Dialect LLC v. Meta Platforms Inc

I. Executive Summary and Procedural Information

• Parties & Counsel:
  • Plaintiff: Dialect, LLC (Texas)
  • Defendant: Meta Platforms, Inc. (Delaware)
  • Plaintiff’s Counsel: BLUE PEAK LAW GROUP Group
• Case Identification: 7:25-cv-00060, W.D. Tex., 02/07/2025
• Venue Allegations: Plaintiff alleges venue is proper in the Western District of Texas because Defendant maintains a regular and established physical place of business in Austin, Texas, and has committed acts of patent infringement within the District. The complaint highlights Meta's Austin office's involvement in augmented and virtual reality development related to the accused products.
• Core Dispute: Plaintiff alleges that Defendant’s Meta AI platform, Quest virtual reality headsets, and Ray-Ban Meta Glasses infringe five patents related to conversational voice recognition and natural language understanding (NLU) technology.
• Technical Context: The patents relate to foundational NLU systems that enable computers to understand and respond to human speech in a natural, context-aware manner, a core technology for modern digital assistants.
• Key Procedural History: The complaint notes that the asserted patents originate from VoiceBox Technologies Inc., and have been licensed to other companies. It also states that in October 2024, the Patent Trial and Appeal Board (PTAB) denied institution of inter partes review petitions filed by Google against two of the asserted patents, U.S. Patent Nos. 7,398,209 and 8,015,006.

Case Timeline

Date	Event
2002-06-03	Priority Date for all Asserted Patents
2008-07-08	U.S. Patent No. 7,398,209 Issues
2011-09-06	U.S. Patent No. 8,015,006 Issues
2013-05-21	U.S. Patent No. 8,447,607 Issues
2016-02-16	U.S. Patent No. 9,263,039 Issues
2017-08-15	U.S. Patent No. 9,734,825 Issues
2024-XX-XX	Accused Product "Meta AI" Introduced
2024-04-XX	Google files IPR petition against the ’209 Patent
2024-04-XX	Google files IPR petition against the ’006 Patent
2024-10-XX	PTAB denies institution of IPR on the ’209 Patent
2024-10-XX	PTAB denies institution of IPR on the ’006 Patent
2025-02-07	Complaint Filed

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 7,398,209 - Systems And Methods For Responding To Natural Language Speech Utterance

Issued July 8, 2008

The Invention Explained

• Problem Addressed: The patent’s background section states that prior art speech recognition systems were based on "highly structured" and "not inherently natural" machine-based queries, which were "fundamentally incompatible" with how humans naturally ask questions by relying on context and domain knowledge (’209 Patent, col. 1:27-35; Compl. ¶36).
• The Patented Solution: The invention proposes a system that uses context, prior information, domain knowledge, and user profiles to create a more natural conversational environment. A central element of this solution is the use of "domain agents," described as autonomous executable modules that organize domain-specific information and behavior, to process user requests (’209 Patent, col. 2:48-59; Compl. ¶36). The system parses a natural language query, determines the appropriate domain, invokes the correct agent, and formulates a machine-processable query for that agent (’209 Patent, col. 3:53-54).
• Technical Importance: This approach aimed to move beyond rigid "Command and Control" voice systems toward more flexible and intuitive human-computer interaction (Compl. ¶24).

Key Claims at a Glance

• The complaint asserts independent Claim 1 (Compl. ¶75).
• Essential elements of Claim 1 include:
  • receiving a user-generated natural language speech utterance containing a request;
  • maintaining a dynamic set of prior probabilities or fuzzy possibilities;
  • recognizing words and phrases using dictionary and phrase tables;
  • parsing the recognized words to determine a meaning and a context for the request;
  • selecting at least one domain agent based on the determined meaning, where the agent is an "autonomous executable";
  • formulating the request in accordance with a grammar used by the selected domain agent;
  • invoking the agent to process the request; and
  • presenting the results to the user.

