Training data only includes regular parallel data. At testing, model performances on unseen idioms are measured. 


University of California, Berkeley

There are three idiom testing conditions.


Idiom Testing Conditions


Zero Testing Condition


Training data includes regular and idiom-train data. At testing, model performances on idioms from different contexts of training data are measured.


Joint Testing Condition


Upsampling is the same as the joint split, but the idiom-train data is up-sampled N times. This measures whether up-sampling targeted training data is necessary for achieving better idiom translation quality. 


Learn Before

Related