Modelling the task directly as a multi-class classification problem where each reading corresponds to one class is impractical, as there are 1,266 unique morphological tag combinations in the corpus. Rather, we cast the problem as arankingof the alternative readings for every ambiguous token, where...