Message-Id: <199511202011.PAA01610@locke.ccil.org>
Date:         Mon, 20 Nov 1995 14:56:31 -0500
Reply-To: Logical Language Group <lojbab@ACCESS.DIGEX.NET>
Sender: Lojban list <LOJBAN@CUVMB.BITNET>
From: Logical Language Group <lojbab@ACCESS.DIGEX.NET>
Subject:      misc. responses to And
To: John Cowan <cowan@LOCKE.CCIL.ORG>
Status: OR

And answering Jorge and me and maybe others:
>Lojbab:
>> >If you turned a grammarian loose on Lojban they'd come up with a
>> >different grammar from the official one, because Lojban has fiddled
>> >the initial data. The current official grammar overgenerates.
>> And natlang "official grammars" undergenerate if they follow Jacques
>> Guy's principle and exclude center-embedding, or overgenerate if they
>> follow Chomsky's and include fully-recursive center-embedding.
>
>The status of centre-embedding is a rare, atypical, and theoretically
>contentious issue.  It is (in practise, at least) of rather marginal
>relevance to how grammaticality is delimited.

It was an example.  Presumably most recursion and repetition in natlangs
has limits, but Chomskyan formal grammars, just as Lojban's formal
grammar, presume no such limits.  (Hey, we're on Chomsky's side of the
issue for once %^)

My understanding is that virtually every word in natlangs is a selma'o
of its own in reality.  Generalizations such as "adjectives", "noun
phrases" etc.  'work' only as 1st order approximations, for every such
grammatical structure you can probably generate some unacceptable
nonsense.  Only a relatively SMALL percentage of Lojban grammar rules an
generate this kind of nonsense, and most of it is embedded in the rules
associated with the lexer rather than with the grammar in general.  So
at worst, your complaint is mostly that the lexer portion of the grammar
may be oversimplified.  Still, it is very hard to categorically state
that any given string generated by the lexer has NO meaning.

Nora's random sentence generator actually does pretty good at avoiding
uninterpretable nonsense caused by wild PA strings, which suggests that
this is actually a fairly well-ordered part of the grammar.  It is far
easier to get such nonsense out of other parts of the grammar (mostly in
terms of iterative structures that make no sense for the random words
that were chosen).

Maybe I'll liven up the list with some random sentences of various
difficulties so people can see for themselves that most of the semantic
nonsense derives from the words themselves and not from the formal
grammar.

>> WhatIS the case is that there exists grammatical Lojban text using PA
>> that has no defined and agreed-upon semantics - that does NOT mean that
>> it is utter gobbledygook
>
>This is false by any standards other than lojbo ones.  This is exactly
>my objection.  In natural languages, any construction with no defined
>and agreed-on semantics is ungrammatical.

1. Jabberwocky

2. Curious green ideas sleep furiously.

3. The US tax code

Let me know when we agree on the semantics of any of these, and the
semantics of each is to some considerable extent undefined (especially
the tax code %^), yet they are considered grammatical.

>My objection to {duu} is that it is always singleton in extension, so
>should have sumti rather than selbri status.

Since du'u is a relation between a bridi and its expressed form, there
are at least as many extensions as there are ways of expressing a bridi
(perhaps infinite).  "lo du'u" refers to the bridi itself as a bridi,
which tenselessly could be said to be a singleton, but it IS a sumti.

But what makes singularities necessarily sumti, anyway?.  I am sensing
the kind of logic that would say that there can be no 1-place predicates.
(munje?)

>My objection to {nu} is that really the event is an argument of the
>bridi, so {jai fau broda} is truer to the meaning.

If the event is an argument of the bridi that is reflected in the event
I see infinite recursion.

And since jai fau is a sumti raising that is defined in terms of the
abstraction in the fai place you get recusrion in that sense too.

Since "fau" is built off of "fasnu", what is the x1 of fasnu, if not a
"nu abstraction".  You MIGHT argue that no other nu abstractions are
needed (though I would argue in the cases of gasnu and zukte as well
this is false, and you might get real wordy in expressing dome of the
others), but that still makes "nu" useful.

>My objection to {ka} was that it should be a sumti tail, not a selbri,
>but now that I get an inkling of its relationship to {duu} I guess it
>should be a sumti.

I don't understand this well, not having followed the recent discussion,
but it is not clear that ka abstractions of a given bridi are singular
in extent so your argument above for du'u doesn't work here either.

back to the beginning for one out-of-place sentence to respond to:
>(3) Similarly, the case for selmao NU is pretty weak.

