US 6246981

I'll search for litigation and docket information on this patent, then summarize from the authoritative full text provided.

Let me run a couple more targeted searches to check for any 2026 litigation or CAFC activity.

Search results: no CAFC 2026 activity found

I searched for US 6,246,981 in PTAB/USPTO filings and Federal Circuit docket material. I found no CAFC 2026 appeal, IPR, or district-court docket in which US 6,246,981 itself is the asserted patent. The only hits that referenced this number were: (a) prior-art citations of the patent (e.g., the WO 2009/059199 search report, which cites "US 6246981 B1 (PAPINENI et al.) 12 June 2001"; Amazon's US 10,891,152 and Unified Patents' US 10,089,072 reference lists), and (b) unrelated hits for a different patent sharing the bare digits JP 6246981 B2 (Qualcomm, "Adapting Blind Reception Duration for Range and Congestion" — Japanese patent, 2017). Note the strict-identifier rule: JP 6246981 B2 is not the same document as US 6,246,981 and should not be conflated with it.

The 2026-era PTAB documents surfaced concern other Nuance-origin patents (e.g., the '209 and '659 patents, Microsoft v. Voice Inventions/Dialect lines), not the '981 patent. Uncertainty flag: I cannot confirm from these results whether US 6,246,981 has ever been asserted or appealed; if it has, I did not locate the docket.

---

Summary of US 6,246,981 (from the authoritative full text)

Discrepancy note: the Google Patents authoritative record lists priority date 1998-11-25; a Unified Patents summary page lists 1998-11-24. Per the instructions I treat the authoritative patent text (1998-11-25) as controlling and flag the one-day difference.

Abstract (verbatim): "A system for conversant interaction includes a recognizer for receiving and processing input information and outputting a recognized representation of the input information. A dialog manager is coupled to the recognizer for receiving the recognized representation of the input information, the dialog manager having task-oriented forms for associating user input information therewith, the dialog manager being capable of selecting an applicable form from the task-oriented forms responsive to the input information by scoring the forms relative to each other. A synthesizer is employed for converting a response generated by the dialog manager to output the response. A program storage device and method are also provided."

Plain-language overview of the independent claims

Claim 1 — System. A conversational system with three cooperating parts: (1) a recognizer that receives input and produces a recognized representation of it; (2) a dialog manager coupled to the recognizer that holds multiple task-oriented forms (form-based slots mapping to user input), and that chooses which form applies to the user's input by scoring the forms against each other rather than by following a fixed state machine; and (3) a synthesizer that converts the dialog manager's response into output. The scoring-based form selection is the core inventive hook.

Claim 11 — Program storage device (software article). A machine-readable storage medium embodying instructions to perform dialog-management steps: (a) assemble (attribute, value) pairs from a semantic representation of input speech; (b) process that representation to determine a set of admissible forms consistent with those pairs; (c) score the admissible forms to find which is most relevant/responsive to the input; and (d) send output messages keyed to that best-matching form and the conversational context.

Claim 24 — Method. The method counterpart to claim 11, reciting the same four steps (assemble attribute/value pairs → determine admissible forms → score to select the most relevant form → send messages consistent with input and context) performed for a speech-recognition dialog system.

Dependent-claim themes (context for the independents)

• Claims 2–10 (from claim 1): control via a backend and a script (2); speech recognizer/text output (3); natural-language understanding module (4); text-to-speech synthesizer (5); per-task forms with data fields (6); scoring based on information needed to complete each form plus user-input context (7); question formulation for prompting/clarification (8); confirmation means (9); inheritance of prior information into a present form (10).
• Claims 12–23 (from claim 11): scoring on filled vs. unfilled slots (12); pre-execution confirmation messages (13); domain-independent input handling (14); threshold comparison for whether speech was properly received (15); rephrase request (16); natural-language I/O (17); attribute/value inheritance across forms (18); natural context shifting to any task at any time (19); dynamic machine-initiative ↔ mixed-initiative switching via the admissible-forms list (20); spoken cancellation of current/prior transactions (21); context-sensitive help (22); list navigation (23).
• Claims 25–36 mirror claims 12–23 in method form.

Specification highlights grounding this: the dialog manager is "application-blind," driven by a declarative script (see Appendix A sample script and the LOGIN/BUY forms), communicates with a backend through labeled messages and return codes, and maintains a list of admissible forms whose expansion/contraction implements the dynamic initiative switching. The patent expressly distinguishes itself from state-table dialogue managers (e.g., Takebayashi U.S. Pat. No. 5,577,165), state/RTN/RL approaches (Pieraccini et al., AMICA; Levin et al.), and the information-based typed-feature approach (Denecke & Waibel), and cites the Ratnaparkhi maximum-entropy parser for the NLU stage.

Confidence: High for all bibliographic and claim content (taken directly from the authoritative full text). Low/uncertain for any litigation or CAFC 2026 proceeding specifically involving US 6,246,981 — none was found, and I will not assert one exists.

I'll search for litigation involving US patent 6246981 across patent litigation databases.

Let me search specialized patent litigation databases and case-law sources for this specific patent.

Litigation Search Results: US Patent 6,246,981

Bottom line: I found no known litigation involving US Patent 6,246,981.

Based on searches across patent-litigation-oriented sources (Unified Patents materials, USPTO/PTAB filings, CAFC documents, Docket Alarm, and general web results), I could not identify any district court case, ITC investigation, CAFC appeal, or PTAB proceeding in which US Patent 6,246,981 was asserted, challenged, or otherwise at issue.

What I did find (and why it is not this patent)

These hits are important to distinguish, because they involve different patents that share similar numbers or subject matter:

• JP 6246981 B2 (Qualcomm Inc.) — Japanese Patent No. 6246981, titled (translated) "Adapting Blind Reception Duration for Range and Congestion," assigned to Qualcomm, registered 2017-11-24. This is a Japanese patent on wireless/WLAN power-saving technology, wholly unrelated to the US patent and appearing frequently in searches because of the identical numeric string "6246981." It is not US 6,246,981.
• US 6,246,981 (Papineni et al.) as prior art / citation only — The patent appears as a cited reference in various prosecution records and international search reports (e.g., the search report for PCT/US2008/082077, where it is cited as "US 6246981 B1 (PAPINENI et al.) 12 June 2001" against claims 1–114; and in USPTO Information Disclosure Statements). Being cited as prior art is not litigation.

