US 7702508

I'll search for authoritative information on this patent number and any related litigation.

Let me get the claim text and check for litigation/docket information.

Let me try to retrieve the actual claim language and check specifically for any CAFC 2026 activity.

Summary of US Patent 7,702,508 B2

Important caveat up front: My tool calls returned a large set of results but I was cut off before I could retrieve the verbatim claim text, and I found no evidence of a Court of Appeals for the Federal Circuit (CAFC) 2026 docket involving this patent. Where I lack authoritative information I say so explicitly below rather than guessing.

Bibliographic data (verified against the Google Patents record and corroborating sources)

Priority chains (per the Google Patents "external-priority" entries): The 1999-11-12 priority is claimed from US 09/439,060; US 09/439,145; US 09/439,174; and US 09/439,173. A 2003-06-25 priority is claimed from US 10/603,998 (issued as US 7,392,185).

Assignment history: 2006-10-26 — assignment to Phoenix Solutions, Inc. (assignor Gururaj, Pallaki); 2009-05-20 — assignment to Phoenix Solutions, Inc. (assignor Bennett, Ian M.); 2013-08-06 — assignment to Nuance Communications, Inc.

Classifications (representative): G06F16/243 (Natural language query formulation); G06F16/24522 (Translation of natural language queries to structured queries); G06F16/3344 (Query execution using natural language analysis); G10L15/22, G10L15/30; G09B7/00.

Abstract

As reproduced by secondary sources (e.g., the Golden knowledge-graph entry):

"Candidate answers responsive to a user query are analyzed using a natural language engine to determine appropriate answers from an electronic database. The system and methods are useful for Internet based search engines, as well as distributed speech recognition systems such as a client-server system. The latter are typically implemented on an intranet or over the Internet based on user queries at his/her computer, a PDA, or a workstation using a speech input interface."

Plain-language overview of the disclosed invention

The specification describes a Natural Language Query System (NLQS) distributed across a client/server architecture. A user speaks a question; the client performs only partial speech processing (extracting MFCC vectors) and transmits them to a server, which completes recognition and converts the query to text. The server then runs a two-step algorithm (FIGS. 11A–11C):

1. Fast "pruning" step: The query text is tokenized, POS-tagged, and grouped to extract its noun phrases (NPs). A customized SQL query (using CONTAINS / FREETEXT / NEAR / AND full-text predicates) is run against a full-text database of stored question/answer pairs, returning a recordset of candidate stored questions that are linguistically similar.
2. Precise "selection" step: Each candidate stored question is likewise parsed into noun phrases; the candidate whose noun-phrase set yields the maximum match with the user's query NPs is selected as the best match. The paired answer file is then retrieved, compressed, and returned to the client, where a text-to-speech engine voices it through an animated agent.

Independent claims — plain-language overview (with uncertainty flagged)

I could not retrieve the verbatim claim set for this patent. My searches surfaced only the abstract, the specification "Definitions" summary, and third-party metadata; the actual printed claims were not returned in the results I obtained. Accordingly, the overview below is inferred from the specification's described embodiment (the FIG. 11A–11C two-step flow, the NLE/DBProcess module descriptions) and the abstract, not read off the claim language. Treat it as a description of the disclosed subject matter and expected claim scope, not as a verified claim-by-claim mapping.

Based on the title ("System and method…") and the specification, the patent appears to contain at least one method claim and at least one system/apparatus claim, each generally covering:

• Receiving/recognizing a user query (optionally a spoken query converted to text);
• Linguistically processing the query — tokenizing, part-of-speech tagging, and grouping tokens to derive noun phrases (and noun sub-phrases) of the query;
• Formulating and executing a query against a database to obtain a recordset of candidate/stored questions (the spec emphasizes full-text SQL predicates such as CONTAINS/FREETEXT/NEAR);
• Linguistically processing each candidate stored question to derive its noun phrases;
• Comparing the candidate questions against the user query using a noun-phrase-count metric to select the single best-matching stored question; and
• Returning the answer paired with the selected best-matching question (optionally compressing and transmitting it back to a client for text-to-speech rendering).

If you need the exact independent-claim wording (which is essential for any infringement/invalidity analysis), it must be pulled from the official USPTO PatentCenter/Patent Full-Text record for US 11/003,055 — I did not have that verbatim text available in this session.

Litigation / CAFC 2026 docket

• The Google Patents record carries a "Family has litigation" flag with a link to the Darts-IP global patent-litigation dataset, indicating litigation exists somewhere in this patent family — but the record does not identify the cases, and I could not confirm any specific suit on the '508 patent itself.
• I found no CAFC 2026 docket, appeal, or decision referencing US 7,702,508. My searches returned unrelated 2025–2026 matters (e.g., a Nuance Communications class-action settlement in D. Mass. with a March 31, 2026 final-approval hearing, and an unrelated Nevada case involving "Nuance Medical, LLC" / "Nuance Intermediary, LLC"). Neither involves this patent.
• Treat the litigation status as unverified. A definitive check would require the USPTO PatentCenter assignment/litigation record, PACER, or Darts-IP directly; the free sources I could reach did not expose a case caption or docket number tied to 7,702,508.

Uncertainty summary

1. Verbatim independent claims not retrieved — the overview above is inferred from the specification, not the actual claim text.
2. No CAFC 2026 activity confirmed — absence of evidence in my searches, not proof of absence.
3. Assignee is confirmed as Nuance Communications, Inc. in the Google Patents record, though one aggregator (Unified Patents portal) shows a slightly different "1999-11-11" priority date than the 1999-11-12 date on Google Patents. I have used the Google Patents dates as primary.

I'll search patent litigation sources for US 7,702,508 specifically.

Let me drill into the specific patent number and known Phoenix Solutions/Nuance litigation.

Let me verify which specific patents were asserted in each Phoenix/Nuance case, and check whether '508 appears anywhere.

Bottom line

I found no litigation anywhere naming U.S. Patent No. 7,702,508 as an asserted or challenged patent. I checked the sources you specified (Unified Patents litigation portal, CAFC/Justia dockets, PACER-derived aggregators, CourtListener), searched the literal strings "7,702,508" and "7702508", and cross-checked the Phoenix Solutions / Nuance family litigation. Nothing ties the '508 patent to a case caption, docket number, complaint, or judgment.

