Entailment Pair Extension Using Rule-Based Transitivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for collecting entailment pairs in question-answering systems lack precision and fail to exhaustively cover necessary paraphrasing patterns, making it difficult to predict and extend entailment pairs effectively.
Innovation Solution
An entailment pair extension apparatus that generates n-term entailment pairs from m-term pairs using stored generation rules, applying modification rules and transitivity to extend existing pairs, ensuring high precision and predictability in the generation of new pairs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a machine-trained determiner is used to collect entailment pairs, then a huge number of entailment pairs can be collected without human labor, but precision is not guaranteed and it is impossible to predict what specific pairs would be obtained
Solution Approach 1:
The patent segments the entailment pair collection process into two distinct phases: (1) using a machine-trained determiner to generate a large volume of entailment pair candidates without human labor, and (2) applying manual verification to a sampled subset of these candidates to assess precision. This segmentation allows the system to achieve both high productivity in pair generation and measurable precision through targeted human evaluation, resolving the contradiction between automated scale and quality assurance.
2Measurement precision
If manual collection of entailment pairs is performed to ensure precision, then high precision can be achieved, but it requires significant human labor and cannot generate a huge number of pairs efficiently
Solution Approach 1:
The patent applies preliminary action by having experts manually verify and evaluate a sampled subset of entailment pairs before full-scale deployment. This preliminary validation establishes precision benchmarks and refines the machine-trained determiner, ensuring that subsequent automated generation maintains high precision while achieving large-scale productivity. The preliminary manual work on samples prevents the need for continuous manual verification of all pairs.
3Device complexity
If necessary paraphrasing pattern pairs are not exhaustively covered, then the system development is simplified, but the question-answering system cannot provide appropriate answers for all query variations
Solution Approach 1:
The patent implements self-service by enabling the system to automatically generate and extend entailment pairs through rule-based transformations and transitivity applications. Once initial entailment pairs are collected and verified, the system autonomously generates additional pairs by applying linguistic rules (such as transitivity: if A entails B and B entails C, then A entails C) and pattern variations. This self-service mechanism enhances adaptability and coverage without requiring proportional increases in manual curation effort.
Data Source
AI summary
An entailment pattern pair extension apparatus 50 extends entailment pairs by generating an n-term entailment pair from an m-term entailment pair, where m and n are integers not smaller than 0 and satisfying m<n. Each entailment pair includes a first language pattern and a second language pattern entailed by the first language pattern. Entailment pattern pair extension apparatus 50 includes: a generation rule storage unit110 storing generation rules for generating n-term entailment pairs from m-term entailment pairs, and a binary pair adding unit 112 receiving m-term entailment pair, determining, for each generation rule stored in the generation rule storage unit 110, whether its condition is satisfied by the m-term pair, and if the condition is satisfied, applying a modification rule of the generation rule to each language pattern constituting the m-term entailment pair.