Skip to Main Content
Table 9: 
Intrinsic and extrinsic experiment results for the ad-hoc LexSub. The Vanilla model here refers to language model embeddings trained on Wikitext-103 without the lexical constraints. Ad-hoc LexSub outperforms the Vanilla embeddings on both intrinsic and extrinsic tasks indicating the gains from post-hoc LexSub can be extended to the ad-hoc formulation.
Relatedness TasksSimilarity Tasks
Modelsmen3k(ρ)WS-353R(ρ)Simlex(ρ)Simverb(ρ)
Vanilla 0.5488 0.3917 0.3252 0.2870 
 
ad-hoc LexSub 0.5497 0.3943 0.3489 0.3215 
(a) Intrinsic evaluation results for ad-hoc models in word similarity and relatedness tasks. 
Relatedness TasksSimilarity Tasks
Modelsmen3k(ρ)WS-353R(ρ)Simlex(ρ)Simverb(ρ)
Vanilla 0.5488 0.3917 0.3252 0.2870 
 
ad-hoc LexSub 0.5497 0.3943 0.3489 0.3215 
(a) Intrinsic evaluation results for ad-hoc models in word similarity and relatedness tasks. 
Close Modal

or Create an Account

Close Modal
Close Modal