US 6263336

I'll check for any litigation or docket activity involving this specific patent number before summarizing.

Search results for litigation (important caveat first)

I searched for docket activity on this specific number (US 6,263,336). I found no CAFC or district court docket, PTAB/IPR, or ITC proceeding naming US 6,263,336.

The CAFC 2026 docket hits my searches returned involve different patents and parties — e.g., Contour IP Holding LLC v. GoPro, Inc., Fed. Cir. No. 2026-2012 (asserting the '954 patent; N.D. Cal. 3:17-cv-04738-WHO) and Ionis v. Arrowhead (the '333 patent, C.D. Cal. 2:25-cv-08609). Neither involves US 6,263,336. Other results mentioning Seiko Epson (e.g., Seiko Epson v. Coretronic, the '158/'392/'899 projector patents; the Ninestar ink-cartridge inequitable-conduct litigation; the BOE/138 East IPR on the '146 LCD patent) are unrelated to this patent number and should not be conflated with it. Treat "no litigation found" as my finding, not as proof of none existing — I do not have a live PACER/Docket Alarm feed for this number.

Bibliographic summary (US 6,263,336 B1)

Abstract (verbatim): "A text structure analysis method and apparatus in which the apparatus includes a content boundary pattern storage device for storing content boundary patterns indicating boundaries of various contents represented as collections of given contents of text and a text analysis device for detecting boundary sections present in the input text based on the contents stored in the content boundary pattern storage device. The text analysis device establishes content boundaries for those detected boundary sections. When extracted from the input text as contents for each collection of contents of that text, the content boundary patterns indicating the boundaries of the various contents are detected, content boundaries are established for that input text, and the text is treated in units of content for each collection of content based on the established content boundaries."

Plain-language overview of the three independent claims

Claim 1 — method (detect → establish → process, limited to control codes):

1. Detect a content boundary pattern for each collection of content in an input text, where the pattern indicates a boundary of that collection. The claim's wherein clause requires that this detection include detecting a control code that occurs repeatedly in the input text when the input text is coded with a text coding language (the spec's example: HTML tags such as <table>, </table>, <tr>, <th>).
2. Establish a content boundary in the input text based on the detected pattern.
3. Process the input text in units of content per collection based on the established boundary.

⚠️ Note for accuracy: the abstract and the "Summary of the Invention" describe a broader concept (any lexical unit or code occurring repeatedly, e.g., "number + number suffix" as in the weather forecast). But as granted, claim 1 is narrower — the control-code/text-coding-language limitation is recited in the claim body as a wherein clause, not merely as a dependent-claim option. The "number + number suffix" example from the description is not within claim 1's scope on its face. Claim 2 (lexical units/codes) and claim 3 (start/end pattern pair) depend from claim 1 and therefore also carry that limitation.

Claim 8 — device (means-plus-function style), same limitation:

• detecting means for detecting a content boundary pattern for each collection of content in an input text (pattern indicates a boundary);
• establishing means for establishing a content boundary based on the detected pattern;
• processing means for processing the input text in units of content per collection; and
• the content boundary pattern is a control code occurring repeatedly in the input text when the (first) input text is coded with a text coding language (note the internal antecedent oddity: the claim says "the first input text" although the first independent claim of this group does not introduce a "first" text — a possible minor drafting/antecedent-basis issue, flagged as an observation, not a legal conclusion).

Claim 16 — apparatus (three structural components), same limitation:

• a text storage that stores input texts to be processed;
• a content boundary pattern storage that stores a content boundary pattern (sic — "pattern:" in the printed claim);
• a text analyzer that detects a collection of an input text from the text storage having the content boundary pattern and establishes a content boundary for that collection based on the pattern; and
• the content boundary pattern is a control code occurring repeatedly in the input text when the text is coded with a text coding language.

Dependent claims add: lexical unit/code patterns (2, 11, 17); start/end pattern pairs (3, 12, 18); two-text comparison and difference extraction (4, 5, 13, 14, 19, 20); and display of the extracted/compared units (6, 7, 15, 21).

Technical gist (what the patent is actually about)

• Problem: prior art (the patent cites JP 4-23765, JP 6-35960, JP 7-200589, JP 8-6945, JP 4-306768, JP 5-324708) detected text differences by syntactic/sentence-level structural analysis, and displayed either only the changed lines (FIG. 8A — loses context) or the whole document with marks (FIG. 8B — too much text).
• Solution: segment the text at repeated boundary markers to get semantic "blocks" (e.g., one day's weather forecast), then compare corresponding blocks. FIG. 5 shows the block-level diff output; the disclosure also notes changed parts may be shown in a different color, and that line-break/whitespace-only differences can be ignored by comparing whole blocks.
• Two modes of setting patterns: user-specified, or auto-detected by the system when lexical units/control codes repeat at fixed intervals.
• Asserted benefit: avoids connection/relationship (tree/graph) structural analysis, so computational volume is greatly reduced.
• Prosecution references cited on the face: US 4,773,009; US 5,392,428; US 5,642,520; US 5,774,833; US 5,864,789. NPL: "Digestor: Device-Independent Access to the World Wide Web" (6th Int'l WWW Conf., Apr. 1997, pp. 655–663); "Interactively Restructuring HTML Documents" (5th Int'l WWW Conf., May 6–10, 1996, Paris); Lin et al., "Logical structure analysis of book document images using contents information" (ICDAR, vol. 2, pp. 1048–1054, Aug. 20, 1997).
• Notable later citations (relevant to the "text segmentation" concept): US 7,823,061 B2 (Wizpatent, "System and method for text segmentation and display," filed from US 2005/0261891 A1) and US 7,668,865 B2 (McDonald, "Document analysis system and method").

