Domain-Specific Question Generation for Syntactic Parsing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing natural language parsing technologies face challenges in accurately processing domain-specific questions due to underrepresentation in general-purpose corpora, leading to poor parsing performance, especially in specialized domains like finance and genomics.
Innovation Solution
A method is developed to generate a large training corpus for syntactic parsing of domain-specific questions by leveraging a domain schema and knowledge base, constructing question templates, and filling gaps with new entities, systematically leveraging domain expertise to create labeled questions for training parsers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If general-purpose corpora are used for training syntactic parsers, then the training data is readily available and easy to obtain, but the parsing accuracy for domain-specific questions deteriorates due to underrepresentation
Solution Approach 1:
The patent creates synthetic training data by copying and adapting the structure of seed-labeled questions through template abstraction and entity substitution. Instead of manually creating new labeled questions, the system generates numerous synthetic examples by replacing entities in abstracted templates with new entities from domain knowledge bases, thereby multiplying the training value from a small set of seed questions while maintaining domain-specific characteristics
Solution Approach 2:
The patent performs preliminary actions by first abstracting templates from seed questions and preparing domain knowledge bases before generating the full training corpus. This preliminary structuring of domain knowledge and template creation enables efficient batch generation of synthetic training data, avoiding the need for manual labeling of each training example while ensuring domain relevance
2Measurement precision
If domain-specific training data is created manually, then the parsing accuracy for domain-specific questions improves, but the time and effort required for data preparation increases significantly
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate its own training data without human intervention in the data creation process. The automated pipeline abstracts templates from seed questions, substitutes entities using domain knowledge bases, and generates synthetic labeled questions autonomously, eliminating the need for manual annotation while maintaining high domain-specific parsing accuracy
Solution Approach 2:
The patent applies parameter changes by systematically varying entities in abstracted question templates to generate diverse training examples. By changing the entity parameters (substituting with different entities from knowledge bases) while maintaining the syntactic structure, the system creates numerous training variations from a single template, dramatically expanding the training corpus size without proportional increases in manual effort
3Reliability
If more seed questions are provided for training, then the quality of generated training data improves, but the initial data collection effort and time investment increase
Solution Approach 1:
The patent applies partial action by demonstrating that a small subset of seed questions (as few as 10) is sufficient to generate a comprehensive training corpus when combined with domain knowledge bases. Rather than requiring extensive manual collection of numerous seed questions, the system leverages the structured domain knowledge to amplify the value of minimal seed examples, achieving reliable training data generation with reduced initial investment
Data Source
AI summary
A computer-implemented method for generating a question from an abstracted template is described. A non-limiting example of the computer-implemented method includes receiving, by a processor, a question. The method parses, by the processor, the question into a parse tree and abstracts, by the processor, an abstracted template from the parse tree. The method receives, by the processor, a domain schema and a domain knowledge base and generates, by the processor, a new question based on the abstracted template, the domain schema, and the domain knowledge base.