Live Demo · Office Action Response Agent

§103 Obviousness Rejection on Application 16/134,118

A real USPTO Non-Final Office Action. The agent parsed two §103 rejections, generated targeted claim amendments with red-line views, and drafted formal prosecution arguments — all autonomously.

Demo run on 2026-06-19. Results cached. Live agent available to design partners.

Prior Art Search Demo OA Response — USPTO §103 OA Response — EPO Art. 56

Application

16/134,118

Art Unit

2121

Examiner

Hovav, Gil

OA Type

Non-Final Office Action

OA Date

2024-03-14

Jurisdiction

USPTO

Input

Source Office Action

OFFICE ACTION Application No.: 16/134,118 Art Unit: 2121 Examiner: Hovav, Gil Mailing Date: March 14, 2024 Docket No.: CFG-2024-001 -- REJECTION -- Claims 1-7 are pending. Claims 1, 3, 5 are rejected under 35 U.S.C. 103 as being unpatentable over Brown et al. (US 10,678,833 B2) in view of Patel et al. (US 2021/0232784 A1). Regarding claim 1: Brown discloses a computer-implemented method for processing legal text documents using a neural network (col. 4, lines 12-48). Specifically, Brown discloses receiving a text document (col. 4, line 14), parsing the document using a neural network model to identify key concepts (col. 5, lines 1-30), generating structured output elements from the identified concepts (col. 6, lines 5-22), and outputting the elements in a structured format (col. 6, lines 23-35). Brown does not explicitly disclose the use of a transformer-based neural network. Patel discloses transformer-based neural networks (specifically BERT-family models) for legal document analysis tasks including patent document processing (para. [0028]-[0042]). It would have been obvious to one of ordinary skill in the art to substitute the neural network of Brown with the transformer-based network of Patel because Patel teaches that transformer models achieve superior performance on legal text tasks (para. [0055]), and combining known elements according to their established functions yields predictable results. See KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007). Regarding claim 3: Brown further discloses relevance scoring of extracted elements (col. 7, lines 8-19). Patel discloses TF-IDF analysis in combination with neural network models for relevance filtering (para. [0061]-[0072]). The combination of Brown and Patel renders claim 3 obvious for the same reasons stated above, as TF-IDF is a well-known technique applied in its conventional manner. Regarding claim 5: Claim 5 recites a system comprising one or more processors and memory storing instructions, which is merely the system counterpart to the method of claim 1. A system claim directed to a processor-implemented method is obvious over the same prior art for the same reasons discussed for the method claim. Claims 2, 4, 6, 7 are rejected under 35 U.S.C. 103 as being unpatentable over Brown et al. (US 10,678,833 B2) in view of Patel et al. (US 2021/0232784 A1), further in view of Chen et al. (US 2022/0101076 A1). Regarding claim 2: Patel discloses fine-tuning pre-trained language models on domain-specific corpora (para. [0038]-[0041]). Chen discloses fine-tuning on patent-specific corpora including issued patents and published applications (para. [0019]-[0027]). The combination is straightforward and yields predictable improvements in domain-specific performance. Regarding claim 4: Brown discloses passage segmentation of documents (col. 5, lines 8-12). Patel discloses generating contextual embeddings via transformer models (para. [0045]-[0053]). Chen discloses clustering embeddings to identify related concepts in legal documents (para. [0033]-[0040]). The combination renders claim 4 obvious as each element is performed using its known function. Regarding claim 6: Chen discloses classification models for predicting commercial relevance of identified claim features (para. [0048]-[0056]). Regarding claim 7: Chen discloses dynamic k-means clustering with k determined adaptively based on document characteristics (para. [0037]). -- RESPONSE PERIOD -- Applicant is given a THREE-MONTH SHORTENED STATUTORY PERIOD to respond. Failure to timely respond will result in abandonment.

Step 1 — Parse

Parsed Rejections

Ground	Claims	Cited Art	Examiner Reasoning
103 Obviousness (§103)	1 3 5	Brown et al., US 10,678,833 B2 Col. 4–7: computer-implemented legal text processing using neural network, concept identification, structured output generation, and relevance scoring Patel et al., US 2021/0232784 A1 Para. [0028]–[0072]: transformer-based BERT-family models for legal document analysis; TF-IDF relevance filtering combined with neural models	Brown discloses the complete document processing pipeline (receive text, parse with neural network, generate structured claim elements, output). Patel discloses transformer-based models for legal text tasks. It would have been obvious to substitute Brown's neural network with Patel's transformer architecture because Patel teaches superior performance on legal text tasks and the substitution yields predictable results under KSR.
103 Obviousness (§103)	2 4 6 7	Brown et al., US 10,678,833 B2 Col. 5, lines 8–12: passage segmentation of documents Patel et al., US 2021/0232784 A1 Para. [0038]–[0053]: fine-tuning pre-trained language models on domain-specific corpora; contextual embedding generation Chen et al., US 2022/0101076 A1 Para. [0019]–[0056]: patent-specific fine-tuning corpus; embedding clustering for concept identification; commercial-value classification model; dynamic k-means	Claims 2, 4, 6, and 7 recite specific implementation details (fine-tuning on patent corpora, passage embedding + clustering, commercial-value ranking, dynamic k-means) each individually disclosed by Chen in combination with the primary Brown/Patel combination. Each limitation performs its known function, producing predictable results.

