Microsoft Corp v. Dialect LLC

1. Case Identification

• Case #: IPR2025-00656
• Patent #: 8,447,607
• Filed: March 7, 2025
• Petitioner(s): Microsoft Corporation
• Patent Owner(s): Dialect, LLC
• Challenged Claims: 12-13

2. Patent Overview

• Title: Interactive Natural Language Processing System
• Brief Description: The ’607 patent discloses systems and methods for mobile devices that use a combination of speech and non-speech interfaces to enable natural language human-machine interactions. The technology is designed to process multi-modal inputs, such as a user's speech and concurrent non-speech actions, to understand and respond to user commands across various domains.

3. Grounds for Unpatentability

Ground 1: Claims 12 and 13 are obvious over Maes.

• Prior Art Relied Upon: Maes (Patent 6,964,023), which Petitioner argued expressly incorporates by reference Coffman (International Publication No. WO 00/20962) and Ittycheriah (Patent 5,937,383).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner asserted that Maes, when read to include its incorporated references, disclosed all limitations of the challenged claims. Maes described a multi-modal data processing system that receives both audio (natural language utterance) and video (non-speech input) signals. Petitioner argued Maes taught generating a non-speech transcription by extracting visual speech features (visemes) from video input and a speech-based transcription by processing audio signals to produce decoded text. Maes allegedly disclosed using user-specific models based on prior interactions (a "cognitive model"), as taught by the incorporated Ittycheriah reference, which described a user-specific cache database of previously decoded words. Petitioner contended that Maes taught generating a "merged transcription" by aligning the decoded text script with the viseme sequence. Maes further described using a "context stack" to store historical event data and searching this stack to determine user intent based on the merged transcription, thereby identifying an entry matching the merged information. Finally, Petitioner argued that Maes's "application programs" or "dialog manager," which determines the target application to handle user input, met the "domain agent" limitation.
  • Motivation to Combine (for §103 grounds): The primary argument for this ground was that Coffman and Ittycheriah were effectively part of Maes due to express incorporation by reference. Maes stated that Coffman could be employed to provide a framework for its I/O manager, recognition engines, and context stack, and that Ittycheriah disclosed methods for detecting incorrectly recognized words.
  • Expectation of Success (for §103 grounds): Petitioner argued a POSITA would have had a reasonable expectation of success because combining the teachings was a simple integration of analogous systems to add functionality, improve accuracy, and enhance system responsiveness, consistent with Maes's own stated goals.

Ground 2: Claims 12 and 13 are obvious over Maes in view of Coffman and Ittycheriah.

• Prior Art Relied Upon: Maes (Patent 6,964,023), Coffman (WO 00/20962), and Ittycheriah (Patent 5,937,383).
• Core Argument for this Ground: This ground was presented as an alternative, asserting that even if the references were not considered incorporated, a POSITA would have been motivated to combine them.
  • Prior Art Mapping: The mapping of claim limitations to the combined teachings of the references was substantially the same as in Ground 1. Maes provided the foundational multi-modal system. Ittycheriah was argued to provide the use of a user-specific cache database storing keywords and scores from prior interactions, fulfilling the "cognitive model" limitations. Coffman was cited for its teachings on searching a context stack to find entries that match transcribed user input and for merging speech and non-speech (e.g., keyboard) transcriptions.
  • Motivation to Combine (for §103 grounds): Petitioner asserted multiple motivations to combine. All three references shared a common inventor (Stephane Maes) and a common assignee (IBM), and Maes expressly cross-referenced the other two. A POSITA would have been motivated to integrate Ittycheriah's user-specific cache ("cognitive model") into Maes's system to improve the speed and accuracy of its decoding process. Similarly, a POSITA would have combined Coffman's teachings on organizing and searching a context stack to provide more efficient retrieval of the most recent context, improving the responsiveness of the Maes system.
  • Expectation of Success (for §103 grounds): Petitioner contended that a POSITA would have expected success in combining these analogous natural language processing systems, as it would have been a predictable implementation to improve system accuracy and functionality.

4. Arguments Regarding Discretionary Denial

• Petitioner argued against discretionary denial under §325(d), asserting that the primary reference, Maes, was never cited or considered by the examiner during prosecution. Although a "relative" of Coffman was listed on an IDS, it was not substantively evaluated. Therefore, Petitioner contended the examiner materially erred by not considering these highly relevant disclosures.
• Petitioner also argued against discretionary denial under §314(a) based on the Fintiv factors. Petitioner argued the factors weighed against denial because the trial date in the parallel district court litigation involving a real party-in-interest was still distant and likely to slip, the investment in the court case was minimal, and the parties stipulated to not pursue the same grounds in district court if the IPR was instituted.

5. Relief Requested

• Petitioner requested institution of an inter partes review and cancellation of claims 12 and 13 of Patent 8,447,607 as unpatentable.