lujvo place structure

From Lojban
Jump to: navigation, search

By la nitciion:

Your place structure or mine? lujvo place structure paper.

Version 3.0.

1. Introductory.

1.1. How meaning works.

What is the place structure of a lujvo? A good question; to answer it, we need to establish what a lujvo means. To establish that, we need to ask what a tanru means. To answer that, we need to ask first what a gismu means. To answer that, in turn, we'll have to establish what a selbri means. And to answer *that*, we'll need to detour even further, getting somewhat philosophical (this is, after all, Lojban), and ask: what does it mean to 'mean'?

Since late last century, philosophers and linguists have accepted there are two components to meaning. The first is denotation (smuni se cuxna), and is the set of all entities in the world meant by an expression. The second is sense (smuni ve cuxna), and is akin to the dictionary definition of a word: it gives us the criteria by which to pick out the denotation of an expression. Thus, the sense of "the guy who wrote this paper"/"le prenu poi finti ledei skeprosa" is something involving notions of "guy", "writing", "time", "paper", and "this". The denotation of the same expression is the singleton set {Nick Nicholas} (lu'i la nitcion. NIkolas. lu'u).

Note that it is possible for two expressions to have different senses, but the same denotation. The classic example of this is "Evening Star" and "Morning Star". They are defined differently --- to find one, you look up in the evening, and to find the other, you look up in the morning --- but both refer to the same entity: {Venus}. If, however, two expressions have different denotations, they must have also (with an exception we'll come to shortly) had different senses --- and by extension, different meanings.

So let's build this up gradually. What is the meaning of "ti"? If I say "ti", and I'm pointing at my window, the denotation of "ti" is {Nick's window}. If my niece Elena says it and is pointing at her new Clam Chowder CD, the denotation of "ti" is {Elena's Clam Chowder CD}. In both cases, though, the sense of "ti" is "that which the speaker is pointing at". (So the denotation of "ti" changes according to the real-world circumstances in which it is used: the kind of expression linguists call 'deictic'. This is a can of worms which does not concern us directly here, so I'll discretely run away from it.)

What is the meaning of a selbri like "blue"/"blanu"? Well, the sense of "blanu" has to do with light wave frequencies; the denotation, however, of "blanu" is taken as "the set of all things which are blue"/"le selcmi be ro da poi ke'a blanu". Thus, the denotation of "blanu" is the set {my National Bank of Greece deposit book, my Collins Gem pocket Greek dictionary, my jeans, my CD of Gustav Mahler's early lieder... etc.}.

There's a *lot* of fine print to the sense of "blanu" --- how much of the surface of something has to be blue before the whole thing is considered blue? What's the borderline between blue and purple? and so on --- which I'm not going to go into. In fact, some more recent thinking on meaning, known as Prototype Semantics, argue for shades of grey (or in this case, blue) in denotation: some things are really blue, others are sorta blue, and "blue" itself has no fixed set membership. Although this approach makes a lot of sense, it makes it very difficult to talk about things like the denotation of gismu at all; so for the time being, for simplicity's sake, I will stick to the old-fashioned approach of 'something's either blue or it ain't'.

Still, for all the fine print of "blanu", its meaning is relatively straightforward, because it's a predicate with only one argument. What about gismu selbri, in general? Well, let's start small: take "house"/"zdani", a predicate with two arguments. What is its meaning? The way Lojban views the world, the sense of "zdani" is a relationship between a home ("le zdani") and its dweller ("le se zdani"). Both have to be present for the former to be called a home; if some building never had dwellers, never could have dwellers, and was never intended to have dwellers, then it's no home. Your local post office is probably a good example of this. (Inevitably, shades of grey come into play here too. What if a homeless person takes up residence at the steps of the post office at night? What happens to a house when the people living in it go on vacation? What about a display home? Once again, I'm going to sweep these difficulties under the carpet.)

The denotation of "zdani", then, as a relation (selbri), is the set of all pairs of entities for which the "zdani" relation holds. More explicitly, it's the set of all things, such that there's at least one being, such that the thing houses the being. In logic:

1.1) {<x, y> | \E y: zdani(x, y)}

And in Lojban:

1.2) lu'i ro da ce'o ro de vu'o poi da zdani de

So the meaning of any gismu in Lojban is tied up integrally with its place structure: the number of argument places it has, and the role assigned to each of those places. Adding or subtracting a place changes the nature of the relationship described by the selbri, and therefore its denotation as well. (Readers might want to ponder the difference between "klama" and "litru" as an instance of this.)

