Scalable Statistical Language Understanding System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Natural Language Understanding (NLU) systems face challenges in scaling to support multiple domains and intents, requiring the addition of numerous new models for each new domain, leading to increased complexity and training data requirements.
Innovation Solution
A scalable statistical language understanding system uses a fixed number of models that are updated to incorporate new domains, reusing existing information and training data, including a domain detector, intent action detector, intent object detector, and slot/entity tagging model, allowing for detection of multiple domains and intents within a single utterance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a new domain is added to the NLU system, then the system's domain coverage is improved, but the number of models increases
Solution Approach 1:
The patent applies universality by creating a single intent model that can handle multiple domains through domain-agnostic intent detection. The intent model is trained to recognize intent actions across different domains (e.g., finding movies, booking flights, ordering food) without requiring separate intent models for each domain, thereby reducing overall system complexity while maintaining broad domain coverage
Solution Approach 2:
The patent segments the NLU system into distinct functional components: a domain detector that identifies the domain from user input, an intent model that detects intent actions independently of domain, and a slot model that extracts entities. This segmentation allows each component to be optimized independently and reduces the need for creating entirely new models when adding domains
2Adaptability or versatility
If a new domain is added to the NLU system, then the system's domain coverage is improved, but the training data requirements increase
Solution Approach 1:
The intent model serves as a universal component that can detect intent actions across multiple domains using a single training corpus. By training the intent model on diverse data from different domains simultaneously, the system avoids the need to collect and train separate intent models for each new domain, thereby reducing overall training data requirements while expanding domain coverage
Solution Approach 2:
The patent recovers and reuses existing training data from previous domains to train the intent model for new domains. Instead of discarding old training data and starting fresh, the system leverages the universal intent detection capabilities learned from previous domains, reducing the amount of new training data needed for each additional domain
Data Source
AI summary
A scalable statistical language understanding (SLU) system uses a fixed number of understanding models that scale across domains and intents (i.e. single vs. multiple intents per utterance). For each domain added to the SLU system, the fixed number of existing models is updated to reflect the newly added domain. Information that is already included in the existing models and the corresponding training data may be re-used. The fixed models may include a domain detector model, an intent action detector model, an intent object detector model and a slot/entity tagging model. A domain detector identifies different domains identified within an utterance. All/portion of the detected domains are used to determine associated intent actions. For each determined intent action, one or more intent objects are identified. Slot/entity tagging is performed using the determined domains, intent actions, and intent object detector.