Per your strict rule, I am reporting that as "no litigation known," not padding the list with the family cases or the near-miss number I describe in §3.

1. Searches performed

2. Litigation involving this patent family — NOT the '508 patent

These are the Phoenix Solutions / Nuance cases that are frequently confused with the '508 patent. I list them only to rule them out, and I flag the asserted patents explicitly.

Why the '508 almost certainly was never in these cases: US 7,702,508 issued 2010-04-20. The West complaint was filed in 2009, and DirecTV in 2008 — before the '508 existed as an enforceable patent. A patent cannot be asserted before it grants. This is a structural reason, not just an absence of evidence.

3. A false positive you should be aware of (do not conflate)

One search returned an IPR petition document (P‑TACTS, IPR2022-00055) containing the sentence:

"According to assignment records, the '508 patent is assigned to MCOM IP…"
followed by a long list of MCOM IP, LLC v. Vantage Bank Texas, MCOM IP, LLC v. First Tech Federal Credit Union, MCOM IP, LLC v. IBM, etc.

This is not US 7,702,508. The surrounding facts are irreconcilable with it:

• The patent in that IPR is assigned to MCOM IP, whereas the '508 patent's recorded chain is Phoenix Solutions → Nuance Communications (assignment 2013-08-06).
• The MCOM IP campaign is a banking/transaction-processing assertion (credit unions, banks, Fiserv, FIS, Diebold), wholly different subject matter from the '508's speech-recognition/NLQS claims.
• A "'508" in that docket is a different patent number.

I am applying your literal-interpretation rule and therefore explicitly refusing to fold this into the '508 answer. I could not verify which patent number the MCOM IP "'508" actually is, and I will not guess. If you need it resolved, pull IPR2022-00055 Paper 1 from P‑TACTS and read the cover page patent number.

4. Current status of the '508 patent (relevant to why no cases exist)

• Expired – Lifetime; adjusted expiration 2022-05-31 (Google Patents record). Any assertion in 2026 would reach only pre-expiry conduct within the § 286 six-year lookback.
• No IPR, PGR, CBM, or reexamination naming it (consistent with the earlier PTAB section of this analysis).
• The litigation campaign that generated the family's case law peaked 2008–2011 — before IPR existed (2012-09-16), and the '508 did not issue until 2010-04-20, near the tail of that campaign.

5. Honest limits

1. I could not directly query the Unified Patents litigation caselist at https://portal.unifiedpatents.com/litigation/caselist filtered by this patent, nor run a PACER docket search, in this session. My "no litigation" finding is based on open-web search of the docket aggregators, CAFC/Justia, CourtListener, and the Unified portal case page for the related West case.
2. Absence of evidence is not proof of absence. A short-lived case filed and voluntarily dismissed, or a state-court/ITC action, might not appear in the free sources I reached.
3. I did not retrieve a single document asserting 7,702,508. I am not aware of any, and I will not manufacture one.

To close this out definitively, pull three records and preserve them:

• USPTO Patent Center "Reviews"/"Litigation" tab for 7,702,508 (application 11/003,055);
• Unified Patents litigation portal, searched on patent number 7,702,508: https://portal.unifiedpatents.com/litigation/caselist;
• PACER / CourtListener docket search on the literal string 7,702,508.

If any of those surfaces a caption, give me the case number and I will pull the claim-level disposition.

I'll verify the structured "no PTAB proceedings" result against public sources before reporting.

Let me check the Unified Patents patent page and PTAB-specific sources for the '508 patent directly.

Let me do a final targeted check for any institution decision or FWD citing this patent's number.

PTAB / AIA Trial Analysis — U.S. Patent No. 7,702,508 ("System and method for natural language processing of query answers")

Note on sourcing: The canonical structured source for this task — the "PTAB proceedings on file" block sourced from the USPTO Open Data Portal — reports no AIA trial proceedings. I ran independent web searches to test that result (targeting IPR/PGR/CBM petitions, institution decisions, and FWDs naming "7,702,508" / "7702508"). My searches surfaced only this patent's appearance as a cited reference in unrelated later patents' IDS/reference lists (e.g., Sonos, Apple Siri-related patents) and one noisy PTAB exhibit table that happens to list "7702508" alongside unrelated patent numbers. Nothing indicated an AIA trial proceeding in which 7,702,508 is the patent under review. That is absence of evidence, not proof of absence, but the two sources are consistent.

Proceedings overview

Total AIA trial proceedings on U.S. 7,702,508: 0 (active: 0; claims invalidated by PTAB: 0; claims sustained: 0; settled: 0; institution denied: 0).

Bottom line for a defendant: because no IPR/PGR/CBM was ever filed against this patent, there is no PTAB-issued claim cancellation, no statutory estoppel, and no Board record to lean on. Unlike a "claims 1–5 canceled" situation, you cannot cite an FWD to kill the patent — any invalidity attack must be built from scratch (district court, a fresh IPR/PGR you would file, or an ex parte reexam). The offsetting fact is that the patent is expired (adjusted expiration 2022-05-31), so the practical exposure is historical and damages-limited, and any new AIA trial would be decided under Phillips claim construction as an expired patent.

Per-proceeding detail

None. There are no proceedings to detail. No IPR, PGR, or CBM number can be reported for this patent without fabrication, so none is reported. (Per the operating rules, I do not invent proceeding numbers.)

For completeness, the nearest things my search did find — and why each is not a proceeding on this patent:

If any of these are in fact connected to '508, it is not visible in the free sources I reached, and I flag that rather than assert it.

Strategic summary

Claim status. With zero PTAB proceedings, every claim of 7,702,508 remains as issued — there is no claim canceled, narrowed, or confirmed by the Board. In PTAB terms the whole claim set is UNTESTED. Anyone asserting the patent gets the full issued claim scope (for whatever pre-expiration conduct is at issue); anyone defending has no Board precedent, no FWD findings, and no representative-claim construction to borrow. This is the opposite of a "hardened" patent (which would have survived an IPR) and also the opposite of a "troll with no case" patent (whose claims were canceled).

Estoppel landscape. Because no AIA trial was instituted, 35 U.S.C. § 315(e)(2) estoppel is inapplicable — no petitioner, and no privy of any petitioner, is barred. There is likewise no § 325(e)(2) PGR estoppel and no § 315(b) one-year bar running against anyone. Practically, this means the entire prior-art universe is still available to a current defendant: you are not limited to art that "could not reasonably have been raised" by some earlier petitioner, because there was no earlier petitioner. That is a clean slate — but a blank one, so the search-and-invalidity work is entirely yours.

