LangFIR: Discovering Sparse Language-Specific Features from Monolingual Data for Language Steering
Not on the map. Why?
Abstract (arXiv)
Large language models (LLMs) show strong multilingual capabilities, yet reliably controlling the language of their outputs remains difficult. Representation-level steering addresses this by adding language-specific vectors to model activations at inference time, but identifying language-specific directions in the residual stream often relies on multilingual or parallel data that can be expensive to obtain. Sparse autoencoders (SAEs) decompose residual activations into interpretable, sparse feature directions and offer a natural basis for this search, yet existing SAE-based approaches face the same data constraint. We introduce LangFIR (Language Feature Identification via Random-token Filtering), a method that discovers language-specific SAE features using only a small amount of monolingual data and random-token sequences. Many SAE features consistently activated by target-language inputs do not encode language identity. Random-token sequences surface these language-agnostic features, allowing LangFIR to filter them out and isolate a sparse set of language-specific features. We show that these features are extremely sparse, highly selective for their target language, and causally important: directional ablation increases cross-entropy loss only for the corresponding language. Using these features to construct steering vectors for the multilingual generation control task, LangFIR achieves the best average accuracy x BLEU among steering methods across three models (Gemma 3 1B, Gemma 3 4B, and Llama 3.1 8B), three datasets, and twelve target languages, outperforming the strongest monolingual baseline by up to 4.7x and surpassing methods that use parallel data. Our results suggest that language identity in multilingual LLMs is localized in a sparse set of feature directions discoverable with monolingual data. Code is available at https://github.com/JaMussCraft/LangFIR
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 31 of its 39 references that are in the map. First posted on arXiv 4 Apr 2026.
Builds on
31 of 39 references.
References · 31
- 2024The Llama 3 Herd of Modelscited by 14,192
- 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,115
- 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
- 2022No Language Left Behind: Scaling Human-Centered Machine Translationcited by 1,290
- 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,009
- 2024Refusal in Language Models Is Mediated by a Single Directioncited by 960
- 2022Toy Models of Superpositioncited by 928
- 2016Bag of Tricks for Efficient Text Classificationcited by 869
- 2023Steering Language Models With Activation Engineeringcited by 824
- 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 673
- 2019On the Cross-lingual Transferability of Monolingual Representationscited by 575
- 2021The FLORES-101 Evaluation Benchmark for Low-Resource and Multilingual Machine Translationcited by 556
19 more
- 2024Scaling and evaluating sparse autoencoderscited by 530
- 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 381
- 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 283
- 2024Do Llamas Work in English? On the Latent Language of Multilingual Transformerscited by 275
- 2019WikiMatrix: Mining 135M Parallel Sentences in 1620 Language Pairs from Wikipediacited by 221
- 2024A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoderscited by 148
- 2024Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoderscited by 133
- 2020The Tatoeba Translation Challenge -- Realistic Data Sets for Low Resource and Multilingual MTcited by 127
- 2024Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Modelscited by 124
- 2024Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Modelscited by 114
- 2024Automatically Interpreting Millions of Features in Large Language Modelscited by 106
- 2024On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neuronscited by 83
- 2024Understanding and Mitigating Language Confusion in LLMscited by 75
- 2024Separating Tongue from Thought: Activation Patching Reveals Language-Agnostic Concept Representations in Transformerscited by 42
- 2024The LLM Language Network: A Neuroscientific Approach for Identifying Causally Task-Relevant Unitscited by 34
- 2025Unveiling Language-Specific Features in Large Language Models via Sparse Autoencoderscited by 12
- 2025Causal Language Control in Multilingual Transformers via Sparse Feature Steeringcited by 12
- 2026CLaS-Bench: A Cross-Lingual Alignment and Steering Benchmarkcited by 11
- 2025Language Lives in Sparse Dimensions: Toward Interpretable and Efficient Multilingual Control for Large Language Modelscited by 2
Built on by
Citing papers per year: 2026: 1.
1 paper.