Skip to Main Content
Table 2: 
Similarity and relatedness results for baselines and LexSub. The results indicate that LexSub outperforms all the baselines on relatedness tasks and is competitive on the similarity tasks. This indicates that our model retains the distributional information better than the other models while also learning synonymy and antonymy relations.
Relatedness TasksSimilarity Tasks
Modelsmen3k(ρ)WS-353R(ρ)Simlex(ρ)Simverb(ρ)
Vanilla 0.7375 0.4770 0.3705 0.2275 
 
Retrofitting 0.7437 0.4701 0.4435 0.2976 
Counterfitting 0.6487 0.2497 0.4870 0.4119 
LEAR 0.6850 0.3385 0.5998 0.5637 
 
LexSub 0.7493 0.4956 0.5044 0.3983 
Relatedness TasksSimilarity Tasks
Modelsmen3k(ρ)WS-353R(ρ)Simlex(ρ)Simverb(ρ)
Vanilla 0.7375 0.4770 0.3705 0.2275 
 
Retrofitting 0.7437 0.4701 0.4435 0.2976 
Counterfitting 0.6487 0.2497 0.4870 0.4119 
LEAR 0.6850 0.3385 0.5998 0.5637 
 
LexSub 0.7493 0.4956 0.5044 0.3983 
Close Modal

or Create an Account

Close Modal
Close Modal