[notational note: I looked again at some of the formal semantics
literature, and it seems that 'kind' is preferred to 'generic' for the
abstract individuals like 'chihuahuas' we've been talking about - which
I think corresponds to what are called 'kinds' in e.g. Chierchia
"References to Kinds across Languages" 1998; 'generic' seems to be
reserved for the {lo'e} idea of "typical individuals". I use 'kind'
below.]
* Saturday, 2011-09-10 at 15:04 -0300 - Jorge Llambías <jjllambias@gmail.com>:
> On Sat, Sep 10, 2011 at 12:10 PM, Martin Bays <mbays@sdf.org> wrote:
> > * Saturday, 2011-09-10 at 10:43 -0300 - Jorge Llambías <jjllambias@gmail.com>:
> >>
> >> 1L: ro prenu cu se prami su'o tciuaua
> >> 1E: Everyone is loved by some chihuahua.
> >>
> >> 2L: ro prenu cu se prami zo'e noi tciuaua
> >> 2E: Everyone is loved by chihuauas.
> >>
> >> 3L: zo'e noi tciuaua cu prami ro prenu
> >> 3E: Chihuahuas love everyone.
> >>
> >> 4L: su'o tciuaua cu prami ro prenu
> >> 4E: Some chihuahua loves everyone.
> >>
> >> We also have two domains of discourse:
> >>
> >> D1 = {lo prenu ku xi pa, lo prenu ku xi re, lo prenu ku xi ci, .., lo
> >> tciuaua ku xi pa, lo tciuaua ku xi re, ...}
> >> = {person_1, person_2, person_3, ...., chihuahua_1, chihuahua_2, ...}
> >>
> >> D2 = {lo prenu ku xi pa, lo prenu ku xi re, lo prenu ku xi ci, .., lo tciuaua}
> >> = {person_1, person_2, person_3, ...., chihuahuas}
> >>
> >> D1 and D2 are not the same domain. Sentences 1 and 4 "put us" in
> >> domain D1, while sentences 2 and 3 "put us" is domain D2. By that I
> >> mean that those are the natural domains in which to interpret those
> >> sentences without any more context. Do we agree so far?
> >
> > Not entirely. I think 1E-4E could just as well be interpreted in the
> > union D12 of D1 and D2 - because a sentence involving "some chihuahua"
> > can't have the generic "chihuahuas" as an witness, and although
> > (as in Carlson) a predication involving "chihuahuas" is ambiguous
> > between being about the generic and about its
> > manifestations/stages/whatever, that doesn't mean the domain of
> > discourse has to be different for different interpretations.
>
> The English situation is additionally complicated by the
> singular/plural morphology. You say that "some chihuahua" can't have
> chihuahuas as a witness, even when chihuahuas are in the domain of
> discourse. But why is that? Is it because only chihuahuas can be a
> witness, and chihuahuas are not chihuahuas? That can't be the reason
> because chihuahuas are indeed chihuahuas. I think it has to do with
> something like the witness has to satisfy "...is a chihuahua", and not
> just "...are chihuahuas". So those two are different predicates in
> English, at least when the domain of discourse is D12. In Lojban we
> have to make do with "tciuaua" for both.
The situation in English is rather strange. The singular does indeed
seem to refer specifically to individuals. Meanwhile the plural is
ambiguous between pluralities of individuals and pluralities of
strict subkinds - "some chihuahuas" can't be witnessed by the kind
'chihuahuas', but it *can* be witnessed by 'black chihuahuas' or 'dead
chihuahuas'.
So it seems English does differentiate between kinds and mundanes,
but it confuses the two in plurals.
This does seem to really be a binary ambiguity, though. I've tested
a few native english speakers on the phrase "some dogs love everyone;
indeed, chihuahuas do", and they report understanding the intention, but
experiencing some surprise on reaching the second clause and having to
re-evaluate the first clause to refer to kinds rather than individuals.
Further evidence for its binary nature:
*"many dogs love everyone; indeed Barney does, chihuahuas do..."
is, I think, semantically anomalous.
I don't see why we should import this ambiguity to lojban. Even apart
from all the problems it causes which it doesn't cause in english ('I
hate dogs' doesn't imply 'I hate one or more dogs', even when
interpreted in the same domain of discourse), I would think it
unlojbanic to have a binary ambiguity - especially one which can't be
straightforwardly and clearly disambiguated.
> > You seem to be saying that D12 is an intrinsically unnatural domain for
> > lojban. That seems to be a difference from english.
>
> I'm saying a domain like D12 needs extra work. For example, instead of
> a single predicate "tciuaua" we need two predicates, to go with the
> English "... is a chihuahua" and "... are chihuahuas", which we would
> have to use instead of the plain "tciuaua" to properly restrict the
> quantifier.
Yes, I think something like this might be the solution. We allow domains
like D12 as a matter of course, and have a new predicate corresponding
to the kind sense of "... are chihuahuas" - i.e. meaning "is
a (non-strict) subkind of the kind 'chihuahuas'". It and {tciuaua} should be
mutually exclusive.
If we have a way of getting explicitly at the kind 'tciuauas', we can
then use {klesi} to get at subkinds.
So we need something corresponding to Chierchia's down operator. This
shouldn't be {lo'e}, because that's about genericity. I'm wondering
whether it could indeed be {lo} - using Chierchia's type-shifting (which
is similar to Carlson's quantification over stages) to get back to
existential quantification over instances. The crucial change from what
you've been suggesting would be that although {lo broda cu broda} would
hold, and {lo broda} would be referring to an individual in our
universe, it would *not* follow that this individual satisfies the x1 of
{broda} in the usual sense - rather, {lo broda cu broda} would transform
by type-shifting to {su'o da poi [instance-of] lo broda cu broda} (where
{me} might or might not work for [instance-of]).
I'll read some more of Chierchia and see if I can come up with
a proposal along these lines which would satisfy us both (and hopefully
everyone else).
> > (In fact, I *would* like to claim that 1L logically implies 2L, because
> > I would still like to analyse {zo'e} (but not {lo}) as in the subject
> > line of this thread. But that's beside the point.)
>
> So you would like to claim
>
> D1 |= 1L
> implies D1 |= 2L
>
> and that D1 is a natural/preferred domain for 1L.
>
> If that's how you want to analyse "zo'e", you still have to account
> for the "obvious" as opposed to the "irrelevant" sense of "zo'e". As
> in:
>
> - xu do djica lo nu klama lo zarci
> - u'u mi na kakne lo nu klama .i ei mi klama lo drata
> "Do you want to come to the market?"
> "Sorry, I can't go. I have to go somewhere else."
>
> That should not come out as "I can't go anywhere."
Hmm. Yes, it isn't a simple existential quantifier. But how about just
having the domain of the existential quantification be contextually
determined - i.e. the quantifier is "for some xs such that context
suggests I would likely be talking about xs here". The domain
'everything' would always be plausible; other domains like {the market
we just mentioned} or 'all markets' would be plausible in certain
contexts.
Martin
Attachment:
pgpA9OZ6Tr6wg.pgp
Description: PGP signature