Google LLC v. Dialect LLC

1. Case Identification

• Case #: IPR2024-00748
• Patent #: 8,015,006
• Filed: April 5, 2024
• Petitioner(s): Google LLC
• Patent Owner(s): Dialect, LLC
• Challenged Claims: 1-6 and 9

2. Patent Overview

• Title: Systems and Methods for Processing Natural Language Speech Utterances with Context-Specific Domain Agents
• Brief Description: The ’006 patent discloses a method for processing natural language speech requests using a system that includes a speech recognition engine, a parser, and context-specific "domain agents." The system recognizes words in a user's utterance, parses them to determine a meaning and context, formulates a request, and processes it with the appropriate domain agent to generate a response.

3. Grounds for Unpatentability

Ground 1: Claims 1-3 are obvious over Coffman in view of Kanevsky, Kneser, and Zadrozny

• Prior Art Relied Upon: Coffman (WO 00/20962), Kanevsky (Patent 5,897,616), Kneser (Patent 6,157,912), and Zadrozny (1994 publication).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Coffman disclosed the foundational conversational computing system, including a speech unit, processing device, parser, and conversationally-aware applications that function as the claimed "domain agents." To implement Coffman’s speech recognition, a POSITA would have looked to Kanevsky, which Coffman expressly incorporated by reference. Kanevsky taught using n-gram language models (e.g., trigram models) which Petitioner asserted are the "dictionary and phrase tables" of claim 1. To improve performance, a POSITA would have incorporated the teachings of Kneser, which disclosed adaptive n-gram language models, thereby meeting the "dynamically updating" limitation. Petitioner further argued Coffman disclosed speaker identification ("determining an identity"), and that Zadrozny, cited by Kanevsky as an example of a semantic analyzer, confirmed the obviousness of using a grammar to parse recognized words and formulate a request for the domain agent.
  • Motivation to Combine: A POSITA would combine these references because Coffman expressly incorporated Kanevsky. The references represent a collective body of work, sharing common inventors and assignees (IBM), and address the same technical problems. A POSITA would have been motivated to use the known techniques in Kneser and Zadrozny to implement and improve the functionality described at a high level in Coffman.
  • Expectation of Success: A POSITA would have had a reasonable expectation of success because the combination involved applying well-known, conventional techniques for speech recognition and natural language understanding to achieve predictable improvements.

Ground 2: Claim 4 is obvious over Ground 1 Prior Art in further view of Maes ’101 and Franceschi

• Prior Art Relied Upon: Coffman, Kanevsky, Kneser, Zadrozny, and in addition, Maes ’101 (Patent 6,073,101) and Franceschi (Patent 6,321,196).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that this combination rendered claim 4 obvious, which adds limitations for handling speech recognition failures. Specifically, claim 4 requires requesting a user to spell unrecognized words and updating the dictionary tables based on the spelling. Petitioner asserted that Maes ’101 taught the utility of prompting a user to spell unrecognized words to improve the system's vocabulary and adapt to the user. Franceschi was cited to show that phonetic alphabet spelling was a routine, commercially available feature in speech recognition systems used to resolve ambiguity.
  • Motivation to Combine: A POSITA would combine Maes ’101 and Franceschi with the Coffman system to implement a known, effective error-resolution strategy. Prompting for spelling is a logical solution to the problem of low-confidence recognition results, which Coffman’s system would encounter. The references share a common inventor (Stephane Maes) and assignee (IBM), making them a natural source for compatible solutions.

Ground 3: Claim 5 is obvious over Ground 1 Prior Art in further view of Chai

• Prior Art Relied Upon: Coffman, Kanevsky, Kneser, Zadrozny, and in addition, Chai (Patent 6,829,603).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued this combination rendered claim 5 obvious, which adds limitations related to the step of formulating a request. These steps include determining required and optional values, extracting criteria and parameters from keywords, and inferring missing information from context. Petitioner contended that Chai taught these exact concepts through its use of "action templates" with defined "slots," some of which are "required" and others optional. Chai also explicitly disclosed inferring values for empty slots based on the discourse context.
  • Motivation to Combine: A POSITA would combine Chai's flexible slot-filling techniques with the system of the primary references to create a more robust and user-friendly natural language interface. This would allow the system to handle more complex user utterances and reduce the need for follow-up clarification prompts. Chai and Zadrozny share a co-inventor, linking Chai’s teachings directly to the core prior art.

Ground 4: Claims 6 and 9 are obvious over Ground 1 Prior Art in further view of Redfern

• Prior Art Relied Upon: Coffman, Kanevsky, Kneser, Zadrozny, and in addition, Redfern (Patent 6,078,914).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued this combination rendered claims 6 and 9 obvious, which relate to processing a request containing both a command and a question. The claims require submitting multiple asynchronous queries to distinct information sources, receiving multiple results, and scoring the relevance of the results to determine a best response. Petitioner asserted that Redfern disclosed a meta-search application that performed these exact functions. Redfern taught receiving a single natural language query, formulating and submitting asynchronous queries to multiple web search engines, and then scoring, scraping, and evaluating the results to select the best answer for the user.
  • Motivation to Combine: A POSITA would integrate Redfern's efficient meta-search backend with Coffman's conversational front-end to enhance the system's search capabilities. Redfern provided an express motivation for its natural-language-based meta-search approach as being more efficient for the user, a goal consistent with the Coffman system.

4. Arguments Regarding Discretionary Denial

• §314(a) (Fintiv): Petitioner argued that discretionary denial would be inappropriate because the co-pending district court litigation is in a very early stage. At the time of filing, no substantive discovery or claim construction had occurred, no trial date was set, and the case had recently been transferred to a new venue.
• §325(d): Petitioner contended that denial would be improper because, while the U.S. counterpart to Coffman and Chai were before the Examiner during prosecution, they were part of a large IDS of approximately 300 references and were not substantively discussed. Crucially, the Examiner did not have the benefit of the other cited references (e.g., Kanevsky, Kneser, Redfern), which Petitioner argued are necessary to provide context and demonstrate how the claims were obvious combinations of known elements.

5. Relief Requested

• Petitioner requests institution of IPR and cancellation of claims 1-6 and 9 as unpatentable.