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 .
Claims 1-6 are pending and have been examined. Claims 1, 5, and 6 are independent.
Apparent priority: June 30, 2022.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 12/26/2024 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-6 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 (Statutory Category). Claim 1 is directed to a generation apparatus comprising a processor and a memory (a machine), claim 5 is directed to a generation method (a process), and claim 6 is directed to a non-transitory computer-readable recording medium storing a program (an article of manufacture). Each independent claim is therefore directed to a statutory category of invention.
Step 2A, Prong One (Recitation of a Judicial Exception). Independent claims 1, 5, and 6 recite, in pertinent part, receiving a constraint information list as an input and outputting, as a lexical constraint, each element of a subset of one or a plurality of pieces of constraint information included in the constraint information list; generating one or more candidates for the second sequence by using the first sequence and the lexical constraint; and calculating a score indicating appropriateness as the second sequence for each of the one or more candidates. Under their broadest reasonable interpretation, these limitations recite mental processes, that is, concepts performed in the human mind, including observation, evaluation, judgment, and opinion, and the calculation of a score recites a mathematical concept. A translator who is handed a source sentence together with a list of words that the translation is expected to contain, chooses some or all of the words on that list as the words the translation must include, writes out one or more candidate translations of the source sentence that contain the chosen words, and then judges how good each candidate translation is by assigning it a score, performs each recited step mentally or with the aid of pen and paper. Reciting that the steps are carried out by a processor executing program instructions held in a memory (claim 1), by a generation apparatus (claim 5), or by a computer executing a program stored on a recording medium (claim 6) does not remove the limitations from the mental process grouping, because a claim that recites a judicial exception performed by a generically recited computer component still recites the judicial exception. The claims therefore recite an abstract idea.
Step 2A, Prong Two (Integration into a Practical Application). The judicial exception is not integrated into a practical application. Beyond the mental processes and the mathematical concept, claim 1 recites a processor and a memory storing program instructions that cause the processor to perform the recited operations. Claim 5 recites that the method is executed by a generation apparatus. Claim 6 recites a non-transitory computer-readable recording medium storing a program for causing a computer to execute the method. Each of these additional elements is recited at a high level of generality. The specification describes the hardware on which the apparatus is realized as a general-purpose computer: "Any of the apparatus or the device (the generation apparatus 100 or the extraction device) described in the present embodiment can be implemented by causing a computer to execute a program, for example. This computer may be a physical computer, or may be a virtual machine on a cloud." (specification, paragraph [0123]), the program "being executed by use of hardware resources such as a CPU and a memory built in the computer" and being "stored and distributed by being recorded in a computer-readable recording medium (portable memory or the like)" (specification, paragraph [0124]; see also paragraph [0125] and Fig. 14), and neither the claims nor the specification recite any particular processor, memory, or medium. These additional elements amount to mere instructions to apply the mental processes using generic computer components, and to generally linking the use of the judicial exception to the technological environment of computer-implemented sequence generation. The claims do not recite an improvement to the functioning of a computer or to any other technology or technical field; the asserted advance lies in which words from the list are required and in how the resulting candidates are judged, not in any improvement to the way the recited processor, memory, or medium operates. Accordingly, the additional elements do not integrate the abstract idea into a practical application, and the claims are directed to the abstract idea.
Step 2B (Inventive Concept). The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration, the processor, the memory storing program instructions, the generation apparatus, and the non-transitory computer-readable recording medium are recited at a high level of generality and perform the well-understood, routine, and conventional functions of receiving data, executing stored instructions on that data, and outputting the result. That these components are conventional in this field is evidenced by Chousa and Post of record, discussed below: each describes lexically constrained sequence generation carried out by a neural machine translation model implemented in a general-purpose deep learning toolkit executed on ordinary computing hardware, and neither describes anything more particular. Considered individually and as an ordered combination, the additional elements amount to no more than mere instructions to apply the abstract idea using generic components and do not provide an inventive concept. The claims are not patent eligible.
