Lujvo paper part 1 of 4

[Nick Nicholas <nsn@MULLIAN.EE.MU.OZ.AU>]

"Doing the {belenu} blues": Lujvo place structure Paper.
Version 2.

1. Introductory.

1.1. What tanru do.

Tanru in Lojban are a combination of two or more predicates. Since each
predicate expresses a relation between its arguments, tanru express a
relation between the arguments of all their components.

Take for example
the tanru {gerku zdani}. {gerku} expresses a relation between an individual
dog {le gerku}, and a dog breed {le se gerku}. {zdani} expresses a relation
between a house {le zdani}, and someone housed {le se zdani}. The tanru
{gerku zdani}, which denotes something to do with both dogs and houses,
expresses a relationship between the dog, its breed, the house, and the
house occupant. Though this relationship is inherently ambiguous, all these
places can be named explicitly in a sentence. If we want to somehow relate
Spot, a Saint Bernard, with the White House, home of the U.S. President, we
*can* speak of:

1.1) {la blabyzdan. cu gerku befa la spat. bei la sankt. bernard. be'o zdani le
merli'e}

Of course, a tanru is not just a jumble of gismu: not all the sumti of its
component gismu need be present in the tanru, and not all of them need have
the same interpretation
as they do in the original gismu. Consider of the possible interpretation
of {gerku zdani} as dog kennel. In speaking of dog kennels,
we are expressing a relation between a dog and a house; it is not as clear
that we are also expressing a relation involving a dog breed, or a
resident of the house *distinct* from the dog. If we are to speak of
Spot's kennel, called Mon Repos, we may say:

1.2) {la monrePOS. cu gerku befa la spat. bei la sankt. bernard. be'o zdani la
spat.}

But obviously {le gerku} is the same as {le se zdani}, so we lose no
precision by saying:

1.3) {la monrePOS. cu gerku bei la sankt. bernard. be'o zdani la spat.}

Furthermore, it could be argued that a kennel is a kennel, regardless of
what breed of dog lives in it; the breed is not important to the
*definition* of a kennel, and only arguments important in such a way should
be present in a bridi. If we accept this argument, the fact that Spot
is a Saint Bernard does not add any important information to the claim that
Mon Repos is its kennel. Thus there are only two important sumti left:

1.4) {la monrePOS. cu gerku zdani la spat.}

1.2. What lujvo do.

A lujvo is equivalent to a tanru, but it unambiguously expresses a specific
chosen relation between its arguments, out of the many an ambiguous tanru can
convey. Which one it chooses depends on the relative usefulness of the
concept, though for nonce use, this is equivalent to 'whatever the speaker
decides'. (In writing a dictionary,  we at least in theory have to be prepared
to override someones nonce usage because we see a more obvious interpretation
for a lujvo.)

As distinct from a tanru, a lujvo does not nest its sumti: all its
component sumti appear in sequence, as they do for a gismu. Compare:

1.5) {.i (mi) zmadu (do) [leka (mi) barda (leka xadycla kei) (lo'e remna) kei]
(lo mitre be li pimu)}

1.6) {.i (mi) barda [be (leka xadycla kei) bei (lo'e remna) be'o] zmadu (do)
(zo'e) (lo mitre be li pimu)}

1.7) {.i (mi) bramau (do) (leka xadycla kei) (lo mitre be li pimu)}

"I exceed you in height by half a metre". ("I'm half a meter taller than you";
see below for the reason the {lo'e remna} place is omitted.)

The place structure of a lujvo is extremely important. In Lojban, a
predicate language, place structure is essential to the *definition* of
a word, whether gismu or lujvo. As we saw with {gerku zdani},
not all the places of the component gismu will be as important in this
definition. Some may be omitted altogether in the lujvo place structure;
some may be merged with other places, and some may have their meaning
shift. The ability to select, from the places of the original gismu,
those that should go into the final lujvo is critical for making lujvo which
clearly and succinctly express the desired relation.

In using selbri in Lojban, it is important to remember the right
order of the sumti. This becomes critical with lujvo: the set of places
selected must be ordered in such a way that a reader, unfamiliar
with the lujvo, is able to tell which place is which. The ordering of
places should somehow be reproducible, and follow a consistent pattern.