Pattern signals. There is no repeat-petitioner pattern and no defensive-aggregator (e.g., Unified Patents) proceeding on this patent, because there are no petitions at all. The only litigation signal in the record is the Google Patents flag "Family has litigation — First worldwide family litigation filed" (Darts-IP family 46299868) — a family-level flag that does not identify the case or the family member, and therefore does not establish that '508 itself was ever asserted. The current listed assignee is Nuance Communications, Inc. (Microsoft subsidiary post-2022). The prior-generated sections noted a D. Mass. class-action matter involving Nuance with a 2026-03-31 final-approval hearing — unrelated to this patent.

Why the absence matters. The prior sections' observation stands: well-asserted patents eventually attract IPRs. The lack of any AIA filing against a 1999-priority speech/NLP patent held by the largest speech-recognition portfolio owner is a meaningful negative signal that this specific patent was not a front-line assertion asset, and/or that it was expired by the time modern challengers cared. Combined with expiration, the defensive posture is: low ongoing risk, clean-slate invalidity options if historically asserted, but no ready-made PTAB kill shot.

Recommended next steps

1. Treat "no PTAB activity" as the finding, and verify it at the primary source before relying on it. Check:
   • USPTO PTAB E2E / PTAB Center: https://ptab.uspto.gov (search patent owner "Nuance Communications" and patent number 7,702,508).
   • USPTO Open Data Portal (ODP) API — the source of record for the "no proceedings" result.
   • CourtListener (RECAP): https://www.courtlistener.com — full-text docket search for "7,702,508" and "7702508".
   • Federal Circuit docket / CourtListener for any appeal referencing the patent number.
2. Given no FWD exists, an invalidity position cannot be shortcut via estoppel or prior Board findings. Build it as a first-instance attack — but note the patent is expired (2022-05-31), so (a) damages are confined to the statutory look-back window on pre-expiration conduct, and (b) if you file an IPR/PGR, the Board will construe claims under Phillips, and an expired patent is still eligible for AIA review on § 102/§ 103 grounds.
3. If the patent-holder's demand letter relies on issued claim language, you cannot say "the claims are canceled." Instead, the leverage points are expiration (damages cutoff), the 1999 priority date (rich prior art from 1995–1999 speech/NLP literature), and the fact that no tribunal has ever construed these claims — so claim construction is genuinely open.
4. Do not represent, in any filing, that '508 has been litigated or invalidated without a confirmed case caption/docket from a primary source; the family-level Darts-IP flag is not sufficient to establish either.

Confidence / limits: High confidence that no AIA trial proceeding appears on file for this patent (two consistent sources). Lower confidence about litigation generally — the family-litigation flag is unverified as to '508 specifically, and free sources did not expose a case caption. I did not retrieve a verbatim claim set in this session, so any claim-number references should be confirmed against the official patent record before use.

Primary links for citation:

• Patent record: https://patents.google.com/patent/[US7702508](/patent/US7702508)/en
• PTAB E2E (verify no proceedings): https://ptab.uspto.gov
• CourtListener (docket full-text search): https://www.courtlistener.com
• USPTO PatentCenter (app 11/003,055): https://patentcenter.uspto.gov

I'll research the assignment chain, correspondents, reel/frame data, and corporate history for this patent.

Let me pull the specific assignment records and corporate history for Phoenix Solutions and Nuance.

Let me find the specific recorded assignment reel/frame data and the Nuance-to-Voice-Inventions transfer.

Assignment & Ownership Analysis — US 7,702,508 B2

Scope/limitation note up front. I hit the tool-call ceiling before I could open the USPTO Assignment Center record for US 11/003,055 directly. Everything below is reconstructed from the patent's own Google Patents legal-events block (given as authoritative in the source text), the litigation/assignment documents my searches did surface, and corporate-history reporting. I did not retrieve verified reel/frame numbers or corresponds for the '508 patent itself, and I will not invent them. Where a field is unknown I say so. I flag every inference.

Inventors

Related-party assignor worth naming: the recorded legal events also show an assignment from "GURURAJ, PALLAKI" to Phoenix Solutions (recorded 2006‑10‑26). Pallaki Gururaj does not appear as a named inventor on the '508 patent; the Gururaj→Phoenix recording is therefore an ownership cleanup of rights (founder/employee or prior-entity rights), not an inventorship event. I could not verify Gururaj's title at Phoenix Solutions from primary sources — treat the founder/principal inference as unconfirmed.

Unusual-pattern check: No evidence that all inventors departed the original assignee within 12 months of filing. The filing (2004‑12‑03) is itself a late continuation of a 1999‑priority family, and Bennett remained the assignor of record into 2009, so there is no inventor-departure/fire-sale signal on the invention side. What is unusual is the timing of the recorded assignments relative to the enforcement campaign — see the timeline and signal #5 below.

Original assignee

Entity on the issued patent: Phoenix Solutions, Inc. (Palo Alto, CA).

• Primary line of business: Phoenix Solutions was the proprietor of the "NLQS" (Natural Language Query System) distributed speech-recognition/NLP technology described in the '508 specification — a client/server natural-language voice query system marketed for e‑commerce, e‑support and e‑learning. On the evidence I found, Phoenix Solutions functioned principally as a technology owner/licensor rather than a volume product vendor: its public-facing activity in 2007–2010 was a licensing-and-litigation campaign, conducted through outside counsel.
• Did it ship a product embodying the claims? Unclear / probably not at volume. I found no filed product literature, no SKU catalogue, and no SEC-registered revenue from Phoenix Solutions. The company's visible commercial act was licensing the portfolio. This is a meaningful data gap for the NPE analysis and I am not treating it as established either way.
• Licensing/assertion channel: Phoenix Solutions' portfolio-licensing contact was publicly advertised as Mr. R. Joseph Trojan, Trojan Law Offices, P.C., Beverly Hills, CA (per the firm's Feb‑2009 announcement of the Wells Fargo license). A contingency/outside licensing firm as the front door to monetization is a classic assert-and-license posture — see signal #2.
• Current status of the original assignee: Not operating as an independent entity. Phoenix Solutions' portfolio passed to Nuance Communications, Inc. (recorded 2013‑08‑06); the 2008–2010 campaign and the Microsoft DJ action were the last visible Phoenix activity. I could not confirm a formal dissolution/bankruptcy record, so I state "no longer the owner and no current operating footprint found," not "dissolved."

