Bilingual Word Alignment Using Greedy Inversion Transduction Grammar

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The high computational cost of searching for word alignments that satisfy the inversion transduction grammar constraint in bilingual word alignment makes existing methods impractical for efficient use in translation and cross-language retrieval.

Innovation Solution

A method and apparatus that preprocess bilingual documents, compute probability gains for word links, apply a greedy algorithm iteratively to find alignments satisfying the inversion transduction grammar constraint, and output the best alignment, reducing computational costs while maintaining alignment quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If exact search algorithms are used to find word alignments satisfying the inversion transduction grammar constraint, then alignment quality is improved, but computational cost becomes too high for practical use

Engineering Contradiction:

Improvealignment qualityVSAvoidcomputational cost

Core Design Contradiction:

Measurement precisionVSLoss of energy

Solution Approach 1:

The patent changes the search strategy parameter from exhaustive search to greedy iterative search. Instead of evaluating all possible alignments, the algorithm iteratively builds alignments by greedily selecting the best local improvements, transforming the computational approach from O(n!) to a much more efficient iterative process while maintaining alignment quality

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent applies partial action by searching over only part of all potential word alignments rather than all possible alignments. The greedy algorithm explores a subset of alignments that are most likely to be optimal, avoiding the computational burden of exhaustive search while still finding high-quality alignments

Inventive Principle:

Principle #16Partial or excessive action

2Productivity

If approximating search algorithms are used to reduce computational cost, then processing speed is improved, but alignment quality deteriorates

Engineering Contradiction:

Improveprocessing speedVSAvoidalignment quality

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent ensures continuity of useful action by iteratively improving the alignment through greedy selection. Rather than performing a single-pass approximation, the algorithm continuously refines the alignment by adding links that provide positive probability gain, ensuring that each iteration contributes meaningfully to the final alignment quality

Inventive Principle:

Principle #20Continuity of useful action

Solution Approach 2:

The patent implements feedback through the iterative greedy algorithm that evaluates the probability gain of adding each potential link. The algorithm uses this feedback to selectively add only those links that improve the alignment, continuously refining the result until no further improvements can be made, thus maintaining quality while improving efficiency

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS9460086B2Method and apparatus for performing bilingual word alignment

Publication Date: 2016.10.04 THE BOEING CO

AI summary

A method for performing bilingual word alignment on source and target text in bilingual documents includes computing probability gains of adding a link (i.e. the corresponding relation) between any pair of source and target words in the source and target text; applying a greedy algorithm to iteratively search for a plurality of word alignments that satisfy an inversion transduction grammar constraint; and outputting the best word alignment among the plurality.