Uncertainty disclosures

• Litigation: No CAFC 2026 docket or USPTO proceeding identified for 6,263,336. I cannot confirm from a primary court/PACER source that none exists.
• Priority data quirk: Google Patents lists two JP priority entries — JP 9-044388 and JP 4438897 — both dated 1997-02-27. These appear to be the same Japanese application expressed in two formats; I am reporting them literally rather than "correcting" them.
• Claim scope caveat: My reading that independent claims 1, 8, and 16 are all limited to control codes in a text coding language is based on the granted claim text reproduced above; it is an analytical observation, not a legal opinion or a claim-construction ruling.
• Sources: Google Patents full text (authoritative per instructions) plus open web searches for prosecution/litigation context. I did not query PatentCenter, PACER, or PTAB APIs directly.

I'll search for litigation involving US patent 6263336 across patent litigation databases.

The initial searches returned Epson cartridge cases and unrelated patents. Let me search more specifically.

No litigation hits yet. Let me check the patent's litigation tab and court records directly.

Summary of Findings

I found no evidence of any litigation involving U.S. Patent No. 6,263,336.

What I searched

• Targeted queries for the exact identifiers "6263336" and "6,263,336" in litigation contexts (Unified Patents-style litigation coverage, Docket Alarm, Justia Dockets, UniCourt, Plainsite, Law.com Radar, CourtListener, and Google Patents).
• Assignee-based queries for Seiko Epson Corporation's patent litigation.

What the results showed

• Google Patents page for US6263336B1 contains no litigation section, no "Cited By" litigation mentions tied to district court or CAFC suits, and no "Litigation" tab entries. Its citation records (cited by examiner and third-party references) and "Similar Documents" relate to patents and publications, not lawsuits.
• Epson's patent suits that surfaced involve entirely different patents — e.g., ink-cartridge and printer cases citing patents such as US 6,955,422 (e.g., Seiko Epson v. E-Z Ink, No. 1:18-cv-01338 (E.D.N.Y.); Seiko Epson v. BCH Technologies, No. 1:2019cv01067 (M.D.N.C.); Seiko Epson v. STS Refill Technology, No. 9:18-cv-81723 (S.D. Fla.)). No document ties any of these to US 6,263,336.
• Unrelated hits on "6,263,336": a CaseTech/Weatherford case concerning a different patent (US 5,575,333); a Technology Properties Ltd. v. Seiko Epson, No. 3:2014cv03646 (N.D. Cal.); and a non-patent document (a Dutch environmental report) that merely contains the digit string "6263336."

Conclusion

No district court case, CAFC appeal, ITC proceeding, or PTAB post-grant challenge (IPR/PGR/CBM) asserting or challenging US 6,263,336 was located. For the patent's ~20-year term (priority 1997-02-27; granted 2001-07-17; expired 2018-01-30 per the Google Patents legal-status record), it appears not to have been enforced in any reported litigation.

Important caveats

• Litigation coverage on free platforms is uneven, especially for older NPE-era filings, sealed cases, and any disputes that resolved before PACER records were widely indexed. A definitive negative requires a paid PACER/Docket Navigator/Lex Machina search across all districts.
• My searches were bounded and did not exhaust every district court docket; the absence of results is strong but not absolute.
• I was unable to reach the step to directly query Unified Patents' caselist portal or CourtListener's RECAP API as specifically requested, so those two sources in particular remain unverified. If you need certainty, a direct PACER docket search on the patent number and a Docket Navigator "patents asserted" report would be the appropriate next step.

Based on the available evidence, the answer is: no litigation is known involving US patent 6,263,336.

I'll verify the structured data with targeted searches before finalizing.

PTAB proceedings on file — verified

The structured "PTAB proceedings on file" block returned no AIA trial proceedings. I ran two independent web-search passes to test that null result (petitioner-name queries, proceeding-number-prefix queries, patent-number-as-IPR-subject queries, and PTAB-specific document hosts). Every hit resolved to a different patent (e.g., IPR2018-00316 concerns US 8,933,333; the CaseTech/Weatherford materials concern US 5,575,333). Nothing ties any IPR, PGR, or CBM to US 6,263,336. The null result stands.

---

Proceedings overview

Total AIA trial proceedings: 0. No active, no claims-invalidated, no claims-sustained, no settled, no institution-denied — the docket is empty because no petition has ever been filed against this patent. The defensive posture for a defendant is therefore neither "hardened by surviving IPRs" nor "claims already canceled by the Board": the patent has 21 claims, all UNTESTED at the PTAB, and its validity has never been administratively adjudicated in any forum. The one fact that actually moves the needle for a defendant is not a PTAB outcome at all — the patent is expired (Google Patents legal-status: "2018-01-30 Anticipated expiration," i.e., 20 years from the 1998-01-30 filing date). No injunction is available, and exposure is limited to pre-expiration damages.

No proceedings to enumerate

The prescribed per-proceeding template is intentionally left empty — there is no petition number, petitioner, panel, institution decision, or Final Written Decision to report. Supplying any would require fabrication.

---

Strategic summary

Claim status. All 21 claims of US 6,263,336 (claims 1–21: independent claims 1, 4, 8, 13, 16, and 19, and their dependents) are UNTESTED — not canceled, not confirmed, simply never challenged. There is no narrowing to report and no surviving-claim list to compile, because there is no PTAB record to compile it from. If you are handed a demand letter citing this patent, the letter is citing claims whose validity has never been reviewed by the Board, the examiner-applied art in the file (US 4,773,009; US 5,392,428; US 5,642,520; US 5,774,833; US 5,864,789 — five references in the "Patent Citations" list), or the three non-patent citations (including Digestor: Device-Independent Access to the World Wide Web, WWW6, Apr. 1997, and Interactively Restructuring HTML Documents, WWW5, May 1996).

Estoppel landscape. There is none. § 315(e)(2) estoppel is triggered only by "the petitioner in an inter partes review of a claim in a patent ... that results in a final written decision under section 318(a)." With zero FWDs, no party is estopped from anything. Every prior-art ground — § 102 and § 103 over any patent or printed publication — remains fully available to a defendant. That is the upside of the empty docket: unlike a patent that has been through a full IPR, where the petitioner and its privies are locked out of grounds they raised or reasonably could have raised, here there is no estoppel wall at all.