Assignment timeline

Reel/frame numbers: NOT RETRIEVED. The Google Patents legal-events block gives dates, parties, and conveyance types but not reel/frame, and I was unable to open the Assignment Center abstract-of-title for 11/003,055 before the tool limit. The entries below are the recorded legal events; the recording dates shown by Google Patents are the events' recorded dates (they may differ from the underlying execution dates, which I could not verify). Correspondent fields are unknown for all three entries — I will not fabricate them.

• 2006‑10‑26 (recorded; execution date unverified) — Reel not retrieved/unknown

  • Conveyance: Assignment (assignment of interest)
  • Assignor: GURURAJ, PALLAKI
  • Assignee: PHOENIX SOLUTIONS, INC.
  • Correspondent: unknown — not retrieved
  • Context: Internal ownership consolidation — a non-inventor assignor conveying rights into the original assignee; consistent with founder/employee rights cleanup.

• 2009‑05‑20 (recorded; execution date unverified) — Reel not retrieved/unknown

  • Conveyance: Assignment (assignment of interest)
  • Assignor: BENNETT, IAN M. (sole named inventor)
  • Assignee: PHOENIX SOLUTIONS, INC.
  • Correspondent: unknown — not retrieved
  • Context: Inventor-to-company assignment recorded mid-enforcement-campaign — Phoenix was by then suing Wells Fargo (N.D. Cal. 3:08‑cv‑00863, filed 2008‑02‑08) and had filed against PG&E (C.D. Cal., Jan‑2009). Recording the inventor's assignment during active litigation is a standing-chain cleanup, not a market transfer.

• 2013‑08‑06 (recorded) — Reel not retrieved/unknown

  • Conveyance: Assignment (assignment of interest)
  • Assignor: PHOENIX SOLUTIONS, INC.
  • Assignee: NUANCE COMMUNICATIONS, INC.
  • Correspondent: unknown — not retrieved
  • Context: Exit/portfolio sale — the entire Phoenix NLQS family passes to Nuance, an operating speech-technology company. This is the terminal link in the chain and matches the Google Patents "Current Assignee: Nuance Communications, Inc."

If you need the reel/frame and corresponds, pull the abstract of title at https://assignmentcenter.uspto.gov/ (search patent number 7702508) or the indexed mirror https://assignment.uspto.gov/patent/index.html. I could not complete that step in this session.

Timeline diagram

timeline
    title Ownership of US 7702508
    1999 : Priority date for NLQS family
    2004 : Application 11 003 055 filed
    2006 : Gururaj assigns rights to Phoenix Solutions
    2009 : Bennett assigns rights to Phoenix Solutions
         : Recorded during enforcement campaign
    2010 : Patent US 7702508 issues
    2013 : Phoenix Solutions assigns portfolio to Nuance
    2021 : Nuance sells some voice patents to Voice Inventions
    2022 : Microsoft acquisition of Nuance completes
         : Patent term adjusted expiration 31 May 2022

NPE / troll-pattern signals

1. Shell-entity transfer — NOT PRESENT. The chain runs inventor/related-party → Phoenix Solutions, Inc. (operating-named corporation) → Nuance Communications, Inc. (large operating company). There is no "IP / Holdings / Ventures / Licensing LLC," no single-member LLC, and no registered-agent-service address anywhere in the recorded chain (per the Google Patents legal events). The name "Phoenix Solutions, Inc." is a corporate operating name, not an obvious shell designator, and I found no evidence of a registered-agent drop-box.

2. Known asserter in the chain — PRESENT (moderate). Phoenix Solutions is not on the classic high-frequency NPE rosters (Acacia, Marathon, IV, Wi‑LAN, Round Rock, etc.), but it functioned as an assertion/licensing entity: in 2008–2010 it sued Wells Fargo (N.D. Cal. 3:08‑cv‑00863), PG&E (C.D. Cal., Jan‑2009), West Interactive (C.D. Cal., Nov‑2009), and DirecTV (C.D. Cal. No. 08‑984), and it sent American Express license demands covering 15 patents (the demand that triggered Microsoft v. Phoenix Solutions, a declaratory-judgment action that survived Phoenix's motion to dismiss — Law360, 2010‑08‑19). The target set — banks and end-user/IVR deployers, not platform competitors — is the single most probative NPE tell in this record. Nuance, by contrast, is an operating company.

3. Repeat correspondent across the chain — UNCLEAR / NOT RETRIEVED. I obtained no correspondent-of-record data for any of the three recorded assignments on the '508 patent, so I cannot assess recurrence, and I will not guess. Two adjacent data points that would matter if corroborated on the reel/frame faces: (a) the Phoenix portfolio's public licensing contact was R. Joseph Trojan, Trojan Law Offices, P.C. (Beverly Hills, CA) — an assertion/monetization correspondent, not a prosecution one; and (b) Nuance filings of this era were signed by in-house Leanne J. Fitzgerald, Associate General Counsel – IP, Nuance Communications, Inc. Neither is confirmed as the correspondent on the '508 recordings. Do not treat either as a finding until the assignment faces are pulled.

4. Cascading transfers — NOT PRESENT. Only three recorded events across 2006 → 2009 → 2013 (approx. 7 years end-to-end). No sub-24‑month LLC chain, no shared-correspondent relay, no common-principal clustering.

5. Pre-litigation transfer — PRESENT (weak-to-moderate), on the recording side. The inventor's assignment to Phoenix was recorded 2009‑05‑20, squarely during the enforcement campaign (Wells Fargo suit filed 2008‑02‑08; PG&E filed Jan‑2009). That timing is consistent with a standing/chain-of-title cleanup done to support assertion, which is a recognized pre-assertion pattern — though here it is a completion of title, not a transfer to a new litigating shell. The 2013 sale to Nuance is post-campaign and carries no assertion taint.

6. Bankruptcy fire-sale — NOT PRESENT / UNCLEAR. No bankruptcy proceeding involving Phoenix Solutions appears in anything I retrieved. The 2013 Phoenix→Nuance transfer is not documented by me as a §363 sale; I found no Chapter 7/11 record. Unverified, not affirmatively disproven.

