cleft place strcutures - an open issue that we want opinions on

[lojbab (Bob LeChevalier)]

I'm going to raise a 'real' undecided issue in Lojban - one that affects
many place structures in the ongoing place structure review.  I welcome
ideas, opinions, and proposals on this issue.  This is long because I
intend to present all info ABOUT LOJBAN needed to express an opinion, so
you can rest assured that you probably know about as much on this topic
as anyone.

Th problem is somewhat akin to jimc's binxo problems, and it was that
discussion which brought this to mind.  Some of the things posted in
that discussion may or may not be relevant.

>From that discussion, and/or elsewhere, you may know that Lojban has at
least two major kinds of sumti (arguments) in its bridi (predications) -
(only two are relevant here).  I will call these simple and abstract
sumti.

A simple sumti is comparable to what in English are 'common nouns' -
objects that you can point to.  Examples include "le stizu" (the chair),
"le zarci" (the market).  But because Lojban doesn't distinguish nouns,
verbs, and adjectives, we can also have "le blanu" (the blue thing), or
"le kurji" (the one taking care of ...) as simple sumti.  All of the
examples so far are what we call 'descriptions' in the Loglan/Lojban
project.  In a description, a selbri (the predicate word or phrase that
defines the relationship) is converted into a sumti, omitting the x1
place, using a descriptor word like "le" or "lo".  The description then
refers to something intended that would fill that x1 place.  Thus "le
klama" is something that would fill the x1 place of "x1 comes/goes to x2
from x3 via x4 using mode x5".  (As many may know, other places besides
the x1 place may be filled in using "be" and "bei").

An abstract sumti looks and acts different.  In an abstract sumti, you
take a whole bridi predication (including the x1), i.e., a whole
sentence, and turn it into a sumti.  That sumti then represents the
abstract state or event of the predicate relation occurring ("nu"), the
characteristic property(ies) of that relation ("ka"), or one of several
other abstractions, including "du'u" (a predication about the relation),
and "jei" (the truth value of such a relation).  Others may be found in
selma'o ('lexeme') NU.

What is hidden in most usage of these abstract sumti is that we have
created an entirely new selbri relationship encompassing x1 through xn
of the abstracted bridi.  For most abstractions, this new selbri has
only one place, though "jei", which talks about truth values, has an x2
place for epistemology.  When you use one of these abstractions in a
sumti, you are again filling in the x1 place of >A< selbri, but at a
higher, more abstract level than for a simple sumti.  Thus there is a
parallel between these different levels of sumti such that both are tied
back to a bridi relation with one unspecified place.  An example is:

x1 is the event ("nu") of (xk1 comes/goes (klama) to xk2 from xk3 via
xk4 using mode xk5)


This parallel allows us to grammatically treat these two kinds of sumti
alike in many ways.  An abstraction has the abstract marker from NU on
the front and an elidable terminator "kei" on the back, to keep the
language unambiguous, allowing you to know whether a selbri is part of
the abstraction or is the main selbri of the sentence, or whether a
sumti is a sumti of the main sentence bridi, or of the abstraction
bridi.

On way to commonly treat these is to substitute a symbol for them.
These descriptions and abstractions are long, possibly comples in
grammar, and generally a pain to repeat when you are saying a lot about
them.  So we have symbols or 'anaphora' to stand for them.  From
postings, you will be familiar with "ko'a" which can be assigned to
attach to any sumti, whether abstract or simple, as well as "mi" (me)
and "do" (you).  There are a lot of others.  One other kind of anaphora
is names.  When you use the name "lojbab." for me, the name represents
me for discussion in a sentence.  "la lojbab." is thus grammatically
equivalent to "do" and "le nu la djan. klama le zarci".

We can also use names as anaphora for events and other abstraction sumti.
"The Rennaissance" is an important historical period.

By equating simple sumti and abstract sumti grammatically, we achieve
some of the power of Lojban's grammar.  Lojban allows the manipulation
of both types of arguments using its grammar as predicate logic does -
you ignore the representation when manipulating the symbols.

Now we can turn to the problem.  Here are some relations expressed as
English sentences:

I know about John.                                                    (1)
I know about John sleeping with Susan.

The cooking is done. (The cooking has been completed.)                (2)
I'm done cooking     (I have completed the cooking.)

I turn the water into steam by boiling it.                            (3)
My boiling it turns the water into steam.

