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 .
This action is responsive to the claims dated 12/29/2023.
Claims 1-20 are presented for examination.
Drawings
The drawings are objected to as failing to comply with 37 CFR 1.84(p)(5) because they include the following reference character(s) not mentioned in the description: 850 (FIG. 8, designating the "MODEL" block). Corrected drawing sheets in compliance with 37 CFR 1.121(d), or amendment to the specification to add the reference character(s) in the description in compliance with 37 CFR 1.121(b) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either "Replacement Sheet" or "New Sheet" pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Specification
The disclosure is objected to because of the following informalities:
(i) [0026] recites that the model "generates two distributes," which should read distributions;
(ii) [0043] refers to "the other tokens in the vocabular," which should read vocabulary; (iii) [0039] recites "which prevents nay leaking," which should read any leaking;
(iv) the unnumbered paragraph following [0074] refers to "the Lare Language Models," which should read Large Language Models;
(v) the brief description of FIG. 5 at [0012] and again at [0049] reads "FIG. 5 illustrating creation and pre-processing of a training dataset," which is missing a main verb and should read "FIG. 5 illustrates";
(vi) [0029] recites that "the model can provide condition generation based on a country of residence," which should read conditional generation;
(vii) [0065] recites "The model generates encode outputs," which should read encoder outputs; and
(viii) [0091] recites "a basic software system 1100 that may be employed for controlling the operation of computer system 1100," which appears to be an incorrect element number because the software system is numbered 1100 and the controlled computer system is numbered 1000 (see [0078], [0092]), so the recitation should read computer system 1000.
Appropriate correction is required.
The use of the terms JavaScript ([0027]), Python ([0068]), and OpenAI ([0002]), each of which is a trade name or a mark used in commerce, has been noted in this application. The terms should be accompanied by the generic terminology; furthermore each term should be capitalized wherever it appears or, where appropriate, include a proper symbol indicating use in commerce such as (TM), SM, or (R) following the term. Although the use of trade names and marks used in commerce (i.e., trademarks, service marks, certification marks, and collective marks) is permissible in patent applications, the proprietary nature of the marks should be respected and every effort made to prevent their use in any manner which might adversely affect their validity as commercial marks.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 8-12 and 18-20 are rejected under 35 U.S.C. 112(b) as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor regards as the invention.
Claims 8 and 18 each recite, within the target-processing limitation, removing the copy tag from the target. There is insufficient antecedent basis for the singular limitation the target in the claim. The target-processing limitation first recites compressing elements in each target (a distributive, plural reference to the corresponding set of targets of parent claim 7 / claim 17) and separately recites a particular target that includes a copy tag. Because the claim introduces both each target and a particular target, it is unclear whether the target in removing the copy tag from the target refers to the single particular target that includes the copy tag or to each target of the set. Consequently, one of ordinary skill in the art cannot determine with reasonable certainty whether the copy-tag-removal step is performed once on the particular target or iteratively on every target, and the metes and bounds of the step are rendered uncertain. For the purposes of examination, the claims 8 and 18 limitations ending with “and removing the copy tag from the target” is interpreted as “and removing the copy tag from the particular target”.
Dependent claims 9-12 do not cure the claim 8 deficiencies, thus claims 9-12 are also rejected under 35 U.S.C. 112(b) for at least being dependent on a rejected base claim.
Dependent claims 19-20 do not cure the claim 18 deficiencies, thus claims 19-20 are also rejected under 35 U.S.C. 112(b) for at least being dependent on a rejected base claim.
Claim Rejections - 35 USC § 102
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 the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1 and 13 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by See et al. "Get To The Point: Summarization with Pointer-Generator Networks" (2017) (hereinafter See 2017).
