# Pronoun Disambiguation on Winograd Schema Challenge

Source: [Pronoun Disambiguation on Winograd Schema Challenge](<https://www.sota2.com/research/sota/pronoun-disambiguation-on-winograd-schema-challenge>)

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

Data updated: 2026-02-26T05:27:54.372Z

Evaluation of model performance on resolving pronoun ambiguities using the standard Winograd Schema Challenge dataset, typically measured by accuracy.

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 | Pronoun Disambiguation |
| Dataset (including any stated split) | Winograd Schema Challenge |
| Metric | Accuracy |
| 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

- [Accuracy](<https://www.sota2.com/research/sota/pronoun-disambiguation-on-winograd-schema-challenge.md?metric=Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| RoBERTa-WG | 90.1 | Not recorded | 68fa6a031e209131cea256d1 |
| CorefRoBERTa LARGE | 83.2 | Not recorded | 68fa6a031e209131cea256d1 |
| RoBERTa-DPR | 83.1 | Not recorded | 68fa6a031e209131cea256d1 |
| MNPP | 79.5 | Not recorded | 68fa6a031e209131cea256d1 |
| AMS-CSQA-DPR | 75.5 | Not recorded | 68fa6a031e209131cea256d1 |
| UDSSM-MASKEDWIKI-DPR | 75.1 | Not recorded | 68fa6a031e209131cea256d1 |
| BERT-MASKEDWIKI-DPR | 72.5 | Not recorded | 68fa6a031e209131cea256d1 |
| BERT-WIKICREM-DPR | 71.8 | Not recorded | 68fa6a031e209131cea256d1 |
| CorefBERT LARGE | 71.4 | Not recorded | 68fa6a031e209131cea256d1 |
| BERT\_NSP-DPR | 71.1 | Not recorded | 68fa6a031e209131cea256d1 |
| WSC Know. Hunting | 71.1 | Not recorded | 68fa6a031e209131cea256d1 |
| GPT-2 | 70.7 | Not recorded | 68fa6a031e209131cea256d1 |
| CSS | 69.6 | Not recorded | 68fa6a031e209131cea256d1 |
| Ensemble LMs | 63.7 | Not recorded | 68fa6a031e209131cea256d1 |
| BERT-WIKICREAM | 63.4 | Not recorded | 68fa6a031e209131cea256d1 |
| UDSSM | 62.4 | Not recorded | 68fa6a031e209131cea256d1 |
| BERT-MASKEDWIKI | 61.9 | Not recorded | 68fa6a031e209131cea256d1 |
| Maximum Attention Score (MAS) | 60.3 | Not recorded | 68fa65dc1e209131cea0ce64 |
| Modified Skip-Gram | 60.3 | Not recorded | 68fa6a031e209131cea256d1 |
| BERT Inner Attention | 60.3 | Not recorded | 68fa6a031e209131cea256d1 |
| Know. Hunter | 57.1 | Not recorded | 68fa65dc1e209131cea0ce64 |
| Bi-LSTM-DPR | 56 | Not recorded | 68fa6a031e209131cea256d1 |
| Single LM | 54.5 | Not recorded | 68fa65dc1e209131cea0ce64 |
| Transformer | 54.1 | Not recorded | 68fa65dc1e209131cea0ce64 |
| USSM + Supervised DeepNet + KB | 52.8 | Not recorded | 68fa65dc1e209131cea0ce64 |
| USSM + KB | 52 | Not recorded | 68fa65dc1e209131cea0ce64 |
| Random guess | 50 | Not recorded | 68fa65dc1e209131cea0ce64 |

### RoBERTa-WG

Accuracy: 90.1. Result ID: 698dfe2cffc3901de09a01d3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66b21e209131cea11e5a.md>)

[Implementation](<http://winogrande.allenai.org/>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"WinoGrande","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### CorefRoBERTa LARGE

Accuracy: 83.2. Result ID: 698dfe2cffc3901de09a01d2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa686b1e209131cea1bff4.md>)

[Implementation](<https://github.com/thunlp/CorefBERT>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### RoBERTa-DPR

Accuracy: 83.1. Result ID: 698dfe2cffc3901de09a01d1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66b21e209131cea11e5a.md>)

[Implementation](<http://winogrande.allenai.org/>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"DPR","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### MNPP

Accuracy: 79.5. Result ID: 698dfe2cffc3901de09a01dd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### AMS-CSQA-DPR

Accuracy: 75.5. Result ID: 698dfe2cffc3901de09a01d0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa677a1e209131cea16a6d.md>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"CSQA, DPR","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### UDSSM-MASKEDWIKI-DPR

Accuracy: 75.1. Result ID: 698dfe2cffc3901de09a01cf.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa661c1e209131cea0e664.md>)

