I've forgotten most of what I knew about ConllDependencyNode - do you need headNode to have access to attributes specific to the ConllDependencyNode type or would a BaseToken suffice, to generalize the solution? If a ConllDependencyNode is what's needed, then what you proposed sounds good to me.
-- James On Fri, Jun 10, 2016 at 7:08 AM, Pei Chen <chen...@apache.org> wrote: > I don't see any issues with adding the additional optional > attribute... I think we already did the same for other items like > relations for similar reasons. The only catch is probably that the > dependency will need the dictionary lookup to be run first (assuming > that the logic will be added to the DP to iterate through all NE's in > the CAS) if they want to use that attribute. > > > On Thu, Jun 9, 2016 at 5:13 PM, Miller, Timothy > <timothy.mil...@childrens.harvard.edu> wrote: > > How do people feel about modifying the typesystem? I'm finding that > > grabbing the dependency headword is something very useful for feature > > extraction. But it is a bottleneck if every feature extractor that uses > > it has to recompute it. So I propose adding a field to the > > IdentifiedAnnotation type of "headNode" with type ConllDependencyNode. > > > > Any thoughts or good reasons to avoid this? > > > > Thanks > > Tim > > >
