Lexical Graph Semantic Space for Accurate Word Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computational approaches for semantic representation of words either rely on hand-built lexical ontologies with errors and inconsistencies or distributional semantics that mix contextual and lexical similarities, failing to preserve precise lexical relationships.
Innovation Solution
A system that generates a real-valued semantic vector space representation by determining synonym and antonym relations, using Laplacian embedding and independent component analysis to preserve lexical relationships without mixing contextual or distributional similarities, and projects words into a metrical vector space for easier computation and interpretation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If hand-built lexical ontologies are used, then semantic relationships are consciously constructed and highly interpretable, but they contain errors, inconsistencies, and ad-hoc decisions
Solution Approach 1:
The patent segments the semantic representation task into two distinct components: lexical semantics (synonym/antonym relationships) and distributional semantics (contextual relationships). By using separate vector spaces for each component, the system avoids mixing these different types of semantic information, thereby preserving the precision of lexical relationships while maintaining interpretability.
Solution Approach 2:
The patent extracts lexical semantic relationships (synonymy and antonymy) from the mixed distributional semantic space and places them in a separate lexical vector space. This extraction allows the system to preserve precise lexical relationships without the contamination of contextual similarities that occur in distributional approaches.
2Ease of operation
If distributional semantics are used, then real-valued vectors are produced allowing mathematical operations, but contextual and lexical similarities are mixed together
Solution Approach 1:
The patent creates two separate vector spaces: a distributional semantic space for contextual relationships and a lexical semantic space for synonym/antonym relationships. This segmentation allows mathematical operations to be performed on each space independently, preserving the precision of lexical relationships while maintaining the computational advantages of real-valued vectors.
Solution Approach 2:
The patent uses a projection mechanism as an intermediary to map words from the distributional semantic space to the lexical semantic space. This projection allows the system to leverage the mathematical properties of distributional vectors while ensuring that lexical relationships are preserved in the final representation.
3Extent of automation
If distributional semantics are used, then automatic construction from large corpora is achieved, but large corpora are required and resulting basis is semantically uninterpretable
Solution Approach 1:
The patent segments the semantic representation into interpretable lexical components (synonym/antonym relationships in the lexical vector space) and automatic distributional components (contextual relationships in the distributional semantic space). This segmentation preserves interpretability of lexical relationships while maintaining automatic construction capabilities for the distributional aspect.
Solution Approach 2:
The patent extracts the interpretable lexical semantic information from the uninterpretable distributional semantic space and places it in a separate lexical vector space. This extraction recovers the lost interpretability while maintaining the automatic construction advantage of distributional semantics.
Data Source
AI summary
Described is a system for generating a semantic space based on the lexical relations between words. The system determines synonym and antonym relations between a set of words. A lexical graph is generated based on the synonym and antonym relations. Manifold embedding of the lexical graph is determined, and Laplacian coordinates of the manifold embedding are assigned as semantic features of the set of words. A quantitative representation of the set of words is generated using the semantic features.