mlsanityData to 14 Oct 2026

Retrieval Disagreement as a Signal: Diagnosing and Steering Multilingual Dense Embedders

arXiv 2610.18526 ↗First posted 13 Oct 2026cs.CL cs.IRcited by 0

Not on the map. Why?

Abstract (arXiv)

Multilingual dense retrievers place all languages in one shared space so that a question finds relevant documents, regardless of which language they are written in. We show that the same question, asked in different languages, retrieves substantially different documents, and this disagreement is not because of weak language understanding. Each language performs well on its own and ranks candidates consistently with the other languages; divergence appears only when languages share an index. We trace the mechanism: each language keeps its own ranking, but in the shared index, its correct answers are pushed far down, buried under non-relevant documents that are either in the query's own language or in English, depending on the model. We introduce disagreement-guided steering: for each language, we contrast the documents that other languages agree on for a question against those that language retrieves idiosyncratically, and take the difference of their representations as a language-specific direction, added to that language's queries at inference. Against baselines of language-mean removal, subspace removal, and per-language rotation, it most strongly reduces same-language crowding while remaining competitive on recall, trading off against mean removal, which better preserves ranking quality.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 27 of its 37 references that are in the map. First posted on arXiv 13 Oct 2026.

Builds on

27 of 37 references.

References · 27

  1. 2022No Language Left Behind: Scaling Human-Centered Machine Translationcited by 1,312
  2. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,261
  3. 2022Text Embeddings by Weakly-Supervised Contrastive Pre-trainingcited by 1,253
  4. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  5. 2023Steering Language Models With Activation Engineeringcited by 922
  6. 2020Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillationcited by 593
  7. 2019On the Cross-lingual Transferability of Monolingual Representationscited by 578
  8. 2017Word Translation Without Parallel Datacited by 487
  9. 2024Multilingual E5 Text Embeddings: A Technical Reportcited by 372
  10. 2023The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variantscited by 311
  11. 2017A Survey Of Cross-lingual Word Embedding Modelscited by 143
  12. 2018When and Why are Pre-trained Word Embeddings Useful for Neural Machine Translation?cited by 126
15 more
  1. 2020On the Language Neutrality of Pre-trained Multilingual Representationscited by 85
  2. 2020LAReQA: Language-agnostic answer retrieval from a multilingual poolcited by 42
  3. 2025Language-Specific Latent Process Hinders Cross-Lingual Performancecited by 21
  4. 2024Discovering Low-rank Subspaces for Language-agnostic Multilingual Representationscited by 19
  5. 2021A Simple and Effective Method To Eliminate the Self Language Bias in Multilingual Representationscited by 17
  6. 2025Steering into New Embedding Spaces: Analyzing Cross-Lingual Alignment Induced by Model Interventions in Multilingual Language Modelscited by 11
  7. 2023Multi-EuP: The Multilingual European Parliament Dataset for Analysis of Bias in Information Retrievalcited by 5
  8. 2025The Cross-Lingual Cost: Retrieval Biases in RAG over Arabic-English Corporacited by 5
  9. 2026Cross-Lingual Activation Steering for Multilingual Language Modelscited by 5
  10. 2025Language Bias in Information Retrieval: The Nature of the Beast and Mitigation Methodscited by 3
  11. 2026All Languages Matter: Understanding and Mitigating Language Bias in Multilingual RAGcited by 3
  12. 2026LANGSAE EDITING: Improving Multilingual Information Retrieval via Post-hoc Language Identity Removalcited by 2
  13. 2026Latent-Space Intervention for Cross-Lingual Factual Consistency: Consistency Improvements without Accuracy Dropscited by 2
  14. 2026Hubness, Not Anisotropy, Drives Cross-Lingual Retrieval Asymmetry in Multilingual Embedding Modelscited by 1
  15. 2026SHIFT: Semantic Harmonization via Index-side Feature Transformation for Multilingual Information Retrievalcited by 1

Built on by

Not cited yet.