Using INIT in adnominal possession library

Hello all, but mostly Olga :),

I’ve got a situation similar to the one in the clausal comps library where I need to make sure that possessives only go through one order of head-comps in languages where both orders are instantiated. It looks like the way to do this is by use of the INIT feature. I want to make sure I understand how INIT is being used in clausal comps, so I can implement a solution in adnom poss that’s compatible with it. If I could get some input on whether or not I’m understanding your system correctly, I’d appreciate it :slight_smile: This is what I think is going on:

  1. Each instantiation of head-comps has an INIT value specified on the head dtr, according to the order of that rule. I’m assuming [ INIT + ] is appropriate on the head dtrs of head-initial rules?
  2. Lexical items that take comps all must be specified to be either [ INIT + ] or [ INIT - ], or they will go through all head-comps rules. I can see that INIT is added to all complementizers in, but I don’t see where it’s added to any other lexical items, so I may be misunderstanding how this works.


Hi Elizabeth,

This all seems correct!

INIT is also added to other lexitems here (in an ad-hoc manner, as I now see, and I need to check which items should really go on that list; probably not det!):

path = 'SYNSEM.LOCAL.CAT.HEAD'      
 for pos in ['tverb','aux','det','cop']:
        if ch.get(pos) or pos in ['tverb']:
            name = lexbase.LEXICAL_SUPERTYPES[pos]
            mylang.add(name + ' := [ ' + path + '.INIT ' + default_init_value + ' ].'
                    , merge=True)

One thing: ideally, you want to check that you are not adding constraints that are constradicting what auxiliaries and clausal comps want.

Note also this bug: (erroneously filed under AGG)

Please let me know if you have more questions.

1 Like

Thanks, Olga! I hadn’t spotted that section of code, so thanks for pointing me to it! I’ll probably imitate what you did there for adding INIT to other lexitems – it looks like it won’t be possible to call just the relevant bits of code from my library, so I’ll just replicate it in my library so that INIT will be set correctly even if the user hasn’t entered any info for clausal comps.

Just wanted to check – I think there is as yet no equivalent to INIT for head-spec, is that correct? I imagine a separate feature would be required in that case, rather than reusing INIT.

Nope and yes — unless we expected everything that has an opinion about both rules to have the same INIT value for both head-spec and head-comp, a separate feature makes sense.

Hm, I haven’t been thinking of INIT as belonging conceptually to the phrase structure rules; rather to lexical items, and then the phrase structure rules are either sensitive to it or not.

Could you give an example of where two features are required? (For example, in clausal complements library, I need two separate features to license extraposition in VOS languages combined with head-final attachment if complementizers).

With HCR and HSpR, I would imagine that typically we have different lexical types going through them, and so I don’t immediately see why you couldn’t use INIT in both. But of course perhaps I am just not seeing the problem right away.

unless we expected everything that has an opinion about both rules to have the same INIT value for both head-spec and head-comp

What does it mean, “to have the same INIT value for both head-spec and head-comp”?

Do you mean, the same lexical types participate in both HCR and HSpR? Then sure, we can’t assume they will always utilize the same order in all situations.

In my library, the way you would get multiple orderings of head-spec and of head-comps is by implementing more than one possessive strategy. If I’m understanding correctly, the potential conflict is the one you outline, Olga – if a lexical type can act as a specifier and can take a complement (which in fact happens in my library in the case where the possessor is marked by an adposition), then it might have two different opinions about which order it is in for each rule.

1 Like