Span

[sylvainkahane]

Grew-match does not allow us to ask if X dominates Y. (X dominates Y if Y depends on X or Y depends on X1 depending on X and so on.) But we can precompute the maximal projection of X, that is the set of Y dominated by X, X included. And add to the each node some features:

• X.left-span = distance from X to the leftmost Y (dominated by X)
• X.right-span
• X.span = distance between the leftmost and the rightmost Y = weight of the projection of X
• X.continuous = Yes if the projection of X is continuous, No elsewhere. Can the domination relation be precomputed? X ->* Y

[sylvainkahane]

We have to add that several domination relations (and several spans) can be considered, according to the dependency relations we take into account. Three cases seem particularly relevant:

• all: all dependency relations
• micro+: all, minus what we call macrosyntactic relations: parataxis, discourse, dislocated.
• micro: all, minus macrosyntactic relations and paradigmatic lists: conj and reparandum.

The question of punctuation is an additional problem. From the linguistic point a view, it seems reasonable not to consider the punct relations; linking punctuation to their governor.