Here’s some pretty established code from customize.py:
has_question_particles = bool(ch.get('q-part')) if has_question_particles and has_stative_predicate_adjectives: roots.add('root := [ SYNSEM.LOCAL.CAT.HEAD +vjc ].') elif has_question_particles: roots.add('root := [ SYNSEM.LOCAL.CAT.HEAD +vc ].') elif has_stative_predicate_adjectives: roots.add('root := [ SYNSEM.LOCAL.CAT.HEAD +vj ].') else: roots.add('root := [ SYNSEM.LOCAL.CAT.HEAD verb ].')
I noticed now, revisiting my Russian grammar (now obtaining it from Matrix; my old “dev” version had a lot of manual tweaks) that this is problematic because then you get actual complementizers as possible root heads, and then trees like the below tree on the left become possible:
It is pretty bad, right? No reason whatsoever to have that tree on the left?
It will go away if I remove complementizers from possible root heads, but that will break sentences with question particles, presumably.
What does this mean? Could it be that the particles should not be treated as complementizers after all? (Second position ones are treated as modifiers; sentence initial and final ones have been treated as complementizers all this time, but nobody has ever tested them along with optional subordinating complementizers.)
Or could I do something at the level of the subordinator, or the embedded verb, or?..
Or should optional complementizers be a special case somehow…