Multilingual encoders improve cross language search with query aware routing
Query-aware routing for Cross-lingual performance gains in Encoders
Computation and LanguageArtificial IntelligenceInformation Retrieval
Summary
When people search using one language but relevant information is in another, search tools can struggle. The authors improved how search works between Finnish and Swedish by adding a special layer that adjusts queries only when languages differ. This method keeps how well searches work when languages match while making cross-language searches better. Their approach works by deciding when to use the original or new search method based on the query’s language. This gives better overall results on a financial dataset without having to redo stored document data.
What this means in practice
- •For search engine engineers: Improve search accuracy when queries and documents are in different languages without rebuilding document indexes.
- •For financial data platform developers: Enhance retrieval of multilingual financial documents by selectively applying cross-language tuning to queries.
Authors
Akshay Jain, Edward Kim
Abstract
Multilingual encoders can exhibit reduced retrieval effectiveness when queries and relevant documents differ in language, despite strong same-language performance. We investigate whether Finnish and Swedish cross-lingual retrieval can improve while preserving an encoder's existing same-language performance and document index. We combine a query-only low-rank adapter, trained against frozen document embeddings, with deterministic routing based on query and index languages. Cross-language queries use the adapter, while same-language queries use the original encoder. SampoTron, our fine-tuned low-rank (LoRA) adapter alongwith the Nemotron-3-Embed-1B model, improves average retrieval quality across six English, Finnish, and Swedish directions from 0.241 to 0.291 in normalized discounted cumulative gain (nDCG) at rank ten, a 20.9% relative gain on a sampled financial benchmark. All six cross-lingual directions improve, and routing preserves the original same-language performance, including two full-corpus Finnish evaluations. The approach enables selective cross-language specialization with reusable document embedding vectors.