Syntactic Parser Assisted Semantic Rule Inference for NLU Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Natural language understanding (NLU) engines perform suboptimally when trained with limited data, as they lack sufficient annotated samples to generalize effectively.
Innovation Solution
The method generates annotated data by parsing input phrases into syntactic trees and creating alternative versions using transformation rules, preserving alignment between expressions and annotations, which are then validated and used to enhance NLU engine training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If NLU engine is trained with limited annotated data, then training cost is reduced, but NLU engine performance deteriorates
Solution Approach 1:
The system performs preliminary syntactic parsing and transformation rule generation on a small set of annotated sentences before training the NLU engine. By pre-processing the limited data to generate multiple transformed variants, the system amplifies the effective training data volume without requiring additional manual annotation effort, thus maintaining high performance with limited annotated data
Solution Approach 2:
The system creates multiple copies and transformed versions of each annotated sentence through syntactic transformation rules. Each original annotated sentence generates several variant forms that preserve the semantic meaning while varying the syntactic structure, effectively multiplying the training data from a small set of annotated examples
2Reliability
If more annotated samples are collected to improve NLU performance, then NLU engine performance is improved, but annotation cost and time increase
Solution Approach 1:
Instead of manually creating multiple annotated samples, the system automatically generates multiple transformed copies from each annotated sentence using syntactic transformation rules. This copying approach creates diverse training examples without requiring proportional increases in manual annotation time
Solution Approach 2:
The system performs syntactic parsing and transformation rule application as preliminary actions on the annotated data before training. This pre-processing step automatically generates additional training variants, reducing the need for extensive manual annotation while ensuring the NLU engine receives sufficient diverse training examples
3Adaptability or versatility
If syntactic transformation rules are applied to generate alternative versions, then training data diversity is improved, but system complexity increases
Solution Approach 1:
The system segments the complexity by separating syntactic parsing, transformation rule generation, and data generation into distinct modular components. The syntactic parser analyzes sentence structure independently, transformation rules are generated based on parsed structures, and alternative versions are created by applying these rules - each step is independent and can be developed and maintained separately
Solution Approach 2:
The system generates its own transformation rules automatically based on the syntactic structure of the annotated sentences. Rather than requiring manual creation and curation of transformation rules by linguists, the system learns and generates appropriate transformation rules from the data itself, reducing the complexity burden of rule management
Data Source
AI summary
Natural language understanding (NLU) engines perform better when they are trained with large amounts of data. However, a large amount of data is not always available. Embodiments of the present invention overcome this problem by generating annotated data for use in a NLU system. An example embodiment generates annotated data by parsing an input annotated phrase, generating a syntactic tree reflecting a grammatical structure of the parsed phrase, and generating one or more alternative versions of the input annotated phrase based on the syntactic tree. Alignment between expressions and corresponding annotations in the annotated phrase are preserved in the one or more alternative versions generated to ensure intention of the input annotated phrase is maintained.