U.S. Patent No. 8,015,006 - Systems And Methods For Processing Natural Language Speech Utterances With Context-Specific Domain Agents

Issued September 6, 2011

The Invention Explained

• Problem Addressed: The patent addresses the "difficult problem" of a machine's ability to communicate naturally with humans, noting that natural language queries are often "incomplete in their definition," "ambiguous or subjective," and difficult to translate into a machine-processable format (’006 Patent, col. 1:33-41, col. 9:11-21; Compl. ¶43).
• The Patented Solution: The invention claims a method that processes natural language by dynamically updating dictionaries based on probabilities, determining a user's identity from voice characteristics, and associating the recognized words with that identity (’006 Patent, Claim 1; Compl. ¶46). This user identification is described as useful for "security and other purposes" by allowing recognized words to be tagged with an identity in further processing (’006 Patent, col. 16:60-17:4; Compl. ¶45). The complaint references Figure 1 of the patent as showing an overall diagram of the system architecture (Compl. ¶47).
• Technical Importance: The claimed combination of context-based parsing, confidence-level checks, and user identification through voice characteristics was presented as a novel approach to overcoming barriers in NLU systems (Compl. ¶45, ¶49).

Key Claims at a Glance

• The complaint asserts at least Claim 5 (Compl. ¶95).
• Essential elements of Claim 5 include:
  • receiving a natural language speech utterance containing a request;
  • recognizing words or phrases using dictionary/phrase tables;
  • parsing the utterance to determine a meaning and context;
  • formulating the request according to a domain agent's grammar, which includes the sub-steps of:
    • determining required and optional values;
    • extracting criteria and parameters using context-sensitive procedures;
    • inferring further criteria and parameters using a dynamic set of prior probabilities; and
    • transforming the extracted and inferred criteria into tokens compatible with the grammar.
  • processing the request with the domain agent to generate a response.

U.S. Patent No. 8,447,607 - Mobile Systems And Methods Of Supporting Natural Language Human-Machine Interactions

Issued May 21, 2013

Technology Synopsis

This patent describes a system for processing "multi-modal" natural language inputs, which include both a natural language utterance and a non-speech input. The system generates a "merged transcription" from both inputs and uses a "cognitive model" based on prior user interactions to process the request (Compl. ¶53, ¶54).

Asserted Claims

Claim 12 (Compl. ¶115).

Accused Features

The complaint alleges infringement by Meta's products that receive and process both speech and non-speech user inputs (Compl. ¶115, ¶118).

U.S. Patent No. 9,263,039 - Systems And Methods For Responding To Natural Language Speech Utterance

Issued February 16, 2016

Technology Synopsis

This patent claims a method for processing both speech and non-speech communications by transcribing and merging them into a single query. The system then compares text combinations from the query to a "context description grammar" to generate a "relevance score," which is used to select one or more domain agents to generate a response (Compl. ¶57).

Asserted Claims

Claim 13 (Compl. ¶135).

Accused Features

The complaint accuses Meta's products of infringing by merging speech and non-speech communications and using a relevance scoring system to select domain agents or equivalent functionality (Compl. ¶135, ¶138).

U.S. Patent No. 9,734,825 - Methods and Apparatus for Determining a Domain Based on the Content and Context of a Natural Language Utterance

Issued August 15, 2017

Technology Synopsis

This patent describes a method for determining a domain for a user's utterance by first calculating a score for at least two possible contexts. The scores are based on keywords and "associated prior probabilities or fuzzy possibilities" received from system or domain agents. A domain is then determined based on the recognized words and the context scores (Compl. ¶64).

Asserted Claims

Claim 5 (Compl. ¶155).

Accused Features

The complaint alleges infringement by Meta products that determine scores for possible contexts of a user utterance to select a domain for processing the request (Compl. ¶155, ¶158).

III. The Accused Instrumentality

Product Identification

The accused instrumentalities are Defendant's "Meta AI" product, the Meta Quest line of headsets (Quest 2, 3, 3S, and Pro), and the Ray-Ban Meta Glasses (collectively, "Accused Products") (Compl. ¶7). Meta AI is described as being available through platforms including Facebook, Messenger, WhatsApp, and Instagram (Compl. ¶7, ¶68).

Functionality and Market Context

The complaint alleges that Meta AI is powered by Meta's Large Language Model, LLaMA, and is designed to assist users with daily tasks, information retrieval, and content generation (Compl. ¶67). Users are alleged to interact with Meta AI through voice commands, search bars, and chat interfaces (Compl. ¶68). The Quest headsets and Ray-Ban glasses are alleged to use AI for features like voice commands (Compl. ¶69). The complaint asserts that these products make significant use of natural language processing and understanding, which are central to the inventions of the Asserted Patents (Compl. ¶71). The complaint includes a screenshot of an early VoiceBox desktop application which is described as enabling multi-modal user interactions by accepting both speech and non-speech user input (Compl. ¶26, p. 7).

