Re: Brainstorm! (two years too late)

la nitcion. cusku di'e

> So it's like this. You want to append a relative clause to a conjunction of
> sumti. Like, you're talking about Emma Thompson and Kenneth Branagh who are
> actors. If you say
> 
> 	 la .ebu.ty. .e. la ky.by. noi xe draci

Minor nit:  either "la .ebutys." or ".ebu ty.", preferably the latter.
LA+sumti is not allowed.  (Nick knows this, but I want to reinforce it for
others.)

> you mean
> 
>	 E.T., and [K.B who is an actor].
> 
> You want to say both are actors.
> 
> Until now, the only real option has been to say
> 
> 	lu'a la .ebu.ty. .e. la ky.by. lu'u noi xe draci.
> 
> But you need a "lu'a" before you get started. You can't
> change the scope of the following rel.clause, from the sumti it immediately
> hooks on to, to the conjunction of sumti, as an *afterthought*. You can't
> say the equivalent of "E.T. and K.B..... oh, yeah, who are both of them
> actors", because that "who" is bound to Branagh perforce, once you neglected
> to start the phrase with "lu'a". In Lojban, you're currently forced to say
> the whole phrase from the start --- which doesn't match what you can usually
> do in such cases in the language.

[...]

>  [Consider a]n elidable terminator for EK-joined sumti.
> 
> Let's call it XOI. (It could be a CVhV no problem; won't be used all that
> often.) It would change the existing sumti scheme as follows:
> 
> sumti = sumti-1 [(joik # | ek #) sumti-1 /XOI/] ...
> 
> sumti-1 = sumti-2 [ek [stag] BO # sumti-1 /XOI/]
> 
> You could then say:
> 
>	la .ebu.ty. .e. la ky.by. noi xe draci
>	ET and (KB who is an actor)
> 
>	la .ebu.ty. .e. la ky.by. xoi noi xe draci
> 	(ET and KB) who are actors
> 
>	la .ebu.ty. .e. la ky.by. xoi noi xe draci ku'o .e. 
>		la margrit.tatcer. xoi noi brito:
>	((ET and KB) who are actors, and MT) who are British.

Both lojbab and I thought this idea was indeed worth pursuing, and I've just
been doing some YACCing.  Well, it won't work as specified, unfortunately,
but I think the basic idea is salvageable.

We can indeed add an elidable terminator like XOI above, but it doesn't do
the necessary, because down in sumti-4 we still have only a sumti-5, not
a full sumti.  And we cannot change this without multitudinous reduce-reduce
errors.  Without such a change, the elidable terminator is useless.

I experimented a bit, and finally came up with the following idea.  We
demote all the existing sumti rules by one notch, and add a new top-level
kind of sumti, with rule:

	sumti<90> = sumti-1 [XUhO relative-clauses]

This cmavo XUhO is not an elidable terminator, but simply serves as glue between
a full sumti and one or more relative clauses which are intended to modify
the full sumti.  This allows:

	by .e cy noi broda cu brode
	B and (C who is a thingummy) are whatchacallits

	by .e cy xu'o noi broda cu brode
	B and C (who are thingummies) are whatchacallits

but not

	*by. .e cy. xu'o noi broda ku'o .e dy. xu'o noi brodi cu brode
	(B and C (who are thingummies) and D) (who are all furgles)
		are whatchacallits.

because allowing this "XUhO + relative-clauses" at more than one level
causes shift-reduce errors.  So we can afterthought-modify a whole sumti,
but not part of a sumti.  I think this is an acceptable restriction.
Here's the formal change proposal:

CHANGE 34

CURRENT LANGUAGE:

Relative clauses always attach to the simple sumti (sumti-5) immediately
to the left; the only way to attach a relative clause to a larger construct,
like logically or non-logically connected sumti, is to bracket the construct
with LAhE...LUhU.

PROPOSED CHANGE:

Introduce a new cmavo "xu'o" of selma'o XUhO (note: assign non-experimental
value later) which serves as glue between a full sumti and one or more
relative clauses (joined with ZIhE, if there's more than one).  The new
construct would have lower precedence than any other sumti construction.
Therefore, it would be possible to attach relative clauses to a complete
sumti (with XUhO), or to a simple sumti (with no cmavo), but not to anything
in between.

RATIONALE:

The LAhE...LUhU construction has two limitations.  First, it is inherently
forethought in nature, making it impossible to add long-scope relative clauses
in afterthought.  Second, LAhE cmavo have semantic import, so when used
merely for bracketing purposes, it is important that the correct cmavo
("lu'a" for individuals, "lu'o" for masses, "lu'i" for sets) be chosen
so as to have a null semantic effect; this is annoying.

Comments?

-- 
John Cowan		sharing account <lojbab@access.digex.net> for now
		e'osai ko sarji la lojban.