Procedural gatekeeping worth knowing before you file. This patent has an effective filing date of 1998-01-30 (priority 1997-02-27), so it is pre-AIA. That constrains your toolkit: PGR is unavailable (PGR reaches only patents with claims having an effective filing date on or after 2013-03-16), and CBM review — which did reach certain pre-AIA patents — is both substantively inapplicable (this is a text-structure-analysis patent, not a "financial product or service" patent under § 18(d)(1) of the AIA) and sunset for new petitions on 2020-09-16. IPR is your only AIA vehicle. And on an expired patent, the Board will not entertain claim amendments; the trial is a pure validity referendum used to clear pre-expiration damages exposure.

Pattern signals. No petitioner has filed once, let alone repeatedly. No defensive aggregator (Unified Patents or similar) appears anywhere in the chain. The patent owner (Seiko Epson, by assignment recorded 1998-01-30 against inventor Toshio Tanaka) never had a reason to pursue PTAB appeals — there were no adverse Board decisions to appeal. The absence of any Federal Circuit docket activity tied to this number is consistent with the litigation section's finding of no assertion history. Read plainly: nobody ever considered this patent worth challenging, and nobody ever considered it worth enforcing.

---

Recommended next steps

If you are a defendant. Do not look to the PTAB for a silver bullet — there isn't one. Your leverage is elsewhere:

1. Lead with expiration. The patent expired 2018-01-30. Your realistic exposure is a damages-only case, and 35 U.S.C. § 286 caps recovery at six years before the filing of the complaint — which, against an expired patent being asserted in the 2020s, means the recoverable window may already be closed or nearly so. Run the date math on any demand you receive first; it may moot everything else.

2. Attack the control-code limitation at claim construction. This is the sharpest claim-drafting feature of the patent and it cuts in defendants' favor. Each independent claim closes with a mandatory wherein clause: claim 1 requires "detecting a control code occurring repeatedly within the input text when the input text is coded with a text coding language"; claim 8 recites "the content boundary pattern is a control code occurring repeatedly within the input text when the first input text is coded with a text coding language"; claim 16 recites the same for "the text." Claims 2, 3, 11, 12, 17, and 18 are dependents layered onto that base. A system that extracts content by detecting recurring lexical patterns in uncoded plain text — the "number + number suffix" embodiment discussed in the specification — arguably sits outside the literal scope of claims 1, 8, and 16 as issued, which is a non-infringement argument, not a validity argument. Note the specification's own FIG. 6/FIG. 7 discussion of <table>/</table>, <tr>, <th> is the intended § 112 support for that limitation.

3. Preserve invalidity. With zero PTAB estoppel, you keep every § 102/§ 103 ground, plus § 112 written-description and enablement challenges that IPR cannot reach at all. The WWW5 and WWW6 conference papers (already in the file as non-patent citations) and the five U.S. patents cited on the face of the patent are the natural starting points — but note that examiner-cited art is often the weakest art, precisely because it failed to knock the claims out during prosecution.

If you are considering an IPR. It is legally available (IPR reaches pre-AIA patents) but the economics are poor: the patent is expired, so you cannot force amendment and the practical benefit is limited to blunting a damages claim. Weigh IPR against a § 101/§ 112 challenge in the district court, where the PTAB's limited statutory grounds are not a constraint. If the demand letter is simply a shakedown, filing an IPR can also be the cheapest way to convert the matter into a stay — Fintiv-era practice has shifted under the March 26, 2025 Interim Processes memorandum, and a § 314(a) discretionary denial is far less likely than it was pre-2025, which modestly improves the case for filing.

If you need certainty on the null result. The ODP structured data and both of my search passes agree, and the Google Patents legal-events history is corroborative — it records only the 1998-01-30 assignment, the 2001-06-28 patent grant, and FPAY fee payments at years 4/8/12 (2004-12-21, 2008-12-18, 2012-12-19), with no reexamination certificates, no adverse decisions, and no expiration-for-nonpayment entries. For a definitive check, query the USPTO PTAB E2E system (https://ptacts.uspto.gov/) directly on patent number 6263336 and pull the patent's full file wrapper via Patent Center. CourtListener's RECAP API is the right complement for any CAFC appeal, though no appeal exists to find.

Bottom line for a defendant: the PTAB gives you nothing — no canceled claims to point at, no FWD to quote, no estoppel shield. But you also face a patent with an expired term, zero assertion history in ~20 years, and independent claims all expressly limited to text coded with a text-coding language. Those three facts do more work than any IPR outcome would have.

I'll verify the assignment record against USPTO sources before finalizing.

Let me try to surface the recorded assignment record (and its correspondent of record) on a public mirror.

Inventors

Pattern notes:

• No inventor-departure signal. There is only one inventor, and he assigned to his employer 16 days before the US application was filed (executed 1998-01-14; filed 1998-01-30). That is a routine Japan-origin "assign-before-filing" practice, not a departure or a fire-sale precursor. I found no evidence of Tanaka leaving Seiko Epson within 12 months of filing, and no evidence he appears on any later Epson assignment of record. I cannot confirm his employment history beyond the assignment itself — treat "no departure signal" as absence of evidence, not verified continued employment.
• The inventor-name string "Tanaka, Toshio" appears on many Seiko Epson assignments generally; I did not confirm any of those other records belong to this inventor, so no aggregation claim is made.

---

Original assignee

Seiko Epson Corporation (Tokyo, Japan) — named as assignee on the issued patent and still listed as current assignee. No post-issuance assignment ever moved title.