It is good.  (It representing "the cat")                              (4)
It is good.  (It representing "the long romantic walk to the park")

Mary hit it.  (It representing "the cat")                             (5)
Mary hit it.  (it representing "the long romantic walk to the park")

(1) through (5) of course all involve the same relationship using in one
case a simple sumti, and in the other case an abstract sumti.  (4)
serves to remind that abstract sumti and simple sumti are equivalent in
Lojban bridi.  You need to be able to manipulate them using their
symbols, without worrying about what the symbols represent, or whether
you end up with nonsense, as in (5).

All 5 sentences involve relationships that are traditionally represented
by a single word root ('primitive' or "gismu") in Lojban; respectively these
are "djuno" (know), "mulno" (complete), "galfi" (modify), "xamgu" (good),
and "darxi" (hit).  The place structures of the first three have been
subject to much debate.  I have phrased (1) to demonstrate the current
consensus about "djuno":

x1 knows about x2 (a simple or abstract sumti)            (1a)

There is no solid consensus about (2) and (3):

x1 is complete in aspect x2                               (2a)
x1 completes doing/being x2 (in aspect x3?)               (2b)

x1 modifies x2 into x3 by doing/being x4                  (3a)
x1 modifies x2 into x3                                    (3b)

2b and 3a are the more familiar usages based on English and many other
languages, but 2a and 3b in many ways make more sense for Lojban.

If all this makes perfect sense, then I have successfully slipped the
"cleft sumti" problem in on you.  The corresponding problem version
of "djuno" is:

x1 knows about x2 doing/being x3                          (1b)

In all three of these sentence pairs, we have one version where an
abstract sumti fills a single place in one version, whereas the other
version has the same abstract sumti, but also repeats the "actor" of
that event in a separate place.  Thus, in (2), I may have fooled you in
that the x1 place in one case is the actor, and in the other is the
event/state being completed.  In effect, a simple sumti and an abstract
sumti are interpreted to play different semantic roles in the x1 place,
and this is unacceptable in Lojban, which requires simple and abstract
sumti to be treated alike.  If you use "ko'a" in the x1 place, you do
not know which kind of sumti it represents, until you substitute.  To
logically manipulate the sentences, you do not substitute.  You can
easily end up with nonsense, if you manipulate an actor in the way you
manipulate an event.

We have to choose one or the other place structure for a couple more
reasons.  First is the learning problem.  The place structure represents
the meaning of these words, and if there are two place structures you
have 2 meanings, and twice the amount of learning.  It is actually worse
than that, because you have to learn WHICH words have two meanings.

The other problem occurs when you turn one of THESE words into a simple
sumti.  Is "le mulno" an action that is complete, or the actor that
completed it?  Is "le galfi" a modifier, or the modification process?
You clearly want to be able to access the actor, since he/she/it is more
likely to be used in such a sumti.

In older versions of Loglan, there were many more problem words of this
sort.  Jim Brown basically argued that place structures should be what
is 'natural' for speakers, including all information that is needed to
determine the truth conditions of the relation.  Both of these place
structures do, so he typically chose the more English-like version of
the place structure.  This led to all manner of subtle difficulties.
Since the actor is specified in one place, then in the event sumti, you
typically will elliptically omit the actor, as well as other places.

?mi mulno    le  nu       [mi] jukpa   [le cidja]     (6a)
 I  complete the event-of [me] cooking [the food].
 I finish cooking.

?le  nu       mi jukpa   [le cidja] cu mulno          (6b)
 The event-of me cooking the food      is-complete.

Note that the second sentence is often 'shortened' in a couple of other
ways in English.  For "I'm done cooking." and "The food is done
cooking", (6b) reveals that in English we are merely condensing the
abstract event in a different way, by ellipsizing a different sumti of
the event bridi.  This is a common action in language, and it leads to
logical errors.  In effect, one place of a relationship is standing for
the whole relationship, but this representation is not noticeable in the
grammar.  We don't want this in a logical language.

In Loglan, something else happened.  Since most words had an actor in
the x1 place, when one did try to express one of these bridi using the
given place structures, one usually ellipsized the first sumti of the
event, which was just a repetition of the actor.  The large number of
ellipsized x1 actors LED TO A SEPARATE GRAMMATICAL STRUCTURE, that not
only had no basis in logic, but caused all manner of ambiguities in
Loglan, ambiguities that were solved by cheating in the machine grammar.

