quantifier query response forwarded from Larry Horn

[Georgia]

Re: Fw: queries about quantifiers
----- Original Message ----- 
From: Laurence Horn 
To: Georgia ; Ash Asudeh ; Emily Bender ; howard.gregory at phil.uni-goettingen.de 
Cc: hpsg-l at lists.Stanford.EDU 
Sent: Monday, January 19, 2004 2:52 PM
Subject: Re: Fw: queries about quantifiers


Hi, all.


If I can jump in here on both questions, forwarded to me by Georgia:


On (1), the standard semantic account of Ladusaw (1979 et seq.; see for example his "Negation and Polarity" article in the Blackwell Handbook of Contemporary Semantic Theory, Lappin, ed., 1996) correctly predicts that the restrictor of "all" should be a downward entailing environment.  Carl Pollard writes:


> Hi Howard.
>
> I have no idea about the second point. Please let us know what you
> learn. But in connection with the first point, I don't believe MOST DOES
> license NPI's:
>
> Everybody ate something/*anything.
>
> Nobody ate *something/anything.
>
> Most people ate something/*anything
>
>
> *Everybody/*most people/nobody gives a rat's ass about NPI licensing.
>
> Note also:
>
>
> (i)  Everyody who has ever given it any thought supports the euro.
> (ii) All English people who have ever given it any thought support the e=
> uro.
>
> These are just as good as your examples with MOST, so it seems to
> be the restrictive relative clause environment, not MOST, that is
> responsible for the NPI licensing.
>
> At least so it seems to me at first blush.
>
> All best,
>
> Carl


But there are two different issues here.  Consider statements of the form (DET A) B.   
Does the DETERMINER establish a downward entailing (DE) environment?  ALL does, MOST doesn't, MANY doesn't, SOME doesn't, NO does, etc.  If so, A (the restrictor) should accept negative polarity items, as it does in "{Everybody/Nobody} who has ever eaten raw squid remembers it".  If not, it shouldn't, as it doesn't in "*{Some/Many} people who have ever eaten raw squid remember it."  Thus "ALL" and "NO" are DE determiners. 
Now does the QUANTIFER (DET A) establish a downward entailing environment (i.e. in the predicate or nuclear scope)?  The only quantifier in the above group that does is "No CN", where CN is the common noun that "No" combines with.  So


{No/*Some/*Many/*Most/*All} people who eat raw squid could ever be friends of mine.


One of the motivations of Ladusaw's theory was in fact that ALL licenses polarity items (i.e. in the restrictor) but ALL PEOPLE/EVERYBODY does not (i.e. in the predicate).  But, as "*Some people who have ever eaten raw squid remember it" shows, it's not restrictors in general that license NPIs, but only restrictors of DE determiners (like ALL and NO).   So indeed, "Most people give a rat's ass about NPI licensing" is correctly predicted to be bad, but "Most people who have ever eaten raw squid could be friends of mine", with the NPI in the restrictor, *should* be bad (because MOST is not a DE determiner) but it *isn't*.  This is a problem for classical DE licensing, and it's one that has been discussed in the literature.  I think Irene Heim, in her NELS 14 paper "A Note on Negative Polarity and Downward Entailingness", around pp. 100-04, is the first to deal with this explicitly; she proposed a notion of "limited DE-ness" that may or may not strike you as a satisfactory solution.  (Similar questions arise for bare plurals:  "People who have ever eaten raw squid will know what I mean"--not that bad, right?)


Now, as for (2)--


At 10:01 PM -0600 1/18/04, Georgia wrote:
  ----- Original Message -----
  From: "Ash Asudeh" <asudeh at csli.stanford.edu>
  To: "Emily Bender" <bender at csli.stanford.edu>
  Cc: <hpsg-l at lists.Stanford.EDU>
  Sent: Sunday, January 18, 2004 9:19 PM
  Subject: Re: queries about quantifiers


  >
  > Dear Emily, Howard, and list members,
  >
  > I think (Larry) Horn would make an excellent starting point for Howard's
  > second point. I'm not sure where the relevant stuff is published, though.
  > I'm sure it does appear somewhere, because I heard him give a related talk
  > at CLS a couple of years ago (maybe check the proceedings for CLS 38 when
  > they come out).
  >
  > I think the gist of Horn's LSA talk was that since there is an implicature
  > from "some" to "not all", "not all" is unnecessary (according to Gricean
  > pragmatics).
  >
  > I remember wondering at the time, though, whether this really explains
  > things. Since "not all" equally implicates "some" (otherwise the speaker
  > should use "no" according to Gricean principles), why isn't "not all"
  > lexicalized at "some"'s expense?
  >
  > I probably missed some detail of the presentation.
  >
  > Best,
  > Ash
  >
  > On Sun, 18 Jan 2004, Emily Bender wrote:
  >
  > > Dear Howard,
  > >
  > > I believe that Horn addressed your second point (the lack of
  > > lexicalization of "not all") in his talk 'Lexical Pragmatics: H. Paul,
  > > Grice and beyond' at the LSA earlier this month.  He gave a pragmatic
  > > explanation for it, although the details escape me now...
  > >
  > > Emily


Right; as Emily notes, this is something I've been working on--in fact, since my 1972 dissertation, Chapter 4 of which is devoted to attempting to answer it.  (I address the same issue in section 4.5 of my 1989 book, A Natural History of Negation, reissued in 2001 by CSLI.  There's also a summary of my arguments in Levinson's 2000 book, _Presumptive Meanings_ (MIT Press).  The basic idea is stated in terms of the traditional post-Aristotelian square of oppositions.  The fact that no natural language lexicalizes "nall" (for 'not all'), which was noticed first, as far as I know, by Thomas Aquinas, is attributed in my argument to Gricean reasoning--basically, the two subcontraries ("some F are G", "not all F are G") essentially implicate each other, ceteris paribus, so lexicalizations for both are not necessary.  Given that we're dealing with closed class items, such unnecessary determiners/quantifiers are not tolerated.  This finding extends to modals ("can't" lexicalizes for 'not-possible' or 'not-permitted', but not for 'possible not' or 'permitted-not'), for binary connectives (we have both...and, either...or, neither...nor, but no lexicalization noth...nand corresponding to the Sheffer stroke), and so on.  It's not that simple, and there's some reason to believe there are other factors involved in the quantifier case as opposed to that of other operators that can be mapped onto the square, but that's the basic idea.  And why, given that the I ("some") and O ("not all") vertices don't both need to be lexicalized, is it always the I and never the O that is?  I attribute that to the fundamental informational asymmetry between negation and affirmation, and the markedness of the former in natural language--this is also discussed in great (maybe TOO great) detail in my negation book. 


best,
Larry
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/hpsg-l/attachments/20040119/32b53dff/attachment.htm>