Named Entity Normalization for Spoken Dialog Input Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Intelligent automated assistants face errors in natural language input due to mispronunciation, transcription errors, and other inaccuracies, which affect task performance and user-device interaction efficiency.
Innovation Solution
The method involves determining a domain for the natural language input, using a named entity model to identify and correct errors, and generating a mapping to train the model for improved natural language processing, allowing for accurate task execution despite input errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If speech recognition is used to interpret user input, then natural language interaction is enabled, but errors in recognized input occur due to mispronunciation and transcription errors
Solution Approach 1:
The system generates feedback mappings by comparing recognized values against domain properties and expected value types. These mappings are used to train the speech recognition model iteratively, creating a feedback loop that continuously improves recognition accuracy based on actual performance data
Solution Approach 2:
The system performs preliminary domain determination and value type classification before final task execution. By pre-processing the recognized input to identify its expected format and constraints, the system can detect and correct errors before they affect task performance
2Reliability
If error correction mechanisms are added to improve input accuracy, then task performance improves, but device complexity increases
Solution Approach 1:
The system performs self-correction by automatically comparing recognized values against domain constraints and generating correction mappings without requiring external intervention. The model trains itself using the generated mappings, eliminating the need for manual error correction processes
Solution Approach 2:
The patent introduces an intermediary mapping generation process that bridges speech recognition and task execution. This intermediary layer translates recognized input into domain-compliant formats, isolating the complexity of error correction from both the recognition and execution components
3Measurement precision
If multiple value determination methods are used (direct recognition and named entity model), then accuracy improves, but processing time increases
Solution Approach 1:
The system applies partial verification by checking only critical domain constraints and value types rather than performing exhaustive validation. This selective approach maintains high accuracy for essential parameters while reducing overall processing time
Data Source
AI summary
Systems and processes for processing natural language input are described. An example process for processing natural language input includes receiving a natural language input and determining a domain corresponding to the natural language input. The example process further includes, in accordance with determining the domain corresponding to the natural language input, determining, based on the natural language input, a first value for a first property of the domain and determining, based on a named entity model and the natural language input, a second value for the first property of the domain, where the second value defines a parameter for a task corresponding to the natural language input. The example process further includes performing the task based on the parameter and providing a result based on the performed task.