Regarding independent claim 1, See 2017 discloses a method comprising: generating a text output based on an input context using a machine learning model, wherein the input context comprises a set of one or more context elements (Abstract and Section 2 describes a sequence-to-sequence attentional machine learning model (a machine learning model) that produces a summary (a text output) from a source document (an input context) whose source words (a set of one or more context elements) are read by the model, the model copying source words while retaining the ability to produce novel words, "we use a hybrid pointer-generator network that can copy words from the source text via pointing, which aids accurate reproduction of information, while retaining the ability to produce novel words through the generator"),
wherein generating the text output comprises: generating a set of next tokens for the text output (Section 2.1 at each decoder timestep the decoder produces a target word for the summary, the sequence of which constitutes a set of next tokens for the text output, "On each step t, the decoder ... receives the word embedding of the previous word ... and has decoder state st," the decoder state and context vector being "fed through two linear layers to produce the vocabulary distribution Pvocab");
generating a generation distribution that represents, for each generated token in the set of next tokens, a probability the generated token is to be added to the text output (Section 2.1 the vocabulary distribution Pvocab (a generation distribution) is a probability distribution over all words in the vocabulary, providing, for each word (a generated token), the probability that word is produced and added to the summary (the text output), "Pvocab is a probability distribution over all words in the vocabulary, and provides us with our final distribution from which to predict words w");
generating a copy distribution that represents, for each context element in the input context, a probability the context element is to be copied to the text output (Section 2.2 the attention distribution at over the source positions (the copy distribution) provides, for each source word (a context element), the probability that the source word is copied to the summary, the copy term being the summed attention over source positions whose source word equals the candidate word, "copying a word from the input sequence by sampling from the attention distribution at");
and determining, based on a copy weight, whether to (a) use the generation distribution to add a generated token from the set of next tokens to the text output or (b) use the copy distribution to copy a context element from the input context to the text output (Section 2.2 Eq. 8 and Eq. 9 the generation probability pgen (a copy weight), calculated from the context vector, the decoder state, and the decoder input, is used as a soft switch to choose between generating a word from the vocabulary by sampling from Pvocab (the generation distribution) or copying a word from the source by sampling from the attention distribution (the copy distribution), the final extended-vocabulary probability being pgen times Pvocab plus (1 minus pgen) times the summed attention distribution, "pgen is used as a soft switch to choose between generating a word from the vocabulary by sampling from Pvocab, or copying a word from the input sequence by sampling from the attention distribution at"),
and wherein the method is performed by one or more computing devices (Abstract and Section 2 the neural network model is trained and run on computing hardware, the model being implemented and executed by one or more computing devices).
Regarding independent claim 13, it is a one or more non-transitory storage media claim that is substantially the same as the method of claim 1. Thus, claim 13 is rejected for the same reason as claim 1. In addition, See 2017 discloses one or more non-transitory storage media storing instructions which, when executed by one or more computing devices, cause (Abstract and Section 2 the model is implemented as instructions executed on computing hardware, the program instructions necessarily residing on non-transitory storage media (one or more non-transitory storage media) read and executed by the hardware (one or more computing devices)).
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 2-5, 14, and 15 are rejected under 35 U.S.C. 103 as being unpatentable over See 2017, as applied in the rejections above of claims 1 and 13, in view of Zhao et al. "Improving Grammatical Error Correction via Pre-Training a Copy-Augmented Architecture with Unlabeled Data" (2019) (hereinafter Zhao 2019).
Regarding dependent claim 2, See 2017 teaches all the elements of claim 1.
See 2017 does not expressly teach wherein the machine learning model is a transformer model comprising one or more encoder layers and one or more decoder layers.
However, Zhao 2019 teaches wherein the machine learning model is a transformer model comprising one or more encoder layers and one or more decoder layers (Section 2.1 uses the attention based Transformer (a transformer model) as its base architecture, the Transformer encoding the source sentence with a stack of L identical blocks (one or more encoder layers) and a decoder block having an extra attention layer over the encoder's hidden states (one or more decoder layers), "We use the attention based Transformer (Vaswani et al., 2017) architecture as our baseline. The Transformer encodes the source sentence with a stack of L identical blocks ... the decoder block has an extra attention layer over the encoder's hidden states").
Because See 2017 and Zhao 2019 are analogous art and within the same field of endeavor, specifically neural sequence-to-sequence text generation with a copy mechanism, they address the same problem solving area of accurately reproducing source content in generated text while retaining the ability to produce novel tokens, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to combine Zhao 2019's Transformer encoder-decoder base architecture with See 2017's copy-or-generate framework, with a reasonable expectation of success, such that the bidirectional-LSTM encoder and LSTM decoder of See 2017 are replaced by the well-established Transformer encoder and decoder layers to teach wherein the machine learning model is a transformer model comprising one or more encoder layers and one or more decoder layers. This modification would have been motivated by the desire to leverage the parallel multi-head attention of the Transformer to improve modeling of long-range dependencies in the source while retaining the copy mechanism (Zhao 2019: Section 2.1).
Regarding dependent claim 3, See 2017, in view of Zhao 2019, teach the method of claim 2, wherein the copy distribution is generated from cross-attention scores generated by the one or more decoder layers (See Zhao 2019 Section 2.2, Eqs. 6-8, the copying score over the source input tokens (the copy distribution) is calculated with an attention distribution between the decoder's current hidden state and the encoder's hidden states (cross-attention scores generated by the one or more decoder layers), the copy distribution being the softmax of those attention scores, "The copying score over the source input tokens is calculated with a new attention distribution between the decoder's current hidden state htrg and the encoder's hidden states Hsrc. ... Pcopy t (w) =softmax(At)").
