diklujvo (Regular Compounds)

[cbmvax!uunet!math.ucla.edu!jimc]

			diklujvo (Regular Compounds)
			    Jim Carter,  91.5.4

  Abstract: -gua!spi has been successful with three forms of diklujvo
  (regular compounds).  Events: the sub-word forms a "le nu" abstraction.
  Parallel: typically the mover and the destination are also related by a
  compound directional property; a noun-adjective (like diklujvo) is also
  parallel.  Transitive: the sub-word is the object of an action. The
  synthesized arguments get main bridi sumti replicated into them; this
  is useful in explicit sumti as well.  It helps a lot if gismu cases are
  regular.  In "The Welding Shop" all but two of the tanru / lujvo were
  rendered as diklujvo.

First a point about terminology.  JCB (L1 p. 44) introduced the concept
of "N-place predicates" and Lojban continues this tradition.  However,
in languages like Latin or Russian with very noticeable case structure,
the case tags act as "place tags" to show which case / "place" each
argument goes in.  The ancient concept of case fits perfectly our need
to talk about the various formal parameters (cases, places) of a 
predicate.  And it is familiar to linguists new to Lojban even if the
more mathematical L1 terminology is not.  For these reasons I recommend
that we say "case" rather than "place".

In the same place JCB describes predicates as relations between
arguments, and again Lojban continues this tradition.  To my mind, the
emphasis on predicate relations is the foundation feature of Loglan and
Lojban.  This is sufficient justification for a campaign to interpret
(what now are) non-predicate grammatical structures as abbreviations or
surface structures that can be transformed to the predicate form. 
Another justification is that predicate relations are tractable
theoretically whereas other meaning classes, such as metaphors and
paralinguistic grunts, are so unclear as to be beyond the reach of
logical analysis, including analysis by a machine or a beginning human
student.

Already it is said that all sumti tcita can be understood as fi'o
<gismu>, putting the containing bridi and the tagged sumti (if any) in
the relation predicated by the gismu.  Another target for the campaign
(not yet accomplished) is the <UI> indicator class.  Of late there has
been considerable discussion about tanru, lujvo, and their case
structures. This essay is about how to interpret many of them as
predicate combinations of gismu.  It turns out to be quite simple. 

A review of the lujvo in L4 (the Loglan dictionary) showed that about
90% of them could be interpreted according to some simple rules, and
substantially more could be made regular if a few gismu cases were
reordered.  In designing -gua!spi I set a goal to make diklujvo
(regular compounds) even more productive.  This would mean that a
speaker could create such compounds on the fly and could expect any
listener to understand, without either one needing to memorize large
numbers of dictionary lujvo.  Only the most figurative meanings would
have to be rendered by a true metaphor.  What I describe here are the
-gua!spi rules, modified to fit Lojban conventions.  They may be 
considered a starting point for defining Lojban diklujvo.

The most common lujvo type (in -gua!spi) involves an event argument.  
Some gismu have a case obviously intended for an event argument, such
as binxo (x1 becomes [event] x2 under conditions [event*] x3), that is,
(x1 changes so [event] x2 becomes true...).   With this main word the
lujvo or tanru partner or sub-word is to be taken out and inserted as an
abstraction in the main word's case.  For example:

	la kiras. cu ja'ibi'o le falnu
	la kiras. cu jgari binxo le falnu	
		Kira takes hold of the sail
	la kiras. goi ko'u cu binxo le nu jgari ko'u le falnu
		Kira changes so (he holds the sail)

Several points:  (1) Students and machines know that this rule applies
because the dictionary says binxo x2 is for an event, and also that x3
is ineligible.  

(2) A bridi must include sumti (arguments).  If you want the
synthesized event argument to mean something you must copy into its
first case the first case of binxo.  

(3) Automatic replication is very useful even when the event is said
explicitly (le nu).  Even cases ineligible for compounding, like x3 of
binxo, may want auto-replication.

(4) Where do following sumti go, on the event or on the main bridi? 
Usually for proper meaning they should go on the event.  -gua!spi case
structures tend to rely on sumti tcita for details like x3 of binxo,
but if there is a following case it's important, so the rule is that
following cases (if any) of the main selbri are filled first, and
leftover sumti are dumped into the event.  But Lojbanists may prefer to
retain detail cases and to fill the event first.

