PTAB

IPR2025-01334

Meta Platforms Inc v. Dialect LLC

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Method Responsive to a User Generated Natural Language Speech Utterance
  • Brief Description: The ’209 patent describes a system for processing spoken user requests. The system recognizes words and phrases from a natural language utterance using dictionaries and phrase tables, determines the user's intent to select a corresponding "domain agent" (an autonomous executable), and invokes that agent to process the request.

3. Grounds for Unpatentability

Ground 1: Claims 1 and 4 are obvious over Ross in view of Ittycheriah.

  • Prior Art Relied Upon: Ross (Patent 7,085,723) and Ittycheriah (Patent 5,937,383).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Ross discloses the overall framework of the claimed method, including a "speech center" that receives a spoken utterance, determines a target application (a "domain agent") based on context, and invokes it. Petitioner asserted Ross is agnostic about the specific speech recognition engine used. Ittycheriah, an IBM patent like Ross, allegedly supplies the missing details by teaching a speech recognition engine that uses a cache database of frequently used keywords and a larger "classical" vocabulary. Petitioner contended these databases function as the claimed "dictionary and phrase tables." Ittycheriah's use of Hidden Markov Models (HMMs) to generate "likelihood scores" for keywords allegedly teaches the limitation of "maintaining a dynamic set of prior probabilities."
    • Motivation to Combine: A POSITA would combine these references because Ross expressly contemplates using off-the-shelf speech engines. Ittycheriah was presented as a well-known, contemporaneous IBM system designed to improve speech recognition speed and accuracy via its caching mechanism, making it an obvious and desirable choice to implement in Ross's more general framework.
    • Expectation of Success: Petitioner asserted a high expectation of success, as the combination involves integrating a specialized component (Ittycheriah's engine) into a compatible system (Ross's framework) for its intended purpose. The output of Ittycheriah (decoded text) is directly usable as the input for Ross's context manager.

Ground 2: Claim 6 is obvious over Ross and Ittycheriah in view of Ross018.

  • Prior Art Relied Upon: Ross (Patent 7,085,723), Ittycheriah (Patent 5,937,383), and Ross018 (Patent 7,249,018).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground builds on the Ross/Ittycheriah combination to address the specific parsing and formulating limitations of claim 6. Petitioner argued that Ross018, which shares a provisional application with Ross and is described as an "expansion" of it, teaches a sophisticated parser that satisfies these limitations. Ross018 allegedly discloses a "semantics analysis module" that processes an utterance to determine required and optional values, extracts criteria and parameters into "attribute-object-value triples," infers further parameters using a goal-directed reasoning facility, and transforms this information into a script call (a "token") compatible with the selected domain agent's grammar.
    • Motivation to Combine: A POSITA would add Ross018's teachings to the Ross/Ittycheriah system because Ross018 was explicitly designed as a complementary enhancement to Ross. Its purpose was to add domain knowledge and reasoning to better understand user utterances, making it a natural and motivated addition to improve the base system's functionality.
    • Expectation of Success: Success would be expected, as Ross018 was designed to integrate with and expand upon the functionality of Ross's conversation manager.

Ground 3: Claims 1, 4, and 6 are obvious over Hartono in view of Baker.

  • Prior Art Relied Upon: Hartono (WO 00/11571) and Baker (Patent 5,680,511).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground presents an alternative basis for invalidity. Petitioner argued that Hartono discloses an agent-based architecture for processing natural language commands that maps to the overall framework of the challenged claims, including parsing speech, selecting task agents (domain agents), and invoking them. Like Ross, Hartono is generic regarding the specific speech recognition module. Baker, a system from leading developer Dragon Systems, allegedly provides the detailed recognition engine. Petitioner asserted that Baker teaches maintaining dynamic probabilities through its use of a "choice list" for ambiguous words, acoustic probability signals, and context matching using a "word list" derived from reference databases (e.g., dictionaries, thesauruses), which function as the claimed "dictionary and phrase tables."
    • Motivation to Combine: A POSITA would combine these references because Hartono explicitly states that any "suitable off the shelf speech recognition system" could be used. Baker was presented as a well-known commercial system that offered advanced features like recognizing words "on the fly" using context, making it an obvious and advantageous choice to implement as Hartono's recognition module.
    • Expectation of Success: Petitioner argued for a high expectation of success because Hartono's system was designed to be modular. Baker's system produces a standard speech-to-text output that could be predictably integrated with Hartono's natural language parser.
  • Additional Grounds: Petitioner asserted additional obviousness challenges based on combinations including Aridor (Patent 6,636,848) to address claims 7-9, which relate to processing questions and querying network information sources. These grounds argued for adding Aridor's metasearch and result-ranking "knowledge agents" to the base systems taught by Ross/Ittycheriah or Hartono/Baker.

4. Relief Requested

  • Petitioner requests institution of an inter partes review (IPR) and cancellation of claims 1, 4, and 6-9 of the ’209 patent as unpatentable.