Patent identity confirmation (for the record)

• US 6,246,981 B1 — "Natural language task-oriented dialog manager and method"
• Inventors: Kishore A. Papineni, Salim Roukos, Robert T. Ward
• Original assignee: International Business Machines Corp. (filed 1998-11-25; granted 2001-06-12)
• Reassigned to Nuance Communications, Inc. on 2009-03-06
• Legal status: Expired – Lifetime (anticipated expiration 2018-11-25)

Notably, the patent expired in November 2018, which is relevant context: any assertion litigation would necessarily predate expiration, and I found none.

Caveats and confidence level

• I have high confidence that no well-known or widely reported litigation exists for this specific patent, and that the "6246981" hits circulating online overwhelmingly concern the unrelated Japanese Qualcomm patent.
• I have lower confidence that this is an exhaustive negative. Complete coverage would require querying paywalled/comprehensive litigation databases directly — e.g., Unified Patents Portal (portal.unifiedpatents.com/litigation), RPX Insight, Lex Machina, Docket Alarm, PACER, and Westlaw/Lexis. My searches did not surface any consolidated docket specifically listing US 6,246,981 as an asserted patent.
• Per the operating rule, I am not auto-correcting or conflating numbers: I am treating US 6,246,981 and JP 6246981 B2 as distinct instruments, and I am not attributing the Qualcomm Japanese patent's activity to the US patent.

Recommended next step for definitive confirmation

To conclusively rule litigation in or out, run the patent number as an "asserted patent" filter in Lex Machina or RPX Insight, and check PACER for the original assignee (IBM) and current assignee (Nuance Communications) as plaintiffs, since any pre-2018 assertion would most plausibly have been brought by Nuance or a successor/NPE. If those sources also return nothing, the conclusion that no litigation is known can be stated with high confidence.

I'll verify the structured block's "no proceedings" default against public sources before reporting.

PTAB / AIA Trial Landscape — U.S. Patent 6,246,981

"Natural language task-oriented dialog manager and method" — filed/priority 1998-11-25; granted 2001-06-12; original assignee International Business Machines Corp.; assignment to Nuance Communications, Inc. recorded 2009-03-06; anticipated expiration 2018-11-25 (status: Expired - Lifetime). Source: Google Patents US6246981B1.

Proceedings overview

Total AIA trial proceedings on file: 0. The USPTO Open Data Portal structured block for this patent returns no AIA trial proceedings as of the most recent ingest, and independent web searching surfaced no IPR, PGR, or CBM petition naming U.S. 6,246,981 — so the breakdown is 0 active / 0 claims invalidated / 0 claims sustained / 0 settled / 0 institution denied, and there is no claim-level PTAB outcome to build a defense on. The defensible bottom line for a defendant is not "the patent is hardened by surviving IPRs" but rather: every one of the 36 claims is untested at the PTAB, and the patent expired on 2018-11-25, so the only live exposure is pre-expiration past damages — but with no IPR on file, there is also no § 315(e)(2) estoppel and no PTAB record to borrow from.

No proceedings to report

There is no proceeding number, panel, institution decision, FWD, settlement, or appeal to describe. I am deliberately not populating the per-proceeding template, because doing so would require inventing a proceeding number, and the operating rules forbid that.

One confusion to pre-empt: searching for "IPR '981 patent" surfaces IPR2019-01080 (petitioner TripAdvisor, patent owner Minzi, LLC, D. Del. 1:18-cv-01768), in which all challenged claims were held invalid and the patent owner declined to appeal (Rothwell Figg case note). That is a different '981 patent (a location-ranked mobile search patent), not U.S. 6,246,981. Do not cite IPR2019-01080 as an outcome for this patent.

Strategic summary

Claim status. Claims 1–36 of U.S. 6,246,981 are all UNTESTED — none canceled, none sustained, none construed by the Board. For the record, the claim set is: claims 1–10 (system), 11–23 (program storage device), 24–36 (method). Claim 1 recites the recognizer / dialog-manager-with-task-oriented-forms-selected-by-scoring / synthesizer combination, and claims 2–10 add the backend-and-script control, speech recognizer, NLU module, TTS synthesizer, form/field structure, scoring-by-information-needed-and-context, prompt/clarification, confirmation, and inheritance limitations. Since nothing was canceled, a demand letter citing claim 1 cannot be answered with "that claim is dead" — the answer has to come from district-court invalidity, § 101, or the expiration/past-damages math.

Estoppel landscape. Because no IPR/PGR was ever instituted against this patent, § 315(e)(2) estoppel is a non-issue — no petitioner or privy is barred from any ground, and a defendant today retains the full universe of § 102/§ 103 art, plus § 112 and § 101 challenges. That cuts both ways: there is no prior petitioner's work product to free-ride on, but also no risk that a prior petitioner's unsuccessful grounds were already run. Two timing points matter far more than estoppel here:

• The patent expired 2018-11-25, so the recoverable window under 35 U.S.C. § 286 is roughly 2012→2018-11-25. Any accused conduct post-dating expiration cannot be direct infringement.
• § 315(b)'s one-year bar from service of an infringement complaint is irrelevant absent an active suit; § 315(a)(1) is only triggered if you file a DJ action first.

Pattern signals. The patent is from IBM's 1998 dialog-manager work (inventors Papineni, Roukos, Ward) and was swept into Nuance's portfolio. Nuance's documented enforcement campaigns — e.g., Nuance Communications, Inc. v. Omilia Natural Language Solutions, Ltd., D. Mass. No. 1:19-cv-11438 (CourtListener docket 15874727) — asserted the '905, '993, '839, '534, '804, '532, '688 and '925 patents, not the '981; the Nuance/MModal N.D. Ga. litigation asserted the '786, '040, '524 and '829 patents, again not the '981. I found no defensive aggregator (e.g., Unified Patents) challenge and no Federal Circuit appeal involving this patent. Separately, Nuance has been divesting patents out of this portfolio to monetization entities (in the 2025 IPR filings on U.S. 7,398,659, the patent had passed from Nuance to Voice Inventions, LLC). Google Patents still lists the current assignee as Nuance Communications Inc., but re-verify the chain of title before assuming who the real party in interest is.

A note on why the IPR vacuum exists. U.S. 6,246,981 is heavily cited as prior art against others (it appears in the ISA search report for PCT/US2008/082077, cited against claims 1–114, and in numerous prosecution IDS listings such as U.S. 10,890,672 and U.S. 10,906,664). It was never itself the target. Given that it expired in 2018 and that IPR practice (post-AIA, available from 2012-09-16) overlapped with its last six years of life, the absence of petitions most plausibly reflects low assertion value rather than demonstrated strength — but I cannot prove that; treat it as inference, not record.

