Semantic Syntactic Graphs for Utterance Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current spoken dialog systems face high classification error rates due to data sparseness and the variability of expressing the same intent in short utterances, requiring extensive labeled data for reasonable performance.
Innovation Solution
Incorporating lexical, semantic, and syntactic information into a graph for each utterance, extracting n-grams from these graphs, and using them for classification to improve the accuracy of intent recognition in spoken language understanding systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If only lexical (word-based) information is used for classification, then the system is simple to implement, but classification accuracy deteriorates due to data sparseness
Solution Approach 1:
The patent combines lexical, semantic, and syntactic information into a unified graph representation for each utterance. This merging of multiple information types creates a richer feature set that improves classification accuracy while maintaining system manageability through integrated processing.
Solution Approach 2:
The patent transitions from one-dimensional lexical features to multi-dimensional features by incorporating semantic roles and syntactic structures. This dimensional expansion allows the classifier to capture relationships between words and their meanings, resolving the data sparseness problem without excessive complexity.
2Reliability
If extensive labeled data is collected to train the classification system, then classification accuracy improves, but the loss of time and resources for data collection increases
Solution Approach 1:
The patent performs preliminary action by constructing semantic and syntactic graphs that enrich the feature representation of each utterance. This preprocessing transforms limited labeled data into more informative representations, allowing the system to achieve better accuracy with less training data required.
Solution Approach 2:
The patent changes the parameters of the feature representation by incorporating semantic roles and syntactic structures. This transformation increases the information content of each feature, enabling the classifier to learn more effectively from smaller datasets and reducing the time needed for data collection.
3Reliability
If semantic and syntactic information is incorporated into graphs for each utterance, then classification accuracy improves, but device complexity increases
Solution Approach 1:
The patent segments the complex task of utterance classification into distinct components: lexical analysis, semantic role labeling, and syntactic parsing. Each component processes specific aspects of the utterance independently, then their results are integrated into a unified graph representation, managing complexity through modular organization.
4Reliability
If n-grams are extracted from semantic and syntactic graphs, then the feature set becomes richer, but the processing time increases
Solution Approach 1:
The patent extracts n-gram features specifically from the semantic and syntactic graphs rather than processing the entire utterance text. This selective extraction focuses computational resources on the most informative portions of the data, enriching the feature set while minimizing unnecessary processing time.
Data Source
AI summary
Disclosed herein is a system, method and computer readable medium storing instructions related to semantic and syntactic information in a language understanding system. The method embodiment of the invention is a method for classifying utterances during a natural language dialog between a human and a computing device. The method comprises receiving a user utterance; generating a semantic and syntactic graph associated with the received utterance, extracting all n-grams as features from the generated semantic and syntactic graph and classifying the utterance. Classifying the utterance may be performed any number of ways such as using the extracted n-grams, a syntactic and semantic graphs or writing rules.