transformer for paraphrasing a first sequence of tokens as a second sequence of tokens, and wherein the first sequence of tokens and the second sequence of tokens are distinct)
generate, based at least on the one or more machine learning models processing second text data representing second text, second prompt data representing one or more second prompts (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts, and using a natural language model (second model), and using the received natural language utterance, and in [0021] teaches the input words are converted to text [here using the second text generates a second prompt], and in [0018] teaches the semantic parser may be a machine learnt semantic parser, which uses a trained machine learning model to generate the canonical utterance);
generate, based at least on the one or more language models processing the second text data and the second prompt data, a fourth canonical form associated with the second text (Van Durme in [0020] teaches generating a canonical utterance, using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences [here using the second text and second prompt generates a fourth canonical utterance]);
update the one or more
Van Durme, as modified above, teaches the third canonical form.
Van Durme, as modified above, does not teach, however Baidsy teaches
determine one or more second losses based at least on comparing [the fourth canonical form] to the first canonical form represented by the ground truth data (Baidsy in [0052] teaches comparing the output [here output maps to third canonical form] to the corresponding ground-truth canonical speech sample of the training output audio data to generate the loss);
and
based at least on the one or more second losses (Baidsy in [0052] teaches comparing the output to the corresponding ground-truth canonical speech sample of the training output audio data to generate the loss).
Baidsy is considered to be analogous to the claimed invention because it is in the same field of using ground truth canonical speech. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Baidsy to allow for using a ground-truth canonical speech sample to generate a loss. Motivation to do so would allow for automatically determining the breakpoints, to provide for a more natural user-experience as the turn-by-turn nature of the system mimics a normal conversation (Baidsy [0027]).
Claims 16, 18 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson.
Regarding claim 16, Van Durme teaches a system (Van Durme in [0017] teaches the aspects of the disclosure may be practiced as methods, systems or devices)
comprising:
one or more processors to (Van Durme in [0028] teaches an instruction processor which processes or executes the instructions)
to:
generate, by one or more prompt models processing data representing a textual input, first output data representative of one or more vectors associated with one or more prompts (Van Durme in [0049] teaches a prompt generator (first model) which generates prompts (output data), and using a natural language model (second model) , and using the received natural language utterance, and in [0021] teaches the input words are converted to text, and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations (vectors));
the one or more prompts associated with converting the textual input into a textual structure that [one or more dialogue managers] are able to use to interpret [at least a novel intent for the one or more dialogue managers] (Van Durme in [0018] teaches the prompt has a textual structure, and in [0057, Fig. 4] teaches using a prompt generator to create input into a natural language machine where the prompt includes examples of input/output pairs);
generate, by one or more language models and based at least on the textual input and the output data, second output data representative of a canonical representation that includes the textual structure (Van Durme in [0020] teaches generating a canonical utterance, using as input both the prompt (output data) and the natural language utterance, and that the prompt includes exemplary pairs of sentences);
that includes the textual structure [and represents at least the novel intent for the one or more dialogue managers] (Van Durme in [0018] teaches the prompt has a textual structure).
Van Durme does not teach, however Johnson teaches
[the one or more prompts associated with converting the textual input into a textual structure that] one or more dialogue managers are able to use to interpret at least a novel intent for the one or more dialogue managers (Johnson in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager), and in [0006] teaches determine an intent (novel intent) based on the canonical translation)
[generate, by one or more language models and based at least on the textual input and the output data], second output data representative of a canonical representation that includes the textual structure and represents at least the novel intent for one or more dialogue managers (Johnson in [0006] teaches determine an intent based on the canonical translation, and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots for date, time, and text, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager))
generate, [based at least on the textual input and the second output data], third output data representing at least the novel intent represented [using the canonical representation and one or more parameters related to the novel intent] (Johnson in [0006] teaches determine an intent based on the canonical translation, and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots for date, time, and text, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager) [this can be repeated to obtain a third output])
cause, using the one or more dialogue managers and based at least on third output data, one or more actions associated with the novel intent from the textual input, the one or more dialogue managers able to interpret the novel intent based at least on the canonical representation [and the one or more parameters being related to one or more second intents for which the one or more dialogue managers were trained to interpret] (Johnson in [0006] teaches determine an intent (second input data) based on the canonical translation, and in [0004] teaches determining an intent of the speaker by the automated assistant, so that one or more responsive actions can be performed based on the speaker's intent, and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots for date, time, and text, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager)).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme further in view of Johnson to allow for determining intent based on canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Regarding claim 18, Van Durme, as modified above, teaches the system of claim 16.