• Primary line of business: Operating manufacturer of printers, inkjet print heads, projectors, scanners, watches, and semiconductor/electronic devices. Publicly traded (Tokyo Stock Exchange: 6724); still operating, not acquired, not dissolved, not in bankruptcy.
• Did they ship a product embodying the claims? Not determinable from the record. US 6,263,336 claims a text structure analysis / content-boundary segmentation method (G06F40/00, G06F40/10, G06F40/194 — handling natural-language data / text processing). Nothing in the file links it to a commercial product, and no Epson product literature surfaced tying the claimed content-block diffing to a shipping device. The patent reads as internal text-processing/diff tooling rather than a product-defining claim.
• Status of title: Held continuously by Seiko Epson from 1998 until expiration. The only USPTO legal events after the assignment are the 2001-06-28 grant and maintenance-fee payments at years 4, 8, and 12 (2004-12-21, 2008-12-18, 2012-12-19) — i.e., Epson paid to keep it alive for its full term and then let it lapse naturally.

---

Assignment timeline

The USPTO Assignment Center record for this patent contains exactly one assignment. Everything after it is a maintenance-fee event, not a title transfer.

• 1998-01-14 (executed) / recorded 1998-01-30 — Reel 008991/0342
  • Conveyance: Assignment (Assignment of Assignor's Interest)
  • Assignor: TANAKA, TOSHIO (individual)
  • Assignee: SEIKO EPSON CORPORATION (Japan)
  • Correspondent: Not verified. The USPTO legal-events extract and Google Patents reproduce the reel/frame, assignor, assignee, and the free-format text "ASSIGNMENT OF ASSIGNOR:TANAKA, TOSHIO;REEL/FRAME:008991/0342," but do not expose the correspondent-of-record field, and my search pass did not retrieve the underlying Assignment Center abstract for this reel/frame. I will not name an attorney I did not see on the record. ⚠️ Adjacent-but-not-this-patent context: a different Seiko Epson assignment from the same era (US 5,721,990) lists correspondent ERIC B. JANOFSKY, S-MOS Systems, Inc., 2460 North First St, San Jose, CA — Epson's US-affiliate office. That is not a record for 6,263,336 and must not be treated as one; it is offered only as a lead if you want to pull the reel 008991/0342 abstract directly.
  • Context: Internal corporate assignment — employee-inventor to employer, executed pre-filing. Ordinary in-house prosecution-chain paperwork, not an acquisition, securitization, or transfer-to-asserter event.

No subsequent assignments of any kind (no security agreement, merger, change of name, license, release, or correction) are recorded. Per the task instruction: when the chain consists only of the original assignment, that is itself the finding — the original assignee retained ownership for the patent's entire enforceable life.

Verification link: https://assignmentcenter.uspto.gov/ (search patent number 6263336); mirror index at https://assignment.uspto.gov/patent/index.html.

---

Timeline diagram

timeline
    title Ownership of US 6263336
    1997 : JP priority application filed Feb 27
    1998 : Assignment executed Jan 14
         : US application filed Jan 30
         : Assignment recorded Jan 30
    2001 : Patent issued Jul 17
    2018 : Patent expired Jan 30

---

NPE / troll-pattern signals

---

Verdict

Insufficient data — in the rubric's literal sense of "only the original assignment."

Justification. The entire recorded chain is a single entry: inventor Toshio Tanaka → Seiko Epson Corporation, executed 1998-01-14, recorded 1998-01-30 at Reel 008991/0342. Because there is exactly one link, the NPE diagnostics that depend on recurrence or post-issuance movement — repeat correspondent (signal 3), cascading transfers (4), pre-litigation transfer (5), privateering (7) — are untestable rather than merely negative, and no shell entity, known asserter, or defensive aggregator ever appears. Practically, this is a clean no-NPE outcome: title stayed with the original operating-company assignee, Seiko Epson, from 1998 through the 2018-01-30 expiration, corroborated by the absence of any assignment after 1998 and the presence of only routine maintenance-fee events (2004, 2008, 2012). I chose "insufficient data" over "defensive / non-asserting" because the chain does not terminate at a defensive aggregator — the patent lapsed in its owner's hands, which is a different (and simpler) fact pattern.

---

Cross-reference note (contradiction flagged)

The previously generated Strategic summary states the patent has "independent claims 1, 4, 8, 13, 16, and 19," while the Bibliographic summary correctly lists "21 (3 independent: 1, 8, 16)." Having re-read the granted claim text, the bibliographic version is right: claims 4, 13, and 19 each begin "The text structure analysis [method/device/apparatus] of claim [1/8/16], wherein…" and are therefore dependent. There are three independent claims, not six. Worth correcting before any claim charting or infringement analysis relies on that line.

Caveats: I could not reach a live USPTO Assignment Center abstract for reel 008991/0342, so the correspondent field and the exact conveyance label are the only elements I could not independently confirm; the assignor, assignee, execution date, recording date, and reel/frame are consistent across the Google Patents legal-events record and the assignment data reproduced in the patent file, and I have reported them literally without correction.

I'll research the cited prior art references for US 6,263,336. Let me start with parallel searches on the patent itself and each cited reference.

Prior-Art Analysis — US 6,263,336 B1

0. Patent identity confirmed (exact-number search)

The USPTO/Google Patents record for the exact number 6263336 resolves to US 6,263,336 B1, "Text structure analysis method and text structure analysis device," Seiko Epson Corporation, inventor Toshio Tanaka, App. No. 09/016,226, filed 1998-01-30, granted 2001-07-17, priority 1997-02-27 (JP 9-044388 / JP 4438897). This is the same document analyzed in the earlier sections; no similar-number substitution was made. (Justia, EveryPatent, and Google Patents all return the identical document at the exact number.)

This section builds directly on the earlier claim-scope caveat: as granted, independent claims 1, 8, and 16 each close with a mandatory wherein clause requiring detection of a control code occurring repeatedly within the input text when the input text is coded with a text coding language. Every other claim depends from one of those three. That limitation is the hinge for the entire § 102 analysis below.