The dependent claims have been considered and do not cure the deficiencies of the independent claims. Each dependent claim is addressed separately below.
Claim 2 recites that the score is calculated on a basis of at least one of a likelihood output from a model used for generating each candidate or a likelihood obtained from each candidate by a re-ranking model. Judging a candidate by the likelihood that a model assigns to it is a mathematical calculation, which is itself a judicial exception, and the model that supplies the likelihood is recited generically as a tool for performing that calculation; the claim supplies no additional element beyond those addressed above and is neither integrated into a practical application nor significantly more.
Claim 3 recites that, where the constraint information list includes constraint information having an ambiguity, the candidates are generated by performing a lexically constrained beam search in consideration of the ambiguity. Generating candidate sentences that contain one of several alternative forms of a required word is a step a translator performs mentally or with pen and paper, and naming the search as a lexically constrained beam search recites a known search technique at the level of its class without any recited implementation detail; the claim supplies no additional element and is neither integrated into a practical application nor significantly more.
Claim 4 recites that at least one piece of constraint information is received in a format in which two or more ambiguities are allowed, and that the lexical constraint is generated with the ambiguities retained. Receiving a required word together with its alternative forms and carrying those alternatives forward describes the content of the information on which the mental process operates; the claim supplies no additional element and is neither integrated into a practical application nor significantly more.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 2, 5, and 6 are rejected under 35 U.S.C. 103 as being unpatentable over Chousa and Morishita, "Input Augmentation Improves Constrained Beam Search for Neural Machine Translation: NTT at WAT 2021," arXiv:2106.05450v1, June 10, 2021, https://arxiv.org/abs/2106.05450v1, hereinafter Chousa, in view of Post and Vilar, "Fast Lexically Constrained Decoding with Dynamic Beam Allocation for Neural Machine Translation," Proc. NAACL-HLT 2018, pp. 1314-1324, https://aclanthology.org/N18-1119, arXiv:1804.06609, https://arxiv.org/abs/1804.06609, hereinafter Post.
Regarding claim 1, Chousa discloses:
1. A generation apparatus for generating, from constraint information and a first sequence that is a sequence of information, a second sequence that is a sequence of different information, the generation apparatus comprising: (Chousa discloses systems that output translated sentences containing given word constraints: "This paper describes our systems that were submitted to the restricted translation task at WAT 2021. In this task, the systems are required to output translated sentences that contain all given word constraints." (Chousa, Abstract, first two sentences), the task being translation from a source sentence X to a target sentence Y under constraints C: "Suppose we have a source sentence X = (x1, x2, ..., xS) with S tokens and a target sentence Y = (y1, y2, ..., yT) with T tokens." (Chousa, Sec. 2, first sentence) and "where C = (C1, C2, ..., CN) is the provided word constraints with N phrases, and the constraints are given in random order." (Chousa, Sec. 2, the sentence following Eq. 2). The submitted system of Chousa is the generation apparatus; the source sentence X is the first sequence; the target sentence Y is the second sequence; and the provided word constraints C are the constraint information.)
a processor, and a memory storing program instructions that cause the processor to:
receive a constraint information list as an input and output, as a lexical constraint, each element of a subset of one or a plurality of pieces of constraint information included in the constraint information list; (The recited subset is interpreted, under the broadest reasonable interpretation, to include the constraint information list itself, a set being a subset of itself; the limitation is therefore met by receiving the list and outputting each of its members as the lexical constraint, and is also met by outputting the members of a proper subset of the list. Chousa receives lists of target words as the input word restrictions: "In the restricted translation task, lists of target words are provided to represent word restrictions, and systems are required to output translations that contain all of the target words in each list." (Chousa, Sec. 2, second para.). Chousa outputs the members of the received list as the constraints enforced during generation, by concatenating each phrase Ci of C onto the source sentence: "The input is constructed by concatenating the source sentence X and each phrase Ci in the constraints C with a separator symbol (sep), as follows: [X, (sep), C1, (sep), C2, ..., CN, (eos)]," (Chousa, Sec. 4.3, second para. and Eq. 3), and, when that input is constructed, selects a proper subset of the available constrained words: "Here, we sampled the number of constrained words k from 0 to 14 following the distribution that is p = 0.4 for 0 and p = 0.6/14(= 0.04) for the other ones." (Chousa, Sec. 4.3, third para.). The list of target words is the constraint information list; each phrase Ci on it is a piece of constraint information; and the phrases concatenated onto the source sentence and enforced at decoding are the lexical constraint.)
