Skip to Main Content
Table 11: 
Results for Compositional Comparison. Accuracy over three answer candidates (random is 33%).
ModelZeroMLPMLMLinearLangSense
shotWSMaxWSMaxpertnolang
RoBERTa-L 29 36 49 31 41 
BERT-WWM 33 41 65 32 36 
BERT-L 33 32 35 31 34 
 
BERT-B 32 33 35 33 35 
RoBERTa-B 33 32 40 29 33 
 
Baseline 34 35 48 
ModelZeroMLPMLMLinearLangSense
shotWSMaxWSMaxpertnolang
RoBERTa-L 29 36 49 31 41 
BERT-WWM 33 41 65 32 36 
BERT-L 33 32 35 31 34 
 
BERT-B 32 33 35 33 35 
RoBERTa-B 33 32 40 29 33 
 
Baseline 34 35 48 
Close Modal

or Create an Account

Close Modal
Close Modal