Recommended next steps

1. Confirm the negative with primary sources before relying on it. Check PTAB E2E / PTAB Center (https://ptacts.uspto.gov/) and the USPTO patent's "Patent Trial and Appeal Board" tab for any recently filed petition the ODP ingest may not yet reflect. Re-run the search for IPR2025- / IPR2026- numbers tied to this patent before finalizing any freedom-to-operate or invalidity opinion.
2. Do not cite IPR2019-01080 as this patent's outcome. It belongs to a different '981 patent (Minzi).
3. If you are a defendant: the absence of any FWD means there is no disposition to quote and no estoppel to leverage. Build the defense on (a) expiration — no post-2018-11-25 damages; (b) § 286 — a six-year lookback floor; (c) § 101 — the claims are old-economy form-filling/speech plumbing and would face a serious Alice challenge; and (d) § 102/§ 103 — the patent's own Background section cites GUS: A Frame-driven Dialog Manager (Bobrow et al., 1977), A Form-based Dialog Manager for Spoken Language Applications (Goddeau et al., ICSLP 1996), Denecke & Waibel (Eurospeech-97), Pieraccini et al. (Eurospeech-97), Levin et al. (ICASSP 1998) and Takebayashi U.S. 5,577,165 — i.e., the applicant admitted much of the form-based, mixed-initiative prior art on the face of the patent. That admitted art is free to use in district court with no § 315(e)(2) risk.
4. If you want a PTAB record and one does not exist: a petition is theoretically available even against an expired patent (the Board has instituted on expired claims in other matters), but the value is low because only past damages remain, and claims are construed under Phillips in that posture. A third-party ex parte reexamination request remains an option that is unaffected by expiration and, critically, carries no § 315(b) clock and no estoppel — worth pricing if the client needs an Office-side validity determination rather than a litigation defense.
5. Re-check ownership and RPI status. Because Nuance has been moving patents from this vintage into NPE hands, verify the current recorded assignee and any exclusive-licensee before assuming the asserting party is Nuance/Microsoft.

Confidence note: I have high confidence in the absence of AIA proceedings based on the ODP structured block plus two negative web searches. If a petition was filed in the last few weeks, the ODP ingest would not yet show it — step 1 is the safeguard.

I'll research the assignment record for this patent before reconstructing the chain.

Interesting — I found a signal about a 2021 Nuance-to-"Voice Inventions" transfer. Let me dig into that and the original IBM→Nuance record.

Let me check whether '981 is in the Voice Inventions / Dialect LLC portfolios and look at the RPX coverage.

Ownership & Assignment Reconstruction — US 6,246,981 B1

Data-provenance caveat up front (important). I could not open the USPTO Assignment Center / Patent Assignment Search UI directly with the tools available to me; the two recorded links below are taken from the Google Patents legal-events record for US 6,246,981 (which mirrors the USPTO assignment index but, as scraped, does not expose reel/frame numbers), plus PTAB exhibit documents that quote the relevant assignment agreements. I did not retrieve a reel/frame for either link, and I did not retrieve a correspondent of record for either link. I am stating that plainly rather than inventing reel/frame values. To complete the two missing fields, search patent number 6246981 at:

• https://assignmentcenter.uspto.gov/ (Assignment Center)
• https://assignment.uspto.gov/patent/index.html (legacy Patent Assignment Search)

Everything below that is not sourced to those two portals is flagged as unverified.

---

Inventors

Unusual-pattern check: Not present. The Google Patents reassignment record shows all three inventors executing an assignment of interest to IBM on 1999-01-11, six weeks after filing — a standard employment/confirmation assignment, not a departure signal. The invention is IBM Yorktown dialog-management research (the specification's Appendix B transcript is an IBM Mutual Fund Voice System demo). I found no evidence that any inventor left IBM within 12 months of filing, which is the classic precursor to a portfolio fire-sale. (I did not run individual personnel histories for Ward; treat that one cell as unverified.)

---

Original assignee

International Business Machines Corporation (Armonk, NY) — named on the issued patent (granted 2001-06-12).

• Primary line of business: diversified enterprise computing, software and research; the relevant unit here is IBM Research's natural-language / speech dialog program.
• Product embodying the claims: Partially. The specification describes the dialog manager as a working system (Appendix B is a real telephone transcript of an IBM Mutual Fund Voice System), and IBM commercialized speech/telephony dialog technology in its WebSphere voice and IVR lines. The '981 dialog manager is a software component rather than a branded product, so "shipped a product embodying the claims" is probable but not documented to product-part-number level.
• Current assignee of record per Google Patents: Nuance Communications, Inc. (Burlington, MA).
• Current status: IBM remains an operating company. Nuance was acquired by Microsoft Corporation (announced April 2021, closed March 2022) and operates as a Microsoft subsidiary; the Nuance brand persists in Microsoft's clinical-speech products (e.g., Dragon Copilot).
• Patent status: Expired — anticipated expiration 2018-11-25 (20 years from the 1998-11-25 filing). This matters a great deal below: the patent was already expired before the 2021 Nuance speech-portfolio transfer that dominates this technology space.

---

Assignment timeline

Two recorded links are verifiable. Neither reel/frame nor correspondent was obtainable from the sources I reached.

• 1999-01-11 (executed and recorded; execution date not separately stated) — Reel not retrieved

  • Conveyance: Assignment of assignors' interest (employment/confirmation assignment)
  • Assignor: Kishore A. Papineni; Salim Roukos; Robert T. Ward
  • Assignee: International Business Machines Corporation
  • Correspondent: not retrieved from source
  • Context: Internal — inventors' rights vest in the employer 47 days after filing; no third party involved.

• 2009-03-06 (recorded; execution date not stated on the indexed record) — Reel not retrieved

  • Conveyance: Assignment
  • Assignor: International Business Machines Corporation
  • Assignee: Nuance Communications, Inc.
  • Correspondent: not retrieved from source. (I could not confirm whether this correspondent recurs elsewhere in Nuance's IBM-origin acquisition filings — that comparison requires Assignment Center access I did not have.)
  • Context: Corporate portfolio transfer — IBM divested speech/voice patents to Nuance during the 2008–2010 period in which Nuance was consolidating speech-IP assets. Exact deal structure (sale vs. license-back) not confirmed by the sources I located.

No further recorded assignment of US 6,246,981 was found after 2009-03-06, and no litigation asserting this patent was found.

