On 15 October 2015 at 09:45, R A Brown <[log in to unmask]> wrote:
> On 15/10/2015 10:03, R A Brown wrote:
> [snip]
>> Um - I must return to my Glossopoeia page and redefine
>> 'loglang', I think.  but it will have to wait as I am
>> actually in the process of updating stuff on Britainese
>> ;)
> I think it is not only Britainese holding me up on this, but
> also some other points that need rethinking.
> On my 'Glossopoeia & Glossopoeic Languages' page, I wrote:
> {quote}
> ... Thus we find in the early Conlang period loglang and
> loglan used interchangeably to mean "engelang." This is all
> very unsatisfactory.  In this section I used Loglan to mean
> JCB's language as defined by The Loglan Institute, and
> Lojban to mean the language, originally derived from JCB's
> early Loglan, and developed by the Logical Language Group.
> The term 'loglang' is now properly confined to those
> languages which are based on formal logic - in practice
> predicate logic. The term 'loglan' is sometimes used to mean
> any conlang ultimately descended from JCB's original
> language. Thus Voksigid (1991-1992) may be termed a loglan,
> but many feel it has departed too far from formal logic to
> be termed a loglang.
> {unquote}
> I no longer feel that using loglan as a generic term for any
> language that is ultimately descended from JCB's original
> Loglan helpful.  I think the only unambiguous generic
> noun is _loglang_.  Therefore, the above will be changed
> and, as I wrote in my previous email, my definition of
> 'loglang' itself will be changed.
> Let us accept for the sake of argument what And wrote, namely:
> "Natural languages encode [predicate--argument structure]s
> of limited complexity and do it ambiguously."
> and:
> "A loglang ... is a language that unambiguously encodes
> predicate--argument structure of unlimited complexity."

Personally, while I like And's definition for it's simplicity and
clarity, I think it a bit too narrow.
I am uncertain just how much it can be broadened without losing all
practical meaning, however. If, for example, we were to drop the
"unlimited complexity" part, then nearly all natlangs automatically
qualify! But, nevertheless, it feels like a language that allows, for
example, expression predication graphs of arbitrary size but finite
maximal degree (i.e., you can have as many distinct referents as you
want, but each one can only be an argument to a finite number of
predicates/relations) ought to count. Or, flip it around the other
way- allow a finite set of referents and infinite relations on them;
and if the finite set of referents is small enough, then most natlangs
qualify, which seems wrong, but it still feels like there is a
qualitative difference if the set of referents that can be
unambiguously distinguished is, say, 50, rather than 3 or 4.

Broadenings that I am more comfortable with would be to say that a
loglang is a language which *can* encode arbitrary structure
unambiguously (thus allowing that a loglang may also allow ambiguity
if you want it, which is probably important for usability), and that a
loglang is a language which is *intended* to encode arbitrary
structure unambiguously. Although, it may after all be a good idea to
leave intent out of it and just refer to "failed loglangs" (using an
anti-intersective adjective!) for those languages which were intended
to be loglangs but aren't, under the remainder of the definition.

> AIUI modern predicate grammar sees predicates as relations
> or functions over arguments.  Predicates are placed on the
> left outside of brackets, whereas the predicate's arguments
> are placed inside the brackets; and the number and position
> of the arguments is determined by the valency of the predicate.

I don't think I have seen more than two sources that ever use exactly
the same notation for all the More Complicated Stuff (logical
connectives, quantifiers, higher-order operators and reification,
etc.), but that does seem to be a pretty standard baseline.