Rough notes from discussion on 30-04-2012

• Kim tried and failed to leave

• Kim can and in fact must solve this problem

To create a logical form that is a tree you have to copy the argument. Proposal: keep the structure as the one that is already generated (with two holes pointing to the same label), and then interpret with a second order variable.

We're higher-order anyway, so there shouldn't be a problem with the meta-variable technique

A related problem when the argument is non-scopal, given the algebra's requirement that such verbs identify their label with that of their complements:

• Kim hired and fired every consultant.

hire and fire both identify their labels with the label of the argument.

[[[*If* we are ditching the quantifiers on pronouns.

But maybe we don't want to do that: "You on the left of the room ..."]]]

-> Doesn't matter about pronouns or not. If the verbs share COMPS values in coordination, then they'll try to identify their labels with the same label, even if it's not the label of anything else (because the LTOP of the NP is underspecified).

Why does the algebra say that? What are the non-pronoun cases?

Kim slipped into the room. _slip_v_2 (arrive sneakily)

• -> PP complement

.... but in fact the problem isn't ordinary head-nexus rules, but because Kim hired and fired every consultant is treated as RNR. Subjects aren't problematic because the label of the subject is identified with the label of the coord_rel.

The issue has to do with extraction in the RNR analysis.

Taking as an example (assuming that down the hill is the complement of slid and stumbled):

Kim slid or stumbled down the hill

At the point of extraction of down the hill there is no conj rel for the label of down the hill to be identified with.

Proposed target representation: label of down the hill is equated with the label of the conj rel anyway.

Why link it to anything? If it's not linked to anything, to many possible scopes, including possibly some where it's in the restriction of some quantifier. (Basically anywhere where the variables are inside the scope of their associated quantifiers.) Some spurious some incorrect.

So why not have the RNR-attaching rule do the identification? -> would require two such rules, one for clausal/scopal complements, one for label sharing ones.

Or a "GRAB-ME" feature, which the constituents use to expose LTOP or INDEX depending on their type (nominal or clausal or PP). Would also help with subject: currently two subj rules, one for NP and one for clausal subjects.

The LF also requires that the type of the arguments be fixed for a given predicate symbol, like the arity. (But again, that's an LF-level constraint and not an LF-description level constraint.) There are lots of verbs that do this ... diathesis? NP -> CP rule? (as a type of metonymy)

Interim summary

• Kim can and must leave => MRSs as descriptions/meta-variable

  • Kim hired and fired every consultant

    • Kim slid or stumbled down the hall => different RNR rules

      • Unlinked labels -> too many scopes

      • "GRAB ME" probably still not a good solution (or not??)

      • target representation: h1:down, h1: conj, that is the complement's label should be shared with the conjunctions and not the conjuncts themselves; conjuncts shouldn't share labels with each other

Down the stairs we think Kim fell. (assuming this is grammatical for the sake of the example)

In solving this problem, need to delay identifying the labels until we know coordination has/hasn't happened (in case of extraction). (Don't do it lexically or on the complement extraction rules, but on RNR or head-filler. In head-filler case in particular, need to find a way to pass up the label of the verb along with the slash value.)

Further examples to worry about:

• Every consultant, Kim hired and fired. => coordination rule switches out the label being carried along.

• The consultant Kim hired and fired was unhappy.

In fact: the place to bifurcate is probably the coordination rule, treating the ATB case separately from the other cases.

Side note from over waffles:

Context: in searching for the PP complement examples, we talked briefly about put-type verbs, but these are treated as taking the PP argument as a scopal argument so were not directly relevant to that point of the discussion. (Though The book was placed or hidden on the shelf would have the same problem as Kim can and must leave.)

Why treat put as a resultative? It's definitely three place so there are only two choices for what the value of ARG3 is: e from the locative or h from the locative. In other resultatives, there is evidence that it has to be h, because scopal operators can intervene:

• Kim made the room not warm.

not doesn't work for put, but that's arguably because not doesn't attach easily to PPs (but cf: not in a million years, not on my watch, not in my house ...).