Google LLC v. Dialect LLC

1. Case Identification

• Case #: IPR2024-00746
• Patent #: 7,398,209
• Filed: April 5, 2024
• Petitioner(s): Google LLC
• Patent Owner(s): Dialect, LLC
• Challenged Claims: 1-9 and 12

2. Patent Overview

• Title: Systems and Methods for Responding to Natural Language Speech Utterance
• Brief Description: The ’209 patent discloses a system for processing natural language speech from a user to retrieve online information or execute commands. The system uses a speech recognition engine, a parser, and selectable "domain agents" to interpret a user's utterance, determine its context, and respond to the embedded request.

3. Grounds for Unpatentability

Ground 1: Obviousness over Coffman, Kanevsky, Kneser, and Zadrozny - Claims 1-2, 4-5 are obvious over Coffman in view of Kanevsky, Kneser, and Zadrozny.

• Prior Art Relied Upon: Coffman (International Publication No. WO 00/20962), Kanevsky (Patent 5,897,616), Kneser (Patent 6,157,912), and Zadrozny (a 1994 publication on natural language understanding).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that the primary reference, Coffman, discloses a conversational computing system that meets most limitations of independent claim 1. Coffman teaches receiving a speech utterance containing a request, parsing it to determine context, and selecting a "conversationally-aware application" (the claimed "domain agent") to process it. To supply missing details, Petitioner combined Coffman with secondary references. The combination with Kanevsky was argued to teach using n-gram language models, which constitute a "set of prior probabilities" for speech recognition. The further addition of Kneser was argued to render this set "dynamic" by teaching adaptive language models where probabilities are updated with use. Finally, Zadrozny, cited by Kanevsky as a semantic analyzer, was argued to teach using a "grammar" associated with an application to formulate a user's request into a format the application can execute, meeting the final limitations of claim 1.
  • Motivation to Combine: Petitioner contended a person of ordinary skill in the art (POSITA) would combine these references because they represent a collective body of work from the same field and, in many cases, the same assignee (IBM). Coffman expressly incorporates Kanevsky by reference. A POSITA implementing Coffman's system would have looked to well-known techniques, such as Kneser's adaptive models, to improve performance. Because Kanevsky cites Zadrozny for semantic analysis, a POSITA would have consulted Zadrozny for implementation details regarding grammar-based parsing.
  • Expectation of Success: A POSITA would have had a reasonable expectation of success, as the combination involved applying well-known and predictable technologies—such as n-gram language models, adaptive learning, and grammar-based parsing—to improve the functionality of Coffman's foundational system.

Ground 2: Obviousness over Ground 1 Art in view of Maes - Claim 2 is obvious over the combination of Coffman, Kanevsky, Kneser, and Zadrozny in further view of Maes.

• Prior Art Relied Upon: Coffman, Kanevsky, Kneser, Zadrozny, and Maes (International Publication No. WO 00/21075).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground focused on the limitation in dependent claim 2 requiring association of recognized words with user identity and context only "when the recognized words and phrases satisfy a predetermined confidence level." While the base combination discloses the association, Petitioner argued Maes teaches the confidence level aspect. Maes describes rejecting speech recognition results that do not meet a "given threshold" based on "confidence measures."
  • Motivation to Combine: Petitioner asserted that Coffman explicitly incorporates Maes by reference and they share common inventors and priority applications. While Coffman mentions "confidence and ambiguity results," it lacks implementation details. A POSITA would have been motivated to look to the incorporated Maes reference for its specific teachings on using confidence scores to improve the system's reliability and robustness, which was a stated goal in Coffman.
  • Expectation of Success: Using confidence thresholds in speech recognition was a conventional technique, and integrating Maes's specific teachings into the Coffman system would have been a straightforward and predictable improvement.

Ground 3: Obviousness over Ground 1 Art in view of Redfern - Claims 7-9 and 12 are obvious over the combination of Coffman, Kanevsky, Kneser, and Zadrozny in further view of Redfern.

• Prior Art Relied Upon: Coffman, Kanevsky, Kneser, Zadrozny, and Redfern (Patent 6,078,914).

• Core Argument for this Ground:

  • Prior Art Mapping: This ground targets claims related to processing a request with both a question and a command, and querying multiple information sources. Petitioner argued that Redfern teaches a meta-search application that would serve as the "invoked domain agent" in Coffman's system. Redfern discloses receiving a natural language query (the "question") and an explicit or implicit "command" to search, then submitting asynchronous queries to multiple, distinct search engines (e.g., Lycos, AltaVista). Redfern further teaches receiving the results, scoring their relevance, and extracting the best segments to present to the user, mapping to the limitations of claims 7-9 and 12.
  • Motivation to Combine: A POSITA would have been motivated to combine Redfern's meta-search engine with Coffman's conversational front-end to enable users to perform powerful, multi-source web searches via spoken commands. This would enhance the search capabilities already described in Coffman by leveraging a known, efficient method for querying multiple sources.
  • Expectation of Success: Integrating a meta-search backend as taught by Redfern with a speech-enabled interface like Coffman's was argued to be a predictable combination of existing technologies to achieve the known benefit of voice-controlled, comprehensive searching.

• Additional Grounds: Petitioner asserted additional obviousness challenges for claim 3 (based on adding Maes ’101 and Franceschi for phonetic spelling error correction) and claim 6 (based on adding Chai for parsing requests with required and optional parameters).

4. Arguments Regarding Discretionary Denial

• Petitioner argued against discretionary denial under §314(a) by stating that the co-pending district court litigation is in a very early stage, with no discovery, claim construction, or trial date set.
• Petitioner argued against discretionary denial under §325(d), acknowledging that a U.S. counterpart to the Coffman reference was before the Examiner during prosecution but contending it was never substantively discussed. Furthermore, Petitioner argued the Examiner did not have the benefit of the other secondary references, which provide the necessary context and teachings to render the claims obvious.

5. Relief Requested

• Petitioner requests institution of an inter partes review and cancellation of claims 1-9 and 12 of the ’209 patent as unpatentable.