generate one or more candidates for the second sequence by using the first sequence and the lexical constraint; and (Chousa generates candidate translations by lexically constrained decoding of the source sentence under the constraints: "We used the lexically constrained decoding (LCD) technique (Hokamp and Liu, 2017; Post and Vilar, 2018) to incorporate constraints at decoding time." (Chousa, Sec. 4.2, first para.), the technique maintaining a beam of candidates for each number of satisfied constraints: "At each decoding step, a beam is allocated to each number of constraints, and the top-k candidates that contain n constraints are selected for the nth beam." (Chousa, Sec. 7, first para.). The candidates carried in the beams are the one or more candidates for the second sequence, generated from the source sentence and the constraints.)
calculate a score indicating appropriateness as the second sequence for each of the one or more candidates. (Chousa scores each candidate by the conditional probability the model assigns to it, the translation being found "by finding the best target sentence that maximizes the conditional probability" (Chousa, Sec. 2, first para., and Eq. 1 and Eq. 2); Chousa computes averaged scores across an ensemble and generates the hypotheses from those scores: "We then computed the average scores of these models and generated hypotheses based on these scores using beam search decoding." (Chousa, Sec. 4.6, first para.); and Chousa states that the constrained decoder scores the candidates: "This result indicates that the output of LeCA is helpful for LCD to score the candidates and that LeCA can save inference time." (Chousa, Sec. 6, third para.). The conditional probability assigned to each candidate is a score indicating its appropriateness as the second sequence.)
Chousa does not expressly disclose the processor and memory aspects of the claim, that is, that the recited operations are performed by a processor executing program instructions stored in a memory.
Post discloses:
A generation apparatus for generating, from constraint information and a first sequence that is a sequence of information, a second sequence that is a sequence of different information, the generation apparatus comprising: (Post discloses a neural machine translation decoder that produces an output sequence from an input sequence under pre-specified lexical constraints: "Recent work, however, has introduced a new capability: lexically constrained or guided decoding, a modification to beam search that forces the inclusion of pre-specified words and phrases in the output." (Post, Abstract, second sentence) and "Inference in statistical machine translation seeks to find the output sequence, y, that maximizes the probability of a function parameterized by a model, θ, and an input sequence, x" (Post, Sec. 2, first sentence).)
a processor, and a memory storing program instructions that cause the processor to: (Post discloses its decoding as software executed on GPU hardware: "This produces a large matrix of dimensions k x |VT|, that can be computed quickly with modern GPU hardware." (Post, Sec. 2, second para.) and "We compare it to GBS and demonstrate empirically that it is significantly faster, making constrained decoding with an arbitrary number of constraints feasible with GPU-based inference." (Post, Sec. 1, last para.), the implementation being released as part of a software toolkit: "Our implementation is available as part of SOCKEYE." (Post, Abstract, last sentence). A GPU executing a released software implementation is a processor executing program instructions held in memory.)
receive a constraint information list as an input and output, as a lexical constraint, each element of a subset of one or a plurality of pieces of constraint information included in the constraint information list; (Post receives a set of constraints and enforces them during decoding: "This algorithm takes a set of constraints, which are words that must appear in the output, and ensures that hypotheses have met all these constraints before they can be considered to be completed." (Post, Sec. 2, third para.) and "A word constraint provided to the decoder is a single token in the target language vocabulary." (Post, Sec. 3, second para.).)
