Textual Entailment Recognition via Predicate-Argument Vector Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional textual entailment recognition systems fail to determine entailment when the argument structures of predicates are different, as they assume identical argument structures for comparison.
Innovation Solution
A method involving a vector generation unit that creates vectors for predicate-argument structures using words other than argument type indicators, a combination identification unit that compares these vectors, and an entailment determination unit that calculates feature amounts to determine entailment based on identified combinations, regardless of argument structure differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional textual entailment recognition systems assume identical argument structures for comparison, then matching determination can be performed, but entailment determination fails when argument structures of predicates are different
Solution Approach 1:
The invention changes the parameter of comparison from exact argument structure matching to vector-based semantic similarity. By converting predicate-argument structures into vectors that capture semantic meaning rather than rigid structural form, the system can handle different argument structures while maintaining reliable entailment determination. The vector representation allows flexible comparison that adapts to structural variations while preserving the core semantic relationship needed for entailment.
2Measurement precision
If the system performs strict matching on predicate argument structures, then structural consistency is maintained, but entailment determination cannot be performed on texts with different argument structures
Solution Approach 1:
The invention introduces vectors as an intermediary representation between the original predicate-argument structures and the comparison process. These vectors serve as a mediator that translates rigid structural information into flexible semantic representations, enabling the system to compare texts with different argument structures while maintaining precise entailment determination. The vector intermediary preserves essential semantic relationships while allowing structural variations.
Data Source
AI summary
A textual entailment recognition apparatus (2) includes a vector generation unit (21) that generates, for each of first and second texts, a vector for each predicate-argument structure by using a word other than a word indicating a type of argument of a predicate in the predicate-argument structure; a combination identification (22) unit that compares the vector generated for each predicate-argument structure for the first text and the vector generated for each predicate-argument structure for the second text, and identifies combinations of the predicate-argument structures of the first text and the predicate-argument structure of the second text based on a result of the comparison; and an entailment determination unit (23) that obtains a feature amount for each of the identified combinations, and determines whether the first text entails the second text based on the obtained feature amounts.