The near-miss I specifically checked (and why it does not attach to '981)

There is a large, active monetization chain in exactly this technology space, and it would be easy to wrongly graft '981 onto it:

• Nuance → "Voice Inventions, LLC" — Assignment Agreement effective January 1, 2021 (document dated 2020-12-24), signed for Nuance by David Greenbaum, VP, Litigation & IP, transferring "the Assigned Patent Portfolio," including the right to sue for past damages (PTAB exhibit EX2003; https://ptacts.uspto.gov/ptacts/public-informations/petitions/[1558142](/patent/1558142)/download-documents).
• Voice Inventions, LLC is a Delaware LLC; RPX's Patent Watch (2021-07-11, "Duo of Speech Recognition Patent Assignments Bears Watching") characterized the transfer as "more than two dozen assets generally related to speech recognition … from Nuance Communications to a Delaware LLC," noting an earlier mid-October 2020 assignment of two speech patents to the same LLC from a different player.
• Those assets were then asserted by Dialect, LLC, against Google, Amazon, Microsoft, Meta, Samsung, Salesforce and Bank of America (e.g., Dialect, LLC v. Microsoft Corp., E.D. Tex. 2:24-cv-01067; IPR2025-01352; Dialect, LLC v. Salesforce, Inc., W.D. Tex. 7:25-cv-00061). Microsoft's own petitions confirm: "Nuance acquired the '659 patent from VoiceBox Technologies Corporation in January 2015 … and sold it to Voice Inventions, LLC, in January 2021."

Why this does not appear to include '981: (a) the exhibited Voice Inventions portfolio table I could read lists VoiceBox-originated Kennewick patents — 7,398,209, 7,502,738, 7,809,570, 8,015,006, 8,112,275, 8,140,327, 8,155,962, 8,731,929, 9,734,825, 7,693,720, 9,031,845, 7,640,160, 7,917,367, 8,326,634 … — none of them IBM's '981; and (b) US 6,246,981 had already expired on 2018-11-25, two years before that transaction, leaving only residual past-damages value. I could not read the full 24+ item portfolio table, so this is a strong but not conclusive negative. Treat "is '981 in the Voice Inventions / Dialect portfolio?" as the single open question in this file.

---

Timeline diagram

timeline
    title Ownership of US 6246981
    1998 : Filed by IBM
    1999 : Inventors assign interest to IBM
    2001 : Patent issued to IBM
    2009 : Assigned to Nuance Communications
    2018 : Patent term expires
    2022 : Nuance acquired by Microsoft

---

NPE / troll-pattern signals

1. Shell-entity transfer — not present (for this patent); unclear (for the family). No assignment to any LLC appears in the '981 record; the last verified assignee is Nuance Communications, Inc. The 2009-03-06 IBM→Nuance link is an operating-company-to-operating-company transfer. Unclear flag: the Nuance → Voice Inventions, LLC (Delaware), effective 2021-01-01 transfer is exactly the shell-entity pattern, but I found no evidence '981 was among the assigned assets, and '981's 2018-11-25 expiration predates it.

2. Known asserter in the chain — not present (for this patent). No assignee in the '981 chain matches the named NPE lists (Acacia, Marathon, IV, IPNav, Wi-LAN, Mosaid/Conversant, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, DGC, Spangenberg entities). Note the distinction: Dialect, LLC is now a high-frequency speech-NLU plaintiff, and Nuance itself has been publicly criticized as an aggressive enforcer (NYT, 2012-10-08, re: Vlingo) — but Nuance is an operating company and neither has a documented '981 link.

3. Repeat correspondent across the chain — unclear / not assessable. I could not retrieve the correspondent of record for either the 1999 or 2009 recording, so recurrence cannot be tested. Note for follow-up: the 2021 Nuance portfolio transfer was executed for Nuance by David Greenbaum, VP, Litigation & IP — but a signing officer inside the assignor is not a recording correspondent, and he is not tied to '981. This field is a genuine data gap, not a negative finding.

4. Cascading transfers — not present. The '981 chain has two links separated by ten years; there is no <24-month chained-LLC sequence.

5. Pre-litigation transfer — not present. No infringement suit naming US 6,246,981 was located, so there is no suit date to measure a transfer against.

6. Bankruptcy fire-sale — not present (for this patent). The only bankruptcy-adjacent fact in this space is VoiceBox Technologies — whose January 2015 assignment of its portfolio to Nuance (and later Nuance's 2021 sale onward) is documented — but '981 is an IBM patent, never a VoiceBox asset, and neither IBM nor Nuance filed bankruptcy.

7. Privateering — not present. No operating-company-to-NPE transfer of '981, and no SEC 10-K/8-K disclosure linking '981 to a proxy assertion program. (I did not run a full EDGAR search; the absence is based on the assignment record and litigation dockets.)

8. Defensive aggregator — not present. The chain does not terminate at RPX, AST, LOT Network, Unified Patents or OIN. Inverse signal absent: the patent was never neutralized by aggregation; it simply expired.

---

Verdict

Insufficient data — but with an operating-company termination, not an NPE one.

The full verifiable record for US 6,246,981 is two links: inventors → IBM (recorded 1999-01-11) and IBM → Nuance Communications, Inc. (recorded 2009-03-06), after which the patent expired on 2018-11-25 while still held by Nuance/Microsoft. There is no recorded shell-entity, NPE, or litigation-facilitating transfer of this patent, and no suit asserting it was found — so the strong NPE signals that cluster around the adjacent VoiceBox portfolio (the 2021-01-01 Nuance → Voice Inventions, LLC sale, and Dialect, LLC's campaigns) are not tied to '981 by any evidence I could locate. The verdict is driven by missing fields rather than a positive finding: I could not obtain reel/frame or correspondent of record for either link, and I could not read the complete Voice Inventions portfolio schedule to close out whether the expired '981 was swept in as a past-damages tag-along.

Two things to verify before finalizing, both resolvable at https://assignmentcenter.uspto.gov/ and https://assignment.uspto.gov/patent/index.html:

1. Reel/frame + correspondent for the 1999-01-11 and 2009-03-06 recordings (search patent 6246981, then check whether the 2009 correspondents recur across Nuance's other IBM-origin recordings).
2. Whether US 6,246,981 appears in the "Assigned Patent Portfolio" schedule of the 2021-01-01 Nuance → Voice Inventions, LLC agreement (the full schedule is in PTAB petition exhibit EX2003, https://ptacts.uspto.gov/ptacts/public-informations/petitions/1558142/download-documents). If it does, the verdict flips to NPE — high confidence, because the chain would then be IBM → Nuance → Voice Inventions LLC → Dialect LLC, terminating in a litigation vehicle that has sued Google, Amazon, Microsoft, Meta, Samsung, Salesforce and Bank of America.

