On Wed, 14 Nov 2012 13:51:37 -0700, Logan Kearsley <[log in to unmask]> wrote:

>On 14 November 2012 02:41, Alex Fink <[log in to unmask]> wrote:
>> Two important choices already are:
>> - What case system shall we use?  We want one that allows the defaults to be used in the greatest proportion of pragmatically arising sentences.  Among other aspects, this subsumes the question of pivots.  For instance if I want to be able to say
>>     nurse(x) & teacup(y) & throw(x, y) & break(y)
>> "the nurse threw the teacup and it[sic] broke" with default argument--variable mapping on "throw" and "break", I've got to be using ergative alignment.  Or again, is it worth it to treat animate nounish predicates and inanimate ones differently, so that "nurse" and "teacup" might also both be default-marked here?
>If case markings are standing in for variable
>bindings, I would suggest going with a trigger-style system, so
>"throw" has some morphology that says its positional arguments are NOM
>and ACC respectively, and  "break" has morphology indicating that its
>sole argument is ACC.

Yes, I guess triggers would be another natural-language phenomenon that interpolates between case and voice in the way I want (except for the pesky fact that they aren't a natural-language phenomenon, as conlangers understand them).  But yes, notionally I want "break" to have morphology indicating that its two arguments are NOM and ACC.  But I was also aiming for sensible defaulting to minimize the obtrusiveness of all that.  

This requires some lexical specification -- you do have to know that "break" has a NOM breaker and an ACC breakee by default, not some other cases -- but it's nòt more information than e.g. Lojban, where you have to know that "break" has a slot-1 breaker and a slot-2 breakee by default.  And one of the premises of this design experiment is that ordered lists are worse than, and should if possible be eschewed in favour of, associative arrays.  In fact, if the predicates dìdn't have default cases or semantic roles or something, then even the trigger approach would probably need an ordering to make the morphology work.

>At that point, however, I find it much more natural to associate
>variable bindings not with the concept of case, but with pronouns.

I'm not making a big deal over that difference.  Think of nominative as just the name of a variable, if you like. 

>Each predicate would simply be marked with a pronominal clitic string,
>formed from a closed set of possible "variable names" for the language
>(perhaps corresponding to first, second, third proximate, and third
>obviative; the set could be expanded with, say, definiteness, or
>encoding of stuff that could be put into predicates but is common
>enough to warrant a special pronouns, like gender). 

If you had allowed a handful of third-person pronouns with no restrictions on them, you would've just reinvented (the core of) Xorban.  I guess my case-oid language differs from yours wrt Xorban in that, where Xorban's pronouns are undistinguished, I'm distinguishing them by what (verby) semantic role they prefer for their referents, and you're distinguishing them by, er, painting one of them proximate, and further distinguishing them by what (nouny) class of referent they prefer.  

One could add a proximate to my system and get something quite trigger-esque, which is maybe interesting to pursue.  Otherwise, hm, if either of these verby or nouny pronoun schemes were to do better than random at lining up with pragmatically useful things to say, it's probably rèally because they correlate well with _each other_.  So maybe that idea is close kin to mine.  

I think that avoidance of situations where pronouns are reused is slightly better in my verby than your nouny scheme: you can make slightly simpler sentences that contain two human female participants which need distinguishing ("Jane saw Karen"), say, than contain two patients which need distinguishing ("I put_on my shirt and combed my hair").  But, then, proximacy.  Maybe it's a wash.  

>That would limit
>the total number of distinct argument referents you could have per
>clause, but that's plausibly not that big a deal. It could perhaps be
>extended with an incorporation mechanism to say "I only need to use
>this referent once, so I'm just putting a defining predicate for it
>directly in something else's argument slot without bothering with
>variable binding". So, say, "the man in the house" might be encoded as
>"man(a) &&amp; in(house,a)" rather than "man(a) & house(b) & in(b,a)" if
>you don't need to refer to the house elsewhere in the clause.

Mhmm, that's one option.  Palno-ish.  

>There's also the possibility of introducing unambiguous quantifiers
>per pronoun per clause, which is a very loglangy sort of feature. 