7. Privateering — UNCLEAR, and NOT applicable to this patent as recorded. There is a genuine privateering thread in the broader Nuance portfolio: Nuance assigned voice patents out to Voice Inventions, LLC effective 2021‑01‑01 (assignment document executed by David Greenbaum, VP Litigation & IP, Nuance; recorded in connection with Nuance's 2020 sale). Nuance then sold these to entities that became active patent owners — the PTAB papers describe Nuance acquiring VoiceBox patents in Jan‑2015 and selling them to Voice Inventions, LLC in January 2021, with Microsoft acquiring Nuance only in March 2022. However: the '508 patent's Google Patents "Current Assignee" remained Nuance Communications, Inc., and the Voice Inventions schedules I retrieved list VoiceBox-origin patents (e.g., US 7,398,209; 7,502,738; 8,015,006), not US 7,702,508. I therefore mark privateering unclear / likely not applicable to '508 specifically, and flag it as the highest-value item to verify against the Voice Inventions portfolio schedules.

8. Defensive aggregator — NOT PRESENT. The chain terminates at Nuance Communications, Inc., an operating company (Microsoft subsidiary since March 2022). There is no RPX / AST / LOT / Unified / OIN link. The '508 patent was not neutralized by a defensive aggregator — it simply expired (adjusted expiration 2022‑05‑31).

Verdict

NPE — moderate confidence.

The chain itself (inventor → Phoenix Solutions → Nuance) is an ordinary operating-company-to-operating-company assignment sequence with no shell-entity, cascading-transfer, or defensive-aggregator signals; but the original assignee's conduct supplies one strong assertion signal — Phoenix Solutions ran a multi-defendant campaign against end-user/financial-services deployers (Wells Fargo, PG&E, West Interactive, DirecTV) with American Express license demands covering the family, and routed its portfolio licensing through outside counsel (R. Joseph Trojan, Trojan Law Offices) while recording the inventor's assignment to itself mid-litigation on 2009‑05‑20. That combination — non-practicing-style assertion against customers rather than competitors, plus contingency/outside licensing — is why the call is "moderate" rather than "operating-company assertion." It is not "high confidence" because I could not retrieve reel/frame or correspondent data for any of the three records, and I could not confirm that Phoenix Solutions lacked a shipping product.

Verify at: USPTO Assignment Center abstract of title for 11/003,055 — https://assignmentcenter.uspto.gov/ (mirror: https://assignment.uspto.gov/patent/index.html). Also cross-check the Voice Inventions, LLC schedules (2020‑12‑24 document) to confirm whether the expired '508 was in or out of the January‑2021 Nuance carve-out.

Caveats I am holding to

• No reel/frame numbers are reported above because I did not retrieve them; the three legal events are from the patent's own record, not fabricated.
• Correspondent fields are blank (unknown), not omitted for brevity — signal #3 is therefore "unclear," not "not present."
• The Voice Inventions privateering thread is NOT tied to the '508 patent on the evidence I have; it is flagged for verification, not asserted.
• Consistent with the earlier sections in this analysis: no CAFC 2026 activity and no PTAB proceeding were found for this patent, and the patent is expired (2022‑05‑31) — which caps the practical relevance of any transfer after that date.

I'll research the prior art cited on the face of US 7,702,508 and in its specification. Let me run several targeted searches.

Let me try to pull the patent's own "References Cited" section and the USPTO full-text record directly.

Let me retrieve the patent's front-page reference list from the PDF image and FreePatentsOnline.

Prior-Art Analysis — US 7,702,508 B2

Sourcing note and a hard caveat (read first)

I searched for this patent's cited references and was able to retrieve the front-page "(56) References Cited" list (via the USPTO-report rendering of the grant) plus the references the specification expressly incorporates by reference, plus the non‑patent‑literature citations. I did not retrieve the verbatim printed claims in this session — my tooling stopped before I could pull the claim text from Patent Center/patentimages. That matters: § 102 anticipation is a claim‑limitation‑by‑claim‑limitation inquiry, and I cannot honestly say "reference X anticipates claim 3" without the claim language in front of me.

So in the tables below, the "potentially anticipates" column is expressed by claimed theme, and each entry is flagged for confidence. Where I could not independently verify a reference's subject matter in this session, I say so rather than invent a title. I also flag the priority-date problem that dominates this particular prior-art analysis.

Cross-check against the previously generated sections: those sections correctly noted the patent is expired (adjusted expiration 2022‑05‑31) and that no PTAB proceeding exists. Nothing here contradicts that. One correction to build on: the earlier sections flagged that verbatim claims were not retrieved; my searches surfaced the cited-reference list but still not the claims, so the claim-mapping gap remains open.

1. The § 102 framework that actually governs this patent

This is a pre-AIA patent (priority 1999‑11‑12; application 11/003,055 filed 2004‑12‑03). Three things drive everything below:

The trap: most of the cited U.S. patents are dated 1999–2001 — i.e., on or after the 1999‑11‑12 priority date. A reference dated after the priority date is not § 102(a) or (b) art against claims entitled to that priority. It can only be § 102(e) art if the underlying U.S. application was filed before the invention date, or it was cited against claims that lost the 1999 priority. This is exactly the kind of priority/inventorship fight that the Phoenix Solutions family litigated in C.D. Cal. (see the previously generated litigation section re: West Interactive and DirecTV). Any anticipation theory below must be run against the correct reference date, not the face date.

Also note: not every cited reference is § 102 art. Examiners routinely cite references under § 103 (obviousness combinations) or purely as background. A § 102 rejection requires one reference disclosing every limitation. I mark which entries are plausible single-reference § 102 candidates vs. § 103/background.

2. Front‑page U.S. patent references ("(56) References Cited")

These are the U.S. Patent Documents listed on the grant as cited-by-examiner (retrieved from the USPTO-report page for 7,702,508). I reproduce them literally; do not auto-correct.