I'll research this patent and its cited references. Let me run several searches in parallel.

Let me gather details on the remaining cited references.

⚠️ Two important corrections before the analysis

1. "Cited By" ≠ prior art. The Google Patents page for US 6,246,981 lists "Cited By (370)" — e.g., US 6,324,513 (Mitsubishi), US 6,505,162, US 8,762,152, and hundreds of Apple/Voicebox/Nuance patents. Those are forward citations (later patents that cite '981). They cannot be prior art under 35 U.S.C. § 102 against '981, because they all post-date its 1998-11-25 priority date. I have excluded them. If your request intended these, the correct answer is: none of them is § 102 art.

2. This patent is pre-AIA. US 6,246,981 was filed 1998-11-25, so the applicable statute is pre-AIA § 102(a)/(b)/(e) — not the AIA first-inventor-to-file § 102. All § 102 characterizations below use the pre-AIA subsections.

I also note a source discrepancy: the Google Patents full text says priority/filing 1998-11-25, while a Unified Patents family record shows 1998-11-24. I treat 1998-11-25 as authoritative per the patent text you supplied.

---

Bibliographic verification

This matches the USPTO.report record (https://uspto.report/patent/grant/[6246981](/patent/6246981)) and the Google Patents record. A separate live USPTO PatentCenter query was not executed in this session; the front-page "References Cited" list could not be fully retrieved, so the reference set below is reconstructed from the patent's own specification (which cites its prior art by name) plus the U.S. patent reference named on the face of the document.

---

The prior art actually cited by US 6,246,981

A. U.S. Patent references

A1. U.S. Pat. No. 5,577,165 — Takebayashi et al.

• Full citation: U.S. Pat. No. 5,577,165, "Speech Dialogue System for Facilitating Improved Human-Computer Interaction," Yoichi Takebayashi, Hiroyuki Tsuboi, Yoichi Sadamoto, Yasuki Yamashita, Yoshifumi Nagata, Shigenobu Seto, Hideaki Shinchi, Hideki Hashimoto; assignee Kabushiki Kaisha Toshiba / Toshiba Software Engineering Corp.
• Dates: Appl. No. 08/312,541 filed 1994-09-26; issued 1996-11-19; continuation of Ser. No. 07/978,521 filed 1992-11-18 (issued as U.S. Pat. No. 5,357,596 on 1994-10-18); JP foreign priority 3-329475, 1991-11-18.
• Description: A speech dialogue system in which a speech understanding unit extracts a semantic utterance representation, a dialogue management unit determines a semantic response representation using dialogue history/state, and a response generation unit produces a speech response and a visual response (text, a human-character image, and a "content visualizing image") via speech synthesis and display. Dialogue progress is managed by transitions between "user states" and "system states" (a state-transition diagram, FIGS. 9–16), and keyword detection is limited according to dialogue state. It also scores/ranks multiple semantic-utterance-representation candidates (scores D1, D2) against dialogue history/state.
• § 102 relevance (pre-AIA § 102(b)): Granted more than one year before the 1998-11-25 filing date, so it is a statutory bar reference. It is the closest U.S. patent reference and is expressly discussed in the '981 background:
  • Claim 1 — potentially anticipates the generic architecture: recognizer (speech understanding) + dialog manager + synthesizer (speech response generation). The specification itself concedes Takebayashi is "an example of a state-based dialog system" (col. 1). The distinguishing limitation of claim 1 is that the dialog manager selects "an applicable form from the task-oriented forms… by scoring the forms relative to each other." Takebayashi's scoring is of candidate semantic representations of a single utterance, not of competing task/forms — so it is strong art for the system skeleton but weak on the form-scoring limitation.
  • Claims 3 and 5 — speech recognizer / text-to-speech synthesizer elements are disclosed.
  • Claim 4 — semantic (frame-based) understanding of the utterance; relevant, though not identical to the "natural language understanding module" of '981 (which the patent implements with the Ratnaparkhi parser, below).

A2. U.S. Pat. No. 5,357,596 (Takebayashi et al.) — parent of A1, issued 1994-10-18. Same § 102(b) posture; same subject matter. Cited indirectly through A1.

No other U.S. patents appear to be cited as prior art on the face of '981; the remainder of its cited art is non-patent literature (below).

---

B. Non-patent literature cited

B1. Bobrow et al., "GUS, A Frame-Driven Dialog System" (1977)

• Full citation: D. G. Bobrow, R. M. Kaplan, M. Kay, D. A. Norman, H. Thompson, T. Winograd, "GUS, A Frame-Driven Dialog System," Artificial Intelligence, Vol. 8, No. 2 (1977), pp. 155–173. (The patent cites it as "vol 8 (1977), pp. 155-173.")
• Date: April 1977 — well over 20 years before filing.
• Description: The original frame-based ("electronic form") dialog manager. The system fills slots in a frame from user utterances in any order and prompts for unfilled slots. The '981 specification expressly relies on GUS: "A first example of such a system is 'GUS: A Frame-driven Dialog Manager' by Bobrow et al…. This work describes dialog management of just one task; booking an airline ticket. This means there is only one active form."
• § 102 relevance (pre-AIA § 102(b), printed publication): Anticipates the frame/slot/mixed-initiative prompting concepts found in:
  • Claim 6 (forms with fields that correspond to information applicable to the application) — arguably anticipated as a genus.
  • Claim 8 (formulating questions prompting the user for needed information) — anticipated by GUS's unfilled-slot prompting.
  • Claim 10 / Claim 18 / Claim 31 (inheriting information supplied in a different context) — GUS's frame-filling and discourse-context resolution is relevant art, though "different context/other forms" inheritance is not clearly shown.
  • Because GUS has only one active form, it does not anticipate claims 1, 7, 19, 20, 32, 33 (multi-form scoring, context shifting, dynamic mode switching).

B2. Goddeau, Meng, Polifroni, Seneff, Busayapongchai, "A Form-based Dialog Manager for Spoken Language Applications" (1996)

• Full citation: D. Goddeau, H. Meng, J. Polifroni, S. Seneff, S. Busayapongchai, in Proc. Int'l Conf. on Spoken Language Processing (ICSLP), Philadelphia, 1996, pp. 701–704.
• Date: October 1996 — more than one year before 1998-11-25.
• Description: A spoken form-based dialog manager in a used-car price-quote domain, with a single task and a single active form. The '981 specification cites it and notes: "Again, there is only one task and one active form corresponding to that task."
• § 102 relevance (pre-AIA § 102(b), printed publication): Anticipates the generic form-based dialog manager with a speech recognizer combination relevant to claim 1 (structural elements) and claim 6; but, like GUS, it lacks the multi-form competitive scoring of claims 1/7 and the context-shifting features of claims 19/32.