generate one or more candidates for the second sequence by using the first sequence and the lexical constraint; and (Post maintains a beam of hypotheses that is extended at every decoding step: "In beam search (Lowerre, 1976; Sutskever et al., 2014), the decoder maintains a beam of size k containing a set of active hypotheses (Algorithm 1)." (Post, Sec. 2, second para.), and allocates that beam dynamically across the constraint banks: "Instead of maintaining C + 1 beams, we maintain a single beam of size k, as with unconstrained decoding. We then dynamically allocate the slots of this beam across the constraint banks at each time step." (Post, Sec. 2, last para.).)
calculate a score indicating appropriateness as the second sequence for each of the one or more candidates. (Each hypothesis is scored by the model probability and the k best are retained by that score: "The beam for the next time step is filled by taking the states corresponding to the k-best items from this entire matrix and sorting them." (Post, Sec. 2, second para.), the matrix being the model's distribution over the vocabulary for each hypothesis: "At each time step t, the decoder model is used to produce a distribution over the target-language vocabulary, VT, for each of these hypotheses." (Post, Sec. 2, second para.).)
Chousa and Post are considered to be analogous to the claimed invention because they are in the same field of lexically constrained neural machine translation. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to implement the restricted-translation system of Chousa as program instructions stored in a memory and executed by a processor, in the form of the GPU-executed constrained decoding implementation of Post. Chousa itself adopts the lexically constrained decoding of Post as its decoder ("We used the lexically constrained decoding (LCD) technique (Hokamp and Liu, 2017; Post and Vilar, 2018) to incorporate constraints at decoding time." (Chousa, Sec. 4.2, first para.)), so the modification amounts to running the decoder Chousa already relies on in the released form in which Post provides it. One of ordinary skill in the art would have been motivated to make this modification in order to obtain the decoding speed that Post expressly teaches its implementation provides, Post stating that its algorithm is "significantly faster, making constrained decoding with an arbitrary number of constraints feasible with GPU-based inference" (Post, Sec. 1, last para.).
Regarding claim 2, Chousa in view of Post teaches the generation apparatus of claim 1 as set forth above, and Chousa further discloses:
2. The generation apparatus according to claim 1, wherein the program instructions cause the processor to calculate the score on a basis of at least one of a likelihood output from a model used for generating each of the one or more candidates or a likelihood obtained from each of the one or more candidates by a re-ranking model. (The claim recites alternatives, of which the first is taught. The score Chousa assigns to each candidate is the conditional probability output by the same model that generates the candidate, the problem of translation with word constraints being defined as the conditional probability p(Y | X, C) of Eq. 2 ("Here, the problem of translation with word constraints can be defined as" (Chousa, Sec. 2, second para., and Eq. 2)), and the averaged model scores are the scores from which the hypotheses are generated: "We then computed the average scores of these models and generated hypotheses based on these scores using beam search decoding." (Chousa, Sec. 4.6, first para.). The conditional probability output by the translation model is a likelihood output from the model used for generating each candidate. Post likewise scores each hypothesis by the probability output by the decoder model performing the generation ("At each time step t, the decoder model is used to produce a distribution over the target-language vocabulary, VT, for each of these hypotheses." (Post, Sec. 2, second para.)).)
No further modification of the combination is required; the score of the combination as set forth for claim 1 is the likelihood output by the generating model.