Van Durme, as modified above, does not teach, however Johnson teaches
wherein the canonical representation associated with the textual input is represented using one or more second vectors (Johnson in [0014] teaches mapping multiple outputs of an encoder portion—e.g., one or more semantic embeddings representing a plurality of semantically-related phrases in a first language—to the same canonical translation [here any one embedding may be the final vector]),
and
the novel intent is determined based at least on comparing the one or more second vectors to one or more third vectors corresponding to one or more intents (Johnson in [0006] teaches determine an intent based on the canonical translation, and in [0076] teaches comparing the embeddings to the canonical translation to determine errors, using techniques such as gradient descent and/or back propagation).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for determining a vector associated with a canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Regarding claim 26, Van Durme, as modified above, teaches the system of claim 16.
Van Durme, as modified above, does not teach, however Johnson teaches
wherein the canonical representation includes one or more words from the textual input that are arranged in an order that is associated with the textual structure (Johnson in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots (parameters) for date, time, and text (second words), and in [0006] teaches determine an intent (first words) based on the canonical translation, and in [0004] teaches determining an intent of the speaker by the automated assistant, so that one or more responsive actions can be performed based on the speaker's intent, and in [0055] teaches using slots and slot values known as parameters, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (dialog manager)).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for using slots and slot values known as parameters. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Elmieh et al. (US Patent Pub. No. 20120117536 A1), hereinafter Elmieh, in view of Partovi et al. (US Patent No. 9305119 B1), hereinafter Partovi.
Regarding claim 17, Van Durme, as modified above, teaches the system of claim 16.
Van Durme does not teach, however Johnson teaches
wherein the novel intent is determined based at least on the canonical representation (Johnson in [0006] teaches determine an intent based on the canonical translation)
the novel intent corresponds to an individual canonical representation (Johnson in [0006] teaches determine an intent based on the canonical translation).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for determining intent based on canonical translation. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, does not teach, however Elmieh teaches
comparing the canonical representation to a set of one or more canonical representations (Elmieh in [0005] teaches comparing the first canonical representation and the updated canonical representation).
Elmieh is considered to be analogous to the claimed invention because it is in the same field of using canonical representations. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Elmieh to allow for canonical representation comparisons. Motivation to do so would allow for facilitating flexibility and upgradeability for the authoring system by allowing various canonical forms to be selected for use in the authoring environment based on a particular target graphics engine and its associated performance characteristics (Elmieh [0053]).
Van Durme, as modified above, does not teach, however Partovi teaches
individual canonical representation of the set of one or more canonical representations that is most similar to the canonical representation (Partovi in [col 15 ln 12 – col 16 ln 27, claim 11] teaches determining a similarity or difference measure between the first and second canonical forms, and in [col 5 ln 46-57] teaches selecting the most likely correct option based on similarity);
Partovi is considered to be analogous to the claimed invention because it is in the same field of using canonical forms. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Partovi to allow for determining a difference between the first and second canonical forms. Motivation to do so would allow for correcting data or resolving ambiguities in data, and more specifically, to correcting metadata used to describe content that is submitted by a relevant community or network of users (Partovi [col 1 ln 15-18]).
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Mundie, in view of Bayomi et al. (US Patent Pub. No. 20230306201 A1), hereinafter Bayomi.
Regarding claim 19, Van Durme, as modified above, teaches the system of claim 18.
Van Durme, as modified above, does not teach, however Mundie teaches
wherein ground truth data used for updating one or more parameters of the one or more prompt models (Mundie in [0036] teaches using a ground-truth label that describes that the corresponding entry is associated with the IVR interaction)
during the training (Mundie in [0036] teaches training the voice prompt classification machine learning models of a voice prompt classification machine learning framework).
Mundie is considered to be analogous to the claimed invention because it is in the same field of using prompt models. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Mundie to allow for using a ground-truth label. Motivation to do so would allow for facilitating efficient and effective automated interactions with IVR systems (Mundie [Abstract]).
Van Durme, as modified above, does not teach, however Bayomi teaches
corresponds to one or more outputs of the one or more language models (Bayomi in [0027] teaches using natural language processing machine learning model are trained in an end-to-end manner and based at least in part on ground-truth natural language outputs (e.g., ground-truth text classifications, ground-truth text translations/summarizations, and/or the like) for a set of training input text sequences, such as ground-truth natural language outputs determined based at least in part on subject matter annotations and/or ground-truth natural language outputs determined based at least in part on historical end-user-assigned labels for particular text documents).
