Skip to Main Content
Table 2: 

Results on the ultra-fine entity typing task. LITE series are equipped with the Taxonomic Statement template. “w/o label dependency” is applied to the “NLI+L” setting. The F1 result by LITENLI +L is statistically significant (p-value ¡ 0.01 in t-test) in comparison with the best baseline result by MLMET.

ModelPRF1
UFET-biLSTM (Choi et al., 2018) 48.1 23.2 31.3 
LabelGCN (Xiong et al., 2019) 50.3 29.2 36.9 
LDET (Onoe and Durrett, 2019) 51.5 33.0 40.1 
Box4Types (Onoe et al., 2021) 52.8 38.8 44.8 
LRN (Liu et al., 2021) 54.5 38.9 45.4 
MLMET (Dai et al., 2021) 53.6 45.3 49.1 
 
LITE NLI 1.5 7.1 2.5 
48.7 45.8 47.2 
D+L 27.5 56.4 37.0 
NLI+D+L 45.4 49.9 47.4 
NLI+L 52.4 48.9 50.6 
–w/o label dependency 53.3 46.6 49.7 
ModelPRF1
UFET-biLSTM (Choi et al., 2018) 48.1 23.2 31.3 
LabelGCN (Xiong et al., 2019) 50.3 29.2 36.9 
LDET (Onoe and Durrett, 2019) 51.5 33.0 40.1 
Box4Types (Onoe et al., 2021) 52.8 38.8 44.8 
LRN (Liu et al., 2021) 54.5 38.9 45.4 
MLMET (Dai et al., 2021) 53.6 45.3 49.1 
 
LITE NLI 1.5 7.1 2.5 
48.7 45.8 47.2 
D+L 27.5 56.4 37.0 
NLI+D+L 45.4 49.9 47.4 
NLI+L 52.4 48.9 50.6 
–w/o label dependency 53.3 46.6 49.7 
Close Modal

or Create an Account

Close Modal
Close Modal