Prompt Model Canonical Form Generation for Dialogue Intent

The patent introduces a prompt model as an intermediary component between the user input and the language model. This prompt model generates specialized prompts that guide the language model to produce canonical forms without requiring extensive training data. The prompt model acts as a mediator that translates diverse user utterances into structured queries that the language model can process efficiently, thereby reducing both development complexity and computational resource requirements while maintaining versatility.

The system segments the dialogue processing task into distinct components: the prompt model that handles intent classification and canonical form generation, and the language model that processes the prompted queries. This segmentation allows each component to specialize in specific functions, reducing the overall complexity compared to training a single large model to handle all aspects of dialogue understanding.

The language model, being a large pre-trained model, possesses inherent knowledge of language patterns and semantics from its extensive pre-training. The prompt model leverages this self-service capability of the language model by crafting prompts that activate the model's existing knowledge to handle novel intents without requiring retraining. The system uses the language model's own pre-learned capabilities to adapt to new intents dynamically.

Instead of changing the model weights through fine-tuning (which requires extensive data), the system changes the input parameters (prompts) to the language model. By modifying the prompt structure and content based on the detected intent, the system can guide the language model to produce appropriate canonical forms for both seen and unseen intents, achieving adaptability without additional training data.

The patent replaces the mechanical approach of manually defining detailed rule-based grammar with a learning-based approach using a prompt model. Instead of manually creating complex grammatical rules to extract intents and slots, the prompt model automatically learns to generate prompts that elicit structured canonical forms from the language model, significantly simplifying the development process while maintaining output quality.

The system changes from a static rule-based grammar approach to a dynamic prompt-based approach. The prompt model generates different prompts based on the input context, allowing the language model to adapt its output structure dynamically. This parameter change from fixed rules to flexible prompts makes the system easier to develop while preserving the structured output needed for dialogue management.