Cross-lingual Query Translation via Pseudo-feedback Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cross-lingual information retrieval systems face challenges in accurately translating queries due to polysemic terms and noise in automatically generated dictionaries, leading to unacceptably noisy feedback when adapting pseudo-relevance feedback techniques.
Innovation Solution
A method involving translating a query into a target language, performing a first information retrieval operation to retrieve pseudo-feedback documents, generating an updated query based on these documents, and re-translating it to improve query accuracy, using a translation model derived from pseudo-feedback documents to refine the query translation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If automatically generated dictionaries are used for cross-lingual query translation, then translation coverage is improved, but translation accuracy deteriorates due to noise and polysemic terms
Solution Approach 1:
The patent implements pseudo-relevance feedback by using retrieved documents to identify and correct translation errors. The system retrieves documents using an initial translation, analyzes their language models to identify likely correct terms, and uses this feedback to correct the original translation, thereby improving accuracy while maintaining coverage
Solution Approach 2:
The patent introduces language models as an intermediary between the automatic dictionary translation and the final query. The language model of retrieved documents acts as a mediator to identify and correct translation errors, bridging the gap between coverage-oriented automatic translation and accuracy-oriented manual translation
2Adaptability or versatility
If pseudo-relevance feedback is applied in cross-lingual retrieval, then query enrichment is improved, but feedback quality deteriorates due to translation noise
Solution Approach 1:
The patent implements a feedback mechanism where retrieved documents are used to generate language models that identify translation errors. This feedback loop allows the system to correct its own translation mistakes by analyzing the statistical properties of relevant documents, thereby improving feedback quality despite initial translation noise
Solution Approach 2:
The patent converts the harmful effect of translation noise into a benefit by using the retrieved documents as evidence to identify and correct errors. The noise from automatic translation is transformed into useful feedback signals through language model analysis, turning a disadvantage into an advantage for query enrichment
Data Source
AI summary
Cross-lingual information retrieval is disclosed, comprising: translating a received query from a source natural language into a target natural language; performing a first information retrieval operation on a corpus of documents in the target natural language using the translated query to retrieve a set of pseudo-feedback documents in the target natural language; re-translating the received query from the source natural language into the target natural language using a translation model derived from the set of pseudo-feedback documents in the target natural language; and performing a second information retrieval operation on the corpus of documents in the target natural language using the re-translated query to retrieve an updated set of documents in the target natural language.