B3. Pieraccini, Levin (patent: "E. Levine"), Eckert, "AMICA: the AT&T Mixed Initiative Conversational Architecture" (1997)

• Full citation: R. Pieraccini, E. Levin, W. Eckert, in Proc. Eurospeech-97, Rhodes, Greece, 1997, Vol. 4, pp. 1875–1878. (Independent bibliographic confirmation: Semantic Scholar & researchr both record Eurospeech 1997, Rhodes, pp. 1875–1878.)
• Date: September 1997 — ~14 months before filing.
• Description: A state-based architecture built on a dialog state space, an action set, and a dialog strategy, implemented as a recursive transition network whose arcs are state conditions and whose nodes are actions.
• § 102 relevance (pre-AIA § 102(a), printed publication before the invention): Because it published less than one year before the '981 filing date, it is § 102(a) art, not § 102(b). It is the primary contrast reference for claim 1: it shows a task-oriented dialog manager with a strategy, but its behavior is state-network-specified rather than "selecting an applicable form… by scoring the forms relative to each other." It therefore does not anticipate claim 1's distinguishing limitation; it is § 103 material.

B4. Levin, Pieraccini, Eckert, "Using Markov Decision Process for Learning Dialogue Strategies" (1998)

• Full citation: E. Levin, R. Pieraccini, W. Eckert, in Proc. IEEE Int'l Conf. on Acoustics, Speech, and Signal Processing (ICASSP), Seattle, May 1998, Vol. 1, pp. 201–204.
• Date: May 1998 — ~6 months before filing.
• Description: Formulates dialog-strategy design as an optimization problem solved by reinforcement learning over an MDP.
• § 102 relevance (pre-AIA § 102(a)): This is the latest-dated reference and is very close to the filing date. Its subject matter (learning/optimizing a dialog strategy) is not the form-scoring selection mechanism of claims 1/7, so it is not an anticipation reference; it is best characterized as § 102(a)/§ 103 background art for the "scoring/dialog strategy" concept.

B5. Denecke & Waibel, "Dialogue Strategies Guiding Users to their Communicative Goals"

• Full citation: M. Denecke and A. Waibel, in Proc. Eurospeech-97, Rhodes, Greece, pp. 1339–1342. Note: the '981 specification recites the year as "1998" for this Eurospeech-97 paper; Eurospeech-97 was in fact held September 1997. I am reporting the identifier/dates literally as they appear in the patent while flagging the inconsistency rather than silently correcting it.
• Date: September 1997 (per conference) / 1998 (as recited in '981).
• Description: The "information-based" approach: each user communicative goal is a typed feature structure imposing lower bounds on required data fields; the system generates clarification dialogues by asking questions to fill deficient feature structures. The '981 specification distinguishes itself from this approach.
• § 102 relevance (pre-AIA § 102(a)): Anticipates the clarification-question-generation concept relevant to claim 8 ("clarifying information supplied by the user") and generally to claims 11/24. It does not disclose form scoring or multi-form task selection (claim 1/7), and '981's specification expressly argues the type-hierarchy/maximum-entropy feature-path selection is "entirely dissimilar."

B6. Ratnaparkhi, "A Linear Observed Time Statistical Parser Based on Maximum Entropy Models" (1997)

• Full citation: A. Ratnaparkhi, in Proc. Second Conf. on Empirical Methods in Natural Language Processing (EMNLP), August 1997, Providence, R.I. Incorporated by reference into the '981 patent.
• Date: August 1997.
• Description: A maximum-entropy statistical parser producing the bracketed semantic representation from which '981 assembles (attribute, value) pairs.
• § 102 relevance (pre-AIA § 102(a)): Relevant only to the natural-language understanding module of claim 4, which '981 expressly implements by reference to this paper. It does not reach the dialog-manager claims (1, 7, 19, 20) — it is an enabling-implementation citation, not an anticipation of the dialog-manager subject matter.

---

Bottom line

Honest assessment: None of the cited references is a clean § 102 anticipation of independent claim 1 (system), claim 11 (program storage device), or claim 24 (method). The single limitation that distinguishes '981 from all of the above is the dialog manager's "selecting an applicable form from the task-oriented forms… by scoring the forms relative to each other," with the dependent features of claims 7 (scoring by needed information and context), 19/32 (natural context shifting), 20/33 (dynamic machine-/mixed-initiative switching), and 21/34 (cancellation). GUS, Goddeau, and Takebayashi supply the frame/slot/recognizer/synthesizer elements; AMICA and Levin supply state-based and strategy-learning alternatives; Denecke & Waibel supplies clarification dialogue; Ratnaparkhi supplies the parser. The realistic challenge posture is a § 103 obviousness combination (e.g., GUS or Goddeau form-based manager + AMICA state/strategy framework), not § 102 anticipation.

Caveats:

• I could not retrieve the complete formal "References Cited" / "U.S. Patent Documents" and "Other Publications" list from the '981 front page in this session; the set above is reconstructed from the patent's specification text (authoritative per your instructions) plus the named U.S. patent. There may be additional examiner-cited references not visible here.
• The Google Patents "Cited By (370)" list should not be treated as prior art for § 102 purposes.
• Dates shown as recited in the patent (e.g., Denecke & Waibel "1998") are reported literally, with discrepancies flagged rather than corrected, per your instruction to interpret identifiers literally.

I'll ground the prior-art references cited in this patent before building the § 103 analysis.

Let me verify the remaining references cited in the patent's background and the pre-critical-date patent in the "Cited By" list.

Obviousness Analysis — US 6,246,981 B1 under 35 U.S.C. § 103

Patent: US 6,246,981 B1, "Natural language task-oriented dialog manager and method"
Inventors: Papineni, Roukos, Ward | Original assignee: IBM | Filed/priority: 1998-11-25 | Granted: 2001-06-12
Governing statute: Pre-AIA § 103 (application filed 1998-11-25; AIA § 102/103 does not apply)

Note on scope. The exercise asks me to use "the Prior Art section of this page." Interpreting the page literally, it does not contain a labeled prior-art reference list. What it contains is (a) "Prior art keywords" — a set of auto-generated keyword-frequency tags ("recited, user, forms, dialog manager, input"), which are not references; (b) the Background of the Invention, which expressly enumerates prior-art systems and publications; and (c) a "Cited By (370)" list, which is citing art, not prior art. I therefore treat the Background's enumerated references as the operative prior-art set, and I flag the "Cited By" list separately below because nearly all of it post-dates the critical date.

