Google LLC v. Dialect LLC

1. Case Identification

• Case #: IPR2024-00751
• Patent #: 8,447,607
• Filed: April 5, 2024
• Petitioner(s): Google LLC
• Patent Owner(s): Dialect, LLC
• Challenged Claims: 1-15

2. Patent Overview

• Title: Mobile Systems and Methods of Supporting Natural Language Human-Machine Interactions
• Brief Description: The ’607 discloses systems for processing natural language inputs on mobile devices. The technology uses a combination of speech and non-speech inputs (multi-modal) to enable users to submit questions and commands, which are then processed using context information, user-specific cognitive models, and domain-specific agents to generate a response.

3. Grounds for Unpatentability

Ground 1: Obviousness over Coffman, Kennewick, and Ross - Claims 1-15 are obvious over Coffman in view of Kennewick and Ross.

• Prior Art Relied Upon: Coffman (WO 00/20962), Kennewick (Application # 2004/0044516), and Ross (Application # 2002/0133354).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner asserted that Coffman, the primary reference, discloses the core framework of the challenged claims, including a multi-modal conversational system that receives both speech and non-speech input, generates merged transcriptions, identifies users, and uses a "context stack" to manage interactions with "conversationally-aware applications" (the claimed "domain agents"). Petitioner argued that Kennewick, which shares inventors with the ’607 patent, was added to teach the "cognitive model" limitation. Kennewick explicitly discloses using extensive user profile information—including prior interaction history, session history, and special terminology—to improve speech recognition accuracy. For dependent claims, Kennewick was also cited for its teachings on multi-user "interleaved sessions" and using an "environmental model" (an adaptive filter) to optimize the signal-to-noise ratio based on background noise. Petitioner contended that Ross was added to render obvious the limitation of identifying a matching entry in the context stack. While Coffman describes searching a context stack, Ross provides a detailed methodology for matching a transcribed utterance against grammar information stored in a context list (which functions as a stack) to identify the appropriate application.
  • Motivation to Combine: Petitioner argued a person of ordinary skill in the art (POSITA) would combine Coffman’s conversational system with Kennewick’s user-profiling techniques to achieve the well-understood benefit of improved speech recognition accuracy, a motivation Kennewick expressly provides. A POSITA implementing Coffman's system would have also been motivated to look to a reference like Ross for a detailed, conventional method of matching user input against context entries, a task only generally described in Coffman.
  • Expectation of Success: A POSITA would have had a reasonable expectation of success because combining the references involved applying known techniques (user profiling, grammar-based context matching) to a known type of system to achieve the predictable result of a more accurate and functional conversational interface.

Ground 2: Obviousness over Ground 1 Prior Art and Lee - Claims 5-7, 13-15 are obvious over Coffman, Kennewick, and Ross in further view of Lee.

• Prior Art Relied Upon: Coffman (WO 00/20962), Kennewick (Application # 2004/0044516), Ross (Application # 2002/0133354), and Lee (Application # 2002/0087315).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground specifically targeted claims requiring a "second cognitive model" or "general cognitive model" based on interactions with a plurality of users. Petitioner argued that the primary combination already taught a cognitive model based on a single user's profile (from Kennewick). Lee was added to explicitly teach using information collected from multiple users to enhance speech recognition accuracy for a current user. Lee describes a "popularity engine" that compiles data from multiple users' histories to predict likely requests and improve word recognition, directly corresponding to the claimed multi-user cognitive models.
  • Motivation to Combine: Petitioner asserted that a POSITA seeking to further improve the recognition accuracy of the system disclosed by the primary combination would be motivated to incorporate Lee’s teachings. Lee provides a clear rationale for using multi-user data to improve system performance, and all references are from the same analogous field of art.
  • Expectation of Success: The combination involved the straightforward application of a known data-aggregation technique (using multi-user histories) to improve a known function (speech recognition), which would have yielded predictable improvements.

4. Arguments Regarding Discretionary Denial

• Petitioner argued against discretionary denial under §314(a) by highlighting that the parallel district court litigation is in its early stages. At the time of filing, no trial date was set, no claim construction had occurred, and discovery had not commenced. Petitioner stated its intent to file a motion to stay the litigation if the inter partes review (IPR) is instituted.
• Petitioner further argued against denial under §325(d), asserting that the Examiner made a material error during prosecution. The Examiner considered Coffman and Kennewick but allegedly failed to appreciate their combined teachings, particularly regarding the limitation of identifying an entry in a context stack that matches a merged transcription. Petitioner contended this limitation was made obvious by the combination, especially in view of Ross, which was not before the Examiner. Similarly, Lee, which was relied upon for Ground 2, was also not considered during prosecution.

5. Relief Requested

• Petitioner requests institution of an inter partes review and cancellation of claims 1-15 of Patent 8,447,607 as unpatentable.