Lune

EMNLP Findings2025

ReCoVeR the Target Language: Language Steering without Sacrificing Task Performance

Hannah Sterz, Fabian David Schmidt, Goran Glavas, Ivan Vulic

Abstract

As they become increasingly multilingual, Large Language Models (LLMs) exhibit more language confusion, i.e., they tend to generate answers in a language different from the language of the prompt or the answer language explicitly requested by the user. In this work, we propose ReCoVeR (REducing language COnfusion in VEctor Representations), a novel lightweight approach for reducing language confusion based on language-specific steering vectors. We first isolate language vectors with the help of multi-parallel corpus and then effectively leverage those vectors for effective LLM steering via fixed (i.e., unsupervised) as well as trainable steering functions. Our extensive evaluation, encompassing three benchmarks and 18 languages, shows that Re-CoVeR effectively mitigates language confusion in both monolingual and cross-lingual setups while at the same time-and in contrast to prior language steering methods-retaining task performance. Our data code is available at https://github.com/hSterz/recover.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

2 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models cites The Linear Representation Hypothesis and the Geometry of Large Language Models
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers2

  1. Compositional Steering of Large Language Models with Steering TokensGorjan Radevski, Kiril Gashteovski, Giwon Hong, Carolin Lawrence et al.ACL 2026 · 4 citations
  2. ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language ModelsJiahui Guang, Haiyan Wang, Yingjie Zhu, Cuiyun Gao et al.ICML 2026

Builds on15

  1. Measuring Massive Multitask Language UnderstandingDan Hendrycks, Collin Burns, Steven Basart, Andy Zou et al.ICLR 2021 · 7,905 citations
  2. The Linear Representation Hypothesis and the Geometry of Large Language ModelsKiho Park, Yo Joong Choe, Victor VeitchICML 2024 · 461 citations
  3. In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space SteeringSheng Liu, Haotian Ye, Lei Xing, James Y. ZouICML 2024 · 244 citations
  4. How do Large Language Models Handle Multilingualism?Yiran Zhao, Wenxuan Zhang, Guizhen Chen, Kenji Kawaguchi et al.NeurIPS 2024 · 196 citations
  5. WikiLingua: A New Benchmark Dataset for Multilingual Abstractive SummarizationFaisal Ladhak, Esin Durmus, Claire Cardie, Kathleen R. McKeownEMNLP Findings 2020 · 151 citations

Related papers

  1. Understanding and Mitigating Language Confusion in LLMsKelly Marchisio, Wei-Yin Ko, Alexandre Berard, Théo Dehaze et al.EMNLP 2024 · 10 citations
  2. CLaS-Bench: A Cross-Lingual Alignment and Steering BenchmarkDaniil Gurgurov, Yusser Al Ghussin, Tanja Baeumel, Cheng-Ting Chou et al.ACL Findings 2026
  3. Analysing the Generalisation and Reliability of Steering VectorsDaniel Tan, David Chanin, Aengus Lynch, Brooks Paige et al.NeurIPS 2024
  4. Paths Not Taken: Understanding and Mending the Multilingual Factual Recall PipelineMeng Lu, Ruochen Zhang, Carsten Eickhoff, Ellie PavlickEMNLP 2025 · 15 citations
  5. Don't Trust ChatGPT when your Question is not in English: A Study of Multilingual Abilities and Types of LLMsXiang Zhang, Senyu Li, Bradley Hauer, Ning Shi et al.EMNLP 2023 · 58 citations
Dusk over the sea between two cliffs drawn in fine vertical lines