Inter-language Query Construction via Aligned Target Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search engines can only process queries in a single language, limiting their ability to handle queries across multiple languages effectively.
Innovation Solution
A method for automatically constructing inter-language queries by aligning target vectors from a learning corpus in multiple languages, using intra-language and inter-language cost functions to minimize the sum of cost functions, and applying the Gram-Schmidt ortho-normalization method to filter meanings, allowing search engines to process queries across languages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If search engines process queries in a single language only, then the system complexity remains low, but the adaptability to handle multiple languages is poor
Solution Approach 1:
The patent introduces aligned target vectors as an intermediary representation that bridges multiple languages. By mapping words from different languages into a shared vector space where semantically equivalent words have similar vectors, the system enables cross-language query processing without requiring complex multilingual processing logic in the search engine itself. The alignment step creates this intermediary representation that facilitates language-agnostic search.
Solution Approach 2:
The patent transforms the query processing approach by changing the parameter representation from language-specific text to language-independent target vectors. By representing words as vectors in a shared space and measuring similarity through vector distance rather than text matching, the system achieves multilingual capability through parameter transformation rather than structural complexity.
2Productivity
If word vectors are used for query construction, then the processing speed is high, but the ability to handle multiple languages simultaneously is lost
Solution Approach 1:
The patent creates a universal target vector representation that serves multiple languages simultaneously. By training target vectors on aligned corpora from multiple languages and ensuring that semantically equivalent words across languages have similar vectors, the system achieves multi-functionality where the same vector space handles queries in any of the supported languages without requiring separate processing pipelines.
3Adaptability or versatility
If N words with closest target vectors are recovered for query construction, then the query coverage increases, but the precision of query meaning is reduced
Solution Approach 1:
The patent applies partial action by selectively using only the N closest words from the retrieved candidates to construct the query, rather than using all possible matches. This partial selection balances coverage and precision by including enough variants to handle language variations while limiting the query size to maintain precision. The parameter N controls this trade-off.
Data Source
AI summary
A method for automatically constructing inter-language queries performed by a search engine from a text file containing a learning corpus. The learning corpus includes a set of phrases expressed in a corresponding manner in at least two languages. Each word of each of the two languages being associated with a target vector. The target vectors of the words of the learning corpus in at least two languages aligned. N words in each of the at least two languages having the closest target vectors with respect to a target vector associated with a query word are retrieved. The queries to be performed by the search engine from the N preceding words recovered in the at least two languages are established.