"Only" (again) - comments needed so I don't make a fool of myself

[Logical Language Group <lojbab>]

Following is a posting that appeared last month on Linguist List.  I
think Lojban provides some insights, but wanted to run this by our list
before making a fool of myself in front of the linguist who inspired the
current Lojban design, and who may be the foremost expert on linguistic
negation.  My proposed response follows the quoted message.

>Date:         Wed, 13 Apr 94 14:30:06 EDT
>From: Larry Horn <LHORN@yalevm.ycc.yale.edu>
>Subject:      Query: ONLY vs. *SHMONLY
>
>This query addresses the asymmetry exhibited by English and other
>languages I know of in the lexicalization of exceptive operators.  For
>the sake of simplicity, I'll focus on the distribution of ONLY-type
>items with NP focus; the asymmetry in question is illustrated by the
>following paradigm:
>
>       {Nobody but Homer/Only Homer} ate doughnuts.
>       {Everybody but Marge/!Shmonly Marge} ate doughnuts.
>
>--where ! indicates that no lexical item occupies this slot.  The point
>is that many (most?) languages provide a lexical equivalent (or, a la
>francaise, a frozen semi-opaque collocation) for 'no{body/thing} but X'
>but (hypothetically) no languages provide one for the corresponding
>positive exceptive, 'every{body/thing} but X'.
>
>This asymmetry is interesting in the light of the fact that 'only' is an
>essentially negative operator (triggering negative polarity items and
>subject-aux inversion in appropriate contexts), as of course is
>'no...but...', and ceteris paribus negative operators tend not to
>lexicalize as readily as their positive counterparts:  we have MOST as a
>determiner but not !LEAST (i.e.  'less than half of the...'), SOME (as
>weak positive/monotone increasing determiner) vs.  !NALL (='not all'),
>etc.  And many (most?) languages don't allow lexical negative
>determiners (corresponding to NO, FEW) at all, relying on the scopal
>disambiguation of negation + unmarked positive operators.  Why then do
>get a lexicalization for 'no...but' but not for 'every/all...but'?  My
>line on this would involve the idea that even stronger than its
>prejudice against negative-entailing/negative-incorporating lexical
>items is natural language's antipathy against negative-presupposing
>lexical items, which SHMONLY would have to be by definition, but ONLY
>isn't.
>
>(Various asymmetries might be trotted out as possible reflexes of this
>antipathy, but I won't trot them out here.)
>
>This in turn would be attributable to the characteristic use of negation
>as a second-order operation on an earlier or contextually plausible
>affirmative, as widely discussed (cf.  Parmenides, Plato, Kant,
>Strawson, Wason, Givon, Horn,...).  Whether or not this analysis is
>sound, what I'm really curious about at this point is the empirical
>generalization:  Are there any languages with candidates for
>SHMONLY-type lexical items (= every...but)?  Are there any particularly
>exotic representations for 'only'?  (Notice that of course I'm not
>saying there's anything particularly unwieldy about expressions for
>positive exceptives--both 'Everyone but Marge ate doughnuts' and 'Only
>Marge didn't eat doughnuts' are perfectly natural things to say--but
>they do seem to be harder to lexicalize.  Impossible?
>
>Please reply to LHORN@YALEVM or LHORN@YALEVM.YCC.YALE.EDU; I'll post a
>summary on the list.


In Linguist 5.432, you wrote:
>Subject:      Query: ONLY vs. *SHMONLY

I just spotted this in rereading old mail, and do not see that you have
posted a summary.  So here is a belated input.

I'm going to be talking about Loglan/Lojban, the artificial language
with a predicate-logic-based grammar.  You may be uninterested in
artificial language "implementations" that relate to this problem.  Even
if you aren't in general interested, you may be interested in Lojban's
implementation and problems with negation, since a great deal of our
design analysis derived directly from your book on negation.  In any
case, looking at your examples as I would try to express them in Lojban
suggested to me an at-least partial explanation for lack of "*SHMONLY"

I can send our general paper discussing negation, written a few years
ago not too long after we went through your book, which in turn was
after some serious semantic and scope issues in negation had cropped up.
Your book resolved these issues, and this is a chance to give you a
belated thanks.

