Disfluency Identifier for Speech Parsing Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Natural language processing systems face difficulties in interpreting disfluent sentences due to filled pauses, parenthetical expressions, and speech repairs, which hinder accurate interpretation of spoken language.
Innovation Solution
A natural language processing system is developed with a disfluency identifier model that filters out edited words by analyzing speech repair distributions and using a boosting algorithm to optimize feature weights, improving parsing performance by ignoring disfluencies such as hesitations and repairs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a computer attempts to interpret all words in spoken instructions including disfluencies, then it captures complete speech content, but interpretation accuracy deteriorates due to filled pauses and speech repairs
Solution Approach 1:
The system segments speech into fluent and disfluent portions by identifying edited words and speech repairs. The parser then processes only the fluent segments, separating meaningful content from disfluencies to improve interpretation accuracy while retaining complete speech content for analysis.
Solution Approach 2:
The system extracts and removes edited words and disfluencies from the speech stream before parsing. By taking out problematic elements like filled pauses and speech repairs, the system prevents them from interfering with interpretation accuracy while preserving the original speech for reference.
2Productivity
If a computer deletes disfluencies from transcribed speech to improve parsing, then parsing performance improves, but information about speech patterns is lost
Solution Approach 1:
The system introduces an intermediary edited-word detector between speech transcription and parsing. This detector identifies disfluencies and marks them for separate handling, allowing the parser to focus on fluent content for improved performance while preserving disfluency information for speech pattern analysis.
Solution Approach 2:
The system segments the speech processing pipeline into distinct stages: transcription, edited-word detection, and parsing. Each stage handles specific aspects of the data, allowing parsing to operate on cleaned data for improved performance while maintaining access to original speech patterns through the segmented structure.
3Device complexity
If a simple edited-word detector is used to remove disfluencies, then system complexity is reduced, but identification accuracy of speech repairs deteriorates
Solution Approach 1:
The system uses dynamic programming to efficiently explore multiple parsing hypotheses and identify edited words. This dynamic approach allows the system to achieve high identification accuracy by considering multiple possibilities and selecting the most likely interpretation, while maintaining computational efficiency through optimized search.
Solution Approach 2:
The system changes parameters such as probability thresholds and feature weights in the edited-word detector to optimize performance. By adjusting these parameters based on training data and evaluation, the system achieves high identification accuracy without requiring excessively complex architecture.
Data Source
AI summary
An advanced model that includes new processes is provided for use as a component of an effective disfluency identifier. The disfluency identifier tags edited words in transcribed speech. A speech recognition unit in combination with a part-of-speech tagger, a disfluency identifier, and a parser form a natural language system that helps machines properly interpret spoken utterances.