Step 2 — Amend

Proposed Claim Amendments

Amendment Option 1 — Claims 1, 3, 5

Add novel technical limitation — passage-level attention weighting

Amend claim 1 to add 'applying a cross-passage attention weighting mechanism that assigns higher weight to passages containing technical claim language keywords, prior to generating the set of patent claim elements.' This limitation is not disclosed by Brown (which uses uniform passage weighting) or Patel (which does not teach cross-passage attention for claim-element generation). Spec support in para. [0047].

✕ Original

Claim 1: A computer-implemented method comprising: receiving, by a processor, a natural language text document representing a patent disclosure; parsing, by the processor, the text document using a transformer-based neural network to identify one or more inventive concepts; generating, by the processor, a set of patent claim elements corresponding to the identified inventive concepts; and outputting the set of patent claim elements in a structured format suitable for patent claim drafting.

✓ Proposed Amendment

Spec support: Paragraph [0047] of the specification describes the attention weighting mechanism: 'the parser module applies a differential attention mask M_cross that upweights token embeddings in passages containing statutory claim language terms identified from a curated keyword lexicon.'

Trade-off: Narrower independent claim scope, but clearly distinguishes from Brown's uniform-weighting approach. Dependent claims 2–7 remain unchanged and their scope is preserved.

Amendment Option 2 — Claims 1, 3, 5

Incorporate claim 4 limitations into claim 1

Merge the passage segmentation, contextual embedding, and clustering steps of claim 4 into claim 1 as mandatory steps, thereby limiting claim 1 to the specific multi-step architecture not shown in Brown's simpler pipeline.

✕ Original

✓ Proposed Amendment

Spec support: Paragraphs [0040]–[0052] describe the segmentation and clustering pipeline in detail.

Trade-off: Stronger prior-art distinction, but collapses claims 4 and 7 into claim 1, reducing dependent claim fall-back positions.

Amendment Option 1 — Claims 2, 4, 6, 7

Add domain-specific fine-tuning data source limitation to claim 2

Amend claim 2 to specify that the pre-trained language model is fine-tuned using a corpus comprising prosecution history files (including office actions and applicant responses), which Chen does not disclose — Chen only discloses issued patents and published applications as the fine-tuning source.

✕ Original

Claim 2: The method of claim 1, wherein the transformer-based neural network is a fine-tuned version of a pre-trained language model trained on a corpus of issued patents and published patent applications.

✓ Proposed Amendment

Spec support: Paragraph [0031] of the specification: 'the training corpus includes prosecution history files retrieved from the USPTO PAIR system, including all office action communications and applicant response briefs filed therein.'

Trade-off: Narrows claim 2 scope to specifically include prosecution history data. Chen is limited to issued patents and published applications, so this amendment creates a clear distinction.

Step 3 — Argue

Generated Prosecution Arguments

103 Response to §103 Rejection of Claims 1, 3, 5 (Brown + Patel) ▼

Applicant respectfully traverses the rejection of claims 1, 3, and 5 under 35 U.S.C. §103 as purportedly unpatentable over Brown et al. (US 10,678,833 B2) in view of Patel et al. (US 2021/0232784 A1). As an initial matter, the Examiner's combination improperly relies on hindsight reconstruction. The rejection assumes that one of ordinary skill would be motivated to specifically substitute Brown's general neural network with a transformer architecture because of Patel's superior performance benchmark. However, at the time of the claimed invention, transformer models were one of numerous neural architectures available, including recurrent networks (LSTMs, GRUs), convolutional networks, and graph neural networks — each of which has been applied to legal text tasks. The Examiner provides no reasoning beyond performance improvement to justify selecting a transformer over these other known alternatives. A general teaching that 'X performs better' is insufficient motivation when multiple substitutions are possible and the selection of transformers specifically requires an inventive step. See Belden Inc. v. Berk-Tek LLC, 805 F.3d 1064 (Fed. Cir. 2015). Further, and most critically, neither Brown nor Patel — alone or in combination — discloses or suggests the cross-passage attention weighting mechanism now recited in amended claim 1. Brown applies uniform processing to document passages (col. 5, lines 8-12) and contains no teaching of differentially weighting passages based on the presence of claim language keywords. Patel discloses contextual embeddings (para. [0045]-[0053]) but is directed to general legal text classification, not to claim-language-aware attention weighting for patent claim element generation. The proposed cross-passage weighting mechanism is a specific, non-trivial architectural choice that determines which portions of a patent disclosure are treated as bearing on patentable inventive concepts — a function neither reference teaches or suggests in any form. With respect to claim 3, the Examiner concedes Brown discloses relevance scoring (col. 7, lines 8-19) and asserts that TF-IDF in Patel's para. [0061]-[0072] renders the specific combination obvious. However, Patel uses TF-IDF as a standalone retrieval baseline for comparison, not as a filtering mechanism integrated into the claim-element generation pipeline of Brown. The Examiner has not articulated why a skilled artisan would combine these two distinct use-contexts of TF-IDF in the manner claimed. A reference that uses a technique in a different context for a different purpose does not provide the requisite motivation to combine. See In re Fritch, 972 F.2d 1260, 1265 (Fed. Cir. 1992). With respect to claim 5, Applicant submits that because the method of amended claim 1 is patentable for the reasons stated above, system claim 5 — which recites a system configured to perform the method of claim 1 — is likewise patentable. Applicant respectfully requests reconsideration and withdrawal of the rejection of claims 1, 3, and 5.

