Skip to Main Content
Table 18 

Performance of IMS and UKB on the naturally distributed test set when varying the size of the training set per pseudoword (MFS for Nat-Nat = 70.5%). Note that we do not report an MFS baseline for Uni-Nat configuration as there is no most frequent sense in the training data.

Config.TrainSystemSize of training data
080160240320400480560640720800
Nat-Nat Nat IMS – 81.984.4 86.4 87.4 88.2 88.7 89.3 89.7 90.0 90.3 
UKB 38.8 56.5 59.2 61.1 62.1 62.8 63.4 63.8 64.2 64.4 64.6 
 
Uni-Nat Uni IMS – 59.8 66.3 69.8 72.2 74.0 75.2 76.3 77.2 77.9 78.6 
UKB 38.8 53.6 55.2 56.4 57.5 58.4 59.1 59.7 60.1 60.5 60.8 
Config.TrainSystemSize of training data
080160240320400480560640720800
Nat-Nat Nat IMS – 81.984.4 86.4 87.4 88.2 88.7 89.3 89.7 90.0 90.3 
UKB 38.8 56.5 59.2 61.1 62.1 62.8 63.4 63.8 64.2 64.4 64.6 
 
Uni-Nat Uni IMS – 59.8 66.3 69.8 72.2 74.0 75.2 76.3 77.2 77.9 78.6 
UKB 38.8 53.6 55.2 56.4 57.5 58.4 59.1 59.7 60.1 60.5 60.8 
Close Modal

or Create an Account

Close Modal
Close Modal