---

1. Critical Date and § 102 Status of Each Reference

Critical date: 1998-11-25. References cited in the patent's own Background, with their § 102 posture:

Two literal-interpretation flags:

1. The patent states Denecke–Waibel was "published in Proceedings of Eurospeech-97… 1998." The Eurospeech-97 proceedings are dated 1997 (conference held 22–25 Sept. 1997, Rhodes). I am not auto-correcting the patent's text — I record that the patent's stated year conflicts with the bibliographic record, and that it does not change § 102 status either way.
2. The "Cited By (370)" list is not prior art. Scanning the earliest entries: US 6,324,513 (1999-06-18), US 6,377,913 (1999-08-13), US 6,415,440 (1999-06-15), US 6,505,162 (1999-06-11), US 6,510,411 (1999-10-29), US 6,587,547 (1999-09-13), US 6,587,818 (1999-10-28), US 6,550,345 (1999-08-26), US 6,598,018 (1999-12-15), US 6,601,029 (1999-12-11), US 6,647,363 (priority 1998-10-09). Only US 6,647,363 (Scansoft) predates 1998-11-25, and I was unable to verify its disclosure (search budget exhausted). It is a lead, not a grounded reference. Everything else in that list is evidence of the patent's technological impact, not § 103 material.
3. JP 6246981 B2 (Qualcomm) is not this patent and is not prior art — it is a 2017 Japanese grant on WLAN power saving. Consistent with the litigation section above, I do not conflate the two.

---

2. Person of Ordinary Skill in the Art (POSITA)

At the 1998 critical date, a POSITA would have: (i) a graduate-level or equivalent working knowledge of spoken-language dialogue systems; (ii) familiarity with frame/slot ("E-form") dialogue management, state-based dialogue management, and ASR/TTS pipelines; (iii) knowledge of the then-contemporary literature — GUS, the MIT Galaxy/WHEELS form-based work, AT&T's AMICA, and DARPA/ATIS-era evaluation work. This is a high-level, cross-disciplinary art in which the primary references are the same people, journals, and conferences (ICSLP, Eurospeech, ICASSP, Artificial Intelligence).

---

3. What the Claims Require

Representative independent claims:

• Claim 1: recognizer → recognized representation; dialog manager having task-oriented forms, capable of selecting an applicable form responsive to input "by scoring the forms relative to each other"; synthesizer for the response.
• Claim 11 / 24 (mirror device/method claims): assemble (attribute, value) pairs from a semantic representation; determine a plurality of admissible forms; score the admissible forms to find the "most relevant and responsive" form; send messages per the input and its context.

Dependents of interest: 2 (backend + script), 3–5 (ASR / NLU / TTS), 6 (one form per task, fields), 7 (scoring per information needed + context), 8–10 (means for questioning / confirming / inheriting), 15–16 (score vs. threshold → rephrase), 20 (dynamic machine-/mixed-initiative switching), 21 (cancel anytime), 22 (context-sensitive help), 23 (list navigation).

The two limitations that carry the case are (1) simultaneous multiple forms with inter-form scoring, and (2) mode switching by shrinking/expanding the admissible-form list. Everything else is essentially accounted for by the art.

---

4. Claim-by-Claim Mapping to the References

4.1 Independent architecture claims (1, 3, 5, and the 11/24 preambles)

4.2 Dependent claims that are near-anticipatorily met

---

5. Grounds of Rejection

Ground 1 — § 103: Goddeau (C) in view of Bobrow (B) and Takebayashi (A)

Covers: claims 1, 3–9, 11–14, 17, 24–27, 30.

Rationale. Goddeau discloses the complete form-based dialogue core: an E-form of slots, per-slot prompts with a priority ordering the information acquisition, mandatory vs. optional slots, and — critically — a user who "is not restricted to follow the system's lead, and is free to ignore the prompts and take the initiative." Bobrow supplies the frame-filling-and-confirmation behavior. Takebayashi supplies the ASR → semantic-representation → dialogue-management → response-generation architecture, plus explicit handling of recognition failure and repetition. All three are in the identical field of spoken-language dialogue.

Motivation. The Background itself identifies the driving problem: prior frame-based managers "describe dialog management of just one task" / "there is only one task and one active form corresponding to that task" (the patent's own characterizations of GUS and Goddeau). Once an application exposes several tasks to one caller, the artisan must decide which form to fill — the disputed scoring step is the natural resolution.

Ground 2 — § 103: Ground 1 + Denecke & Waibel (E)

Covers: claims 7, 8, 10, 18, 31, and strengthens Grounds 1/3.

Rationale. Denecke–Waibel expressly propose to model the task rather than the dialogue, represent each communicative goal by a structure whose feature values impose informational lower bounds on required data fields, and generate clarification questions to fill deficient structures. That is the functional equivalent of the claimed step of scoring "according to information needed to complete each form and the context of information," and the "context" prong is supplied by their cross-turn discourse representation — which also supports the inheritance limitations (claims 10, 18, 31) far better than any single-form system can.

Motivation. Denecke–Waibel and the patent share the same problem statement (disambiguation and deficiency-filling). A POSITA building a multi-task system would look to the reference that explicitly addresses selecting/clarifying among multiple goals.

Ground 3 — § 103: Pieraccini (D) in view of Levin (F) and Goddeau (C)

Covers: the "scoring" core of claims 1/7/11/24 and, most importantly, claims 20 and 33.

Rationale. Pieraccini discloses a mixed-initiative conversational architecture modularized into a dialog state space, an action set, and a dialog strategy that "specifies the action to be performed next for each state." Levin et al. then supply the explicit quantitative evaluation of candidate dialogue actions via a Markov decision process — i.e., computing a value/score over competing actions and picking the best, together with objective evaluation of strategies. Combining that decision-theoretic action-selection framework with Goddeau's multi-slot forms yields "score the admissible forms [candidate actions] and take the most relevant."

Motivation (KSR). Both references are same-field, same-era, same research lineage (AT&T/Lucent; MIT). Levin et al. expressly build on the AMICA state/action/strategy model. A POSITA optimizing dialogue behavior would apply the MDP scoring machinery to the already-known form/state representation as a predictable use of a known technique for its known purpose.

