In-situ wh-words in languages which also do extraction

So if instead of type append-list, we want to propagate an additional boolean value WH +/-, so as to ultimately rule out WH + items in the left periphery of head-adjunct and in-situ-question clauses, where should such a feature live on the lexical items?

Let’s put it parallel to L-PERIPH, since it gets passed through the trees in roughly the same way.

OK; this seems to mostly work but I think I need to do something about lexical rules also, for them to propagate the value.

So perhaps the phrase-or-lexrule type should be doing it.

So I now have a situation that seems parallel to what we discussed about head-adjunct and subject extraction, for which @ebender suggested the new list type which I then called clist, for a list of canonical synsems.

It would be ideal to also rule out this ambiguity in the order of head-adjunct and extracted-complement application.

But adding a [ COMPS clist ] constraint to the head-adjunct’s head daughter leads to ruling out all parses. This is because the basic-head-filler phrase says its arguments’ COMPS are of type olist (maybe; I am actually not clear on why I didn’t hit this with the SUBJ).

Seems like meddling with that would potentially break argument optionality. And I did not have to meddle with anything like that to fix the SUBJ issue. What do you think, @ebender, do you see right away at what level the olist constraints could be changed? Or can we make clist cross-typed with olist somehow?

I thought you might hit something like this eventually. It can be solved by creating mutual subtypes of onull & cnull, ocons & ccons, and olist & clist.

Hmm. Looks like I cannot create a type:

occons := ccons & ocons.

because ocons’s FIRST is of type unexpressed which cannot be unified with a canonical synsem, looks like.

Does it sound like I need a more elaborate hierarchy of synsem? Or did I not understand “mutual subtypes” right?

Actually, simply omitting this particular subtype seems to work at least for the sentence in the example. So, just creating ocnull := cnull & onull. Will see if that seems to work generally…

Hmm – that might work, but it seems brittle…

Yeah… I think I sense what you mean.

Let’s leave this question open then.

Furthermore, I still have this ambiguity due to head-adjunct being able to serve as the gap daughter:

Screen Shot 2020-01-30 at 11.09.15 AM

Again, I am assuming semantically this does not make sense? Or does it? I am actually not sure. If it doesn’t, then can something like MODIFIED help me here? I am noticing that, whether this is intended or not, the value on the head daughter in the left tree, the one licensed by the subj-head, is xmod but in the right tree, where the head daughter is head-adj, it is rmod. But I can’t make enough sense of it to see if it is correct and/or applies in any way.

Sorry, what do you mean by “gap daughter”?

Sorry – the head daughter :). As opposed to the “filler” daughter.

That’s nothing other than the usual ambiguity between VP and S attachment of adverbials. Not your problem…

Thanks for confirming!

OK, then I suppose for now the only open question is what to do about clist and possibly having a subtype of clist and olist, somehow. Which might require doing something about the hierarchy of synsem, specifically the relationship between canonical-synsem and unexpressed. Could we have something unifiable between them while not breaking what currently relies on olists

Right – what you’d need is a type that subsumes gap and unexpressed, but not non-gap expressed synsems.

Trying to wrap my head around this:

so, that type should itself be just avm then? Because right now I think unexpressed is a synsem-min and canonical-synsem is a synsem which is also synsem-min.

No – you need to add to the hierarchy under synsem so that there is such a type, and in fact I think I was wrong with “subsumes”. If you want ocons and ccons to unify, then there needs to be a mutual subtype for unexpressed and gap… which is actually impossible because they have incompatible values for SLASH. So, can we convince ourselves that we’d never need a value to be olist and clist and cons?

How do I reason about this. I mean, on the one hand if we say: this would be a list of (expressed) canonical synsems which are also unexpressed, that just doesn’t make sense but also, that’s just the unification failure that we’d be getting.

From another angle, that would be either an empty list or a list of non-gappy synsems which are also optional? That sounds less nonsensical.

But I think I don’t really know why the filler-gap phrase is using olists in the first place. Is that easy to explain/describe?..

I think the thing is to look at where olist is used. I suspect it might be only on the daughter positions of certain rules. If there’s no reason for those rules to also say clist and the whole value that is being constrained to olist or clist isn’t also being copied up (so that “olist” or “clist” goes higher up the tree) then we’re probably safe.

Given that we resorted to a binary feature in the end, does it sound like we can just use L-PERIPH?

No. L-PERIPH has a separate function (tracking whether the constituent is/must be at the left-edge of something).