Message-Id: <199310040659.AA17464@eli.CS.YALE.EDU>
Date:         Mon, 4 Oct 1993 02:58:11 -0400
Reply-To: Logical Language Group <lojbab@ACCESS.DIGEX.NET>
Sender: Lojban list <LOJBAN%CUVMB.bitnet@YaleVM.YCC.YALE.EDU>
From: Logical Language Group <lojbab@ACCESS.DIGEX.NET>
Subject:      TECH: long, but major topic: lean lujvo and fat gismu
To: Erik Rauch <rauch>
Status: RO

mi'e je'abo la lojbab

There seems to be some back-channel discussion about the concepts of
Lean Lujvo and Fat Gismu, primarily in the context of discussions of
Nick's lujvo paper and comments thereon.

It has been suggested that this is a topic for the list, so I won't wait
for someone else to bring it up.  We will also presume that my few
minutes scan of the comment flow that people have generously kept my
mailbox full with, has given me some idea what is being talked about.

Thus if I understand, "gerku" is an example of a "fat gismu".  It has
the places x1 is a dog of species/breed x2.  It is argued on at two
spearate bases that the x2 is superfluous.

Argument #1 is that, when you use such a gismu in a lujvo, you have the
troubling problem of deciding whether to include the species/breed
place:

gerku ger ge'u dog x1 is a dog/canine of species/breed x2 ai 175 (cf.
lorxu, labno, mlatu)

zdani zda nest 'den' x1 is a nest/house/lair/den of/for x2 4d 280 (cf.
dinju, ginka, kumfa, xabju)

Thus is a doghouse (gerzda)
x1 is a gerzda (house-for-dogs)

or

x1 is a gerzda (house-for-dogs) of species/breed x2

It is argued that while dogs might indeed be categorized by
species/breed, dog-houses aren't, and the place should be omitted.  But
then how do you judge what to include or exclude?

If you argue for the "leaner" lujvo. without the extra place, someone
can argue that a doghouse for a St. Bernard is not likely to be
appropriate for a Chihuahua.

Moreover, neither of these expresses what we USUALLY want to say for a
doghouse, which is that

x1 is a gerzda (house-for-dogs) of-dog (specific) x2

which is combining the x1 of dog (the dog) into the x2 of house (the
resident), and it is perhaps unusual in lujvo-making to want to
recapture the x1 of the modifier in the lujvo.

Thus we have another option, a "really fat lujvo"

x1 is a gerzda (house-for-dogs) of-dog (specific) x2 of species/breed x3

and you could probably come up with SOME circumstantial doghouse where
you want to preserve the x2 of zdani independent of filling in the dog
either by species or individual.  So maybe you need 4 places in the
lujvo place structure.  One can easily imagine in a four-or-five-term
lujvo, that the number of places gets ridiculous rather quickly.

Thus the desire for Lean Lujvo and for relatively predictable place
structures argues for the minimizing of the gismu place structures to
make such fatness in the lujvo impossible.

Argument number 2 against "fat gismu" has to do with use of the gismu
standing alone.  When you say "loi gerku", are you saying "the mass of
dogs" or "the mass of dogs of species/breed x2, where x2 is some
specific value zo'e but happens not to be specified", i.e. only the mass
of dogs of one breed.

If I have misrepresented the arguments against fat gismu, I'm sure I will
be corrected %^).

Argument number 2 seems easy to discard.  I will, however, credit Nora
with the demolishment.  I knew there was something wrong with it, but
couldn't figure out what.

There are actually two intertwined flaws.  First of all, the argument
seems to assume that the zo'e unspecified is some singular identifiable
value.  But this need not be the case.  The zo'e in x2 could expand to
la daxysxund (maybe we oughta permit 'xs' as a permissible medial after
all %^) joi la bigl joi la dobrmn joi ... specifying all possible breeds
of dogs.

This still leaves a queasy feeling in me though, perhaps because we
English speakers don't very well think in terms of masses.  However, the
more solid argument turns out to be that there is nothing special about
dogs and their "fat gismu" places in terms of this problem.

"loi klama" after all is (we hope) "the mass of all go-ers", and NOT
"the mass of all go-ers to a specific though unspecified place", much
less "the mass of all go-ers to a specific though unspecified place from
a specific though unspecified place via a specific though unspecified
route by means of a specific though unspecified mode of transport."  If
this argument could be used against "fat sumti places" in gismu it would
have to be used against all nonspecified places of gismu, thus in fact
arguing that all gismu are essentially one place predicates.