>This query addresses the asymmetry exhibited by English and other
>languages I know of in the lexicalization of exceptive operators.  For
>the sake of simplicity, I'll focus on the distribution of ONLY-type
>items with NP focus; the asymmetry in question is illustrated by the
>following paradigm:
>
>       {Nobody but Homer/Only Homer} ate doughnuts.
>       {Everybody but Marge/!Shmonly Marge} ate doughnuts.

Although we have successfully dealt with a large number of negation
issues as a result of your book, the semantics of "only" has remained a
sore spot.  There are several proposed approaches; some but not all have
seen actual usage, as opposed to hypothetical usage that appears in the
discussions of the words.  Usually people have avoided the question by
rephrasing something they would say in English using "only" to some
other equivalent that avoids "only" and then back-translating to Lojban
from there.

We have specifically tried to avoid assigning a single particle to
"only" having determined that in English at least, "only" has several
meanings that are not logically equivalent and we don't want English
semantics transfer in an area that is so obviously tied up with logical
and quantificational structure of the predicate.  (BTW, a word I haven't
seen you discuss in Linguist List that seems to get embroiled in our
"only" discussions is "just", which has overlapping semantics problems
for us.)

>--where ! indicates that no lexical item occupies this slot.  The point
>is that many (most?) languages provide a lexical equivalent (or, a la
>francaise, a frozen semi-opaque collocation) for 'no{body/thing} but X'
>but (hypothetically) no languages provide one for the corresponding
>positive exceptive, 'every{body/thing} but X'.

Based on your examples here, I chose the obvious "only" equivalent from
Lojban:

>   {Nobody but Homer/Only Homer} ate doughnuts.
     le  pamei       po'u          la xomer citka   lo jintitnanba
     The singularity which is      Homer    eat/ate doughnut(s).
     Only            (restrictive)

where "pa", the Lojban quantificational "1" means "Exactly 1".

     noda         po'u     na'ebo     la xomer citka    lo jintitnanba
     No-something which is other-than Homer    eats/ate doughnut(s).
     Nobody but

is a form representing the "Nobody but" version of the semantics.  "no"
is the logical quantifier "no" or "none", "da" the existential variable.
More exact semantics would have us say something like "No person which
is other-than ..." or "None of the people which are other-than
...", since the existential variable has not semantically restricted in
the sentence, but Lojban users commonly make this 'error' when dealing
with "only".

I should note here that "na'ebo" (other-than) is derived from the scalar
negation operator "na'e" (which is distinguished from predicate negation
in Lojban largely thanks to you) so that most of these Lojban versions
explicitly call forth the negative aspect of "only".

Having given these equivalents to "only", it turns out that Lojban seems
to deal rather easily with your "*shmonly" simply by playing with these

>   {Everybody but Marge/!Shmonly Marge} ate doughnuts.

     roda            po'u     na'ebo     la mardj citka    lo jintitnanba
     Every-something which is other-than Marge    eats/ate doughnut(s).
     Everybody but

where "ro" is the logical quantifier "all" - this is obviously patterned
after the "noda" form above.

     le  da'apamei        po'u     na'ebo     la mardj citka   lo jintitnanba
     The all-but-one-some which is other-than Marge    eat/ate doughnut(s).

>This asymmetry is interesting in the light of the fact that 'only' is an
>essentially negative operator (triggering negative polarity items and
>subject-aux inversion in appropriate contexts), as of course is
>'no...but...',

>and ceteris paribus negative operators tend not to
>lexicalize as readily as their positive counterparts:  we have MOST as a
>determiner but not !LEAST (i.e.  'less than half of the...'), SOME (as
>weak positive/monotone increasing determiner) vs.  !NALL (='not all'),
>etc.

