Natural Language Understanding Processing with Context-Aware FSTs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech processing systems experience increased latency due to domain negotiations when multiple domains need to interpret similar user inputs, as they require separate finite state transducers (FSTs) and statistical models to determine the correct action, leading to inefficient processing.
Innovation Solution
The system pre-configures FSTs with context information to differentiate domain operations, allowing the same user input to trigger different functionalities based on context, and processes user inputs using both FSTs and statistical models regardless of output, with a reranker determining the most likely interpretation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate finite state transducers (FSTs) and statistical models are used for each domain, then domain-specific accuracy is improved, but processing latency increases due to domain negotiations
Solution Approach 1:
The system pre-configures FSTs with context information about multiple domains before processing user inputs. This preliminary setup allows the FSTs to be ready to differentiate domain operations immediately when a user input is received, eliminating the need for time-consuming domain negotiations during actual processing.
Solution Approach 2:
The patent introduces a reranker as an intermediary component that receives outputs from both FSTs and statistical models, evaluates them, and determines the most likely interpretation. This mediator coordinates between multiple domain-specific processors, enabling parallel processing without requiring sequential domain negotiations.
2Adaptability or versatility
If multiple domains are supported to increase system versatility, then adaptability improves, but device complexity increases due to multiple FSTs and models
Solution Approach 1:
The system employs a universal reranker component that can handle outputs from multiple domain-specific FSTs and statistical models. This single intermediary serves all domains, coordinating their outputs and determining the most likely interpretation across different domains without requiring separate coordination mechanisms for each domain pair.
Solution Approach 2:
The patent adds a new dimension to the processing architecture by introducing the reranker layer above the parallel FSTs and statistical models. This dimensional addition allows the system to manage multiple domains through a unified evaluation framework rather than through complex inter-domain negotiation protocols.
Data Source
AI summary
Techniques for processing a user input are described. Text data representing a user input is processed with respect to at least one finite state transducer (FST) to generate at least one FST hypothesis. Context information may be required to traverse one or more paths of the at least one FST. The text data is also processed using at least one statistical model (e.g., perform intent classification, named entity recognition, and/or domain classification processing) to generate at least one statistical model hypothesis. The at least one FST hypothesis and the at least one statistical model hypothesis are input to a reranker that determines a most likely interpretation of the user input.