Derivative works

Defensive Disclosure and Prior Art Derivations for U.S. Patent 9,269,097

Document ID: DPD-2026-0510-001
Publication Date: May 10, 2026
Subject: Derivative Implementations and Obvious Variations of Natural Language Processing Based on Contextual Advertisements

This document describes variations, extensions, and alternative implementations of the system and method disclosed in U.S. Patent 9,269,097. The purpose is to place these concepts into the public domain, thereby establishing prior art against future patent applications claiming these or similar incremental improvements. The core concept involves interpreting a user's natural language utterance, particularly pronouns, by referencing a previously presented advertisement as context.

---

Analysis of Core Claims

The fundamental principle of US 9,269,097, as articulated in independent claims 1 and 23, is a computer system that:

1. Provides an advertisement (the context).
2. Receives a subsequent natural language utterance from a user.
3. Interprets the utterance by determining if a pronoun within it refers to the product, service, or provider from the advertisement.

The following disclosures build upon this foundation.

---

Derivative Disclosures

1. Material & Component Substitution

Derivative 1.1: Neuromorphic Processor Implementation

• Enabling Description: The method of claim 1 is implemented on a system utilizing a neuromorphic processing unit (NPU), such as Intel's Loihi 2 or a similar spiking neural network (SNN) architecture, instead of a traditional von Neumann CPU/GPU. The advertisement's metadata (product ID, service type, provider name) is encoded as a static or slowly changing neural map within the SNN. When the user's utterance is received, the speech-to-text output is tokenized and fed into the SNN as a series of spikes. The pronoun token triggers a search for the most recently activated and strongly weighted entity in the neural map. The NPU's inherent parallelism and event-driven processing allow for extremely low-latency pronoun resolution, making it suitable for real-time, power-constrained edge devices. The system's memory can be phase-change memory (PCM) to retain the context map even in low-power states.

• Mermaid Diagram:

  graph TD
      A[Audio Input: "Buy it"] -- STT --> B{Tokenized Input: ["Buy", "it"]};
      C[Ad Context: {Product: 'X', Provider: 'Y'}] -- Encoded --> D[Neuromorphic Core];
      subgraph D [Neuromorphic Core]
          D1[Neural Map: 'X' & 'Y' Nodes]
      end
      B -- Spikes --> D;
      D -- Pattern Match --> E{Pronoun "it" correlates with highly activated 'X' Node};
      E --> F[Generate Request: purchase(product='X')];
  

Derivative 1.2: Federated Learning for Context Models

• Enabling Description: The system of claim 23 is implemented as a decentralized, federated learning architecture. Instead of a central server selecting advertisements and interpreting utterances, individual user devices (clients) maintain their own context models. An on-device model, trained using TensorFlow Lite or Core ML, processes the advertisement and the utterance locally. To improve the model without compromising privacy, only the model updates (gradients), not the raw data, are sent to a central aggregation server. This allows the global model to learn common pronoun-antecedent patterns (e.g., "call them" usually refers to the provider) across a population, while the specific context (the ad and the utterance) never leaves the user's device. This substitutes the central server architecture with a distributed, privacy-preserving computational model.

• Mermaid Diagram:

  sequenceDiagram
      participant ClientDevice
      participant AggregationServer
      ClientDevice->>ClientDevice: Presents Ad, receives utterance "Call them"
      ClientDevice->>ClientDevice: On-device model resolves "them" to Ad Provider
      ClientDevice->>ClientDevice: Computes model gradient from this interaction
      ClientDevice->>AggregationServer: Send encrypted model gradient
      AggregationServer->>AggregationServer: Aggregates gradients from many clients
      AggregationServer-->>ClientDevice: Sends updated global model weights
  

2. Operational Parameter Expansion

Derivative 2.1: High-Frequency Trading (HFT) Ad Auction and Interpretation

• Enabling Description: The invention is applied to an HFT environment where ad slots are auctioned and user commands are executed in microseconds. The "advertisement" is a fleeting data feed of a stock price or a market opportunity presented to a trader. The trader's subsequent voice command, e.g., "sell half of it," must be interpreted at sub-millisecond speeds. The system operates on a field-programmable gate array (FPGA) co-located with the trading exchange's servers. The context (the stock symbol) is held in the FPGA's block RAM. The speech-to-text and NLP processes are hardware-accelerated logic circuits. This scales the operation from typical human-interaction speeds to the extreme frequencies and low latencies required for algorithmic trading.