Institute Loglan STILL has this problem.  It uses the word "po" where
Lojban uses "nu".  In Jim Brown's versions of Loglan, "po" ("nu") did
NOT change the nature of a bridi, as I discussed above.  "le po blanu"
was a simple sumti:  "le (po blanu)".  To get an abstract sumti, you
wrote "lepo" as a single word, which the computer parser would then
treat as a totally different selma'o (grammatical category) than "le",
turning a whole bridi predication into a sumti:  "lepo (ta blanu)".  But
a human being can't tell "lepo" from "le po" in normal speech, so Jim
Brown introduced an arbitrary rule that to separate the two words, you
had to pause between them - a "lexemic" pause.  No human language uses
pause grammatically; pauses can separate words for breathing or
thinking.  But computer languages often use spaces to avoid ambiguities,
and Jim Brown was in effect treating a space as a pause (there is no
symbolic representation in Institute Loglan that a pause is required in
"le po blanu".  Finally, to complete the complexity, Brown had to
introduce a "poge" construction to make a 'long-scope' abstraction for
use with trailing arguments and logical connectives:  "mi viska le poge
ta blanu" is a possible construction, though one never used because it
is identical to "mi viska lepo ta blanu".  Needless to say, the web of
spaces and "ge"s made a mess of the grammar description, especially
since both were used in other ways in the grammar as well.

When we first started Lojban, we had no interest in changing the
grammar; we just wanted to avoid Jim Brown's copyright claims on the
words.  When he started treating his grammar as a trade secret, and
changing it without open review (as he still does), we had to redo the
grammar on our own.  At first, we merely copied the existing Loglan
structures, and hoped that our very existence would pressure Jim Brown
into changing his copyright policy, but this didn't work.  In 1989, we
started teaching the language, and in Lesson 3 of what is now the draft
textbook lessons, we ran square into this problem.  Old-timer Lojbanists
may remember that at that time we had "le nuke", which exactly matched
"le poge", and we had "lenu" and "nu" as distinct.  The much distributed
'February 1988' machine grammar has these fossils in it.

Trying to explain the rats' nest of spaces and "ke"s, and SPEAKING THE
LANGUAGE ALOUD, demonstrated that there was only one real construct
involving "nu" and that was a bridi turned into a selbri.  John
Parks-Clifford (pc), who is a professor of logic, linguist, and editor
of Brown's former publication "The Loglanist", noted that he and others
had unsuccessfully argued for this in the 1970's with Jim Brown. pc, an
early supporter and contributor to the Lojban design, also had discussed
the cleft place structures problem with Brown, but with no resulting
change.

In 1989, both the abstraction grammar and cleft place structures issues
came up; however, it was not until writing this that I realized that the
two problems were so closely related. pc's reasoning on the cleft place
structures was convincing, we changed all place structures where x1 was
the actor and x2 was the event to a single place. tanru (metaphors) and
lujvo (complexes) were used to access the actor "mulri'a" (mulno rinka)
is the actor form of "complete" in baseline Lojban.

But the problem is not really resolved.  We missed several cleft place
structures, and have discovered them intermittently while doing the
place structure reviews.  "galfi" was the most recent discovery - the
cleft place was discovered when I twisted the place structure around the
other day while discussing jimc's "binxo" in a posting.  Meanwhile, all
us native English speakers trying to speak Lojban continue to use "mi
mulno" for "I'm done", usually with humorous results when called on it.
(In Lojban, "you" are presumably not done until the "event" represented
by the word "you" is complete i.e. when you are dead, or even later,
depending on your religion.)

pc, Nora, and I all believe the 'right' way to go in Lojban is to
thoroughly scour the language of cleft place structures.  When you want
to access the actor, there are a variety of ways to do so:  with "rinka"
as mentioned above, or more vaguely with "gasnu", which now has the
extremely relevant place structure:  "x1 is the actor-place in
abstract-event x2".  In effect, "gasnu" can be seen to be the ultimate
cleft place relationship.  We would also retain "zukte", which last year was
re-place-structured to "x1 does x2 with purpose/goal x3".  "zukte" and
"gasnu" in effect translate the English "do", but with a very Lojbanic
flavor.

The negative side of this is that natural language speakers will have a
little harder time learning to fluently use these place structures,
simply because they are unlike English and the other languages.  On the
other hand, place structures will be shortened and systematized,
something both jimc and John Cowan have argued for, among many others
(including myself).  The language will be more amenable to logical
manipulation, and will more explicitly reveal the logical structure of
activities, but it may seem less natural.  I should note by way of
fairness that there are solutions that allow retention of cleft place
structures; these all retain the problems of redundant expression of the
actor (or ellipsis and possible loss of the logical structure as in old
Loglan), and the more serious problem of place structure learning - which
words have cleft structures with actors, and which have events.

The floor is open, doi prenu.  Speak up, or ask further.

----
lojbab = Bob LeChevalier, President, The Logical Language Group, Inc.
         2904 Beau Lane, Fairfax VA 22031-1303 USA
         703-385-0273
         lojbab@snark.thyrsus.com