So good place selection ensures the lujvo meaning is clearly and
economically expressed. Good place ordering ensures the lujvo place
structure can be reproduced and understood without relying on context.
Neither principle is absolute: there are cases where they need to be
violated to accomodate lujvo meaning better. In general, they
should be adhered to, but we should recall that these are guidelines
only; they cannot be expressed as rigid rules. (To quote John Cowan:
"Tanru mean whatever the speaker wants [constrained by
the fact that the place structure of this tanru is that of the final term],
gismu mean what the dictionary
says. Lujvo, on the other hand, have meanings that are constrained by the
place structures of the underlying gismu, but not fully determined by them.")
For this reason, this paper does not
have the same force as other papers published by the LLG, and does not
represent official LLG policy.

But the guidelines are still important. If a desired place
structure does not match that obtained by these guidelines, the base tanru
(veljvo) chosen may not be the best for the job,
and may not actually express what is desired. In general, there
should be no need to introduce places external to the component gismu,
or to radically reorder the lujvo places (though this cannot be ruled out).

Consider as an example the concept "meeting", which involves, at the least,
people meeting (or workshopping), and a topic for the meeting. The gismu
{penmi} (to meet) has
three places: someone doing the meeting, someone met with, and a place at
which the meeting occurs. Is {nunpenmi} a good lujvo for "meeting" as we
have defined it? Not really: there is nothing in the place structure of {nu}
or {penmi} that could give us a topic of discussion; in lacking that place,
all {nunpenmi} can properly mean is "an act of meeting someone, an encounter".
For the type of meeting in which the business of discussing is conducted,
you need a veljvo which will make the appropriate places available. A topic
of discussion is a {se casnu}, so {snununpenmi} better expresses
the desired meaning.

2. Lujvo place selection.

There are two ways to determine the arguments a lujvo should have. The first
is to initially postulate a place structure, and try to derive it by
elimination of places from the component gismu. The second is to start without
any preconceptions of the final place structure, and eliminate places
according to some notion of redundancy. If the result of eliminating places
doesn't convey the information desired (remember that the place structure *is*
the meaning of a bridi in Lojban), try a new veljvo.

The few place structure
derivations published in _ju'i lobypli_ have followed the first approach. In
this study, we will follow the second approach (though we may still be guided
by our preconceptions in eliminating places). We believe this approach
encourages better-made lujvo, with their intended meaning more closely and
predictably matching to the veljvo, and without relying on external knowledge.

There are two criteria for eliminating places from the component gismu of
a lujvo:

* The place conveys redundant information.

* The place conveys irrelevant information.

Usually at least one place will be eliminated by redundancy. To explain why,
we digress to consider the three main classifications of lujvo according
to the relation understood between the seltanru and tertanru.

2.1. Lujvo classifications.

2.1.1. Lujvo classifications: {be}, {je}.

We postulate the following hypothesis: all concepts expressable as tanru or
lujvo in Lojban can also be expressed as bridi in which each selbri is a
single gismu. We
will call this hypothesis the Gismu Deep Structure Hypothesis --- GDSH.)

This hypothesis has been tacitly assumed in much Lojban work. For example,
consider the frequently made claim that we can disambiguate tanru in Lojban
--- namely, that we can identify and present the different possible meanings
of a Lojban tanru (and by extension, of a lujvo) in Lojban. This is why the
cmavo {ta'unai} exists. How would this be done? We can either use
tanru, lujvo, or gismu to express the various bridi necessary. But tanru are
inherently ambiguous, and lujvo, though notionally unambiguous, are derived
from ambiguous tanru. Either case gives a chicken-or-the-egg dead-end. The
only possible way of unambiguous paraphrase must use gismu and cmavo only. But
this is exactly the claim that the GDSH makes.

Of course, the hypothesis is making rather strong claims, and languages
tend to evolve away from that level of formality. But the results discussed
in this paper are still valid if we use a weaker version of the hypothesis
--- that the meaning of a lujvo is *approximately* defined by a GDS, but
enough for its place structure to be determined from it.