Bayomi is considered to be analogous to the claimed invention because it is in the same field of using ground truths. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Bayomi to allow for using ground-truth natural language outputs. Motivation to do so would allow for addressing the efficiency and reliability shortcomings of existing natural language processing solutions (Bayomi [0002]).
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Lester.
Regarding claim 20, Van Durme, as modified above, teaches the system of claim 16.
Van Durme, as modified above, does not teach, however Johnson teaches
wherein the system is comprised in at least one of:
an infotainment system for an autonomous or semi-autonomous machine (Johnson in [0036] teaches using an in-vehicle entertainment system);
an entertainment system for an autonomous or semi-autonomous machine (Johnson in [0036] teaches using an in-vehicle entertainment system);
a system for performing simulation operations;
a system for hosting real-time streaming applications;
a system for generating content for one or more of virtual reality (VR), augmented reality (AR), or mixed reality (MR) (Johnson in [0036] teaches using a virtual or augmented reality computing device);
a system for performing digital twin operations;
a system for performing light transport simulation;
a system for performing collaborative content creation for 3D assets;
or
a system implemented at least partially using cloud computing resources (Johnson in [0032] teaches using a “cloud” computing system).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for using a “cloud” computing system. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, does not teach, however Lester teaches
a system for performing deep learning operations (Lester in [0067] teaches using a Deep Neural Network);
a system implemented using an edge device;
a system implemented using a robot;
a system for performing conversational Al operations;
a system for generating synthetic data (Lester in [0222] teaches using synthetic datasets);
a system incorporating one or more virtual machines (VMs) (Lester in [0071] teaches using a virtual keyboard);
or
a system implemented at least partially in a data center (Lester in [0248] teaches using datacenters).
Lester is considered to be analogous to the claimed invention because it is in the same field of using machine learning models for prompt tuning. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Lester to allow for using deep neural networks, synthetic data sets, virtual machines, and data centers. Motivation to do so would be that prompt tuning can allow for the circumvention of retraining the full pre-trained machine-learned model (Lester [0062]).
Claim 23 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Chen, in view of Baidsy.
Regarding claim 23, Van Durme, as modified above, teaches the method of claim 1.
Van Durme further teaches
wherein the one or more machine learning models are trained (Van Durme in [0043] teaches the training the natural language model using triplet data (a natural language utterance, a canonical utterance, and an instruction)),
at least, by:
one or more second canonical forms determined using the one or more language models processing one or more third vectors output by the one or more machine learning models (Van Durme in [0020] teaches generating a canonical utterance [this can be repeated to create the second canonical form], using as input both the prompt and the natural language utterance, and that the prompt includes exemplary pairs of sentences, in [0049] teaches a prompt generator which generates prompts, and using a natural language model, and using the received natural language utterance, and in [0034] teaches the encoder encodes a set of words or tokens in the prompt into multi-dimensional vector representations (one or more third vectors), and in [0018] teaches the semantic parser may be a machine learnt semantic parser, which uses a trained machine learning model to generate the canonical utterance)
one or more fourth canonical forms (Van Durme in [0020] teaches generating a canonical utterance [this can be repeated to create the fourth canonical form),
and
updating, [based at least on the one or more losses], one or more parameters of the one or more machine learning models (Van Durme in [0043] teaches the training the natural language model using triplet data (a natural language utterance, a canonical utterance, and an instruction)).
Van Durme, as modified above, does not teach, however Johnson teaches
[one or more fourth canonical forms] that the one or more dialogue managers are able to interpret (Johnson in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (fulfillment engine maps to dialog manager)).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for using dialogue managers. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, does not teach, however Baidsy teaches
determining one or more losses based at least on comparing [one or more second canonical forms] to the [one or more fourth canonical forms] (Baidsy in [0052] teaches comparing the output [here output maps to first canonical form] to the corresponding ground-truth canonical speech sample [here output maps to second canonical form] of the training output audio data to generate the loss);
and
based at least on the one or more losses (Baidsy in [0052] teaches comparing the output to the corresponding ground-truth canonical speech sample of the training output audio data to generate the loss).
Baidsy is considered to be analogous to the claimed invention because it is in the same field of using ground truth canonical speech. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Baidsy to allow for using a ground-truth canonical speech sample to generate a loss. Motivation to do so would allow for automatically determining the breakpoints, to provide for a more natural user-experience as the turn-by-turn nature of the system mimics a normal conversation (Baidsy [0027]).