(5) Lujvo and tanru are analysed exactly the same way.

Some more examples:

	ko'u ja'ifi'i ko'o lo xance be ko'u
	ko'u jgari friti ko'o lo xance be ko'u
		He offers it his hand
	ko'u friti ko'o le nu ko'o jgari lo xance be ko'u
		He offers it (it [may] hold his hand)

friti x3 is the event argument and x2 is replicated into it. Most event
arguments need the case before the event to be copied -- usually the
event is x2 but sometimes it is x3 and needs a copy of x2, and there
are a few irregular replications that remain in -gua!spi.  When a
comparison or sort order is involved, the event's x2 may also have to
be copied in.  Incoming sets may have to be strung out in extension. 
All these details are noted in the dictionary and are extracted and
acted on by the parser.

	ko'u carnri'a ko'o
	ko'u carna rinka ko'o
		He turns it over
	ko'u rinka le nu ko'o carna
		He causes (it rotates)

Here no auto-replication is wanted.  I believe this is the standard
transitive conversion pattern in Lojban analogous to "he makes it
rotate".

	xu selcapstidji do
	xu se ckape sisti djica do
		Don't you want to be rescued?
	xu do djica le nu do sisti le nu do se ckape
		Don't you want (you stop (you in peril))

In nested tanru outer sumti are replicated recursively to inner event
arguments. Standard Lojban associativity fits with JCB's practice of
putting the main word last in a lujvo or tanru, though I find the
-gua!spi order more comfortable with the main word first.  I doubt even
the Oxford English Dictionary has a natural language equivalent as
specialized as selcapstidji "desires safety".  Isn't it wonderful to
offer this kind of wordmaking power in a form that any student can
handle?

	ko'u krici le nu ko'o ckape
		He believes that it is dangerous
	ko'u krici le nu ko'o ckape kei ko'o
		He believes of it that it is dangerous

In -gua!spi "believe" is defined "x1 believes that x2 satisfies [event]
x3" with x2 replicated to x3 -- this order is typical for a main word
productive of infinitive compounds.  On the other hand, Lojban has "x1
believes x2 about x3".  There is considerable utility in changing case
structures to fit the majority pattern because students only need to
learn one pattern and a few exceptions.  With either definition, in
this example the event argument has its x1 explicitly, but of whom is
this bridi believed?  The main predicate lacks that sumti. Just as
sumti can be replicated from the main bridi into the event bridi, they
can be retro-replicated from the event to the main bridi to complete
it.

In the second most common lujvo form (in -gua!spi) two words share
sumti in parallel.  The major but not only application is with motion
words and directional properties.  

	ko'o gargrelimna lei djacu
	ko'o gapru pagre limna lei djacu
		It swims down through the water
	ko'o limna lei djacu .ije ko'o gapru lei djacu 
	.ije ko'o pagre lei djacu 
		It swims to the water; it is above (the same) water;
		it penetrates the water

The gismu opposite the English meaning often turns up: it swims from
above.  This pattern works best when one word has exactly two cases, or
when both words have identical case structures.

	ko'u ti'eja'i leko'o stedu 
	ko'u trixe jgari leko'o stedu 
		He holds its head from behind
	ko'u trixe je jgari leko'o stedu 
	ko'u trixe leko'o stedu  .ije ko'u jgari leko'o stedu 
		He is behind its head; he holds its head

This parallel compound has no motion word.

	ko'u ci'acpu ko'o lei vacri
	ko'u cnita lacpu ko'o lei vacri
		He pulls it up to the air
	ko'u lacpu ko'o lei vacri .ije ko'o cnita lei vacri
		He pulls it to the air; it is below the air

For transitive motion words the object, not the actor, relates
directionally to the destination.

	ko'u batlafti ko'o fo lei vacri
	ko'u bartu lafti ko'o fo lei vacri
		He lifts it out of the water
	ko'u lafti ko'o fo lei vacri .ije ko'o bartu lei vacri
		He lifts it from the water; it is out of the water

Specially for bartu "outside" and sepli "separate", what counts is the
mover's relation to where it was, not where it is going.  These two words
are therefore watched for and treated specially.

	lo diklujvo
	lo dikni lujvo
	lo lujvo poi dikni
		A regular compound word

An adjective-noun combination can be translated very neatly as a parallel
compound; the shared x1 case is occupied by the placeholder through which
the sumti referent set is exported.  