This would rather spoil the whole idea of Lojban.  So it can't be right %^)

The answer has to be that zo'e means something a little different than
is being assumed in the above.

Nora chooses to argue on the basis of "lo" rather than "loi" since we
can wrap our mind around individuals and groups thereof rather more
easily than masses.  She also uses mensi (x1 is the sister of x2) as her
example gismu.  After all it is hard to argue that one can be a sister
without being a sister OF someone.

Yet we would not argue that there exists "ro lo mensi" means "ro lo
mensi be da" (all sisters of some specific individual).

Nora suggests that the quantificational expansion must be something like

da mensi ==

su'oda de zo'u da mensi de
For at least one x, there exists a y such that x is the sister of y.


lo mensi ==

su'oda poi [roda] de zo'u da mensi de

At least one x such that for each such x there exists some y such that x
is a sister of y

I believe that you can't actually put the "roda" in this expansion under
the rules of logic since this would redefine "da".  But I'm showing how
I intend the ordering of the terms in the inner "poi" clause to be taken.
If the ordering were reversed:

su'oda poi de [roda] zo'u da mensi de At least one x such that there
exists some y that for each x, she is the sister of y.

or "at least some of all of the sisters of some 'de'"

which if you think about it gives exactly the paradoxical problem
suggested above in argument #2, that you cannot talk about a mass of
sisters, or even a group of sisters, unless they are sisters of the same
person.

But of course if two people were talking and one said
"mi mensi .ije do mensi .iseni'ibo mi'o mensi
I am a sister and you are a sister therefore we are both sisters.

We would NOT presume that they have to be sisters of some third
individual.  Indeed, they MIGHT be sisters of each other.

Similarly, if "I go to France" and "you go to France", we can say "we go
to France" without implying that we started from the identical origina,
used the identical route, as well as the identical means.

I will let all you logicians tear this one apart and/or reformulate it
in lambda calculus or whatever %^)

Returning now to argument #1, my answer is that you CANNOT
algorithmically exactly which places gerkyzdani "needs".  All of the
possibilities mentioned above are distinct and plausible interpretations
of the tanru "gerku zdani".  The one that you choose as the place
structure for gerkyzdani ought to be the one that pragmatically turns
out to be most useful.  Nick has proposed some 5 or 6 tanru
interpretation schemes that are MOST COMMONLY the basis for tanru that
are in turn the most useful basis for lujvo.  I have not entirely
accepted that these interpretations are exhaustive, though I will accept
his argument that most if not all lujvo that have been proposed may be
explained using these schemes.

But I would contend that the determination of place deletion once you
have selected one particular scheme needs a totally separate analysis
(which Nick has more or less discussed in his paper, but I get the
impression from the followup discussions that it is not treated as a
totally separate problem).

Thus the person who proposes the lujvo "gerkyzdani" has to decide just
which concept they want to represent:

x1 is a doghouse of/for dogs x2 of species/breed x3
x1 is a house of/for dogs x2
x1 is a doghouse of/for dogs of species/breed x2
x1 is a house of/for dogs

All are plausible, and each of the above in succession is a bit more
general than the previous one; i.e. there will exist some slightly
broader category of things x1 that will fit the predicate (given
specific values for any other places.

Which one is "correct?"  That has to be decided by the person who
proposes or uses the lujvo.  Each is a slightly different concept,
as is implied by the existence of different sets of values which
could meet each description.

Which should be chosen for the dictionary?  A little easier question,
but not much.  First of all, we have to make it ABSOLUTELY CLEAR that
all lujvo place structures at this point are ONLY proposals, and that
actual usage may cause them to (need to) be changed.  This is no
different than what I say for the gismu place structures and why I
refuse to baseline them even when the dictionary is published.  The
place structures are the heart of the meaning of each word of the
language, and I do not believe that we can analyze each thoroughly
enough to know the perfect place structure for usages that have yet to
be seen.

And if we formally prescribe the semantics of the language to this level
(which is fruitless, since even specifying all of the places of a gismu
doesn't fully accurately specify the semantic relationship between the
places; we are limited in the latter with defining the relationships in
English - or some other natlang - and are thus subject to all the
semantic ambiguity and cultural biases of that natlang if we DO insist
on the prescription), we rather spoil the whole point of devising a new
language.