You keep saying this, but I think that NU, and indeed "nu" is one of the
major genius strokes of the language.  The fact that the number of
places in gismu that are abstractions alone shows how useful it is to
recognize multiple levels of abstraction in analyzing semantics.  I like
"nu" itself simply for the the capability of bundling up an occurance of
a bridi relationship as a semantic unit, and the subcategories allow me
even to define a substructural way of looking at that occurance.

I don't think my mind ever "expanded" so much from working on Lojban, as
when pc finally got across to me the distinction between
states/achievements/processes/activities and we started to explore the
semantic effects of assigning different words to the different
distinctions.  These distinctions are probably not used much yet simply
because there IS nothing remotely like them in English and perhaps in
other natlangs.

And the whole paradigm behind the set of perfectives derives DIRECTLY
from the 4-way subcategorizing of "nu" abstractions, as well much of the
rest of the tense system tied to intervals and substructures.  Again,
these are not much used because of the lack of natlang equivalences.

>So my objections are not only that most members of NU are not useful,
>but also that only {sio} really warrants the syntax of NU (i.e. selbri
>containing bridi).
>...
>> (As an aside, I don't think that something like {mi nelci le ka do melbi}
>> makes any sense. It should be {mi nelci le nu do melbi}.
>
>I agree.

Then "nu" is useful?

>> {za'i}, {zu'a}, {mu'e} and {pu'u} are all kinds of {nu}. I don't think
>> I've seen any example where using any of them makes anything clearer than
>> simply using {nu}. ({za'i} is the most popular, perhaps because the
>> word "event" makes people think that {nu} has to be something dynamic,
>> but this is not really the case. {nu} is perfectly fine for even the
>> most static of states.)
>
>I agree. Someone had been reading linguistics textbooks when these were
>added, and gave them too much weight.

Actually, I hadn't read ANY linguistics textbooks when I added them.
The reason for adding them was that "nu" is sometimes too vague,
especially when dealing with tenses.  (And since tenses are optional,
the use of the right event abstraction may carry more information than
the use of a lot of tense words.)

In my mind, process abstractions are more 'opposite' to states than
'activity' abstractions, since states and activities are presumably
uniform while a process probably is not.  But the
event/achievement/state and processes of running a race are all to my
mind quite different concepts.

"nu" seems to focus on the nounal aspects of an event, whereas its
subcategories focus more on the verbal properties.

>> {jei} has two meanings: by official definition it is a truth value. By
>> usage it is the yes/no indirect question "whether", equivalent to
>> {du'u xukau}.
>
>If every truth-value is unique to the proposition it is truth value of,
>then {jei} works as "whether".  But I don't know anyone who thinks that
>truth-values are thus unique.

How about the converse - every proposition has a unique truth value?

>> I find the definition as a truth value totally out of place among
>> the NUs. There should be a lujvo that means "x1 is the truth value
>> of x2". The usage definition is ok, but redundant to {du'u xukau}
>> and not so frequent that it is worth the trouble to have a short form.

I could accept the euqivalence to such a brivla, with x2 being a du'u
abstraction.  But jei has been around a LOT longer than du'u, which is
itself an exile from the first attempt at a MEX grammar (it meant an
equation in the way that me'o now refers to expressions, which in turn
has many similarities to the current "se du'u")

>Well "unspecified abstraction" is a pretty useless meaning.  It could
>mean "a person who has asserted that p", or anything.

Well, it is OBVIOUSLY the inverse of tu'a and jai.  Probably when people
have fully mastered the semantics of those two, they'll find uses for
su'u as well %^)

>Goran:
>> > {zilxru}: I meant {zil} to delete x1 of {xruti} - i.e. go from
>> > "x1 returns x2 to x3 from x4" to "z1=x2 returns to z2=x3 from z3=x4".
>> xorxes was right: zilxru *was* an abomnination.
>
>Well what are {zio} and {zil-} for, if not for cases exactly like
>these?

Well some of us contend that they are abominations, to be tolerated in
others but never used %^)

In any event, deletion of the x1 of xruti has become consensus (it was
proposed over a year ago, but Cowan wasn't around and it became lost in
my 15 month mail backlog.  Hmm, maybe when my mail backlog reaches 5
years we will have achieved de facto a 5 year baseline %^).

lojbab