To illustrate the hypothesis, consider again the tanru {gerku zdani}. The
following strings of cmavo and gismu can be considered possible
disambiguations (GDS's) of the tanru:

{.i ta zdani be lo gerku}

{.i ta gerku gi'e zdani loimu'a civla}

{.i ta zdani neta'i lo'e gerku .iva'i ta gerku tarmi zdani}

If the GDSH is true, the meaning of a tanru can be expressed as a bridi where
the seltanru and tertanru are in some *syntactic* relation ({broda je brode},
{broda be lo brode}, {broda pe lo brode}). Then at least
one place of the tertanru will contain information shared with a place in
the seltanru. Consider the following two tanru and their GDS's:

2.1) {ta gerku zdani}
{ta zdani da poi ke'a gerku}
{ta zdani be lo gerku}
{z1 zdani z2=g1}

2.2) {ta banli sonci}
{ta banli gi'e sonci}
{ta banli je sonci}
{ge da banli gi da sonci}
{b1=c1 banli gi'e sonci}

(In the above examples, and in other examples in this paper, a letter x
followed by a number n denotes the nth place of the gismu starting with x.
Thus z1 is the first place of {zdani}. The equals sign means that the two
places typically denote the same thing. This might not always be the case,
though. If a tortoise lives in a doghouse, we can say that {le resprtestudine.
cu se gerzda}; but the tortoise, while being a {se zdani}, isn't a {gerku}!
But this issue is complicated: we might not even want to say that {le
gerzda cu gerzda le resprtestudine.}, but simply that {le gerzda cu zdani
le resprtestudine.})

Most (though not all) tanru and lujvo can be analysed by one of the two
paradigms exemplified above. The seltanru and tertanru may be related
in a GDS by the word {be}. In that case, the seltanru
gismu contains information on a place (usually x2) of the tertanru. Such tanru
are called asymmetrical in the tanru paper. Example (2.1) shows this:
the bridi {gerku} gives further information on the x2 place of the tertanru
{zdani} --- {le se zdani cu gerku}. In effect, the place z2 denotes the same
entity as the place g1, and they shouldn't both be included in the place
structure of {gerzda}. Rather than have the place structure of {gerzda} as:

{gerzda}: ... ... is the entity housed in the doghouse; ... is the dog...

one has:

{gerzda}: ... ... is the dog, which is the entity housed in the doghouse...

The entity housed and the dog are the same thing, in this definition of
"doghouse". (There is some fine print to this, which we'll return to.)

On the other hand, the two gismu may both be making claims that are true of
a given entity. In that case, the relationship may be expressed using the word
{je}, or a word with a similar semantic function (like {joi}). Such tanru are
called symmetrical in the tanru paper. Example (2.2) shows this: it is
true of entity {da} that {da banli}, as well as that {da sonci}. So the
places b1 and s1 denote the same object, and they shouldn't both be included
in the place structure of {balsoi}. Rather than having the place structure:

{balsoi}: ... ... is great; ... is a soldier...

one has:

{balsoi}: ... ... is a great-soldier...

To stress the importance of GDS (which will become more apparent when we
discuss place ordering), we will term the paradigm exemplified by (2.1)
"be-tanru", and that exemplified by (2.2) "je-tanru".

2.1.2. Lujvo classifications: {belenu}.

As a special case of be-tanru, consider (2.3):

2.3) {mi gasnu lenu le gerku cu citka loi guzme}
I make the dog eat melons
{mi nu lo gerku cu citka loi guzme kei gasnu}
{mi nunctikezgau le gerku loi guzme}
I feed the dog melons

This is an example of a causative, a predicate in which an agent brings
about an action described by the seltanru. Thus 'feed' is the causative version
of 'to eat'; the transitive verb 'sink' (eg. "I sink boats")
is the causative of the intransitive 'sink' (eg. "I sink in the ocean").
While all languages have causatives, there are different ways to express
them. English only occasionally forms the causative verb form directly
from its non-causative counterpart (eg. 'to sit' --- 'to seat' = 'to make
someone sit'). The corresponding causative is usually an unrelated word,
or an entire phrase (eg. 'to make someone lie down' is the causative for
'to lie down'). The derivation of causatives in Turkish and Esperanto, on
the other hand, proceeds quite regularly.  'To eat' in Esperanto is 'mangxi',
and 'to feed' is 'mangxigi'; 'to sink (intransitive)' is 'sinki', and
'to sink (transitive)' is 'sinkigi'. In Turkish, these verbs are respectively
'yemek', 'yedirmek', 'batmak', and 'batIrmak'.