Ground 4 — § 103: Takebayashi (A) + Goddeau (C) + Denecke–Waibel (E) for mode switching (claims 20, 33)

Denecke–Waibel's dialogue-goal state machine — goals move through Neutral → Selected → Deselected → Determined → Finalized, and the strategy constrains behavior when exactly one goal is Selected — is a disclosed mechanism for narrowing the field of candidate tasks and then tightening/serializing the dialogue. That is the substance of "dynamically switching between machine-initiative and mixed-initiative modes by managing a list of admissible forms." Combining it with Takebayashi's directed (machine-initiative) dialogue behavior and Pieraccini's mixed-initiative architecture is a routine engineering combination of the two known dialogue regimes.

Ground 5 — § 103: domain-independent features (claims 21–23, 34–36)

The specification's own admission (Background §, quoted above) that help, cancellation, list navigation and repetition-request handling are "common for all applications and domains" is an admission against interest that these are known, non-inventive add-ons. Pair with Goddeau's list-returning used-car search (list presentation) and Takebayashi's failure/repetition handling.

---

6. Where the Obviousness Case Is Weak (and the patent likely survives)

I want to be candid rather than manufacture a clean rejection. Four limitations resist the art:

1. Inter-form scoring as such. The art is bifurcated: GUS and Goddeau have a single active form; Pieraccini and Takebayashi have a global state/strategy. Neither computes a comparative score across a set of simultaneously live task forms from the same semantic input. Denecke–Waibel's selection is by typed-feature-structure subsumption/compatibility, not by a numeric scoring function, and the patent expressly disclaims that route ("The scoring of forms is entirely dissimilar to choosing feature paths based on maximum entropy"). Amending "scoring" to cover subsumption-based selection would invite the applicant to argue that the prior art teaches away from scoring per se. Whether the score-based selection is a patentable difference or a mere design choice / predictable implementation of goal selection is the pivot of the whole case — and it is genuinely arguable.

2. Threshold-plus-rephrase (claims 15/16, 28/29). I found no grounded reference on this page disclosing comparing the best form's score against a minimum acceptable threshold and emitting a "please rephrase" message. ASR confidence thresholds and "please repeat" prompts were well known, which supports obviousness-by-analogy, but I could not verify a specific reference disclosing this for form selection. Treat as moderate-to-low confidence for invalidity.

3. The "means for inheriting information previously supplied in a different context" (claims 10, 18, 31). Cross-form donor inheritance is more than ordinary within-turn slot persistence. Denecke–Waibel's cross-goal discourse representation is the closest disclosure, but mapping it onto explicit "donor forms/donor slots" is an argument, not a clean read.

4. Claim 2's script. A run-time-read declarative script that fully determines dialog-manager behavior is closest to Denecke–Waibel's Prolog-style rule program with a domain-independent/domain-dependent split and a specification wizard. This is a reasonable § 103 combination (obvious use of a declarative configuration file), but the "application-blind dialog manager + script + backend separation" framing may survive as a system-level combination.

Confidence: Structural/architectural claims (1 in significant part, 3–9, 21–23, 34–36) — high confidence of obviousness. Scoring-based selection and mode-switching (20/33) — moderate. Threshold/rephrase and cross-form inheritance — low-to-moderate. I could not read the full texts of GUS, Goddeau, Pieraccini or Denecke–Waibel (paywalled; abstracts plus the patent's own summaries only), so each read is indicative, not verbatim.

---

7. Motivation to Combine, Framed for KSR

• Same field of endeavor: every reference is a spoken-language or frame/form-based dialogue system; several appear in the very same proceedings (Eurospeech-97 contributes D and E; ICSLP contributes C and, thematically, F).
• Explicit problem-and-solution linkage: Denecke–Waibel critique fixed-dialogue-state approaches and propose task modeling — the same critique the patent levels at Takebayashi/Pieraccini. Levin et al. are the acknowledged optimization layer over Pieraccini's architecture.
• Predictable results, known techniques: form filling (B, C) + speech I/O (A) + mixed initiative (D) + scoring/optimization (E, F) — each element performs its own known function; the combination yields nothing more than the expected aggregation.
• Applicant admissions (highly useful to a challenger): the Background states that (i) GUS and Goddeau each used a single active form; (ii) the systems demand "a more versatile" manager to handle "a plurality of topics"; (iii) query, clarification, inheritance and confirmation are "common steps involved in dialog management no matter what the task is"; and (iv) help/cancel/list-navigation are "common for all applications and domains."

---

8. Secondary Considerations

On the evidence I have, none weigh in the patentee's favor — and there is no countervailing record either:

• Litigation: per the earlier section of this analysis, no assertion, IPR, or district-court case involving US 6,246,981 was located; the patent expired 2018-11-25. Absence of litigation cuts both ways (no proof of copying, no proof of commercial success attributable to the claims).
• Licensing/industry adoption: IBM → Nuance assignment (2009) indicates value in the portfolio, but I found no nexus evidence tying that value to claims 1/11/24 specifically.
• Long-felt need: the patent's own Background argues a need (multi-topic interaction), but the references D/E already articulated the same need in 1997, which weakens rather than strengthens nonobviousness.
• Unexpected results: none recited; the specification asserts benefits (versatility, natural-language friendliness) without comparative data against GUS/Goddeau.

---

9. Bottom Line

1. Strongest invalidity theory: Goddeau (C) + Bobrow (B) + Takebayashi (A), further in view of Denecke–Waibel (E) — which knocks out the architecture, form/slot, prompt, confirmation, clarification and "information needed" limitations, plus the domain-independent features by admission.
2. Add Pieraccini (D) + Levin (F) to reach the scoring/optimization and mode-switching limitations — this is the only path that meaningfully reaches claims 1/11/24's inter-form scoring and claims 20/33.
3. The patent's realistic line of defense is narrow: that no reference teaches (a) simultaneous multi-form competition resolved by a comparative score, and (b) thresholding that score to trigger a rephrase request. If that defense holds, the surviving claims would be the scoring-dependent ones, subject to a § 112 ¶ 6 construction of the "means for" claims (8, 9, 10) and a § 101 abstract-idea overlay on the scoring step.
4. Confidence: moderate overall that a § 103 rejection could be sustained; high that the dependent domain-independent claims (3–9, 21–23) and much of the architecture are obvious. This is an analyst assessment, not a legal opinion, and it presumes a challenger who can obtain full texts of the primary references, which I could not.

Key caveats: I read abstracts, bibliographic records, and the patent's own characterizations of references B, C, D, E, and F — not their full texts. The one pre-critical-date item in the "Cited By" list (US 6,647,363, Scansoft, priority 1998-10-09) is unverified and should be pulled and evaluated; it is the single most likely additional § 102/103 reference hiding on this page.