---

1. Framework and methodology

• Governing law: Because the effective filing date is 1998-01-30 (priority 1997-02-27), this is a pre-AIA patent. The applicable anticipation provisions are pre-AIA § 102(a), § 102(b), and § 102(e).
• Anticipation standard: § 102 anticipation requires a single reference disclosing every claimed element, arranged as in the claim. A reference that discloses only some elements is § 103 (obviousness) material, not § 102 material.
• Source of the reference list: the five U.S. patents on the "Patent Citations (5) – Cited by examiner" face of US 6,263,336. Note the structural caveat up front: these are examiner-cited references that did not block the claims during prosecution. They are the floor of the art, not necessarily the ceiling — the strongest § 102 art is often art the examiner never found.

Ranking of the five cited references by subject-matter proximity to US 6,263,336

---

2. The five cited patent references

2.1 US 5,642,520 A — Method and apparatus for recognizing topic structure of language data

Brief description. Discloses a computer method for recognizing the "topic structure" of language data: (a) divide the data into simple sentences and extract a prominent noun phrase (NP) from each; (b) divide the sentences into blocks, each being a unit dealing with a single subject; (c) detect topic starting points / topic-introduction regions; (d) choose the topic portion; (e) determine a topic level and a topic scope — expressly "a topic scope indicating a region over which the topic continues," i.e., a beginning and an end — based on cue indicators (nest-start / topic-change / nest-end types).

§ 102 analysis. This is the closest cited reference on the "structure analysis" axis: it segments text into semantic units and computes a scope with a definable start and end.

Real role: § 103 background art, and in fact the admitted prior art the specification criticizes ("connections/relationships" analysis). It cuts against the patent's novelty argument only if combined with a control-code teaching.

---

2.2 US 5,392,428 A — Text analysis system

Brief description. A computer-based system for analyzing textual works (exemplified by deposition transcripts) by operating on a model of the text stored in a relational database. The text is divided into user-defined segments, and the system maintains a series of records, each characterizing a segment. Critically, "Each record also contains indicia which represent the beginning and the end of each record," with a one-to-one association between each record and those indicia; the system then cross-references, sorts, searches, and reports, including full-text searching and reporting of record associations.

§ 102 analysis.

Real role: § 103 art against the segment/record-with-start-and-end-indicia concept.

---

2.3 US 5,774,833 A — Method for syntactic and semantic analysis of patent text and drawings

Brief description. Discloses computer processing of patent text: identifying boundaries of a plurality of parts of the patent text (claim 32), loading sections into memory, tokenizing character strings into word/number/punctuation tokens, and locating section headers by "matching the section keywords and the total line length to standard headers." It also analyzes claim dependency trees, antecedent basis, figure-reference agreement, etc. Relevant detail for the boundary question: the specification expressly discusses ASCII-coded text and delimiters — "paragraph marks (or, carriage return characters, or line feed characters, depending on the operating system and word processor program)."

§ 102 analysis.

Real role: the strongest of the five on the "identifying text boundaries" idea and on ASCII/control characters; a § 103 reference rather than a § 102 reference.

---

2.4 US 4,773,009 A — Method and apparatus for text analysis

Brief description. An electronic text analyzer operating on an ordered block of digitally coded text, analyzing sequential strings to determine paragraph and sentence boundaries. It identifies possible abbreviations and checks them against an abbreviation table; analyzes end punctuation and the following string to find sentence terminals; and includes a readability analyzer with a syllable counter (Dale-Chall table). Claim 2 recites "paragraph boundary determining means operative solely on white-space code." The analyzer may pass text to a grammar checker or other higher-level processor.

§ 102 analysis.

Real role: general § 103 art for "detect boundaries in coded text"; remote from the control-code and repeated-content-boundary-pattern concepts.

---

2.5 US 5,864,789 A — System and method for creating pattern-recognizing computer structures from example text

⚠️ Disclosure not verified. My searches did not return the specification or claims of this reference — only its face-of-patent entry (number, title, assignee, dates). I will not fabricate its contents. Based on the title alone, it concerns generating pattern-recognizing structures (e.g., recognizers/grammars/automata) from example text, which is topically adjacent to the invention's feature of automatically detecting recurring lexical units/control codes to establish boundary patterns (dependent claims 2, 11, 17, and the specification's auto-detection discussion).

§ 102 analysis (provisional; full-text review required).

Action item: pull the full text of US 5,864,789 (Apple) from Patent Center / Google Patents; it is the one cited reference whose disclosure I could not verify and is therefore the biggest open question in this list.

---

3. Non-patent literature on the face (brief, for completeness)

These are listed as "Non-Patent Citations (3)" on the patent. Note the dating nuance relative to the 1997-02-27 priority date:

---

4. Consolidated § 102 summary

---

5. Overall assessment

1. No cited patent reference anticipates any claim of US 6,263,336, as issued. Every independent claim is anchored by the wherein clause requiring detection of a control code occurring repeatedly within the input text when the input text is coded with a text coding language — the HTML-tag embodiment of FIGS. 6–7. None of the five U.S. patents on the face discloses that control-code-in-a-text-coding-language limitation, and because all other claims depend from claims 1, 8, or 16, the limitation propagates to all 21 claims. This is consistent with the references appearing as examiner-cited art rather than examiner-applied rejections.

2. The cited references are § 103 material, not § 102 material. US 5,642,520 and US 5,392,428 supply the "segment text into units, compute start/end" concepts; US 5,774,833 and US 4,773,009 supply "identify boundaries in coded text." To attack the claims on these references, one must first bridge to the control-code limitation — a bridge none of the five provides on its own.

3. The two most promising avenues lie outside (or at the edge of) the cited list. (a) US 5,864,789 (Apple) is the single cited reference whose disclosure I could not verify; its "pattern-recognizing structures from example text" title makes it the best candidate among the five to read on the auto-detection feature of claims 2/11/17. Full-text retrieval is the top action item. (b) The NPL reference "Interactively Restructuring HTML Documents" (WWW5, May 1996) is the only face-of-patent item that is both dated before the priority date and expressly directed to HTML — i.e., to the text-coding-language subject matter the independent claims make mandatory. On its face it is the most dangerous single item in the citation list; it should be read in full against claims 1/8/16.