• Mermaid Diagram:

  graph LR
      A[Market Data Feed: {Symbol: XYZ, Price: 100.50}] -- Pushed to --> B(FPGA Block RAM);
      C[Trader Utterance: "Sell half of it"] -- A/D & STT --> D{FPGA Logic};
      B -- Context --> D;
      D -- Pronoun Resolution --> E{Resolved Command: sell(symbol='XYZ', qty=0.5*position)};
      E -- Executed in <1ms --> F[Trade Execution Engine];
  

Derivative 2.2: Nanoscale Molecular Assembly Control

• Enabling Description: The concept is scaled down to control nanoscale robotic assemblers. The "advertisement" is a visual feed from an atomic force microscope displaying a specific molecule or binding site. The operator issues a natural language command, such as "apply the catalyst to it." The system, running on a quantum-annealing processor, interprets "it" as the specific molecular coordinate highlighted in the microscope feed. The processor resolves the ambiguity by analyzing the quantum state representing the visual context and the utterance, selecting the most probable antecedent. The output is not a web request but a series of precise voltage changes to manipulate piezoelectric nano-actuators.

• Mermaid Diagram:

  stateDiagram-v2
      [*] --> DisplayingContext
      DisplayingContext: AFM shows Molecule_A at [x,y,z]
      DisplayingContext --> WaitingForInput: Operator sees context
      WaitingForInput --> Processing: Receives command "Activate it"
      Processing --> Action: System resolves "it" to Molecule_A
      Action --> [*]: Sends signal to nano-actuator
  

3. Cross-Domain Application

Derivative 3.1: Aerospace - Cockpit Fault Response

• Enabling Description: In an aircraft cockpit, the "advertisement" is a specific warning from the Engine Indicating and Crew Alerting System (EICAS), such as "HYDRAULIC PUMP 1B FAIL." A pilot's subsequent voice command, "run the checklist for it," is interpreted by the avionics system. The system uses the EICAS message as the context to resolve "it" to the "Hydraulic Pump 1B Failure" event. It then automatically displays the correct non-normal checklist on the multi-function display, bypassing the need for manual lookup under high-stress conditions.

• Mermaid Diagram:

  flowchart TD
      A[EICAS Alert: "HYDRAULIC PUMP 1B FAIL"] --> B{Avionics Voice Control System};
      C[Pilot Utterance: "Run the checklist for it"] --> B;
      B -- Resolves "it" against EICAS context --> D[Identify Checklist: "QRH HYD 1B PUMP FAIL"];
      D --> E[Display checklist on MFD];
  

Derivative 3.2: AgTech - Precision Irrigation and Pest Control

• Enabling Description: An agricultural drone surveying a field identifies a specific patch of crops suffering from dehydration or pest infestation. This visual data, with GPS coordinates, is presented to a remote farm operator as a "context alert." The operator issues the command, "dispatch a sprayer to it." The farm management system interprets "it" as the specific GPS-tagged patch of crops. It then generates a mission plan for an autonomous ground vehicle to navigate to the coordinates and apply the appropriate amount of water or pesticide, directly linking a high-level natural language command to a specific, geolocated agricultural action.

• Mermaid Diagram:

  sequenceDiagram
      participant Drone
      participant Operator
      participant FarmMgmtSystem
      participant SprayerBot
      Drone->>Operator: Pushes Alert: "Pest Infestation at [Lat, Lon]"
      Operator->>FarmMgmtSystem: Voice Command: "Dispatch a sprayer to it"
      FarmMgmtSystem->>FarmMgmtSystem: Resolve "it" to [Lat, Lon]
      FarmMgmtSystem->>SprayerBot: Generate & send mission plan
      SprayerBot->>SprayerBot: Navigate and execute spraying
  

Derivative 3.3: Consumer Electronics - Smart Home Ambiance Control

• Enabling Description: A smart home hub displays an interactive advertisement for a movie on a connected TV. The ad includes metadata about the movie's genre (e.g., "Horror"). After the ad plays, the user says to a nearby smart speaker, "set the lights for it." The smart home hub uses the advertisement's genre metadata as context to resolve "it." It then executes a pre-defined lighting scene named "Horror Movie," which dims the lights, lowers the color temperature, and perhaps closes the smart blinds, creating an immersive experience without the user needing to specify the exact scene name.

