# Language Steering on FLORES+, WikiMatrix, and Tatoeba (Averaged) Source: [Language Steering on FLORES+, WikiMatrix, and Tatoeba (Averaged)]() Collection index: [llms.txt]() Data updated: 2026-04-07T08:01:17.826Z Evaluation of Language Steering performance averaged across results derived from the FLORES+, WikiMatrix, and Tatoeba datasets, reporting composite (ACC x BLEU) scores for multiple target languages. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Language Steering | | Dataset (including any stated split) | FLORES+, WikiMatrix, and Tatoeba (Averaged) | | Metric | ACC x BLEU (ES) | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 27 of 27 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [ACC x BLEU (ES)](): Higher is better; unit: SCALAR. - [ACC x BLEU (FR)](): Higher is better; unit: SCALAR. - [ACC x BLEU (DE)](): Higher is better; unit: SCALAR. - [ACC x BLEU (IT)](): Higher is better; unit: SCALAR. - [ACC x BLEU (ID)](): Higher is better; unit: SCALAR. - [ACC x BLEU (PT)](): Higher is better; unit: SCALAR. - [ACC x BLEU (VI)](): Higher is better; unit: SCALAR. - [ACC x BLEU (PL)](): Higher is better; unit: SCALAR. - [ACC x BLEU (CS)](): Higher is better; unit: SCALAR. - [ACC x BLEU (TR)](): Higher is better; unit: SCALAR. - [ACC x BLEU (ZH)](): Higher is better; unit: SCALAR. - [ACC x BLEU (NL)](): Higher is better; unit: SCALAR. - [Average ACC x BLEU](): Higher is better; unit: SCALAR. ## Recorded results | Method | ACC x BLEU (ES) | Result date | Report paper | | --- | --- | --- | --- | | LangFIR | 38.4 | Not recorded | 69d48fa67d0172140ec74423 | | LangFIR | 38.3 | Not recorded | 69d48fa67d0172140ec74423 | | LDA | 37.5 | Not recorded | 69d48fa67d0172140ec74423 | | DiffMean | 36 | Not recorded | 69d48fa67d0172140ec74423 | | Gated-DM | 33.2 | Not recorded | 69d48fa67d0172140ec74423 | | DiffMean | 32.5 | Not recorded | 69d48fa67d0172140ec74423 | | SAE-DM | 32.2 | Not recorded | 69d48fa67d0172140ec74423 | | LDA | 31.5 | Not recorded | 69d48fa67d0172140ec74423 | | DiffMean | 28.2 | Not recorded | 69d48fa67d0172140ec74423 | | SAE-DM | 27 | Not recorded | 69d48fa67d0172140ec74423 | | LangFIR | 25.3 | Not recorded | 69d48fa67d0172140ec74423 | | Gated-DM | 22 | Not recorded | 69d48fa67d0172140ec74423 | | Gated-DM | 18 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Para) | 14.2 | Not recorded | 69d48fa67d0172140ec74423 | | No Steering | 14.2 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Mono) | 14 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Para) | 12.6 | Not recorded | 69d48fa67d0172140ec74423 | | PCA | 12.5 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Para) | 10.8 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Mono) | 9.4 | Not recorded | 69d48fa67d0172140ec74423 | | SAE-DM | 9.2 | Not recorded | 69d48fa67d0172140ec74423 | | PCA | 5.1 | Not recorded | 69d48fa67d0172140ec74423 | | No Steering | 3.3 | Not recorded | 69d48fa67d0172140ec74423 | | LDA | 3.2 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Mono) | 2.3 | Not recorded | 69d48fa67d0172140ec74423 | | PCA | 1 | Not recorded | 69d48fa67d0172140ec74423 | | No Steering | 0.9 | Not recorded | 69d48fa67d0172140ec74423 | ### LangFIR ACC x BLEU (ES): 38.4. Result ID: 69d4b9b047f34ce924bc573b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### LangFIR ACC x BLEU (ES): 38.3. Result ID: 69d4b9b047f34ce924bc5732. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### LDA ACC x BLEU (ES): 37.5. Result ID: 69d4b9b047f34ce924bc5735. [Reporting paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### DiffMean ACC x BLEU (ES): 36. Result ID: 69d4b9b047f34ce924bc573a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### Gated-DM ACC x BLEU (ES): 33.2. Result ID: 69d4b9b047f34ce924bc5739. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### DiffMean ACC x BLEU (ES): 32.5. Result ID: 69d4b9b047f34ce924bc5731. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### SAE-DM ACC x BLEU (ES): 32.2. Result ID: 69d4b9b047f34ce924bc572f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### LDA ACC x BLEU (ES): 31.5. Result ID: 69d4b9af47f34ce924bc572c. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### DiffMean ACC x BLEU (ES): 28.2. Result ID: 69d4b9af47f34ce924bc5728. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### SAE-DM ACC x BLEU (ES): 27. Result ID: 69d4b9af47f34ce924bc5726. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### LangFIR ACC x BLEU (ES): 25.3. Result ID: 69d4b9af47f34ce924bc5729. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### Gated-DM ACC x BLEU (ES): 22. Result ID: 69d4b9b047f34ce924bc5730. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### Gated-DM ACC x BLEU (ES): 18. Result ID: 69d4b9af47f34ce924bc5727. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### Zhong (Para) ACC x BLEU (ES): 14.2. Result ID: 69d4b9af47f34ce924bc5725. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### No Steering ACC x BLEU (ES): 14.2. Result ID: 69d4b9af47f34ce924bc572a. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### Zhong (Mono) ACC x BLEU (ES): 14. Result ID: 69d4b9af47f34ce924bc572d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### Zhong (Para) ACC x BLEU (ES): 12.6. Result ID: 69d4b9af47f34ce924bc572e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### PCA ACC x BLEU (ES): 12.5. Result ID: 69d4b9af47f34ce924bc572b. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### Zhong (Para) ACC x BLEU (ES): 10.8. Result ID: 69d4b9b047f34ce924bc5737. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### Zhong (Mono) ACC x BLEU (ES): 9.4. Result ID: 69d4b9b047f34ce924bc5736. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### SAE-DM ACC x BLEU (ES): 9.2. Result ID: 69d4b9b047f34ce924bc5738. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### PCA ACC x BLEU (ES): 5.1. Result ID: 69d4b9b047f34ce924bc5734. [Reporting paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### No Steering ACC x BLEU (ES): 3.3. Result ID: 69d4b9b047f34ce924bc5733. [Reporting paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### LDA ACC x BLEU (ES): 3.2. Result ID: 69d4b9af47f34ce924bc5723. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### Zhong (Mono) ACC x BLEU (ES): 2.3. Result ID: 69d4b9af47f34ce924bc5724. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### PCA ACC x BLEU (ES): 1. Result ID: 69d4b9af47f34ce924bc5722. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### No Steering ACC x BLEU (ES): 0.9. Result ID: 69d4b9af47f34ce924bc5721. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.