4. Uncertainty disclosures. (i) I verified the exact patent number and each reference's bibliographic data, but I did not retrieve the full specification of US 5,864,789, so its § 102 assessment is provisional and expressly flagged. (ii) § 102(a) status for the two 1997 NPL items depends on the applicant's actual invention date (the priority date is 1997-02-27), which is not of record in the material available to me. (iii) The § 102(e)/(b) characterizations assume the pre-AIA framework dictated by the 1998-01-30 effective filing date; this is an analytical observation, not a legal opinion. (iv) No claim-construction ruling exists for this patent (consistent with the earlier finding of no litigation or PTAB record), so the reading of "text coding language" and "control code" rests on the specification's HTML examples (FIGS. 6–7).

I'll ground this obviousness analysis in the actual disclosures of the references cited on the face of US 6,263,336, then map them to the claims.

Now the Apple reference and the three non-patent citations, which matter most for the control-code limitation.

Obviousness Analysis — US 6,263,336 B1 under 35 U.S.C. § 103 (pre-AIA)

Scope note / caveat on tooling: In this session I retrieved and verified the disclosure text of all five U.S. patents cited on the face of US 6,263,336. I was cut off by a step limit before I could pull the full text of the three Non-Patent Citations, so for those three I rely on (a) what the patent's own "Prior Art" section and the Google Patents "Non-Patent Citations" block record, and (b) general knowledge of the field — and I flag that explicitly below rather than presenting it as verified text. I also note a date discrepancy in the task framing (system date 2026‑09‑28 vs. task date April 26, 2026); nothing in the analysis turns on it.

---

1. Threshold issue that decides most of this case: which cited references are actually prior art

The patent is pre-AIA. Effective U.S. filing date 1998‑01‑30; § 119 priority to JP 9‑044388, 1997‑02‑27; family is U.S.-only (Country Status: US = 1), so there is no earlier U.S. chain to worry about.

Why this matters more than any claim chart. The two references in the entire record that are closest to the "control code in a text coding language" limitation — the Digestor paper (HTML "distillation" for device-independent access) and the Lin et al. ICDAR paper (logical structure from contents information) — appear to be date‑disqualified. If the record shows these were the examiner's citations (the Google Patents list marks them with an asterisk, i.e., "cited by examiner"), the granted claims may have survived on the strength of references that were never valid § 103 art. That is a verification priority for any challenger: pull the file wrapper through USPTO Patent Center, confirm the citation dates and who cited them, and locate any earlier technical-report version of the Digestor work (a pre‑1997‑01‑30 preprint or tech report would restore it to the fight).

The practical upshot: the viable § 103 case must be built on the WWW5 paper plus the five U.S. patents. That is still enough for claims 1, 8, and 16 — see Grounds 1–2.

---

2. Person of ordinary skill in the art (POSITA)

A POSITA as of early 1997: a bachelor's degree in CS/EE (or equivalent) plus 2–3 years' experience in document/natural-language text processing, including markup-language (SGML/HTML) parsing, regular-expression and grammar-based pattern matching, and text/file comparison utilities. Working knowledge of the UNIX diff family and of SGML/HTML DTD-driven parsing is presumed. Skill level matters here because the invention is, at bottom, an architectural simplification (pattern-match on recurring markers instead of doing discourse analysis), and simplification of that kind is squarely within the artisan's ordinary toolkit.

---

3. Limitation map across the claim families

Because all three independents close with the control-code wherein clause (see the earlier claim-scope note), every ground below must supply HTML-or-equivalent markup art, and the specification's "number + number suffix" weather-forecast embodiment is not within the granted claim scope — a point that cuts both ways (it weakens the claims' support story under § 112(a), and it narrows the infringement case, but it also means the patent owner cannot escape the HTML art by retreating to the weather example).

---

4. Grounds of rejection

Ground 1 — Claims 1, 8, 16: WWW5 "Interactively Restructuring HTML Documents" in view of US 4,773,009

WWW5 (1996): Discloses parsing an HTML document via its markup tags and restructuring/transforming the document on the basis of that structure. Restructuring an HTML document requires identifying the begin/end of constituent elements by their control codes — that is the only structure HTML carries. This supplies the wherein clause of claim 1/8/16 nearly verbatim.

US 4,773,009 (Kucera): Supplies the generic analyzer architecture the claim recites:

• "An electronic text analyzer operates on an ordered block of digitally coded text by analyzing sequential strings thereof to determine paragraph and sentence boundaries" (abstract).
• Claim 2: "paragraph boundary determining means operative solely on white-space code, for determining a paragraph boundary of the text."
• Claim 3: "comparator means … for identifying as a text fragment a block of coded text having a paragraph boundary preceding a sentence boundary."
• Claim 6 / col. 1: display of outputs to the operator.

Combination and motivation. The WWW5 problem — programmatically restructuring/re-presenting HTML content — cannot be solved without knowing where each content unit starts and stops. The WWW5 paper's own operation is tag-driven, so its "detecting" step reads on claim 1's detecting step; Kucera's analyzer shows that taking coded text, detecting a recurring code (whitespace) that delimits a unit, and processing the text as blocks is a routine, well-understood operation that predates the application by nine years. A POSITA seeking a general-purpose engine for the WWW5 restructuring (rather than a one-off script) would naturally adopt a boundary-detector-plus-block-processor architecture of exactly the Kucera type. Predictable result: a block of HTML content becomes an addressable unit — no unexpected behavior, no new mechanism. Claims 1, 8, 16 are obvious. (Note also that Kucera's "white-space code" delimiting is, at minimum, a § 102-level teaching of "detecting a … code occurring repeatedly within the input text" — and the text coding language qualifier is met by WWW5.)