• Mermaid Diagram:

  graph TD
      A[TV displays Movie Ad with Genre='Horror' metadata] --> B((Smart Home Hub));
      C[Smart Speaker receives "Set the lights for it"] --> B;
      B -- Resolves "it" using genre context --> D{Action: Execute "Horror Movie" lighting scene};
      D --> E[Control Philips Hue Lights];
      D --> F[Control Lutron Blinds];
  

4. Integration with Emerging Tech

Derivative 4.1: AI-Driven Predictive Intent

• Enabling Description: The system is integrated with a predictive AI model that analyzes user behavior (gaze tracking, ad dwell time, historical interactions). Before the user even speaks, the system pre-computes a probability distribution of likely pronoun referents. For an advertisement for a car, the AI may determine a 70% probability that "it" will refer to the car, 20% to the dealership (provider), and 10% to the financing offer. This allows the system to pre-fetch information or pre-load application logic. When the utterance "what's the price of it" is received, the system can instantly resolve "it" to the car and provide the answer with minimal latency, because the required data was already cached based on the AI's prediction.

• Mermaid Diagram:

  flowchart TD
      subgraph Pre-Utterance
          A[User views Car Ad] -- Gaze data, history --> B(Predictive AI Model)
          B -- Generates --> C(Intent Probabilities <br> P(it=Car)=0.7 <br> P(it=Dealer)=0.2)
          C -- Triggers --> D[Pre-fetch Car Price Data]
      end
      subgraph Post-Utterance
          E[User asks "What's the price of it?"] --> F{NLP Engine}
          C -- Feeds into --> F
          F -- Resolves "it" to Car --> G[Retrieve price from cache]
          G --> H[Synthesize & speak answer]
      end
  

Derivative 4.2: IoT Sensor Fusion for Contextual Disambiguation

• Enabling Description: The advertisement context is augmented with real-time data from IoT sensors. An ad for a smart thermostat is displayed on a user's phone. Simultaneously, an IoT temperature sensor in the room reports a high ambient temperature. The user says, "turn it down." A simple system might be confused between the product (thermostat) and the service (HVAC). This integrated system fuses the ad context with the IoT data. It infers that the user's command is motivated by the physical environment and correctly resolves "it" to the HVAC service controlled by the thermostat, not just the thermostat device itself, and initiates an action to lower the temperature.

• Mermaid Diagram:

  erDiagram
      USER {
          string utterance
      }
      ADVERTISEMENT {
          string product "Thermostat"
          string service "HVAC Control"
      }
      IOT_SENSOR {
          int temperature "80F"
      }
      NLP_ENGINE {
      }
      USER ||--o{ NLP_ENGINE : provides
      ADVERTISEMENT ||--|{ NLP_ENGINE : provides_context
      IOT_SENSOR ||--|{ NLP_ENGINE : provides_context
      NLP_ENGINE }o--|| ACTION : "generates"
  

Derivative 4.3: Blockchain for Verifiable Ad-Based Transactions

• Enabling Description: For high-value transactions initiated via voice command following an ad, the system uses a blockchain for verification. An advertisement for a limited-edition digital art NFT is displayed. The user says, "I'll buy it." The system resolves "it" to the specific NFT. The action generated is a smart contract transaction. The transaction payload immutably records a cryptographic hash of the advertisement content, the resolved user command, and the user's biometric voiceprint signature. This creates an unalterable, auditable record on a public ledger (e.g., Ethereum), proving that the purchase was initiated in direct response to a specific advertisement, preventing disputes about the terms or subject of the sale.