So what to propose?  I suggest that the propensity to use lujvo as
term-bases for longer lujvo should lead us to eliminate places where
practical, i.e. make our lujvo relatively "lean".  We don't want
"gerkyzdanydinjyzbasu" (dog-house-building-maker) to have all of the
possible places suggested by the components:

gerku ger ge'u dog x1 is a dog/canine of species/breed x2 ai 175 (cf.
lorxu, labno, mlatu)

zdani zda nest 'den' x1 is a nest/house/lair/den of/for x2 4d 280 (cf.
dinju, ginka, kumfa, xabju)

dinju dij di'u building x1 is a building/edifice for purpose x2 2k 153
(cf. ginka, zdani)

zbasu zba make x1 makes/assembles/builds/manufactures/creates x2 out of
materials x3 7f 217 (cf. cupra, larcu, rutni, finti, gundi)

gerzdadi'uzba
x1 is a maker of building(s) x2 which contains house(s) x3 for dogs x4 of
species/breed x5 for purpose x6 out of materials x7

I am sure that I can envision a specialist in building dog-apartment
houses that are triplexes for particular Chihuahuas to sleep but not eat
in out of straw and bricks.

But I can't imagine wanting to talk about such oddities enough to
warrant a lujvo even as long as 4-terms for them.

So we eliminate places, in this case most of them.  I would argue that
we focus on the first two places, and eliminate places which are either
a) so rarely relevant to the truth conditions that specifying a value
would seldom constrain the usage and b) places that are less useful but
are easily specified as simple poi clauses on the first two terms.  By
the latter we easily eliminate x3, x4, and x6, which would all go in poi
clauses of x2. x5 is a sub-place of x4, and would seldom be specified
unless x4 is also specified, so it too can be eliminated by implicit
embedding in x2 poi clauses.  Only x7, the materials place cannot easily
be eliminated at least partially because it is operating at the
"outer-most" modification level, that of the final term of the source
lujvo.

So I propose

gerzdazba
x1 is a builds doghousebuildings (for houses, dogs of breed, purposes)
x2 out of materials x3

This happens to resemble the final-term (zbasu) place structure, which
is no coincidence.

(I would probably eliminate the dinju term completely, of course, since
pragmatically one doesn't build nests for dogs, but rather edifices for
them to nest in - thus the dinju is implied in the zbasu term).

Not always will the place structure match the final term.  An example
discussed in commentary was

ponsydjica own-desire

ponse pos po'e possess x1 possesses/owns x2 under law/custom x3
(cf. ckini, ralte, jitro, steci, srana, tutra, turni)

djica dji desire x1 desires/wants/wishes x2 (event/state) for purpose x3
*3l 500 [if desire is for an object, use tu'a in x2]; (cf. taske, xagji,
nitcu, nelci, pacna, prami, rigni, trina, xebni)

for which I suggest

x1 desires that (x1) own x2 under law/custom x3

The law/custom and the thing owned comes from ponse, and not from djica,
and I have dropped the purpose place of djica.  The fact that there is
an abstract sumti (the x2 of djica) being raised to separate sumti
places explains why these have been added.  I eliminated the purpose
place because the number of purposes for wanting to own something are a
rather distinct and small subset of the purposes for wanting some
indeterminate state/event, and indeed are probably the same as the set
of purposes for actually owning x2 (as distinct from wanting to own x2).
The purposes of ownership are NOT part of the place struture of ponse,
but would be specified if appropriate using a BAI sumti tcita place
added in.  Since there is a readily available BAI place, it is
preferable to leave the place out.  (Sometimes a place will show up in
multiple places in a source tanru - you might have a purpose attached to
two different terms - in which case it is harder to delete both of them,
and more necessary to be very clear which "purpose" you mean when you
specify the lujvo place structure.)

I tend to like to delete places when the option exists to add back in
(relatively unambiguously) using a BAI tag.  I will always delete a
place if the abstract concept I have in mind doesn't metaphysically
require the place.  (i.e.  I see no particular reason to use "zi'o", the
new place-deletor to eliminate a place.  I would be more inclined, if I
wanted a "law-independent" desire for ownership above to coin a lujvo:

flalyclaxyponsydjica (law-without-possess-want) or even
terponsyclaxyponsydjica (law-of-possesion-without-possess-want)

and I would feel NO compulsion to analyze the places of either modifier
into the resulting place structure since the ONLY reason they are
included is to clearly delete the semantics of purpose for the compound.)

But this added length in the lujvo means that I will really think twice
before adding in terms to delete the place.

This is long enough.  Let's let people start shooting me down.  Jorge??? %^)

lojbab

lojbab