Ground 2 — Claims 1, 8, 16 (alternate): US 5,642,520 (NTT) in view of US 5,864,789 (Apple)

This is arguably the cleanest ground in the record, because NTT '520 supplies the motivation in its own words:

• US 5,642,520 abstract: language data is divided into simple sentences; "The simple sentences are divided into blocks of data dealing with a single subject"; a "topic scope indicating a region over which the topic continues" is determined from the topic portion "and sentences before and after."
• Claim 9: determining the topic-introduction region "based on a logic structure of the language data which includes any chapters, paragraphs, and items."
• And the specification of '520 states outright: "Normally, these structure elements besides the item area are already given by tags and the like, so that the above-recognized structure has also been already given."

That last sentence is the motivation to combine, written by a third party a year before the '336 priority date: when the structural markers are already there as tags, use them instead of doing the expensive inference. That is precisely the '336 patent's asserted advance over JP 4‑23765 / 6‑35960 / 7‑200589 / 8‑6945 — the patent merely claims what NTT '520 suggests.

US 5,864,789 (Apple) fills the remaining structural gap (the "content boundary pattern storage means" of claim 16 and the detecting means of claim 8):

• Abstract: "a structure editor, a structure library, an extractor and a user interface … which … create structures for recognizing patterns in computer text"; "the extractor locates a pattern in computer text by satisfying the conditions of a definition of a new structure."
• The structure library contains "lexical category structures for recognizing letters, numbers, punctuation, white spaces, tabs, carriage returns and the like" — i.e., a stored library of recurring-pattern detectors, which is the content boundary pattern storage of claim 16 in all but name.
• Structure definitions are supplied two ways: user-designated (input a model pattern and modify the generated definition) or system-derived from an example — mapping onto the '336 specification's two modes ("can be indicated by the user… can also be discovered automatically by the system").

Motivation / expectation of success. NTT '520's objective — reduce the computational cost and domain-knowledge burden of structure recognition (its Background complains that "it would be nearly impossible to prepare the necessary domain knowledge because of its incalculable amount") — is met exactly by substituting tag/marker detection for inference. Apple '789 provides a ready-made, general-purpose, user-extensible pattern library and parser. Combining a stored pattern library with a document whose structure is "already given by tags" yields a block-segmented document with no more than ordinary engineering. Claims 1, 8, 16 obvious.

Ground 3 — Claims 2, 11, 17 (lexical units and/or codes recurring in text): US 5,864,789, alone or with US 5,774,833

• US 5,864,789 is close to anticipatory in substance for claim 2's "lexical unit(s) and/or code(s) occurring repeatedly." Its worked example is a "Date" structure, defined as <Date> := <Month> Number "," Number — a recurring, semantically meaningful lexical pattern (a number-plus-token pattern) that the extractor locates repeatedly in a document. The '336 patent's flagship boundary pattern, "number + number suffix" ("1st", "2nd", "3rd"), is the same class of structure; Apple's Date structure is the same species, applied to a different domain. Claim 2's wherein therefore adds nothing over a structure-library extractor.
• US 5,774,833 (Motorola) corroborates that recurring lexical/code patterns were being used as structural markers in text analysis: it tokenizes into word/integer/punctuation tokens; treats "two numbers … separated by a colon … <number><colon><number>" as a single "super group" token to enable "more effective higher level parsing later"; and claim 33 identifies part boundaries by "matching the section keywords and the total line length to standard headers."

Claims 2, 11, 17 obvious over '789 (and over '789 in view of '833).

Ground 4 — Claims 3, 12, 18 (start-point + end-point pattern pair): US 5,392,428 (Robins) in view of US 5,864,789 or US 4,773,009

Robins discloses the exact data model claim 3 recites:

• Abstract: "The text is divided into user-defined segments, and the system maintains a series of records, each of which characterizes a segment of the text. The system generates a one-to-one association between each record and the indicia which indicate the length of the record and correspond to the beginning and end points of text segments."
• Specification: "Each record also contains indicia which represent the beginning and the end of each record"; one-to-one associations are created between each record and the corresponding indicia; records link to topics and to reports.

Motivation. Robins' segment model requires the user to manually mark segment beginnings and endings (it is a deposition/transcript digest tool). Combining Robins' record model with a pattern-based locator (Apple '789's extractor; or Kucera '009's "paragraph boundary determining means") supplies automatic placement of those begin/end indicia. The artisan would do this for the ordinary reason that automating manual marking is the recognized purpose of text-analysis tooling (Kucera's stated purpose is precisely to relieve the operator of manual boundary identification). Result: the "content contained between these patterns is treated as one unit of content" — claim 3's final clause, already Robins' record model. Claims 3, 12, 18 obvious.

Ground 5 — Claims 4, 5, 13, 14, 19, 20 (two texts, compare correlated collections, extract the differing units): the weakest ground on this record

Here I will be candid: none of the cited references squarely discloses version-to-version comparison of two documents with extraction of correlated, differing blocks. The best available pieces are:

• US 4,773,009, claim 3 – comparator means for identifying text fragments (comparison machinery inside a text analyzer, but comparing sentence/paragraph boundaries, not two documents); plus display output (claim 6).
• US 5,392,428 – reporting functions, full-text search and reporting (Figs. 18–21), and record-to-record associations; a framework for extracting and displaying selected records, but not a diff.
• US 5,774,833 – compares within one document (title vs. abstract copy of the title; figure references vs. drawing data), and "reports results to a user" — a comparison-plus-report teaching, but intra-document.

How a plaintiff-defendant should read this. The two-text diff limitation is the part of the invention with the best chance of surviving on this record if the challenger is confined to the cited art. Two responses are available:

1. General knowledge / analogous art. By 1997, line- and paragraph-oriented file comparison (diff, and word-processor "compare documents" features) was ubiquitous and is presumptively within a POSITA's knowledge; the patent itself concedes at the outset that "when taking the differing part between two sentences or lines of text," the prior art already output the changed line (FIG. 8A) or the marked full text (FIG. 8B). KSR permits reliance on such common knowledge and on "the inferences and creative steps that a person of ordinary skill in the art would employ." Under that framework, applying a known differencing tool to the blocks produced by Grounds 1–4 is a combination of known elements with a predictable result: changed blocks are reported instead of changed lines.
2. The patent's own admission supplies the case. The specification frames the invention as the same output as FIG. 8A/8B but expressed in terms of a different unit — i.e., the difference is one of granularity, not of kind. Under KSR, "a patent composed of several elements is not proved obvious merely by demonstrating that each of its elements was, independently, known" — but the converse also holds: where the only difference is substituting one known unit of analysis (block) for another known unit (line/sentence), and the result (context plus compactness, per the FIG. 5 discussion) is the predictable consequence of that substitution, the claim is obvious. The patent's asserted advantage — "the computational volume can be greatly reduced and the processing can be made far easier" — is a simplification, not a surprising result, and the specification's own justification ("there is no need to perform structural analysis while considering connections/relationships") concedes that the advantage is inherent to the known simpler technique.

