[rmrs] LTOP on NP

[Emily M. Bender]

Replying more quickly this time (in the interests of the empty inbox):

It's intriguing that these two threads should intersect, and I'll be curious to
hear if you are able to work out a solution to the RNR problem using the
rules approach.

Emily

On Thu, Jan 31, 2013 at 3:58 PM, Dan Flickinger <danf at stanford.edu> wrote:
> Hi again, Emily -
>
> [Responding slowly, so you may have to refresh your memory again ...]
>
> The lexical approach to this difference in how heads combine semantically with their complements is mostly convenient, but it is exactly this decision that makes it difficult to implement a clean solution to the Right-Node Raising problem discussed in a separate thread on this list.  If we've already said lexically that a head's LTOP is identified with its complement's LTOP, then it is tricky to avoid the conjunction of two such heads leading to an unwanted identification of their two LTOPs, as in |that building towers and looms over the park|.  I think this would be easier to cope with if instead of the lexical determination of LTOP identity, we did it in two separate syntactic rules, but I haven't tried implementing the idea to confirm this.
>
>  Dan
>
> ----- Original Message -----
> From: "Emily M. Bender" <ebender at uw.edu>
> To: "Dan Flickinger" <danf at stanford.edu>
> Cc: rmrs at delph-in.net
> Sent: Thursday, January 10, 2013 8:38:59 PM
> Subject: Re: [rmrs] LTOP on NP
>
> Thanks for the reply Dan.  I've now reread the thread to remind myself
> of the context of the discussion.  I can see that you could do the
> identification
> lexically or with the rules, but it seems to me that the lexical solution is
> the simpler one in this case:  The lexical item will have to say something
> in either analysis (to do the identification, or to match the rule), so why
> not just have the lexical item do it?
>
> Emily
>
> On Thu, Dec 13, 2012 at 12:20 PM, Dan Flickinger <danf at stanford.edu> wrote:
>> Hi Emily -
>>
>> Picking up on a long-ago thread about the semantic algebra and the need for some heads to identify their LBL with that of their complement, I just wanted to note that even ignoring the issue of pronouns, there are several verb lexical types in the ERG that need to do this identification, namely ones that take contentful PPs or adverb phrases as complements.  Here are some examples:
>>
>> v_pp_dir_le
>> Kim slipped into the room.
>>
>> v_pp_le
>> That building towers over the park.
>> The tablet retails for fifty bucks.
>>
>> v_pp-pp_le
>> Kim lied to her about him.
>> kim applied to the company for a job.
>>
>> v_np-pp_le
>> We raced Kim into the city.
>>
>> v_p-pp_le
>> Kim caught up on elephants.
>>
>> v_adv_le
>> Kim did well.
>> That augers well (for us).
>>
>> v_np-adv_le
>> Kim put it well.
>>
>> While it remains challenging to provide compelling tests for treating optional PPs or adverbs as complements, some of the non-heads above are obligatory, which means they have to be complements, and yet I think we want their semantics to be intersective, not scopal arguments of the verb (different from examples like "We put the book on the shelf", where the PP should be a scopal argument denoting the resulting state of affairs).
>>
>> To respond to another of your questions below, yes, I agree that the determination about whether a complement is treated intersectively or scopally must be determined by each lexical type, though it is a separate implementation question whether that unification of the complement's LTOP is actually done in the lexical type (as currently in the ERG) or in each of two variants of a head-complement rule type (analogous to how modifiers are combined, with distinct rules for intersective and scopal modifiers).
>>
>>  Dan
>>
>> ----- Original Message -----
>> From: "Emily M. Bender" <ebender at uw.edu>
>> To: "Dan Flickinger" <danf at stanford.edu>
>> Cc: rmrs at delph-in.net
>> Sent: Tuesday, September 18, 2012 11:57:43 AM
>> Subject: Re: [rmrs] LTOP on NP
>>
>> Thanks, Dan for the reply!  More below:
>>
>> On Mon, Sep 17, 2012 at 3:26 PM, Dan Flickinger <danf at stanford.edu> wrote:
>>> Hi Emily -
>>>
>>> It is true that the LTOP on a quantified NP should remain unbound, since the LBL of the contained quantifier has to be free to scope as needed.  And it is also true that the semantic algebra constraints seek to ensure that an MRS is fully connected by having heads either unify their LBL with that of their complements/subjects, or take that LBL as a semantic argument of a role in their elementary predication.  This becomes a more visible issue if and when a grammar treats pronouns as not having a separate quantifier, since a verb taking an NP complement or subject then needs to always unify its LBL with that of its argument, expecting that most of the time that unification will be `vacuous' since the argument will be a quantified NP, but the unification will be essential for well-formedness of the MRS when the argument happens to be a pronoun, whose LBL needs to be shared with that of its head.
>>
>> The contained quantifiers are free to scope as needed so long as their LBLs
>> are not identified with the LTOP.  That alone doesn't preclude the LTOP from
>> being used to point to something else (though admittedly, this could be seen
>> as a hack).
>>
>> Regarding the identification of dependent's LTOPs with either the LBL of the
>> head or a hole in the head's EP --- I assume this must be done at a lexical
>> level (and in some cases by specialized constructions), since we can't say
>> that it's always the same for e.g., all head-comp phrases, right?
>>
>> Are there cases other than pronouns without quantifiers where we actually want
>> the SUBJ/COMP's LTOP identified with the head's LTOP?  (If not, maybe I
>> can think of another way around the pronoun thing...)
>>
>>
>>> I think we can take the English apposition case off of your list of three, since we have by now agreed that a better treatment of the semantics of apposition will make use of a new ICONS constraint declaring the equality of the ARG0s of the two NPs, instead of the current two-place EP whose LBL is difficult to anchor anywhere.
>>
>> Remind me: what happens to the scope of the quantifier inside the appositive
>> NP?  Does it matter?
>>
>>> It may also be that an analogous ICONS-based representation will work for at least some of Alex and Katya's gesture MRSs, if it is defensible to represent gestural contributions as quantified expressions.  Up to now, the only tool in the toolbox was the hammer of an additional EP for merging gestural semantic contributions and overtly linguistic ones, but it seems promising to explore the use of ICONS constraints instead.
>>
>> That is an interesting possibility indeed.
>>
>>> This still leaves the Turkish N-Det-Adj structures as potentially problematic, and I don't know enough about the linguistic facts (though probably I should, since it seems to be an important phenomenon challenging our strategies for MRS composition).  If it turns out that what is called a Det here is restricted to definite/indefinite or deictic contrasts, then the Det would not need to introduce a quantifier EP itself, if we factored out these other semantic attributes as we have discussed previously.  But if that Det can be a true generalized quantifier like "every" or "most", this is a more relevant possible obstacle to the current `invisibility' of the N-bar's LBL after the noun has combined with its specifier.
>>
>> Looking back at the paper, it seems we did find a few examples with "every"
>> at least:
>>
>> http://faculty.washington.edu/ebender/papers/ben_egg_tep.pdf
>>
>> Emily
>>
>>
>> --
>> Emily M. Bender
>> Associate Professor
>> Department of Linguistics
>> Check out CLMS on facebook! http://www.facebook.com/uwclma
>
>
>
> --
> Emily M. Bender
> Associate Professor
> Department of Linguistics
> Check out CLMS on facebook! http://www.facebook.com/uwclma



-- 
Emily M. Bender
Associate Professor
Department of Linguistics
Check out CLMS on facebook! http://www.facebook.com/uwclma