There is no training data for this task, it follows the same structure
of the previous (x-lingual) lexical substitution tasks: only a small
trial dataset is made available. People have used this trial data for
training/tuning machine learning algorithms purposes in the past, it
depends on how much data your method needs.
Lucia