2A. Most relevant to the claimed subject matter (NLP search / query answering / speech)

2B. Additional cited U.S. patents (background / § 103 / component-level art)

The face list continues with these (full citation, literal numbering, face date, listed inventor) — I group them because they are more plausibly § 103 or background art than single-reference § 102 anticipation, and I flag that I did not re-verify each title:

US 5,974,413 (Oct 1999, Beauregard et al.); US 5,978,756 (Nov 1999, Walker et al.); US 5,983,190 (Nov 1999, Trower, II et al.); US 5,987,404 (Nov 1999, Della Pietra et al.); US 5,987,410 (Nov 1999, Kellner et al.); US 5,995,918 (Nov 1999, Kendall et al.); US 5,995,928 (Nov 1999, Nguyen et al.); US 6,009,387 (Dec 1999, Ramaswamy et al.); US 6,018,736 (Jan 2000, Gilai et al.); US 6,023,697 (Feb 2000, Bates et al.); US 6,035,275 (Mar 2000, Brode et al.); US 6,038,535 (Mar 2000, Campbell); US 6,044,266 (Mar 2000, Kato); US 6,044,347 (Mar 2000, Abella et al.); US 6,052,439 (Apr 2000, Gerszberg et al.); US 6,052,656 (Apr 2000, Suda et al.); US 6,061,646 (May 2000, Martino et al.); US 6,076,088 (Jun 2000, Paik et al.); US 6,078,886 (Jun 2000, Dragosh et al.); US 6,085,160 (Jul 2000, D'Hoore et al.); US 6,088,692 (Jul 2000, Driscoll); US 6,101,472 (Aug 2000, Giangarra et al.); US 6,105,023 (Aug 2000, Callan); US 6,112,176 (Aug 2000, Goldenthal et al.); US 6,119,087 (Sep 2000, Kuhn et al.); US 6,125,284 (Sep 2000, Moore et al.); US 6,125,341 (Sep 2000, Raud et al.); US 6,131,085 (Oct 2000, Rossides); US 6,138,087 (Oct 2000, Budzinski); US 6,138,088 (Oct 2000, Goeser); US 6,138,089 (Oct 2000, Guberman); US 6,138,091 (Oct 2000, Haataja et al.); US 6,138,100 (Oct 2000, Dutton et al.); US 6,141,640 (Oct 2000, Moo); US 6,144,848 (Nov 2000, Walsh et al.); US 6,144,933 (Nov 2000, …).

Caveat on this block: I confirmed the numbers/dates/inventors as listed, but I did not independently verify each reference's title or subject matter in this session, and the list is truncated at the point my source cut off (US 6,144,933). Treat descriptions as unverified. The face list also includes foreign patent documents (EP 1 094 388, Apr 2001; EP 1 096 471, May 2001; and WO 98/11534, WO 99/48011, WO 99/50830, WO 00/14727, WO 00/17854, WO 00/20962, WO 00/21075, WO 00/21232, WO 00/22610, WO 00/30072, WO 00/30287, WO 00/68823, WO 01/16936, WO 01/18693, WO 01/26093, WO 01/78065, WO 01/95312, WO 02/03380) — note several of these post-date the 1999 priority and are therefore § 102(e)-only or not prior art at all.

3. References expressly incorporated by reference in the specification

These carry unusual weight because the patentee affirmatively adopted them — they are the patent's own admitted state of the art:

4. Non‑patent literature (NPL) cited

The patent's NPL citations cluster into four groups. These are printed publications and therefore § 102(a)/(b) art measured against 1998‑11‑12 / the invention date:

• HMM classics — Baum et al. (late 1960s–early 1970s); Baker (CMU, late 1970s); Jelinek et al. (IBM, late 1970s); Young et al. (Cambridge, 1990s). Relevance: § 102(b)/background on the acoustic-model theme — not, by themselves, anticipatory of a full query-answer system.
• NLP / dialogue — Agarwal, "Towards a Pure Spoken Dialogue System for Information Access" (ACL/EACL Workshop, Madrid, 1997); Ammicht et al., "Knowledge Collection for Natural Language Spoken Dialog Systems" (Eurospeech, Budapest, Sep 1999); Appelt & Pollack (SRI, 1990); DeJong (Yale, 1977); Grosz & Kraus (1996); Grosz & Sidner (1986); Hobbs et al. (1980, 1993); Lascarides & Asher (1991); Lewis (1979); Lochbaum (Harvard PhD, 1994); McRoy & Hirst (1993); Perrault & Allen (1980); Pollack (1986); Rayner & Alshawi (1992); Rieger (Stanford AI Memo AIM‑233, 1974); Grice (1969); Perlis et al. (1997).
• Speech systems — AT&T Watson system documentation (1996, 1998); Gorin, "Processing of Semantic Information in Fluently Spoken Language" (ICSLP, Philadelphia, Oct 1996); Gorin et al., "How May I Help You?" (draft May 21, 1997); Riccardi et al., "A spoken language system for automated call routing" (ICASSP '97, Munich); Prudential News, "Prudential Pilots Revolutionary New Speech-Based Telephone Customer Service System Developed by AT&T Labs" (Dec 6, 1999).
• Semantics/evaluation — Baker, Fillmore & Lowe, "The Berkeley FrameNet project" (COLING‑ACL, Montreal, 1998); Graesser et al., "Using latent semantic analysis to evaluate the contributions of students in AutoTutor" (Interactive Learning Environments, in press); Mohri, "String-Matching With Automata" (Nordic Journal of Computing, 1997).

The spec's own "Definitions" section names Graesser et al. (1999), Baker et al. (1998), and Gildea & Jurafsky (2001) as the hybrid statistical/semantic state of the art. Note the date problem again: Graesser et al. (1999) and Gildea & Jurafsky (2001) are at or after the priority date; only references published before 1998‑11‑12 are § 102(b) art.

5. Bottom-line prior-art assessment

Most relevant prior art for the '508 patent, in order of analytical value:

1. US 5,680,628 — earliest of the patentee-admitted NLP systems; clears the 1998‑11‑12 § 102(b) bar. Best single-reference candidate against the "linguistically process a natural-language query" theme, but almost certainly does not reach the two-step NP-match selection architecture.
2. US 5,963,940 and US 6,006,221 (Liddy et al.) — the most on-point patent references for natural-language query → retrieval; the § 102 analysis turns entirely on whether they disclose noun-phrase extraction used as the matching metric.
3. US 6,021,384 / US 6,044,337 (Gorin et al., AT&T) combined with the AT&T Watson NPL and Gorin (1996/1997) — collectively the strongest art against the spoken-query-to-answer theme; the date question (2000 patents vs. 1999 priority) is the deciding issue.
4. US 5,915,236 — the reference the patentee distinguishes; useful for § 103, weak for § 102 because of the patentee's own distinguishing statement.
5. US 6,029,124 (Gillick), US 6,032,111 (Mohri), US 6,122,613 (Baker) — acoustic-model/finite-state component art; § 102(e)/§ 103 material for the speech-recognition claims only.

