# Language Steering on Tatoeba Source: [Language Steering on Tatoeba]() Collection index: [llms.txt]() Data updated: 2026-04-07T08:01:17.869Z Evaluation of model performance on Language Steering capabilities using the Tatoeba dataset, assessing performance across 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) | Tatoeba | | Metric | Performance (es) | | Unit | PERCENT | | 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 - [Performance (es)](): Higher is better; unit: PERCENT. - [Performance (fr)](): Higher is better; unit: PERCENT. - [Performance (de)](): Higher is better; unit: PERCENT. - [Performance (it)](): Higher is better; unit: PERCENT. - [Performance (id)](): Higher is better; unit: PERCENT. - [Performance (pt)](): Higher is better; unit: PERCENT. - [Performance (vi)](): Higher is better; unit: PERCENT. - [Performance (pl)](): Higher is better; unit: PERCENT. - [Performance (cs)](): Higher is better; unit: PERCENT. - [Performance (tr)](): Higher is better; unit: PERCENT. - [Performance (zh)](): Higher is better; unit: PERCENT. - [Performance (nl)](): Higher is better; unit: PERCENT. - [Average Performance](): Higher is better; unit: PERCENT. ## Recorded results | Method | Performance (es) | Result date | Report paper | | --- | --- | --- | --- | | LangFIR | 50.3 | Not recorded | 69d48fa67d0172140ec74423 | | LangFIR | 49.1 | Not recorded | 69d48fa67d0172140ec74423 | | LDA | 49 | Not recorded | 69d48fa67d0172140ec74423 | | Gated-DM | 41.3 | Not recorded | 69d48fa67d0172140ec74423 | | LDA | 40.3 | Not recorded | 69d48fa67d0172140ec74423 | | DiffMean | 40.1 | Not recorded | 69d48fa67d0172140ec74423 | | DiffMean | 36.6 | Not recorded | 69d48fa67d0172140ec74423 | | SAE-DM | 35 | Not recorded | 69d48fa67d0172140ec74423 | | DiffMean | 33.9 | Not recorded | 69d48fa67d0172140ec74423 | | SAE-DM | 32 | Not recorded | 69d48fa67d0172140ec74423 | | Gated-DM | 28.6 | Not recorded | 69d48fa67d0172140ec74423 | | LangFIR | 27.7 | Not recorded | 69d48fa67d0172140ec74423 | | Gated-DM | 27.3 | Not recorded | 69d48fa67d0172140ec74423 | | No Steering | 19.9 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Para) | 19.2 | Not recorded | 69d48fa67d0172140ec74423 | | PCA | 17.4 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Para) | 16.9 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Mono) | 13.7 | Not recorded | 69d48fa67d0172140ec74423 | | SAE-DM | 13 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Para) | 11.6 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Mono) | 10.3 | Not recorded | 69d48fa67d0172140ec74423 | | PCA | 6.4 | Not recorded | 69d48fa67d0172140ec74423 | | LDA | 5 | Not recorded | 69d48fa67d0172140ec74423 | | No Steering | 2.9 | Not recorded | 69d48fa67d0172140ec74423 | | Zhong (Mono) | 1.7 | Not recorded | 69d48fa67d0172140ec74423 | | PCA | 0.1 | Not recorded | 69d48fa67d0172140ec74423 | | No Steering | 0 | Not recorded | 69d48fa67d0172140ec74423 | ### LangFIR Performance (es): 50.3. Result ID: 69d4b9b047f34ce924bc574d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### LangFIR Performance (es): 49.1. Result ID: 69d4b9b047f34ce924bc5756. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### LDA Performance (es): 49. Result ID: 69d4b9b047f34ce924bc5750. [Reporting paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### Gated-DM Performance (es): 41.3. Result ID: 69d4b9b047f34ce924bc5754. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### LDA Performance (es): 40.3. Result ID: 69d4b9b047f34ce924bc5747. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### DiffMean Performance (es): 40.1. Result ID: 69d4b9b047f34ce924bc5755. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### DiffMean Performance (es): 36.6. Result ID: 69d4b9b047f34ce924bc574c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### SAE-DM Performance (es): 35. Result ID: 69d4b9b047f34ce924bc574a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### DiffMean Performance (es): 33.9. Result ID: 69d4b9b047f34ce924bc5743. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### SAE-DM Performance (es): 32. Result ID: 69d4b9b047f34ce924bc5741. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### Gated-DM Performance (es): 28.6. Result ID: 69d4b9b047f34ce924bc5742. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### LangFIR Performance (es): 27.7. Result ID: 69d4b9b047f34ce924bc5744. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### Gated-DM Performance (es): 27.3. Result ID: 69d4b9b047f34ce924bc574b. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### No Steering Performance (es): 19.9. Result ID: 69d4b9b047f34ce924bc5745. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### Zhong (Para) Performance (es): 19.2. Result ID: 69d4b9b047f34ce924bc5740. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### PCA Performance (es): 17.4. Result ID: 69d4b9b047f34ce924bc5746. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### Zhong (Para) Performance (es): 16.9. Result ID: 69d4b9b047f34ce924bc5752. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### Zhong (Mono) Performance (es): 13.7. Result ID: 69d4b9b047f34ce924bc5751. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### SAE-DM Performance (es): 13. Result ID: 69d4b9b047f34ce924bc5753. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### Zhong (Para) Performance (es): 11.6. Result ID: 69d4b9b047f34ce924bc5749. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### Zhong (Mono) Performance (es): 10.3. Result ID: 69d4b9b047f34ce924bc5748. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction. ### PCA Performance (es): 6.4. Result ID: 69d4b9b047f34ce924bc574f. [Reporting paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### LDA Performance (es): 5. Result ID: 69d4b9b047f34ce924bc573e. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### No Steering Performance (es): 2.9. Result ID: 69d4b9b047f34ce924bc574e. [Reporting paper]() Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction. ### Zhong (Mono) Performance (es): 1.7. Result ID: 69d4b9b047f34ce924bc573f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### PCA Performance (es): 0.1. Result ID: 69d4b9b047f34ce924bc573d. [Reporting paper]() Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction. ### No Steering Performance (es): 0. Result ID: 69d4b9b047f34ce924bc573c. [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.