Lojban has non-specific quantifiers 

so'u  a few
so'o  several (your "LEAST")
so'i  many
so'o  a large number of (your "MOST")
so'a  almost all (your "NALL")

da'a means "all-but #" when followed by a quantifier including these
non-specific values or generally defaults to "one" in ellipsis; it was
designed to handle the logically problematical superlative

"roda          mleca      x"
(everything is less  than x)

which is always wrong since 'x' is not less than itself, and we haven't
restricted the existential variable "da".  So we use"

"da'ada                         mleca      x"
 all-but-(1)/everything-else is less  than x) or using your notation
 !1 something                is less  than x 

And the existence of "da'a" thereby allows easy expression of "!SOME"
"!MANY", "!MOST", etc. corresponding to the earlier sentences:

le  da'aso'imei ...
The !MANY-some

le  da'aso'omei ...
The !LEAST-some

le  da'aso'amei ...
The !NALL-some

The Lojban formulations I gave above present one obvious reason why
natural languages may not have all these quantifiers:  they are
incredibly vague - a restrictive relative phrase/clause that gave useful
additional identifying information, corresponding to the "po'u ..." or
the "po'u na'ebo ..." would be hard to come up with.  Thus, even in
Lojban, I can phrase the quantifier portion of the parallel to "only"
easily above, but I didn't try to give a complete 'argument' because
often the negation on the quantifier would semantically demand that we
specify just which ones are being omitted in the relative clause.  I can
come up with one example though:

le  da'aso'omei poi    nancu li mu  
The !LEAST-some  which  are 5 years (old).

but this relies on the semantic inference that 5-year-olds are a small
portion of most populations to make sense.

>And many (most?) languages don't allow lexical negative determiners
>(corresponding to NO, FEW) at all, relying on the scopal disambiguation
>of negation + unmarked positive operators.  Why then do get a
>lexicalization for 'no...but' but not for 'every/all...but'?  My line on
>this would involve the idea that even stronger than its prejudice
>against negative-entailing/negative-incorporating lexical items is
>natural language's antipathy against negative-presupposing lexical
>items, which SHMONLY would have to be by definition, but ONLY isn't.
>
>(Various asymmetries might be trotted out as possible reflexes of this
>antipathy, but I won't trot them out here.)

My Lojban formulations at the beginning of this message had the obvious
asymmetry above that only the "only" phrasing didn't use a "na'ebo".

My reason just above dealing with the specification of the restrictive
set explains why.  In:

   le  da'apamei        po'u     na'ebo     la mardj citka   lo jintitnanba
   The all-but-one-some which is other-than Marge    eat/ate doughnut(s).

we had to use the other-than operator (na'ebo) to restrictively define
"le da'apamei", and it is difficult to imagine a common restriction that
would avoid the explicit (scalar) negation implied by "na'ebo".

Thus, I see the lack of a lexicalization for "SHMONLY" and its ilk as
being due to the infrequency of it being useful as a single word - if
you are going to have to use a non-trivial relative clause to indicate
the specifics of the implied scalar negation, you might as well use a
longer scalar negation-based phrasing like "all but" to move the
negation up from the relative clause.  Lojban, whose equivalent of "all
but" hides the implied negation, cannot do so, and the "na'ebo" stays in
the relative clause; however, as a result, Lojban has no trouble
lexicalizing the whole spectrum of quantifiers.

>This in turn would be attributable to the characteristic use of negation
>as a second-order operation on an earlier or contextually plausible
>affirmative, as widely discussed (cf.  Parmenides, Plato, Kant,
>Strawson, Wason, Givon, Horn,...).  Whether or not this analysis is
>sound, what I'm really curious about at this point is the empirical
>generalization:  Are there any languages with candidates for
>SHMONLY-type lexical items (= every...but)?  Are there any particularly
>exotic representations for 'only'?  (Notice that of course I'm not
>saying there's anything particularly unwieldy about expressions for
>positive exceptives--both 'Everyone but Marge ate doughnuts' and 'Only
>Marge didn't eat doughnuts' are perfectly natural things to say--but
>they do seem to be harder to lexicalize.  Impossible?
>
>Please reply to LHORN@YALEVM or LHORN@YALEVM.YCC.YALE.EDU; I'll post a
>summary on the list.

We've had several other phrasings for "only" since the issue came up.
I'd be happy to collect them together for you if they would be useful.
I don't know if any of the others formulations necessarily deal with
"SHMONLY" any better than natural languages though.
----
lojbab                           Note new address:    lojbab@access.digex.net
Bob LeChevalier, President, The Logical Language Group, Inc.
2904 Beau Lane, Fairfax VA 22031-1303 USA                        703-385-0273
           Ask me about the artificial language Loglan/Lojban.