List your company
ResearchFormerly WizWand

Linguistic Minimal Pairs on BLiMP 10% (test)

76.1BLiMP 10% AccuracyDTG
72.87673.71374.5575.387May 15, 2026
Updated 5mo ago

Evaluation results

MethodLinks
DTGstructural tokens added=true2026.0576.1
GiLT-DPstructural tokens added=false2026.0576.1
GiLT-PSDstructural tokens added=false2026.0575.7
TXL-Largestructural tokens added=false, number of parameters=334M2026.0575.4
PLM-Maskstructural tokens added=true2026.0575.3
PLMstructural tokens added=true2026.0575.1
TXL (baseline)structural tokens added=false, number of parameters=252M2026.0575.1
GiLT-DMstructural tokens added=false2026.0574.9
Pushdown-LMstructural tokens added=false2026.0574
TGstructural tokens added=true2026.0573.5
GiLT-PASstructural tokens added=false2026.0573