# Language Steering on WikiMatrix

Source: [Language Steering on WikiMatrix](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-07T08:01:17.734Z

Evaluation of language steering performance using the WikiMatrix dataset, reporting aggregated scores across various language pairs/directions.

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) | WikiMatrix |
| Metric | ES Score |
| 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

- [ES Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=ES%20Score>): Higher is better; unit: SCALAR.

- [FR Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=FR%20Score>): Higher is better; unit: SCALAR.

- [DE Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=DE%20Score>): Higher is better; unit: SCALAR.

- [IT Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=IT%20Score>): Higher is better; unit: SCALAR.

- [ID Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=ID%20Score>): Higher is better; unit: SCALAR.

- [PT Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=PT%20Score>): Higher is better; unit: SCALAR.

- [VI Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=VI%20Score>): Higher is better; unit: SCALAR.

- [PL Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=PL%20Score>): Higher is better; unit: SCALAR.

- [CS Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=CS%20Score>): Higher is better; unit: SCALAR.

- [TR Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=TR%20Score>): Higher is better; unit: SCALAR.

- [ZH Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=ZH%20Score>): Higher is better; unit: SCALAR.

- [NL Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=NL%20Score>): Higher is better; unit: SCALAR.

- [Average Score](<https://www.sota2.com/research/sota/language-steering-on-wikimatrix.md?metric=Average%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | ES Score | Result date | Report paper |
| --- | --- | --- | --- |
| DiffMean | 44.2 | Not recorded | 69d48fa67d0172140ec74423 |
| LangFIR | 42.5 | Not recorded | 69d48fa67d0172140ec74423 |
| LangFIR | 40.6 | Not recorded | 69d48fa67d0172140ec74423 |
| SAE-DM | 40.2 | Not recorded | 69d48fa67d0172140ec74423 |
| LDA | 39.5 | Not recorded | 69d48fa67d0172140ec74423 |
| DiffMean | 38.2 | Not recorded | 69d48fa67d0172140ec74423 |
| Gated-DM | 37.7 | Not recorded | 69d48fa67d0172140ec74423 |
| LDA | 36 | Not recorded | 69d48fa67d0172140ec74423 |
| DiffMean | 34.9 | Not recorded | 69d48fa67d0172140ec74423 |
| SAE-DM | 34.8 | Not recorded | 69d48fa67d0172140ec74423 |
| LangFIR | 32.5 | Not recorded | 69d48fa67d0172140ec74423 |
| Zhong (Mono) | 18.7 | Not recorded | 69d48fa67d0172140ec74423 |
| Gated-DM | 18.6 | Not recorded | 69d48fa67d0172140ec74423 |
| Zhong (Para) | 16 | Not recorded | 69d48fa67d0172140ec74423 |
| Gated-DM | 15.1 | Not recorded | 69d48fa67d0172140ec74423 |
| Zhong (Para) | 12.1 | Not recorded | 69d48fa67d0172140ec74423 |
| No Steering | 11.7 | Not recorded | 69d48fa67d0172140ec74423 |
| Zhong (Para) | 11 | Not recorded | 69d48fa67d0172140ec74423 |
| Zhong (Mono) | 10.6 | Not recorded | 69d48fa67d0172140ec74423 |
| PCA | 10.1 | Not recorded | 69d48fa67d0172140ec74423 |
| SAE-DM | 9.7 | Not recorded | 69d48fa67d0172140ec74423 |
| PCA | 5.7 | Not recorded | 69d48fa67d0172140ec74423 |
| No Steering | 4.5 | Not recorded | 69d48fa67d0172140ec74423 |
| Zhong (Mono) | 2.8 | Not recorded | 69d48fa67d0172140ec74423 |
| No Steering | 2.6 | Not recorded | 69d48fa67d0172140ec74423 |
| PCA | 1.8 | Not recorded | 69d48fa67d0172140ec74423 |
| LDA | 1.3 | Not recorded | 69d48fa67d0172140ec74423 |