Claim 5 is a method claim with limitations corresponding to the limitations of Claim 1 and is rejected under similar rationale. Additionally:
Regarding claim 5, Chousa discloses:
5. A generation method executed by a generation apparatus for generating, from constraint information and a first sequence that is a sequence of information, a second sequence that is a sequence of different information, the generation method comprising: receiving a constraint information list as an input and outputting, as a lexical constraint, each element of a subset of one or a plurality of pieces of constraint information included in the constraint information list; a sequence generation step of generating one or more candidates for the second sequence by using the first sequence and the lexical constraint; and a re-ranking step of calculating a score indicating appropriateness as the second sequence for each of the one or more candidates. (Chousa discloses that its restricted translation is a method carried out by its submitted system: "Our system combined input augmentation and constrained beam search algorithms." (Chousa, Abstract, third sentence). The constrained decoding run by the system of Chousa is a generation method executed by a generation apparatus. The receiving and outputting, the sequence generation step, and the re-ranking step correspond to the operations of claim 1 and are mapped above.)
Claim 6 is a computer program product claim with limitations corresponding to the limitations of method Claim 5 and is rejected under similar rationale. Additionally:
Chousa does not expressly disclose the recited non-transitory computer-readable recording medium storing the program.
Post discloses:
6. A non-transitory computer-readable recording medium storing a program for causing a computer to execute the generation method of claim 5. (Post discloses that its constrained decoding program is made available as part of a released software toolkit: "Our implementation is available as part of SOCKEYE." (Post, Abstract, last sentence). A program made available as part of a software toolkit is a program stored on a non-transitory computer-readable recording medium, from which the computer that executes it reads it.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to store the program that causes a computer to execute the constrained generation and scoring method of the combination of Chousa and Post, as set forth for claim 5, on a non-transitory computer-readable recording medium in the manner in which Post makes its implementation available. One of ordinary skill in the art would have been motivated to make this modification in order that the method be executed by a computer from the released toolkit, as Post expressly teaches ("Our implementation is available as part of SOCKEYE." (Post, Abstract, last sentence)).
Claims 3 and 4 are rejected under 35 U.S.C. 103 as being unpatentable over Chousa in view of Post as applied to claim 1 above, and further in view of Anderson et al., "Guided Open Vocabulary Image Captioning with Constrained Beam Search," Proc. EMNLP 2017, pp. 936-945, https://aclanthology.org/D17-1098, arXiv:1612.00576, https://arxiv.org/abs/1612.00576, hereinafter Anderson.
Regarding claim 3, Chousa in view of Post teaches the generation apparatus of claim 1 as set forth above. The combination of Chousa and Post does not disclose the following, which Anderson teaches:
3. The generation apparatus according to claim 1, wherein in a case where the constraint information list includes constraint information having an ambiguity, the program instructions cause the processor to generate the one or more candidates by performing a lexically constrained beam search in consideration of the ambiguity. (Anderson discloses a constrained beam search in which a single constraint is satisfied by any one of two or more alternative words: "To fix ideas, suppose that we wish to generate sequences containing at least one word from each constraint set C1 = {'chair', 'chairs'} and C2 = {'desk', 'table'}. Note that it is possible to recognize sequences satisfying these constraints using the finite-state machine (FSM) illustrated in Figure 2, with start state s0 and accepting state s3." (Anderson, Sec. 3.1, fourth para.), the search maintaining one beam per state of that machine: "Since RNN output sequences are generated from left-to-right, to generate constrained sequences, we take an FSM that recognizes sequences satisfying the required constraints, and use the following multiple-beam decoding algorithm. For each state s in S in the FSM, a corresponding search beam Bs is maintained." (Anderson, Sec. 3.1, fifth para.), and its worked example requiring "Each output sequence must include the words 'chair' or 'chairs', and 'desk' or 'table' from vocabulary V." (Anderson, Fig. 2 caption). A constraint that may be satisfied by any of several alternative words is constraint information having an ambiguity, and Anderson's multiple-beam search, which accepts any one of the alternatives as satisfying the constraint, is a lexically constrained beam search performed in consideration of that ambiguity.)