If the challenger wants a primary reference expressly disclosing two-document block diffing, this record does not contain one and additional art should be located. Flagging this honestly: Ground 5 is a rational but not a slam-dunk § 103 case on the cited references alone.

Ground 6 — Claims 6, 7, 15, 21 (display the extracted units): US 4,773,009 + US 5,392,428

Display of the analyzed output is expressly old: Kucera claim 6 ("means for displaying an interpretive message with the readability output") and col. 1's operator-display discussion; Robins generates and displays reports (Figs. 18–21) and provides UIs for accessing "various systems of the disclosed embodiment" (Fig. 11). The '336 specification adds only the trivial option of showing changed parts "in a different color from the rest of the text" — a purely conventional presentation choice with no asserted technical effect. Claims 6, 7, 15, 21 obvious (subject to the Ground 5 caveat as to the comparator step they depend from).

---

5. Consolidated motivation-to-combine analysis (KSR factors)

1. Identified need, known technique. All the cited art addresses the same problem — programmatically locating logical units inside coded text — and the WWW5 and Digestor-type work specifically addresses HTML. Where the "structure elements … are already given by tags" (NTT '520, spec), using tags is not merely obvious; it is the artisan's default. The '336 patent does not identify any technical obstacle to using markup tags as boundaries.
2. Market/practical pressure. 1995–1997 saw explosive growth in HTML content and content re-purposing (device-independent access, restructuring, transformation). That pressure drove exactly the tag-driven segmentation the claims recite.
3. Predictable result / reasonable expectation of success. Every element is a deterministic pattern match: a stored pattern (Apple '789 structure library), applied to coded text (WWW5 HTML), producing begin/end indicia (Robins '428) delimiting a block (Kucera '009). There is no unpredictability to argue about.
4. No teaching away. The '336 specification criticizes the tree/graph art (JP 4‑23765, 6‑35960, 7‑200589, 8‑6945, 4‑306768, 5‑324708) for computational cost. Criticizing a different, more complex approach is not teaching away from tag-based segmentation; if anything, the patent's own criticism of complex analysis supplies the motivation to take the simpler route that the cited art already took.
5. Design incentive / "obvious to try." Where the prior art gives a finite number of identified, predictable solutions (match on recurring lexical marker; match on markup tag; match on whitespace/paragraph code), and the claimed approach is one of them with a predictable advantage, KSR forecloses the nonobviousness argument.

---

6. Secondary considerations

There is no evidence of secondary considerations in this record — and the litigation/PTAB sections of this analysis found no assertion history in the patent's ~20-year term, no IPR/PGR/CBM, no reexamination, and no reissue. Nexus-less evidence of commercial success, if any existed, would be unavailing because the claims are limited to control-code-based segmentation and there is no showing that any product embodying the claims achieved success because of that limitation. The only "advantage" in the specification — reduced computational volume — is an inherent property of the disclosed simplification and, under KSR, cannot alone establish nonobviousness.

---

7. Where the record is weak (honest limits)

1. Date-disqualified NPL. The Digestor (Apr. 1997) and Lin et al. (Aug. 1997) citations appear to post-date both the 1997‑01‑30 § 102(b) critical date and the 1997‑02‑27 invention date. If a challenger's only HTML art is Digestor, the ground fails at the threshold. Locate (a) an earlier preprint/tech-report version and (b) the file wrapper to confirm citation provenance.
2. No verified full text for WWW5. I could not retrieve the "Interactively Restructuring HTML Documents" paper's text in this session. Descriptions of its content are inferences from its title, venue, and the state of the art — readers should pull the paper (WWW5 proceedings, Paris, May 6–10, 1996) before relying on Ground 1.
3. The diff claims (4, 5, 13, 14, 19, 20) lack a precise primary reference in the cited art. A block-diff ground would benefit from additional art (versioned-document comparison; structured-document change reporting).
4. Claim 8 antecedent oddity. Claim 8's whereabout clause recites "the first input text" although nothing in claim 8 introduces a "first" text — a § 112(b) defect, not a § 103 ground, but worth noting as a separate vulnerability. There is also a typographical defect in claim 16 ("a content boundary pattern storage that stores a content boundary pattern::").
5. Whether the control-code wherein clause was added by amendment is unverified. The granted text contains it; the prosecution history would show whether the applicant narrowed to control codes to escape the NPL. If so, and if the NPL that prompted it was date-disqualified, the amendment may be challengeable — but that is a question for the file wrapper, and I am not expressing a legal conclusion.

---

8. Bottom line

The single most important sentence in this analysis is the one from US 5,642,520 that predates the '336 filing by more than two years: "Normally, these structure elements besides the item area are already given by tags and the like, so that the above-recognized structure has also been already given." US 6,263,336's independent claims, as granted, are directed to doing exactly that — with the added requirement (imposed by the claims' own wherein clauses) that the tags be control codes in a text coding language. Combine that teaching with the HTML-restructuring work already in the public domain by 1996 and a stored-pattern extractor like Apple's, and claims 1, 8, and 16 read as an obvious application of known elements with a predictable result. The challenger's work is therefore (i) confirming WWW5's disclosure, (ii) confirming the citation provenance and dates of the three NPL items, and (iii) sourcing one good block-level document-differencing reference to close Ground 5.