IV. Analysis of Infringement Allegations

The complaint does not include the referenced claim chart appendices. The following is a summary of the narrative infringement theory for the lead patents.

’209 Patent Infringement Allegations

The complaint alleges that the Accused Products practice the method of Claim 1 by receiving natural language speech from a user (e.g., a voice command to Meta AI); processing it using NLU technologies that determine meaning and context; selecting and invoking the appropriate back-end processes (the alleged "domain agents") to handle the request; and presenting results (e.g., answering a question or generating an image) (Compl. ¶75, ¶78). A central point of contention may be whether Meta's LLM-based architecture, which may process queries in an integrated manner, meets the claim limitation of selecting and invoking an "autonomous executable" domain agent associated with a determined context.

’006 Patent Infringement Allegations

The complaint alleges that the Accused Products practice the method of Claim 5 by receiving a user's speech utterance and using NLU to parse it and determine a context (Compl. ¶95, ¶98). The infringement theory centers on the allegation that this process involves inferring criteria and parameters using probabilistic models (a function inherent to LLMs) and transforming the user's request into a structured format ("tokens") compatible with a back-end grammar before processing it with a "domain agent" and generating a response (Compl. ¶97). The dispute may focus on whether the internal data processing of Meta's AI maps to the specific multi-step "formulating" process required by the claim.

V. Key Claim Terms for Construction

The Term: "domain agent"

(e.g., ’209 Patent, Claim 1)

Context and Importance

This term is central to the patented architecture and appears in the independent claims of multiple asserted patents. Its construction will be critical to determining whether Meta's modern, LLM-based AI system, which may handle diverse queries in a unified manner, infringes claims that recite a system of distinct, domain-specific agents.

Intrinsic Evidence for Interpretation

• Evidence for a Broader Interpretation: The specification describes agents as "complete, convenient and re-distributable packages or modules of functionality, typically for a specific domain or application," which could support a broader reading on any software component that handles a specific topic (’209 Patent, col. 2:50-54).
• Evidence for a Narrower Interpretation: The claim language itself requires the agent to be an "autonomous executable that receives, processes, and responds to requests." A defendant may argue this requires a structurally separate and independent software program, potentially distinct from an integrated function within a large language model.

The Term: "a grammar used by the selected domain agent"

(e.g., ’209 Patent, Claim 1)

Context and Importance

This term is key to how the user's natural language request is translated into a machine-processable format. Practitioners may focus on this term because the internal operating logic of Meta's LLaMA-based system may not utilize a pre-defined, agent-specific "grammar" in the way envisioned by the patent, potentially creating a point of non-infringement.

Intrinsic Evidence for Interpretation

• Evidence for a Broader Interpretation: The patent does not appear to provide a specific definition of "grammar," which a plaintiff might argue should be given its ordinary meaning in computer science, potentially covering any set of rules or structures used to process a query.
• Evidence for a Narrower Interpretation: A defendant may argue that the patent's context, which describes parsing a query and formulating it "in accordance with a grammar," implies a more structured, rule-based system that is technically distinct from the probabilistic, pattern-matching approach of a modern LLM.

VI. Other Allegations

Indirect Infringement

The complaint alleges inducement of infringement, stating that Meta provides websites, instructions, and marketing materials that encourage and instruct consumers to use the Accused Products in an infringing manner (Compl. ¶79-84, ¶101-104). It cites several URLs to Meta's and its affiliates' websites as examples of these instructions (Compl. ¶84, ¶104).

Willful Infringement

Willfulness is alleged based on Defendant’s knowledge of the patents "at least as of the filing of the Complaint" (Compl. ¶88, ¶108). The complaint also includes an alternative allegation of willful blindness but does not plead specific facts indicating pre-suit knowledge of the patents or infringement.

VII. Analyst’s Conclusion: Key Questions for the Case

• A core issue will be one of architectural equivalence: can the "domain agent" architecture, described in the 2002-priority-date patents as distinct "autonomous executable" modules for specific topics, be proven to read on the functionality of Meta's modern, integrated Large Language Model-based AI system? The case may turn on whether these two technological approaches are deemed equivalent under the doctrine of claim construction.
• A key evidentiary question will be one of technical mapping: beyond general allegations, what evidence can the plaintiff obtain through discovery to demonstrate that the internal software operations of the Accused Products perform the specific, multi-step processes recited in the claims (e.g., "formulating the... request in accordance with a grammar used by the selected domain agent"), as opposed to merely achieving a similar NLU result via a fundamentally different technical pathway?