### DiffMean

ES Score: 44.2. Result ID: 69d4b9af47f34ce924bc5704.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74051e209131cea5f4ad.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### LangFIR

ES Score: 42.5. Result ID: 69d4b9af47f34ce924bc5705.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### LangFIR

ES Score: 40.6. Result ID: 69d4b9af47f34ce924bc56fc.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### SAE-DM

ES Score: 40.2. Result ID: 69d4b9af47f34ce924bc56f9.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/696722f8e25ad2fa9d1ec049.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### LDA

ES Score: 39.5. Result ID: 69d4b9af47f34ce924bc56ff.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### DiffMean

ES Score: 38.2. Result ID: 69d4b9af47f34ce924bc56fb.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74051e209131cea5f4ad.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### Gated-DM

ES Score: 37.7. Result ID: 69d4b9af47f34ce924bc5703.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78dd1e209131cea7ae1b.md>)

[Implementation](<https://github.com/Aatrox103/multilingual-llm-features>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### LDA

ES Score: 36. Result ID: 69d4b9af47f34ce924bc56f6.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### DiffMean

ES Score: 34.9. Result ID: 69d4b9af47f34ce924bc56f2.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74051e209131cea5f4ad.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### SAE-DM

ES Score: 34.8. Result ID: 69d4b9af47f34ce924bc56f0.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/696722f8e25ad2fa9d1ec049.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### LangFIR

ES Score: 32.5. Result ID: 69d4b9af47f34ce924bc56f3.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### Zhong (Mono)

ES Score: 18.7. Result ID: 69d4b9af47f34ce924bc56f7.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c031e209131cea8bdbc.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### Gated-DM

ES Score: 18.6. Result ID: 69d4b9af47f34ce924bc56fa.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78dd1e209131cea7ae1b.md>)

[Implementation](<https://github.com/Aatrox103/multilingual-llm-features>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### Zhong (Para)

ES Score: 16. Result ID: 69d4b9af47f34ce924bc56ef.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c031e209131cea8bdbc.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### Gated-DM

ES Score: 15.1. Result ID: 69d4b9af47f34ce924bc56f1.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78dd1e209131cea7ae1b.md>)

[Implementation](<https://github.com/Aatrox103/multilingual-llm-features>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### Zhong (Para)

ES Score: 12.1. Result ID: 69d4b9af47f34ce924bc56f8.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c031e209131cea8bdbc.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### No Steering

ES Score: 11.7. Result ID: 69d4b9af47f34ce924bc56f4.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### Zhong (Para)

ES Score: 11. Result ID: 69d4b9af47f34ce924bc5701.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c031e209131cea8bdbc.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### Zhong (Mono)

ES Score: 10.6. Result ID: 69d4b9af47f34ce924bc5700.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c031e209131cea8bdbc.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### PCA

ES Score: 10.1. Result ID: 69d4b9af47f34ce924bc56f5.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 4B"}. This record does not assert independent reproduction.

### SAE-DM

ES Score: 9.7. Result ID: 69d4b9af47f34ce924bc5702.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/696722f8e25ad2fa9d1ec049.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### PCA

ES Score: 5.7. Result ID: 69d4b9af47f34ce924bc56fe.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### No Steering

ES Score: 4.5. Result ID: 69d4b9af47f34ce924bc56fd.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Llama 3.1 8B"}. This record does not assert independent reproduction.

### Zhong (Mono)

ES Score: 2.8. Result ID: 69d4b9af47f34ce924bc56ee.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c031e209131cea8bdbc.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### No Steering

ES Score: 2.6. Result ID: 69d4b9af47f34ce924bc56eb.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### PCA

ES Score: 1.8. Result ID: 69d4b9af47f34ce924bc56ec.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

Recorded evaluation configuration: {"Model":"Gemma 3 1B"}. This record does not assert independent reproduction.

### LDA

ES Score: 1.3. Result ID: 69d4b9af47f34ce924bc56ed.

[Reporting paper](<https://www.sota2.com/research/paper/69d48fa67d0172140ec74423.md>)

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.