It is obvious why single verb causatives like those in Turkish and Esperanto
would be advatageous in a language like Lojban. The alternative, using a
full GDS, is longwinded and counterintuitive. To say {mi gasnu lenu da
citka de} involves sumti nesting, and conceals the fact that we perceive
feeding as a direct relation between three entities (the feeder, the fed,
and the food). {nunctikezgau} has no such disadvantage: as (2.3) shows,
the three entities are related directly by just one bridi.

It is safe to predict that causatives will become quite prevalent in Lojban.
This is because of the small number of gismu in Lojban: many concepts, for
which we have verbs or adjectives in English, fall into causative/non-causative
pairs; more often than not, only the non-causative predicate will be
represented by a gismu. Take the verb "to record". The gismu {vreji} does not
correspond to this verb, but to the predicate "to be a record". If we
want to say "I record data", we need to say the equivalent of "I act so
that X is a record of data", {mi gasnu lenu da vreji loi datni}, or more
elegantly and succinctly, {mi nunveikezgau da loi datni}.

Now, {nunctikezgau} expresses a
relation between an actor and an *event* of eating ({nu citka}); as such,
it is a straight-forward be-tanru ({gasnu be le nuncitka}). In this lujvo,
we would like a place for the actor (g1), for the eater (c1), and
for the food (c2). {nuncitka}, as we will see in section 5,
has the place structure n1 c1 c2: the event of eating is x1,
the eater is x2, the food is x3. g2 is obviously the same as n1 ({le se gasnu
cu nuncitka}). It is possible to omit *both* g2 and n1 from the place
structure: since we already know that we are discussing an act of eating, it
is redundant to leave in a place for the event, when we have already given
all the event's sumti.

So "{nunctikezgau}: agent x1 causes event x2 of x3 eating x4" would be
giving redundant information, when all the *pertinent* information about
the event x2 is given by arguments x3 and x4.

(This is not strictly true: the event of eating is specified by all the
arguments of the bridi {citka}, not just those built into its place
structure. There are time and location, for example: {nu mi citka ca la
pacicac. vi le ctikumfa}. The only argument for keeping a separate place
for {lenu citka} in {nunctikezgau} is that the auxillary places of the event
{nu gasnu} can differ from those of {nu citka}. You can bring about someone
eating at 1 PM by an action of yours done at 9 am. The GDS is: {mi ca la
socac. gasnu lenu da citka ca la pacicac.} If we can keep a place for
{nu citka}, we can transform this to the lujvo: {mi nunctikezgau lenu da citka
ca la pacicac. kei da ca la socac.}

But this lujvo is so redundant (saying nothing that {mi gasnu lenu da ba'e
citka ca la pacicac. kei ca la socac.} doesn't),
and so counterintuitive, that it is best to leave out the
abstract sumti place, and assume it is implicitly specified as much as is
necessary by the remaining lujvo places: {mi nunctikezgau da ca la socac.} If
we need to specify the abstract sumti information more precisely, we can
always recourse to the GDS form. This follows general principles given in
section 2.3, and empirical data from natural langauges. Single words (like
{nunctikezgau}) are used to encode direct relations between a small number
of arguments, while whole phrases (like the GDS, or its English equivalent
"I made the dog eat at 1pm, at 11am") encode less direct relations between
a possibly greater number of arguments.)

For lujvo made with {gasnu}, the seltanru will usually specify an event. We
may thus omit the {nun} rafsi as implicit, and say simply

{ctigau}: agent x1 causes x2 to eat x3. (x1 feeds x3 to x2.)