Claim 24 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Chen, in view of Neale.
Regarding claim 24, Van Durme, as modified above, teaches the method of claim 1.
Van Durme, as modified above, does not teach, however Johnson teaches
wherein the one or more dialogue managers [are not trained] to interpret the novel intent (Johnson in [0011] teaches a machine translation model is trained on training data (second intents) in order to be able to handle any input (novel intent) while in use after being trained, and in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots (parameters) for date, time, and text (second words), and in [0006] teaches determine an intent (first words) based on the canonical translation, and in [0004] teaches determining an intent of the speaker by the automated assistant, so that one or more responsive actions can be performed based on the speaker's intent, and in [0055] teaches using slots and slot values known as parameters, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (dialog manager)).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for using slots and slot values known as parameters. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, does not teach, however Neale teaches
dialogue managers are not trained to interpret the novel intent (Neale in [0073] teaches using a client specific language model which is entirely untrained).
Neale is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Neale to allow for using a client specific language model which is entirely untrained. Motivation to do so would allow for using synonymous terms and phrases to be defined for reducing the number of training variances required, and provide entity detection in a given input token sequence (Neale [0062]).
Claim 25 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Neale.
Regarding claim 25, Van Durme, as modified above, teaches the system of claim 16.
Van Durme, as modified above, does not teach, however Johnson teaches
wherein the canonical representation represents, based at least on the textual structure, at least one or more hat define the novel intenthat the one or more dialogue managers were [not trained] to interpret (Johnson in [0085, Figs. 3-5] teaches the canonical structure includes an intent plus slots (parameters) for date, time, and text (second words), and in [0006] teaches determine an intent (first words) based on the canonical translation, and in [0004] teaches determining an intent of the speaker by the automated assistant, so that one or more responsive actions can be performed based on the speaker's intent, and in [0055] teaches using slots and slot values known as parameters, and in [0059] teaches a NLU engine provides users intent and any slot values to a fulfillment engine (dialog manager)).
Johnson is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Johnson to allow for using slots and slot values known as parameters. Motivation to do so would allow for increasing the language coverage of an automated assistant system, i.e., they may serve to increase the number of queries in one or more non-native languages for which the automated assistant is able to deliver reasonable responses (Johnson [Abstract]).
Van Durme, as modified above, does not teach, however Neale teaches
dialogue managers are not trained to interpret the novel intent (Neale in [0073] teaches using a client specific language model which is entirely untrained).
Neale is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Neale to allow for using a client specific language model which is entirely untrained. Motivation to do so would allow for using synonymous terms and phrases to be defined for reducing the number of training variances required, and provide entity detection in a given input token sequence (Neale [0062]).
Claim 27 is rejected under 35 U.S.C. 103 as being unpatentable over Van Durme, in view of Johnson, in view of Mundie, in view of Nguyen.
Regarding claim 27, Van Durme, as modified above, teaches the system of claim 16.
Van Durme, as modified above, teaches prompt models, intents, and the canonical representation.
Van Durme, as modified above, does not teach, however Mundie teaches
wherein the one or more prompt models are trained using ground truth data representing one or more instances of the canonical representation that represent [the one or more second intents that are different than the novel intent] (Mundie in [0036] teaches using a ground-truth label that describes that the corresponding entry is associated with the IVR interaction, and teaches training the voice prompt classification machine learning models of a voice prompt classification machine learning framework).
Mundie is considered to be analogous to the claimed invention because it is in the same field of using prompt models. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Mundie to allow for using a ground-truth label. Motivation to do so would allow for facilitating efficient and effective automated interactions with IVR systems (Mundie [Abstract]).
Van Durme, as modified above, does not teach, however Nguyen teaches
one or more second intents that are different than the novel intent (Nguyen in [claim9] teaches the second identified intent being different from the first identified intent).
Nguyen is considered to be analogous to the claimed invention because it is in the same field of determining intents. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Van Durme, as modified above, further in view of Nguyen to allow for determining intent which are different. Motivation to do so would allow for a data analysis system which provides suggestions of terms to the user for adding to the prefix to improve accuracy and efficiency (Nguyen [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 PAUL J. MUELLER whose telephone number is (571)272-1875. The examiner can normally be reached M-F 9:00am-5:00pm (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 C. 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.
PAUL MUELLER
Examiner
Art Unit 2657
/PAUL J. MUELLER/Examiner, Art Unit 2657
/DANIEL C WASHBURN/Supervisory Patent Examiner, Art Unit 2657