transitive dominance operator?

[daghaug]

Looking at the documentation and the tutorial, I am unable to find a way to express that node X dominates node Y via one or more edges, i.e. Y is in the subtree rooted by X. Is there a way? Or if not, is this something you could implement? It would be useful, for example, if you are looking for some construction somewhere inside a complement clause (but you don't care how deeply embedded it is).

[bguil]

Thanks for the suggestion.
This feature request was made a long time ago (4 years ago in #12). So it's time to implement it!

With the syntax X ->> Y, you can express the constraint X dominates Y.
It is available in the dev branch of grewlib and in Grew-match for testing.

For now, I haven't implemented large dominance without constraint on edges. We can imagine X -[nsubj|obj]->> Y to restrict the label of the edges to be considered. @daghaug: Would you need that or X ->> Y is sufficient for your usage?

Example

With the following request on bUD_English-EWT:

pattern { X1 [upos=DET]; X2 [upos=NOUN]; X1 < X2; X2 ->> X1 }
without { X2 -> X1 }

We get 8 interesting occurrences.
The 4th and the 5th are clearly annotation errors, and the 7th should be checked (@nschneid?)

[nschneid]

The 4th and the 5th are clearly annotation errors, and the 7th should be checked (@nschneid?)

Will fix. (The 7th from your query is a weird sentence, maybe a dialect thing.)

[nschneid]

Idea for a possible extension of this operator: allowing constraints on the deprels in the dominance chain. E.g. X1 -[!enhanced]->> X2 or X1 -[1=ccomp|xcomp]->> X2.

Does this seem useful or like unneeded complexity?

[daghaug]

X-->Y would be enough for my immediate needs, but I agree that in principle it would be useful to be able to put constraints on the deprels in the chain. One idea (inspired by techniques that are common in LFG) would be to allow regular expressions over the alphabet of edge labels. So for example, taking X to be a verb that is acl:rel and Y to be a node that is PronType=Rel, we can search for particular extraction patterns in relative clauses like ccomp ccomp* or .* advcl or what not. Of course, given that UD labels have internal structure, one might also want to be able to constraint that structures, but allowing both at the same time might end up too complex.

[bguil]

There has been no recent activity. I am closing the issue.
Please feel free to reopen it if needed.