Chousa, Post, and Anderson are considered to be analogous to the claimed invention because they are in the same field of lexically constrained beam search decoding for neural sequence generation. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the lexically constrained decoding of the combination of Chousa and Post such that a constraint whose target word has two or more alternative surface forms is represented, as in Anderson, by a set of alternative words any one of which satisfies the constraint during the beam search. One of ordinary skill in the art would have been motivated to make this modification in order to satisfy a constraint whose required word may appear in the output in more than one form without post-processing, Chousa expressly teaching that its output must otherwise be repaired after decoding where the constraint form and the generated form differ ("For the submission, we need to match the tokenization to the reference constraints. To achieve this, we fixed the terms that are not matched to the constraints due to tokenization issues." (Chousa, Sec. 4.7, first para.)), and Anderson expressly teaching that its constrained beam search guarantees the inclusion of the selected words ("we address the first challenge (guidance) by using constrained beam search to guarantee the inclusion of selected words or phrases in the output of an RNN, while leaving the model free to determine the syntax and additional details" (Anderson, Sec. 1, third para.)), with a reasonable expectation of success because all three references track constraint satisfaction within a beam search.
Regarding claim 4, Chousa in view of Post teaches the generation apparatus of claim 1 as set forth above. The combination of Chousa and Post does not disclose the following, which Anderson teaches:
4. The generation apparatus according to claim 1, wherein the program instructions cause the processor to receive at least one piece of constraint information in a format in which two or more ambiguities are allowed, and generate a lexical constraint with the ambiguities retained. (Anderson receives each constraint as a set of alternative words, "each constraint set C1 = {'chair', 'chairs'} and C2 = {'desk', 'table'}" (Anderson, Sec. 3.1, fourth para.), and carries every alternative through the search in the finite-state machine that enforces the constraint, "Constrained beam search is an approximate search algorithm capable of enforcing any constraints over resulting output sequences that can be expressed in a finite-state machine." (Anderson, Sec. 1, third para.), the machine transitions accepting any of the alternatives (Anderson, Fig. 2, the transitions labeled C1 and C2). A constraint supplied as a set of two or more alternative words is a piece of constraint information received in a format in which two or more ambiguities are allowed, and the finite-state machine transition that accepts any of the alternatives is a lexical constraint generated with the ambiguities retained.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to further modify the combination such that the constraint information is received in the set-of-alternatives format of Anderson and the lexical constraint enforced by the decoder retains every alternative, for the reason given for claim 3.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
a. Chen et al., "Lexical-Constraint-Aware Neural Machine Translation via Data Augmentation," Proc. IJCAI-20, pp. 3587-3593: soft lexically constrained translation by data augmentation, in which the constraint phrases are appended to the source input and the model learns to copy them into the translation.
b. Li et al., "Neural Machine Translation with Noisy Lexical Constraints," IEEE/ACM Trans. Audio, Speech, and Language Processing, vol. 28, pp. 1864-1874 (2020), arXiv:1908.04664: soft lexically constrained translation in which noisy constraint words are supplied to the model and are not required to appear verbatim in the output.
c. Hu et al., "Improved Lexically Constrained Decoding for Translation and Monolingual Rewriting," Proc. NAACL-HLT 2019, pp. 839-850: vectorized dynamic beam allocation with multi-state tries and negative constraints for lexically constrained decoding.
d. Zhang et al. (US 10,114,819 B2): generation of n candidate translations from model-scored, ranked translation hypotheses.
e. Suzuki et al. (US 7,526,423 B2): selection among a plurality of translation word candidates by frequency or co-occurrence evaluation.
f. Le et al. (US 2019/0188268 A1): a bilingual word dictionary used to force target translations into neural machine translation output by token replacement.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to YUVAL H. LEVENTAL whose telephone number is (571) 270-3130. The examiner can normally be reached Monday-Friday, 8:00 AM - 5:00 PM.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, PIERRE-LOUIS DESIR, can be reached at (571) 272-7799. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/YUVAL HAIM LEVENTAL/Examiner, Art Unit 2659
/PIERRE LOUIS DESIR/Supervisory Patent Examiner, Art Unit 2659
Prosecution Insights