The current parser separates reduce direction and nuclearity (e.g., there can be a L:nucleus:span and R:nucleus:span). It would be better to just have one type of binary reduce action per relation type (e.g., just B:nucleus:span) and then infer the head of the resulting subtree from the nuclearity and position of its children. This would reduce the size of the model by a large amount. About 30% of the parser actions (i.e., model classes) are redundant in this way.
The current parser separates reduce direction and nuclearity (e.g., there can be a
L:nucleus:span
andR:nucleus:span
). It would be better to just have one type of binary reduce action per relation type (e.g., justB:nucleus:span
) and then infer the head of the resulting subtree from the nuclearity and position of its children. This would reduce the size of the model by a large amount. About 30% of the parser actions (i.e., model classes) are redundant in this way.