Regarding dependent claim 4, See 2017, in view of Zhao 2019, teach the method of claim 2, wherein the copy weight is generated from a last decoder layer output using a linear and sigmoid layer (See Zhao 2019 Section 2.2, Eq. 9, the balancing factor (the copy weight) is estimated from the copy hidden states of the decoder (a last decoder layer output) as a sigmoid of a linear projection of those hidden states (a linear and sigmoid layer), "αcopy t =sigmoid(WT ∑ (AT t ·V ))").
Regarding dependent claim 5, See 2017, in view of Zhao 2019, teach the method of claim 2, wherein the transformer model further comprises a copy decoder that generates the copy distribution and the copy weight (See Zhao 2019, Section 2.2, Eqs. 5-9, the copy-augmented decoder branch (a copy decoder) computes both the copy distribution as the softmax of the decoder-to-encoder cross-attention and the balancing factor (the copy weight) from the decoder copy hidden states via a sigmoid of a linear projection, "Pcopy t (w) =softmax(At)" and "αcopy t =sigmoid(WT ∑ (AT t ·V ))").
Regarding dependent claims 14-15, these are one or more non-transitory storage media claims that are substantially the same as the method of claims 2 and 5, respectively. Thus, claims 14-15 are rejected for the same reasons as claims 2 and 5.
Claims 6 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over See 2017, as applied in the rejections above of claims 1 and 13, respectively, in view of Vinyals et al. "Pointer Networks" (2017) (hereinafter Vinyals 2017).
Regarding dependent claim 6, See 2017 teaches all the elements of claim 1.
See 2017 does not expressly teach further comprising copying a given context element to the text output directly from the input context based on a set of coordinates indicating a position of the given context element in the input context.
However, Vinyals 2017 teaches copying a given context element to the text output directly from the input context based on a set of coordinates indicating a position of the given context element in the input context (Abstract uses attention as a pointer to select a member of the input sequence (a given context element) as the output, the output tokens being discrete tokens corresponding to positions in the input sequence (a set of coordinates indicating a position of the given context element in the input context), so that the selected input element is taken directly from its input position into the output, "it uses attention as a pointer to select a member of the input sequence as the output" and "an output sequence with elements that are discrete tokens corresponding to positions in an input sequence").
Because See 2017 and Vinyals 2017 are analogous art and within the same field of endeavor, specifically neural sequence-to-sequence models that copy elements from an input sequence into the output, they address the same problem solving area of placing an input element directly into the output by reference to its location in the input, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to perform See 2017's copying by pointing to the input position of the source element as taught by Vinyals 2017, with a reasonable expectation of success, such that the copied element is taken directly from its input position rather than reconstructed from the vocabulary to teach further comprising copying a given context element to the text output directly from the input context based on a set of coordinates indicating a position of the given context element in the input context. This modification would have been motivated by the desire to address variable-sized output dictionaries whose entries correspond to input positions, enabling exact reproduction of out-of-vocabulary source content (Vinyals 2017: Abstract).
Regarding dependent claim 16, it is a one or more non-transitory storage media claims that is substantially the same as the method of claim 6. Thus, claim 16 is rejected for the same reason as claim 6.
Claims 7 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over See 2017, as applied in the rejections above of claims 1 and 13, respectively, in view of Juraska et al., "A Deep Ensemble Model with Slot Alignment for Sequence-to-Sequence Natural Language Generation" (2018) (hereinafter Juraska 2018).
Regarding dependent claim 7, See 2017 teaches the method of claim 1, further comprising: training the machine learning model based on processed training dataset to generate a text output (Sections 2.1 and 4, the sequence-to-sequence model is trained on source-summary pairs (a training dataset comprising a set of training contexts and a corresponding set of targets) to generate the summary (a text output)).
See 2017 does not expressly teach performing context processing and target processing on a training dataset to form a processed training dataset, wherein the training dataset comprises a set of training contexts and a corresponding set of targets.
However, Juraska 2018 teaches performing context processing and target processing on a training dataset to form a processed training dataset, wherein the training dataset comprises a set of training contexts and a corresponding set of targets (Section 5.1, preprocesses the training data by identifying categorical slots in the structured input meaning representation (a set of training contexts) whose values propagate verbatim and replacing the corresponding values in the utterance (a corresponding set of targets) with placeholder tokens, the values being copied back from the input in post-processing (context processing and target processing forming a processed training dataset), "We identify the categorical slots whose values always propagate verbatim to the utterance, and replace the corresponding values in the utterance with placeholder tokens. The placeholders are eventually replaced in the output utterance in post-processing by copying the values from the input MR"). Juraska 2018 describes the delexicalization at the slot level over the paired meaning-representation/utterance training data, the placeholder tokens being aligned with the slots of the input MR so that the values are copied back from the MR in post-processing (Section 5.1); to the extent the express placeholder-substitution sentence addresses the utterance side, performing the corresponding substitution on the slot values of the paired input MR is the predictable preparation of the paired training data required for the post-processing copy-back, and is encompassed by the proposed combination.