• Mermaid Diagram:

  sequenceDiagram
      participant UserDevice
      participant NLP_Service
      participant Blockchain
      UserDevice->>UserDevice: Display NFT Ad (Ad_Hash)
      UserDevice->>NLP_Service: Utterance: "I'll buy it" + Voiceprint
      NLP_Service->>NLP_Service: Resolve "it" to NFT_ID from Ad_Hash
      NLP_Service->>Blockchain: Create Transaction({ad: Ad_Hash, command: "buy", item: NFT_ID, signature: Voiceprint})
      Blockchain->>Blockchain: Execute Smart Contract (Transfer NFT & Funds)
      Blockchain-->>UserDevice: Return Transaction Confirmation
  

5. The "Inverse" or Failure Mode

Derivative 5.1: Graceful Degradation to Disambiguation Dialog

• Enabling Description: The system is designed to fail safely when pronoun resolution confidence is low. An advertisement shows both a product (a smart watch) and a service (a fitness subscription). The user says, "tell me more about it." The system's model computes a low confidence score for the antecedent of "it" (e.g., 55% product, 45% service). Instead of guessing and providing incorrect information, it enters a "limited functionality" mode. It responds with a clarifying question: "Do you mean the smart watch or the fitness subscription?" The system's state machine transitions from INTERPRET to DISAMBIGUATE. This ensures that in ambiguous cases, the system fails to a safe, interactive state rather than executing an incorrect action.

• Mermaid Diagram:

  stateDiagram-v2
      state "Resolve Pronoun" as Resolve
      [*] --> Receiving
      Receiving --> Resolve: Utterance received
      Resolve --> Action: Confidence > 90%
      Resolve --> Disambiguate: Confidence <= 90%
      Disambiguate --> Resolve: User clarifies
      Action --> [*]
      state Disambiguate {
          [*] --> Ask: "Do you mean the product or the service?"
          Ask --> UserResponse
      }
  

---

Combination Prior Art with Open-Source Standards

Combination 3.1: W3C SRGS (Speech Recognition Grammar Specification)

• Description: The system of US 9,269,097 is combined with the W3C's SRGS standard. After an advertisement for a product is shown, the system dynamically generates and activates an SRGS XML grammar. This grammar includes a specific rule for the advertised product, such as <item>Smart Speaker</item>. A generic rule for pronouns like <ruleref uri="#pronoun"/> is linked to this item. When the user says "buy it," the speech recognizer uses this active grammar to constrain the recognition, improving accuracy. The parser's output explicitly tags the pronoun's antecedent based on the grammar structure, simplifying the NLP task to a deterministic lookup rather than a probabilistic inference. This combines the patent's method with a well-established open standard for guided speech recognition.

Combination 3.2: Integration with Hugging Face Transformers

• Description: The pronoun resolution module is implemented using a pre-trained language model from the open-source Hugging Face Transformers library, such as BERT (Bidirectional Encoder Representations from Transformers). The text of the advertisement and the user's subsequent utterance are concatenated into a single input sequence, formatted as: [CLS] ad text [SEP] utterance text [SEP]. This sequence is fed into the BERT model. The model's attention mechanism is inherently trained to understand relationships between tokens. By fine-tuning the model on a dataset of ad/utterance pairs, the system can learn to resolve the pronoun in the utterance to the correct entity in the ad text with high accuracy, leveraging a powerful, open-source NLP foundation.

Combination 3.3: OAuth 2.0 and OpenID Connect for Action Authorization

• Description: When the user's command, resolved in the context of an advertisement, requires an action on a third-party service (e.g., "book it" for a hotel advertised), the system uses the open standards OAuth 2.0 and OpenID Connect to perform the action securely. The system, acting as an OAuth 2.0 client, requests an access token from the third-party provider (e.g., Marriott's booking API). The user authenticates using OpenID Connect. The resulting action (the booking) is executed using an access token with a narrowly defined scope (e.g., create_booking), ensuring the system only has permission to perform the requested action and cannot access other user data. This combines the patented NLP method with open standards for secure, delegated authorization.