Geometric Context Vectors for Bilingual Lexicon Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for extracting bilingual lexicons from comparable corpora face challenges such as coverage issues with rare words and polysemy/synonymy problems due to the limitations of bilingual dictionaries, where dictionary entries with multiple meanings are not adequately handled, and similarities between synonyms are not properly considered.

Innovation Solution

The method involves building source and target context vectors, projecting them into dictionary spaces, and computing similarity measures to identify bilingual pairs, using techniques like canonical correlation analysis (CCA) and probabilistic latent semantic analysis (PLSA) to account for synonymy and polysemy, and iteratively augmenting the dictionary with probable translations from the corpora.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a bilingual dictionary is used to extract bilingual lexicons from comparable corpora, then the extraction process can be performed, but coverage is limited for rare words and polysemy/synonymy problems occur

Engineering Contradiction:

Improveaccuracy of bilingual lexicon extractionVSAvoidcoverage of rare words and handling of polysemy/synonymy

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent introduces context vectors as an intermediary representation between source and target words. These context vectors capture the semantic environment of words, enabling the system to handle rare words and polysemy/synonymy by comparing contextual patterns rather than relying solely on dictionary lookups. The context vectors serve as mediators that bridge the gap between limited dictionary coverage and comprehensive lexical extraction.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transforms the extraction problem by changing the parameter representation from direct word-to-word mapping to context vector space comparison. By representing words in a geometric vector space where semantic relationships are encoded as spatial relationships, the system can handle polysemy/synonymy through angular similarity and rare words through contextual pattern matching, overcoming the limitations of traditional dictionary-based approaches.

Inventive Principle:

Principle #35Parameter changes

2Device complexity

If all dictionary entries are treated as orthogonal in the geometric embodiment, then the vector space model is simplified, but similarities between synonyms are not properly captured

Engineering Contradiction:

Improvesimplicity of vector space modelVSAvoidability to measure similarity between synonyms

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent applies asymmetry by treating synonymous dictionary entries differently from non-synonymous entries. Instead of uniform orthogonal treatment, synonyms are identified through contextual similarity analysis and given special handling in the geometric model. This asymmetric treatment allows the system to capture synonym relationships while maintaining simplicity for other word relationships, resolving the contradiction between model simplicity and measurement precision.

Inventive Principle:

Principle #4Asymmetry

Data Source

PatentUS7620539B2Methods and apparatuses for identifying bilingual lexicons in comparable corpora using geometric processing

Publication Date: 2009.11.17 CONDUENT BUSINESS SERVICES LLC

AI summary

Various methods formulated using a geometric interpretation for identifying bilingual pairs in comparable corpora using a bilingual dictionary are disclosed. The methods may be used separately or in combination to compute the similarity between bilingual pairs.