[Implementation](<https://github.com/namisan/mt-dnn>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"MASKEDWIKI, DPR","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### BERT-MASKEDWIKI-DPR

Accuracy: 72.5. Result ID: 698dfe2cffc3901de09a01ce.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66811e209131cea10bca.md>)

[Implementation](<https://github.com/vid-koci/bert-commonsense>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"MASKEDWIKI, DPR","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### BERT-WIKICREM-DPR

Accuracy: 71.8. Result ID: 698dfe2cffc3901de09a01cd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66811e209131cea10be5.md>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"WIKICREM, DPR","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### CorefBERT LARGE

Accuracy: 71.4. Result ID: 698dfe2cffc3901de09a01cc.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa686b1e209131cea1bff4.md>)

[Implementation](<https://github.com/thunlp/CorefBERT>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### BERT\_NSP-DPR

Accuracy: 71.1. Result ID: 698dfe2cffc3901de09a01cb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65a91e209131cea0bb43.md>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"DPR","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### WSC Know. Hunting

Accuracy: 71.1. Result ID: 698dfe2cffc3901de09a01dc.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### GPT-2

Accuracy: 70.7. Result ID: 698dfe2cffc3901de09a01db.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### CSS

Accuracy: 69.6. Result ID: 698dfe2cffc3901de09a01da.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67791e209131cea169db.md>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### Ensemble LMs

Accuracy: 63.7. Result ID: 698dfe2cffc3901de09a01d9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66791e209131cea10670.md>)

[Implementation](<https://github.com/tensorflow/models/tree/master/research/lm_commonsense>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### BERT-WIKICREAM

Accuracy: 63.4. Result ID: 698dfe2cffc3901de09a01d8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66811e209131cea10be5.md>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### UDSSM

Accuracy: 62.4. Result ID: 698dfe2cffc3901de09a01d7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65891e209131cea0ae90.md>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### BERT-MASKEDWIKI

Accuracy: 61.9. Result ID: 698dfe2cffc3901de09a01d6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66811e209131cea10bca.md>)

[Implementation](<https://github.com/vid-koci/bert-commonsense>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### Maximum Attention Score (MAS)

Accuracy: 60.3. Result ID: 698cce008b517eb4331cdf09.

[Reporting paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Implementation](<https://github.com/huggingface/pytorch-pretrained-BERT>)

Recorded evaluation configuration: {"model":"BERT"}. This record does not assert independent reproduction.

### Modified Skip-Gram

Accuracy: 60.3. Result ID: 698dfe2cffc3901de09a01d4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### BERT Inner Attention

Accuracy: 60.3. Result ID: 698dfe2cffc3901de09a01d5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Implementation](<https://github.com/huggingface/pytorch-pretrained-BERT>)

Recorded evaluation configuration: {"Supervision Type":"Unsupervised","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### Know. Hunter

Accuracy: 57.1. Result ID: 698cce008b517eb4331cdf08.

[Reporting paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa64e41e209131cea07315.md>)

[Implementation](<https://github.com/aemami1/Wino-Knowledge-Hunter>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Bi-LSTM-DPR

Accuracy: 56. Result ID: 698dfe2cffc3901de09a01ca.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6a031e209131cea256d1.md>)

Recorded evaluation configuration: {"Supervision Type":"Supervised","Finetuning Dataset":"DPR","Zero-shot Transfer":"True"}. This record does not assert independent reproduction.

### Single LM

Accuracy: 54.5. Result ID: 698cce008b517eb4331cdf06.

[Reporting paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66791e209131cea10670.md>)

[Implementation](<https://github.com/tensorflow/models/tree/master/research/lm_commonsense>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Transformer

Accuracy: 54.1. Result ID: 698cce008b517eb4331cdf07.

[Reporting paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa703d1e209131cea49db1.md>)

[Implementation](<https://github.com/tensorflow/tensor2tensor>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### USSM + Supervised DeepNet + KB

Accuracy: 52.8. Result ID: 698cce008b517eb4331cdf05.

[Reporting paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6adb1e209131cea2ad7e.md>)

[Implementation](<https://github.com/ptrichel/How-Reasonable-are-Common-Sense-Reasoning-Tasks>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### USSM + KB

Accuracy: 52. Result ID: 698cce008b517eb4331cdf04.

[Reporting paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6adb1e209131cea2ad7e.md>)

[Implementation](<https://github.com/ptrichel/How-Reasonable-are-Common-Sense-Reasoning-Tasks>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Random guess

Accuracy: 50. Result ID: 698cce008b517eb4331cdf03.

[Reporting paper](<https://www.sota2.com/research/paper/68fa65dc1e209131cea0ce64.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
