Skip to Main Content
Table 3: 

Results for fine-grained entity typing. All LITE model results are statistically significant (p-value ¡ 0.05 in t-test) in comparison with the best baseline results by MLMET on OntoNotes and by SEPREM on FIGER.

ModelOntoNotesFIGER
macro-F1micro-F1macro-F1micro-F1
Hierarchy-Typing (Chen et al., 2020b) 73.0 68.1 83.0 79.8 
Box4Types (Onoe and Durrett, 2020) 77.3 70.9 79.4 75.0 
DSAM (Hu et al., 2020) 83.1 78.2 83.3 81.5 
SEPREM (Xu et al., 2021) – – 86.1 82.1 
MLMET (Dai et al., 2021) 85.4 80.4 – – 
 
LITE pre-trained on NLI+UFET 86.6 81.4 80.1 74.7 
NLI+task-specific training 86.4 80.9 86.7 83.3 
ModelOntoNotesFIGER
macro-F1micro-F1macro-F1micro-F1
Hierarchy-Typing (Chen et al., 2020b) 73.0 68.1 83.0 79.8 
Box4Types (Onoe and Durrett, 2020) 77.3 70.9 79.4 75.0 
DSAM (Hu et al., 2020) 83.1 78.2 83.3 81.5 
SEPREM (Xu et al., 2021) – – 86.1 82.1 
MLMET (Dai et al., 2021) 85.4 80.4 – – 
 
LITE pre-trained on NLI+UFET 86.6 81.4 80.1 74.7 
NLI+task-specific training 86.4 80.9 86.7 83.3 
Close Modal

or Create an Account

Close Modal
Close Modal