Provisional theme-to-§ 102 mapping (NOT verified against claim numbers):

Key takeaway: the cited art is strongest against the "process the natural-language query" and "spoken query → answer" themes, and weakest against the specific NP-count two-step selection that the specification presents as the invention. That asymmetry is why, in the family's district-court history, the invalidity wins came from other patents' claims (the '640, '431, '854) — which aligns with the earlier sections' finding that the '508 itself does not appear to have been adjudicated on the merits.

6. Explicit uncertainty and what you must verify

1. Verbatim claims still not retrieved. Every claim mapping above is by theme, inferred from the abstract and spec — not read off the claim language. This is the single biggest limitation. Pull the printed claims from USPTO Patent Center (application 11/003,055) or the grant PDF before relying on any "anticipates claim N" statement.
2. Reference descriptions are partially unverified. I confirmed numbers/dates/inventors from the face list, but I did not independently verify each title/subject in this session, and the list I retrieved is truncated at US 6,144,933. Re-pull the complete "(56)" block from the grant PDF for the full set.
3. Dates control. Do not treat a 2000-dated patent as § 102(a)/(b) art against a 1999‑11‑12 priority. Confirm the priority chain (US 09/439,060; 09/439,145; 09/439,174; 09/439,173) actually supports each claim — if it does not, the effective date moves later and more of this art becomes available.
4. Methodology limits. I hit my tool-step ceiling mid-search, so this is not a substitute for a formal IPR-quality prior-art search (e.g., a full-text search of the 1997–1999 corpus for "noun phrase" + "query" + "speech"). Given the patent expired 2022‑05‑31, note the earlier sections' point that damages would be limited to pre-expiry conduct within the § 286 six-year lookback.

Sources cited: Google Patents record, https://patents.google.com/patent/US7702508/en (spec's "Definitions" section, incorporated-by-reference patents, NPL); USPTO-report grant page, https://uspto.report/patent/grant/7,702,508 (front-page "(56) References Cited" list); Justia patent page, https://patents.justia.com/patent/7702508 (NPL/other-references listing).

I'll ground this analysis in the patent's own cited/incorporated references and check for any external prior-art listing. Let me run targeted searches.

Let me pull the patent's cited references and identify key ones referenced in the specification.

Let me make one targeted attempt to retrieve the actual claim language, since the prior sections flagged it as unavailable.

Obviousness Analysis — U.S. Patent 7,702,508 under 35 U.S.C. § 103 (pre‑AIA)

0. Scope, sourcing, and two contradictions to flag up front

Contradiction / gap #1 — the "Prior Art section of this page" was not supplied to me. You instructed me to "use the results from the Prior Art section of this page." No such section appears in my inputs. The only prior-art information I actually hold is (a) the patent's own specification, which expressly identifies and discusses prior-art references, and (b) web results I retrieved in this session. I therefore build the analysis on the applicant's own admitted prior art (the strongest kind of § 103 evidence, because it is a binding admission) and flag every non-admitted reference as "needs verification." If a separate Prior Art section exists on your page, my analysis must be reconciled against it — I have not seen it.

Contradiction / gap #2 — claim text. The prior-generated sections of this analysis stated the verbatim claims were unavailable. In this session I did retrieve a full claim set (claims 1–20) for US 2005/0080614 A1 (the pre-grant publication of application 11/003,055, the application that issued as 7,702,508) at https://patents.justia.com/patent/20050080614. Caveat: pre-grant publication claims are not the issued claims; amendments during prosecution are common. Treat the mapping below as provisional and confirm against the printed patent. (Note also that claim 20 was truncated in my source; text stops at "(b) linguistically processing said query text to generate search predicates and recognized phrases.")

1. Legal framework, the person of ordinary skill, and the critical date

Statute: pre-AIA § 103(a) (application filed 2004-12-03; priority 1999-11-12). KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007), governs motivation-to-combine.

Graham factors — (1) scope/content of prior art; (2) differences; (3) level of ordinary skill; (4) secondary considerations. Graham v. John Deere Co., 383 U.S. 1, 17–18 (1966).

POSITA (proposed): a B.S. in CS/EE plus ~2 years, or an M.S., in speech recognition and/or information retrieval, with working familiarity with (i) HMM-based ASR, (ii) client–server/network programming, (iii) NLP tokenizing/tagging/parsing, and (iv) relational databases with full-text search. This is a routine-combination skill set — which favors obviousness.

Critical date — this matters enormously. The claims are entitled to the 1999-11-12 date only if the disclosure supports them under § 112 ¶ 1. The specification's semantic-decoding material (FIG. 19–21; "hybrid techniques," LSA, FrameNet) reads as later-added subject matter tied to the 2003-06-25 / 2004-12-03 filings. Practical consequence:

Several references below (e.g., Graesser et al. 1999) sit right on the 1999 boundary. A real § 103 contention should run under the 2003/2004 date first and the 1999 date in the alternative.

2. The claims, grouped by limitation (from US 2005/0080614 A1)

3. The prior art

3.1 Admitted prior art — from the patent's own specification (binding admissions)

All quotes from https://patents.google.com/patent/US7702508/en:

This is the single most damaging § 103 fact: the patent implements Groups E and F using named, pre-existing SQL Server features and concedes their existence, and it cites the NLP, distributed-ASR, and statistical/semantic components as known art.

3.2 Non-admitted references (flagged — verify before use)

4. Combinations, with motivation to combine

Combination 1 — Distributed ASR (Qualcomm '683 or GTE '399) + Internet full-text retrieval of paired answers

Covers: Groups A, H, I; claim 2, 5, 14, 19 (speech-derived query, Internet server).