Needless to say, compound construction is much easier if motion words
all have the same cases despite slight differences in meaning.  Also
motion words are the most complicated of all, and it helps students
learn them if they have a common pattern.  It is my judgement that
these factors outweigh the utility of having cases individually crafted
to match the meanings, with nonstandard orders or missing cases where
they are deemed irrelevant to the meaning.

The third most common pattern (in -gua!spi) is transitive: an action
upon an object.  This pattern was more common in Old Loglan -- possibly
because of JCB's Anglicisms.  Here is an example with two transitive
lujvo.

	ju'ocu'inai le vijsazri pumo'u jacvasxu
	ju'ocu'inai le vinji sazri pumo'u djacu vasxu 
		Maybe the pilot already drowned
	ju'ocu'inai le sazri be lo'e vinji pumo'u vasxu loi djacu
		Maybe the operator of airplanes already breathed water

The sub-word usually goes in x2 of the main word, though there are a
number of exceptions.  The article (lo'e vs. loi in this example) is
determined mainly by the main word, and in -gua!spi is provided from
the main word's dictionary entry.  How do you know whether a compound
is transitive or parallel?  In -gua!spi the tones tell, but it will
take some research to produce an unambiguous rule for Lojban.  It isn't
adequate just to mark it in the dictionary; there are too many words,
like transitive motion words, for which some sub-words should be
parallel and some should be transitive.

	ji mijfe'o lomi nebtau
	ji minji fenso lomi cnebo taxfu
		I machine-sew my tie
	ji fenso fo lo minji lo taxfu be mi bei lo cnebo be mi
		I sew with a machine my garment for my neck

Here x4 of fenso, the tool, should attract transitive compounds, not
x2.  In -gua!spi I define "x1 is a garment of wearer x2 for its body
part x3", and as in Lojban, body parts are "x1 is a (part) of creature
x2".  x3 of taxfu clearly is productive for transitive compounds, but
whose neck is the garment for?  x2 of taxfu should be replicated into
x2 of the synthesized x3 transitive argument.  A major attraction of
diklujvo (and argument replication generally) is that sumti predicates
can be provided with sub-arguments automatically, producing the
meanings which "common sense" tells a human speaker -- but not a
machine.

Here I use the convention that the post-article sumti (as in "lemi") is
fed to x2 of the main selbri, and only if no case can accept it is "ne
mi" generated.  With body parts and garments this rule produces correct
predicate relations; with most materials, foods and artifacts it yields
the desired "ne mi"; and it rarely prevents use of "lemi" because the
wrong case would be filled.

Here are the only remaining true metaphors from "The Welding Shop"
translated to -gua!spi (with Lojban gismu shown).  All others could be
interpreted as diklujvo.

	tirxu tavla	tiger talk	nag (the metaphor is Chinese)
	cinta gacri	paint cover	painted (when the covering was
					 welding flux, not paint)

In Lojban how will you recognize true metaphors?  The indicator ta'u
(tanru) seems to be for that purpose, but a <JA> infix connective seems
more appropriate.  

In Lojban as in Old Loglan the lujvo case structures are individually
crafted and can be substantially different from either of the component
words.  Thus the cases are more useful, being adapted to the exact
meaning chosen for the lujvo, but are also harder for the student,
since he must memorize lujvo the same as gismu.  For diklujvo this
policy must be abandoned; you must jump over useless cases or use sumti
tcita to hit missing ones, but if you know all the gismu you
automatically know an infinite set of lujvo because the rules tell you
what they mean, including cases.  In practice, if the gismu cases are
set up carefully but regularly there are few useless or missing cases,
and I judge that the benefits far outweigh the costs of having
diklujvo.