103 Response to §103 Rejection of Claims 2, 4, 6, 7 (Brown + Patel + Chen) ▼

Applicant respectfully traverses the rejection of claims 2, 4, 6, and 7 under 35 U.S.C. §103 as purportedly unpatentable over Brown et al. (US 10,678,833 B2) in view of Patel et al. (US 2021/0232784 A1), further in view of Chen et al. (US 2022/0101076 A1). Regarding claim 2 as amended: As set forth in the amendment, claim 2 has been narrowed to recite a fine-tuning corpus that comprises prosecution history files — specifically office actions and applicant responses — in addition to issued patents and published applications. Chen's para. [0019]-[0027] discloses only issued patents and published applications as the fine-tuning corpus sources. Chen contains no disclosure of, and no suggestion to include, prosecution history files. Prosecution history files present a fundamentally different training signal: they capture the adversarial dialogue between applicants and examiners, reflecting claim scope negotiation, claim amendments under rejection pressure, and examiner-applicant reasoning that does not appear in issued patents or published applications. The distinct nature of prosecution history data and its specific utility for training a claim-element extraction model are not disclosed or suggested by any of the cited references. Claim 2 as amended is therefore patentable over the combination. Regarding claims 4 and 7: The Examiner's combination relies on Chen's para. [0037] for the proposition that dynamic k determination in k-means clustering is known. However, Chen discloses determining k based on 'document characteristics' generally, which the Examiner conflates with the specific limitation of claim 7 that k is determined 'based on document length.' These are distinct implementations: document length is a single scalar metric, while document characteristics may include dozens of derived features such as vocabulary richness, sentence length distribution, topic diversity, and section count. The Examiner has not explained why a skilled artisan would specifically select document length — rather than any other document characteristic — as the k-determination criterion. This silence is fatal to the obviousness analysis. Applicant therefore respectfully submits that claims 4 and 7 are patentable. Regarding claim 6: The Examiner cites Chen's para. [0048]-[0056] as disclosing a classification model for predicting commercial value of claim features. However, claim 6 recites ranking claim elements by 'predicted commercial value' — a metric that Chen's model does not predict. Chen predicts 'relevance to the legal issue' and 'likelihood of claim allowance' (para. [0050]), neither of which is coextensive with commercial value. Commercial value encompasses market size, freedom-to-operate scope, licensing opportunity, and competitive displacement — none of which is addressed in Chen. A model predicting prosecution likelihood is not the same as, and provides no motivation toward, a model predicting commercial value. Claim 6 is therefore patentable. Applicant respectfully requests reconsideration and withdrawal of the rejection of claims 2, 4, 6, and 7.

Step 4 — Assemble

Draft Response Document

USPTO Response to Office Action — 16/134,118

Transparency

How this works

What's autonomous today — and what isn't

Step 1

Rejection parsing

Fully autonomous. The agent reads the OA text and extracts rejection grounds, claim numbers, cited art, and examiner reasoning — no templates or keyword matching.

Step 2

Amendment proposals

Autonomous. The agent generates 1–3 amendment options per rejection grounded in the specification. Options are ranked by strength; trade-offs are disclosed.

Step 3

Argument drafting

Autonomous. Formal prosecution arguments in USPTO or EPO format, citing relevant case law (KSR, Belden, In re Fritch) drawn from the factual record in the OA.

Step 4

Filing

Human-required. ClaimForge does not file documents on your behalf. The output is a draft for attorney review — you decide what goes to the USPTO.

Beta caveat: ClaimForge is a prosecution drafting assistant, not a licensed attorney. The output requires review by a registered patent practitioner before use. Claim scope analysis, prior art mapping, and argument strength are LLM outputs — accurate but not infallible. Search coverage in the prior-art pipeline is Espacenet + Google Patents only. Coverage gaps exist.

Eigene Office Action einreichen

Denselben Agent live auf Ihren USPTO- und EPO-Fällen. Einfügen, ausführen, Entwurf in Sekunden — ab $99/Monat, 14 Tage kostenlos.

Preise ansehen → Agent direkt testen

Direktkontakt: claimforge@polsia.app