Because See 2017 and Juraska 2018 are analogous art and within the same field of endeavor, specifically training neural sequence-to-sequence models that copy values from a structured input context into the generated text, they address the same problem solving area of preprocessing paired context and target training data so that the model copies source values rather than memorizing them, accordingly, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention, to preprocess the context and target sides of See 2017's training dataset as taught by Juraska 2018, with a reasonable expectation of success, such that both the training contexts and the corresponding targets are processed into a processed training dataset before training to teach performing context processing and target processing on a training dataset to form a processed training dataset, wherein the training dataset comprises a set of training contexts and a corresponding set of targets. This modification would have been motivated by the desire to enhance the model's ability to generalize the learned concepts to unseen inputs by delexicalizing the training data (Juraska 2018: Section 5.1).
Regarding dependent claim 17, it is a one or more non-transitory storage media claims that is substantially the same as the method of claim 7. Thus, claim 17 is rejected for the same reason as claim 7.
Allowable Subject Matter
Claims 8 and 18 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims and overcoming the 35 U.S.C. 112(b) rejections.
The following is a statement of reasons for the indication of allowable subject matter:
The closest prior arts found when taken individually or in combination do not expressly teach or render obvious the limitations recited in dependent claims 8 and 18 when taken in the context of the claims as a whole.
At best the closest prior arts uncovered, specifically, See 2017 disclose neural machine translation systems have become state-of-the-art approaches for Grammatical Error Correction (GEC) task. In this paper, we propose a copy-augmented architecture for the GEC task by copying the unchanged words from the source sentence to the target sentence. Since the GEC suffers from not having enough labeled training data to achieve high accuracy. We pre-train the copy-augmented architecture with a denoising auto-encoder using the unlabeled One Billion Benchmark and make comparisons between the fully pre-trained model and a partially pre-trained model. It is the first time copying words from the source context and fully pre-training a sequence to sequence model are experimented on the GEC task. Moreover, We add token-level and sentence-level multi-task learning for the GEC task. The evaluation results on the CoNLL-2014 test set show that our approach outperforms all recently published state-of-the-art results by a large margin (Abstract); and Juraska 2018 discloses natural language generation lies at the core of generative dialogue systems and conversational agents. We describe an ensemble neural language generator, and present several novel methods for data representation and augmentation that yield improved results in our model. We test the model on three datasets in the restaurant, TV and laptop domains, and report both objective and subjective evaluations of our best model. Using a range of automatic metrics, as well as human evaluators, we show that our approach achieves better results than state-of-the-art models on the same datasets (Abstract).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wen et al. “Semantically conditioned lstm-based natural language generation for spoken dialogue systems” (2015) (ABSTRACT Natural language generation (NLG) is a critical component of spoken dialogue and it has a significant impact both on usability and perceived quality. Most NLG systems in common use employ rules and heuristics and tend to generate rigid and stylised responses without the natural variation of human language. They are also not easily scaled to systems covering multiple domains and languages. This paper presents a statistical language generator based on a semantically controlled Long Short-term Memory (LSTM) structure. The LSTM generator can learn from unaligned data by jointly optimising sentence planning and surface realisation using a simple cross entropy training criterion, and language variation can be easily achieved by sampling from output candidates. With fewer heuristics, an objective evaluation in two differing test domains showed the proposed method improved performance compared to previous methods. Human judges scored the LSTM system higher on informativeness and naturalness and overall preferred it to the other systems).
Gehrmann et al. “Bottom-Up Abstractive Summarization” (2018) (ABSTRACT Neural summarization produces outputs that are fluent and readable, but which can be poor at content selection, for instance often copying full sentences from the source document. This work explores the use of data-efficient content selectors to over-determine phrases in a source document that should be part of the summary. We use this selector as a bottom-up attention step to constrain the model to likely phrases. We show that this approach improves the ability to compress text, while still generating fluent summaries. This two-step process is both simpler and higher performing than other end-to-end content selection models, leading to significant improvements on ROUGE for both the CNN-DM and NYT corpus. Furthermore, the content selector can be trained with as little as 1,000 sentences making it easy to transfer a trained summarizer to a new domain).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KUANG FU CHEN whose telephone number is (571)272-1393. The examiner can normally be reached M-F 9:00-5:30pm ET.
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, Jennifer Welch can be reached on (571) 272-7212. 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.
/KC CHEN/Primary Patent Examiner, Art Unit 2143
Prosecution Insights