Incidentally one might distinguish "loglang" as And uses it, "permitting unambiguous extraction of the logical form", and "loglang" = "language which flaunts the trappings of FOL".  UNLWS, for instance, is not all that far from a loglang in the former sense, in its predicate-basedness, but it's far from the latter: for instance, logical operators / Boolean connectives are not a class in it and basically only "not" is present non-compositely.

>direct-from-predicate-calculus way of doing that would probably be to
>split your clause structure into three parts- assertions, quantifiers,
>and qualifiers for the quanitifiers. E.g.,
>  "All teacups thrown by nurses break" -> Assertions: break(y)
>Quantifiers: For-all(y) Qualifiers: nurse(x) & teacup(y) & throw(x,y)
>With existential quantifiers being assumed by default for any pronouns
>But I just had an interesting new idea: quantifiers as
>morphology applied to the asserted predicates. So, if you introduce a
>quantifier on any predicate with an argument 'x', it would mean "this
>is true given x's that satisfy the unquantified parts of this clause",
>thus allowing mix-and-matching without the three-way clause structure
>division, which seems rather un-natural to me. E.g.,
>   nurse(x) & teacup(y) & throw(x,y) & ALL-break(y) - "I assert that
>all y's satisfying the unmarked predicates in this clause also satisfy
>Not sure how you'd handle embedded subject/object clauses.

Your three-way structure is a bit heavy presented like that, sure, but I think it's basically the fact that "all" wants to govern a scope.  And probably some things dò need scopes.  You're fudging it with the quantifiers-as-morphology thing, which (perhaps this is what you're saying) works cleanly if the assertion is a single predicate, but would need an extra dodge if it is multiple.  Actually, though, this is a lot like a syntactic problem UNLWS is having at presenting, regarding scopes of modal operators.  

Anyway, quantifiers are scary.  Donkey sentences and all that... though perhaps your suggestion handles donkey sentences well enough, if ALL knows which of its arguments it goes on.  Thus "... & break(ALL-y)"; "farmer(x) & donkey(y) & own(x,y) & beat(ALL-x,y)".

>> Then there's the problem of subclauses reusing variables, which I can't entirely wriggle out of even with a good case system: take "the cat ate the rat which ate the grain".  So I guess either we have whole series of secondary cases for conflict avoidance in subclauses, as in
>>     cat(n) & rat(a) & eat(n, a) & grain(a') & eat(a, a') ;
>> or we allow some variables to be reused within subclauses, which on the face of it seems to have the better naturalistic analogy, relative pronouns:
>>     cat(n) & rat(a) & eat(n, a) & relpron(a; [n) & grain(a) & eat(n, a) ] .
>> The relpron says that what is  a  on its outside is  n  on its inside; the misnested brackets express uncertainty about how to best realise that in syntax.  Actually I guess I could also just give up and go analytic here and use two clauses, assuming that the scope of a variable binding is a clause:
>>     cat(n) & rat(a) & eat(n, a).  n=a_last & grain(a) & eat(n, a).
>> Herein n=a_last is some kind of short-range anaphoric pronoun.  It feels somehow inelegant to resort to that so soon, but maybe that's an accidental effect of being raised on ultimately Latinate notions of elegance or whatnot.
>If you use pronoun-binding instead of case, that problem goes away.

How?  In a substantially different way to my variant with a' above?

>But I would point out that "n=a_last" is pretty much *exactly* how
>lots of languages actually handle relative pronouns, so hardly
>inelegant. E.g., Russian "kotorij" = "the last thing that was
>masculine singular? Forget its old case, now it's nominative". Use
>double case agreement rather than gender+number for the previous
>clause / case for this clause, and you've just got Suffixaufnahme,
>which everybody thinks is cool anyway.

Yeah, that's kinda what I was getting at with my "Latinate notions of elegance": there might be no good reason to favour the "everything possible should be one sentence" style.  Actually, to swing to the other extreme, I wonder how usable it would be to make as many clauses as possible _one_ predicate, and rely on a set of pronouns whose values default to staying the same from clause to clause unless a pronoun moves them around.  

>*I note that sentence can be interpreted as implying that I myself am
>a grad student. That is not quite yet the case, but unless I'm
>strangely unlucky, it should be next year.

Bonne chance!