... However, the errors are not equally im-portant, since some errors affect the perfor-mance of subsequent natural language pro-cessing (NLP) tasks seriously while others do not. This paper aims ... stillexists room to improve supervised POS tagging interms of error differentiation.It should be noted that not all errors are equallyimportant in POS tagging. Let us consider the parsetrees in Figure ... structure of Figure 2 as a loss,another loss function Lt(yi, yj) is defined asLt(yi, yj) =12[Dist(Pi,j, yi) + Dist(Pi,j, yj)] × γ, (3)where Pi,jdenotes the nearest common parent...