In this case, the place g2, the action performed, is equivalent to an
*abstraction* composed of all the places of {citka}. Rather than having a
be-lujvo, where a single place of the seltanru substitutes for a place in the
tertanru, we now substitute a *number* of places for that single place. This
particular paradigm we will call a belenu-lujvo; it is useful not only for
causatives ({rinka} and {gasnu} tertanru), but most tertanru with an
abstract sumti in x2 or x3. be-lujvo, je-lujvo, and belenu-lujvo are the three
types of lujvo Jim Carter has identified in his earlier work on lujvo.

Remember that belenu-lujvo are really an abbreviation of be-lujvo based on
abstract sumti. {ctigau} is really an abbreviation for {nunctikezgau}, and
is comparable to other lujvo with elided rafsi, considered in section 4.
Furthermore, this is not an abbreviation that can be performed automatically.
There are only a few tertanru for which the rafsi {nun} is obviously implicit,
and to assume it too widely can lead to ambiguity.

For example, the x2 of
{djica} is either an event or a simple sumti. {le soidji}, in the be-lujvo
interpretation, is someone who wants a soldier ({djica be lo sonci}). In the
belenu-lujvo interpretation, it is someone who wants to be a soldier, a wannabe
soldier ({djica be lenu ri sonci}), or perhaps someone who wants someone else
to be a soldier ({djica be lenu zo'e sonci}). In these cases, it is safer
to say {nunsoidji} for the latter case, and {soidji} for the former.
Lojbanists must be careful to use the
abbreviated form only when no reasonable ambiguity will result. This is much
likelier with bridi like {gasnu} and {rinka} than with, say, {djica} or
{nelci}.

Despite these complications, belenu-lujvo are powerful means in the
language of rendering succinct
and manageable concepts out of quite verbose GDS forms, thereby increasing
expressive power.

2.1.3. Lujvo classifications: Summary.

* In je-lujvo, the x1 of the tertanru is semantically "doubled up" by (denotes
an object identical to, or is also described by) the x1 of the seltanru, so
it shouldn't be necessary for both these places to appear in the lujvo place
structure.

* In be-lujvo, the x2 of the tertanru (typically; sometimes, the x3) is doubled
up by the x1 of the seltanru: again, the two places needn't both appear in
the lujvo place structure.

* In belenu-lujvo, the x2 (typically) of the tertanru is an event abstraction,
whose selbri is the seltanru, and whose internal sumti are the sumti of the
seltanru. This place in the tertanru is thus replaced by the places of the
seltanru bridi.

We must stress that this type of analysis, though it can be
made to apply to many lujvo, is not appropriate for all of them.
Elimination of redundancies in lujvo not fulfilling these patterns has to
be done case-by-case. Even with lujvo falling within these paradigms,
there are often extra redundant places that need to be eliminated.

Take the lujvo {ninpe'i}, "to newly meet", which has been used to translate
"to be introduced to someone". We can analyse this as a be-lujvo:
its GDS is {penmi be lo cnino}. The full GDS, though, is {da penmi de poi
ke'a cnino da kei di}. In other words, not only is c1=p2 (the person met is
the person who is new), but also c2=p1 (the person doing the meeting is the
person to whom the other is new). The place structure resulting should take
advantage of this redundancy: {da ninpe'i de di}, x1 is introduced to x2 (who
is new to x1) at x3.

2.2. Eliminating irrelevant places as lujvo definition.

2.2.1. Tertanru versus seltanru place elimination.

By using lujvo classification, we have concentrated on eliminating
redundant lujvo places. It is also important to eliminate places
irrelevant to the *definition* of the relation desired. This is a skill
essential for lujvo makers. Leaving in or omitting places can make a
big difference to the concept the lujvo ends up specifying.

Consider as an example {tciterjai}, the handle of a tool. The gismu involved
have the following place structures:

{jgari}: x1 grasps/holds/clutches/seizes x2 with x3
{tutci}: x1 is a tool/utensil used for doing x2

So there are five places involved; eliminating redundancy (the {se jgari}
is obviously the {tutci}), we have four: j3 j2=t1 j1 t2. It turns out we
can eliminate two more places.

