On 12 August 2015 at 11:42, And Rosta <[log in to unmask]> wrote:
> Logan Kearsley, On 12/08/2015 16:43:
>> On 12 August 2015 at 07:46, And Rosta <[log in to unmask]> wrote:
>> [...]
>>> Logan Kearsley, On 09/08/2015 20:43:
>>>> a usable language must encode two different kinds of information:
>>>> the basic conceptual atoms out of which more complex ideas are
>>>> built, and structural information that tells you how those atoms
>>>> are combined and produces compositional semantics.
>>>> An isolating bicategorial language divides those two functions into
>>>> two parts of speech. And it probably says something about human
>>>> cognition that there exist natlangs with exactly that split,
>>> Really? Are you thinking of Riau Indonesian? If so, how does its
>>> bicategoriality correspond to the two logical kinds of information?
>> I was not thinking of any one language in particular at the time[1],
>> but of the time when one of my professors (who did field work
>> documenting dying languages in South America) made an equivalent claim
>> and listed off a few examples which  have totally forgotten. My entire
>> knowledge of Riau Indonesian comes from papers by David Gil and a few
>> papers by other people that are direct responses to David Gil, but
>> based on that small sampling, it does actually seem to fit pretty
>> well. There is an open class of S's containing content words, which
>> cover the ontological categories of EVENT, STATE, TIME, PLACE, THING,
>> "and possibly others"; and there is a closed class (notated S/S under
>> the conventions of categorial grammar) of function words which
>> contains negators, quantifiers, abstract prepositions, "and some
>> others".
>> There may be an interesting challenge in creating a bicategorial
>> language that somehow divides the vector space of content+function
>> words with wildly different bases.
> I understand that these bicategorial natlang grammars have categories
> Open/Content and Closed/Function, and I'd assume that Open/Content items all
> encode Basic Conceptual Atoms. What would surprise me would be if the
> Closed/Function items all, or even mostly, encoded Structural Information
> (namely, which arguments occuply which argument place of which predicate).
> For example, negators and quantifiers look to me to be Basic Conceptual
> Atoms.

I sorta-kinda see where you're coming from there, but putting on my
Formal Semantics hat, and looking at things through the lense of
predicate calculus, those are the kinds of things that get special
operators assigned to them, which are not encoded as predicates. And
I'm sure they *could* be, It is difficult in my current frame of mind
to think of them that way; it certainly seems like they naturally
belong to separate category.

> On a different but related point, I wonder if that Closed/Function class can
> really validly be treated as a single class (in whichever languages it is
> posited for). It implies that there are certain generalizations that apply
> only to all members of that class. Minoritarianly, I think PoSs can
> cross-classify (-- at least, that's the undisproved null hypothesis),
> allowing for the possibility of intersection between Function and Content,
> so I'd see a difference between generalizations that apply only to all
> Function items and generalizations that apply only to all items that aren't
> Open. It's imaginable; but simply showing that some items are Open and some
> aren't isn't enough to establish bicategoriality.

I think that, for any natural language, this is probably an area in
which there will always be multiple possible analyses, with little
possibility of reaching universal consensus on the proper one.

Categorial grammar is a particularly useful framework for
investigating this kind of thing. By naming "parts of speech" in terms
of syntactic formulas, it helps highlight exactly the distinctions
that you are talking about (assuming I have understood you correctly),
which get muddied when using traditional terminology form traditional
parts of speech. While it is conceivable that there are additional
as-yet undescribed features of the language that don't fit into these
generalizations, Gil's claim about Riau Indonesian is that all items
of category S belong to an open class; there are no open-class items
that do not belong to the category S; and all items that are not part
of an open class belong to the category S/S, where S = "this can form
a sentence by itself" and "S/S" means "this can form a sentence if
combined with something else that can form a sentence by itself". (And
I'll note that while I have previously been using "class" and
"category" essentially interchangeably, for current purposes
"category" means "syntactic category" as defined in the framework of
Categorial grammar.)

Thus, there is one group of words that has all three of the properties
of a) being open-class, b) encoding "content"-ish concepts, c) having
identical syntactic behavior S, and a second group that has all three
of the properties of a) being closed-class, b) encoding "function"-ish
concepts, c) having identical syntactic behavior S/S. And maybe there
actually are some content-ish things that get thrown into the closed
S/S class after all, but that's just because natlangs are messy; the
vast majority of lexical items fit the pattern, and thus it is fair to
say that "this is how the language works", with footnotes and
appendices to explain that, yeah, this is a natlang, and it has some
weird edges that haven't been filed off.

I don't know what the exact arguments are for the remaining list of
dying South American languages whose names I forgot anyway, but I
imagine there are probably good ones.

> And I guess that bicategoriality in a strong sense means that there are no
> more than two classes, so no subclasses. That seems particularly unlikely
> (and correspondingly particularly interesting if true).

Well, my empirical knowledge of actual natlangs runs into a wall here;
I don't actually speak any of the claimed-bicategorial languages, so I
can really just parrot the analyses that I have heard / read from
other people. If I had to bet, I'd bet that all bicategorial natlangs
do have some sorts of natural subcategories, if only because natlangs
are bound to be messy. But Patrick's FL2 (what I was calling "Hodor",
because that was the title of the thread) could easily be fleshed out
in a way that shows no subcategorization at all.