• '683 discloses client-side acoustic feature extraction with remote word decoding; '399 discloses an HMM-based client/server recognizer that "returns certain keywords … to fill out a dedicated form."
• Yang/Ehrlich (Excite/Yahoo) keyword retrieval over an indexed corpus is admitted art.
• Motivation: the patent itself states the goal — thin-client support and real-time latency ("Processing of the speech utterance is divided between the client and server side so as to optimize processing and transmission latencies"); distributed ASR exists precisely to serve low-resource clients (Gould '236's "tailoring … to available computational resources" supplies the express motivation). Combining a remote recognizer with a remote indexed search service is a predictable combination of two server-side functions on the same host — a classic KSR "arrangement of old elements."

Combination 2 — Combination 1 + post-retrieval phrase/NP re-ranking of candidates

Covers: Groups B, C, D; claims 1, 4, 6, 7, 8, 10, 15, 16, 19.

• Strzalkowski NLIR: retrieve with a fast lexical pass, then re-rank by phrase/NP evidence.
• FAQ Finder: compare the incoming question to stored questions and return the paired answer; TF-IDF + semantic.
• Motivation: precision/recall tradeoff — a fast lexical first pass maximizes recall cheaply; an expensive linguistic second pass over a small candidate set maximizes precision affordably. The patent's own words supply the motivation: the first step "acts as an efficient and fast pruning mechanism" reducing the universe "to a manageable set of likely candidates in a very short period of time," and the second step corrects lexical-search error: "even in cases where perfect speech recognition accuracy is not achieved for each word … the entire query itself is nonetheless recognized with sufficient accuracy." Choosing noun phrases as the unit is not an inventive leap — the specification itself reasons that "because of the enormous diversity of nouns … it is much more relevant to consider the noun phrase as a key linguistic metric," which is simply the well-known justification for NP-based indexing in NLIR.
• Note claim 1's breadth: "an evaluation of at least individual words in the user query which are in common with said first set of potential answers" is close to plain TF-IDF word-overlap scoring. Under § 103 that limitation is very weak.

Combination 3 — Combination 2 + off-the-shelf SQL full-text predicates and asynchronous index maintenance

Covers: Groups E, F; claims 11, 17, 18.

• The specification admits SQL Server's full-text engine supplies CONTAINS/FREETEXT/NEAR/AND and that index population/refresh is asynchronous.
• Motivation: using the database's own full-text facility rather than hand-built inverted indices is the obvious engineering choice where an Internet server already runs a relational DBMS — a "predictable use of prior-art elements according to their established functions" (KSR). The latency benefit of deferring index updates is expressly recognized by the applicant as a design goal ("to reduce recognition latency").

Combination 4 — Combination 3 + context/environment-based restriction

Covers: Group G; claims 9, 13.

• The specification concedes that Course/Chapter/Section (or Product Class/Category/Brand) are "environment variables," and that dynamically loading a scoped grammar/dictionary based on context is known.
• Motivation: narrowing the search and the grammar to the user's current page/section reduces latency and improves accuracy — the same rationale stated for dynamic grammar loading.

Combination 5 — Combination 4 + statistical & semantic hybridization

Covers: Group J; claim 3, also claims 1/6 (word vs. phrase).

• LSA (statistical) + FrameNet/semantic roles + Graesser et al.'s hybrid ITS are cited in the specification itself as known.
• Motivation: the applicant's own rationale — statistical methods detect surface overlap, semantic methods tolerate paraphrase; combining them improves question matching (Burke et al.'s FAQ Finder does exactly this with TF-IDF + WordNet).

The "best single reference" candidate

If forced to pick the most efficient attack: Burke et al. (FAQ Finder, 1997) alone or with GTE '399 / SQL Server full-text as secondary references likely reaches most of claims 1–17. Claim 18 (async index update) is best met by the SQL Server admission; claim 2 by '683/'399.

5. Why a POSITA would have combined them (KSR rationales)

1. Predictable combination of known elements, each performing its known function — a client-server recognizer, an indexed text search engine, and an NL parser are three known modules; chaining them is not architecturally inventive.
2. Design/market demand — voice-enabling web commerce/support and real-time response (3–5 s) drove the industry to distribute recognition and to prune search results.
3. Explicit motivation in the prior art to distribute based on resources — Gould '236 expressly teaches tailoring recognition to available computation, which invites client/server partitioning.
4. "Obvious to try" with a finite, predictable set of solutions — using words vs. phrases vs. noun phrases as the matching unit is a small, enumerated design space with predictable trade-offs.
5. Off-the-shelf availability — the entire predicate/indexing layer was a shipping commercial product (SQL Server), which is a powerful KSR "predictable use" argument.

6. Counterarguments the patent owner will raise (and the rebuttals)

Non-obviousness survivability: I would rate claim 18 (async index update) and the specific interaction among Groups B–D–F as the relatively more defensible limitations; claims 1, 8, 10, 17, and 19 are the most vulnerable because they read on ordinary search-then-rank behavior and admitted commercial tools.

7. Confidence and what I could not do

• High confidence: the specification's own admissions (the CONTAINS/FREETEXT/NEAR/AND and asynchronous-indexing passages; the '683/'399/Gould/Kurzweil/NLP-patent citations; the LSA/FrameNet/Graesser citations) — these are quoted from the authoritative patent text and are the strongest § 103 evidence.
• Moderate confidence: the claim text, because it comes from US 2005/0080614 A1 (pre-grant publication), not the issued patent. Claim 20 was truncated in my source; claims beyond 20, if any, are unknown to me.
• Moderate confidence, needs verification: Strzalkowski NLIR and Burke/FAQ Finder citations and their exact content/priority dates. I did not fabricate content for them, but I also could not verify them in this session (my search budget was exhausted).
• Not done: comparison against the actual Prior Art section you referenced (not supplied), and a check of the file-wrapper/IDS of 11/003,055 for examiner-cited art (which would likely contain the most powerful § 102/§ 103 references, e.g., NLIR and FAQ-retrieval patents).

Recommendation: to convert this into a citable contention, (1) pull the issued claims from PatentCenter for 11/003,055, (2) pull the front-page "References Cited" and the IDS, (3) confirm Strzalkowski and Burke with full bibliographic details, and (4) decide the priority date for each claim element in a claim-by-claim priority analysis.

Primary links: patent record — https://patents.google.com/patent/[US7702508](/patent/US7702508)/en ; published claims — https://patents.justia.com/patent/20050080614 ; PatentCenter — https://patentcenter.uspto.gov (app 11/003,055).