The first place is t2, and the kind of elimination it exemplifies is
prevalent amongst be-lujvo. In many lujvo, the tertanru already has an
x1 place in the lujvo place structure. The other places of the tertanru
are supplementary information about that x1. There is usually no reason
why this information shouldn't appear in the x1 place itself, rather
than in the overall place structure. Thus, if we are considering the
handle of a scraping tool, it seems hard to justify the place structure
{ti tciterjai lo smuci lenu citka} --- x1 is the handle, x2 the tool,
x3 what the tool is used for. What the tool is used for is important
to the definition of the tool, but not of the handle.

It makes much
more sense to say {ti tciterjai lo smuci be lenu citka} --- which
corresponds more closely to the actual syntactic structure of its
translation, "this is the handle of a spoon for eating". "for eating"
belongs with "spoon", and not with "handle"; for that reason, you
can't say *"this is the handle of a spoon, I believe, for eating" as
readily as "this is the handle, I believe, of a spoon for eating". You
can't interrupt a phrase of words that belong together (what syntacticians
call a *consistuent*).

The same thing occurs with our original example, the doghouse. It makes
little sense to say {la monrePOS. gerzda la spat. la sankt. bernard.}. The
breed is important to the definition of the dog, but not to the definition
of the doghouse. {la monrePOS. gerzda la spat. noi gerku la sankt. bernard.}
("Mon Repos is the doghouse of Spot, a Saint Bernard") makes much more sense.

The second place of {tciterjai} we would omit is {le jgari}. A handle is
a handle, irrespective of who grasps it. A handle does not become a
different handle just because Buffy rather than Van Helsing grasps it.
To leave in the "grasper" place, we would
have to argue that the person grasping the handle affects the essence of what
a handle is. But this is not the case. Which tool the handle belongs to, on
the other hand, is very important to the definition of the handle.

(It is true that what is a handle
for *humans* may not be a handle for dogs, and vice versa. In that case
the place may be kept, but its meaning will shift. It no longer denotes
the individual handling the tool, but the class of individuals who can do
so. The difference is akin to that between {le citka} and {le se cidja},
and we will return to it shortly.)

Another example of irrelevant place omission is {laurba'u}, as a translation
of "to bellow". Its component gismu have the following place structures:

{bacru}: x1 utters verbally/says/phonates [vocally makes sound] x2
{cladu}: x1 is loud/noisy at observation point x2 by standard x3

Eliminating the obvious redundancy, we have: b1=c1 b2 c2 c3. But how relevant
is c2 to bellowing? A person bellowing in New York will be pretty quiet at
an observation point in Boston. But that person is still bellowing. What
matters is not where the bellow is loud and where it isn't, but that
bellowers intend their speech to be loud. {le bacru cu cladu zu'i}, in
other words --- they are loud where they need to be, and where precisely
they are loud is not important to the definition of bellowing.

As a further example, take {bramau}, "bigger". In lojban, there are three
arguments to "big": the big entity (b1), the way in which it is big (b2),
and what it is big relative to (b3). The last place is included because
"big" is a subjective concept. It doesn't make sense to say things are
big in the absolute: bigness depends on a reference standard. Thus, an elephant
is big to a mouse, but small to a planet: {.i lo xanto cu barda fi lo ratci
gi'e cmalu fi lo plini}.

Now consider {bramau}. Normally, we would
include all the places of {barda} in the place structure of {bramau}. But
there is nothing relative about one thing being bigger than another. Whether
you have the vantage point of a mouse or a planet, an elephant is bigger
than a human. So it makes little sense (in fact, it is confusing) to say that
something is bigger than something else, *relative to* something. For this
reason, a {te barda} place does not belong in the place structure of
{bramau}. Similar arguments can be made for other such comparatives involving

*******************************************************************************
 A freshman once observed to me:         Nick Nicholas am I, of Melbourne, Oz.
 On the edge of the Rubicon,             nsn@munagin.ee.mu.oz.au (IRC: nicxjo)
 men don't go fishing.                   CogSci and CompSci & wannabe Linguist.
   - Alice Goodman, _Nixon In China_     Mail me! Mail me! Mail me! Or don't!!