1.2. What tanru mean.

So what do tanru mean? To go on one more detour: an idea as old as that of sense and denotation in the philosophy of meaning is that of compositionality. The idea is very simple: it states that the meaning of a complex expression is built out of the meanings of its components. For example, the meaning of "dog-house" is built out of the meaning (sense and denotation) of "dog" and "house". The meaning of "the dog sat in the house" is a combination of the meanings of "dog", "sit", and "house". What is important is that each component adds information to the meaning, and does not subtract. The more we know about something, the more we can narrow down the alternatives on what it is; therefore, a complex expression will typically have a smaller denotation than its components. (There are less dog-houses in the world than there are generic houses --- including human houses, dog-houses, and bird nests. Yet another shade of grey --- the fact that "house" in English by default refers to human houses --- will join the other shades of grey under the carpet.)

In Lojban, a tanru is a combination of two selbri. Each of those selbri can themselves be tanru; so we'll take the simplest case of a tanru, in which both components are gismu (relgi'utanru).

Now, tanru consist of a left element (seltanru), and a right element (tertanru). The tertanru is the head (smuselga'i) of the tanru, and the seltanru is the modifier (smujazga'i). (In linguistics, 'head' refers to the 'major' component of the expression, which bears the main part of the meaning, while 'modifier' refers to the 'minor' component, which modifies the meaning. Thus, in "the dog", "dog" is the head and "the" is the modifier; in "blue car", "car" is the head and "blue" is the modifier. In English and Lojban, the modifier usually precedes the head, but this is not always so with all languages; thus cf. French "voiture bleu" ("car blue"/"karce co blanu").)

As to its meaning, the tanru is a modified version of its head. Therefore, as a selbri, the tanru will have the place structure of its head, and will have a denotation that is a subset of that of its head. Take for example "klama zdani". The gismu "zdani" has two arguments, while "klama" has five. As a selbri, the tanru "klama zdani" will also have two arguments, with the same roles as those of "zdani"; and since a "klama zdani" is a type of "zdani", its denotation will be a subset of that of "zdani". (The set of all goer-houses --- whatever that may be --- is a subset of the set of all houses.) Thus:

1.3) ti klama zdani la spot.

       This is a goer-house for Spot,


1.4) ti zdani la spot.

       This is a house for Spot,

and not

1.5)  ??la spot. klama zdani la vin. la berLIN. la poi banli .AUtoban.

       lo la trabant. karce
       Spot is a goer-house to Vienna from Berlin via The Great Autobahn in
       a Trabant car,


1.6) la spot. klama la vin. la berLIN. la poi banli .AUtoban. lo la

       trabant. karce
       Spot goes to Vienna from Berlin via The Great Autobahn in a Trabant car,

since a "klama zdani" is a type of house, and not a type of goer.

But are the places of the tertanru all that are involved in the meaning of a tanru? No. To see why, let us try to express the denotation of the tanru "gerku zdani", where "dog"/"gerku" is in Lojban a two-place predicate relating a dog to its breed. (Mongrels can join display homes under the carpet). Informally said, a tanru expresses a very loose relation: a "gerku zdani" is a house that has something to do with some dog. What the precise relation is is left unstated. Thus, the denotation of "gerku zdani" can include all of the following: houses housing dogs, houses shaped by dogs, houses which are also dogs (e.g. for fleas), houses named after dogs, and so on.

Let's build up a formal expression of this. The denotation of "gerku zdani" is the same as "ro gerku zdani". Since "gerku zdani" is a type of "zdani", we start with the denotation of "zdani", to narrow it down later. Thus:

1.7a) {<x, y> | zdani(x,y)}

We're looking at all the things (x) such that there's at least one animal (y), such that x is a house for y. More informally: take all things which are houses for something.

Now, "gerku zdani" denotes houses associated with dogs in some way. So let's define variables for the dog:

1.7b) \E z \E w: gerku(z,w)

There's some dog out there (z), and since it's a dog, there's some breed out there associated with that dog (w). (There might be more than one dog, associated with more than one breed.)

Then the denotation of "gerku zdani" is the set of all houses somehow associated with some dog. Tying all the strings together, we get:

1.7c) {<x, y> | \E z \E w: zdani(x,y) & gerku(z,w) & \E v: ckini(x,z,v)}

We're talking about all the things (x), for which there is, out there: something these things house (z --- making x a house), plus at least one animal (z) and a breed associated with that animal (w --- making that animal a dog), plus some sort of relation (v) between the house (x) and the dog (z).

Let's go through that again. For something (x) to qualify as a "gerku zdani", it's got to be a house, first of all. For it to be a house, it's got to house someone (y). Furthermore, there's got to be a dog out there (z). For z to count as a dog in Lojban, it's got to have some breed as well. (w). And finally, for x to be a "gerku zdani", as opposed to any old kind of "zdani", there's got to be some relation (v) between the house (x), and the dog (z).

Doubtless to the relief of the reader, here's an illustration. We want to find out whether the White House counts as a "gerku zdani". We go through the variables. The White House is the x. It houses Bill Clinton (y) as of this writing, so it counts as a "zdani". Let's take a dog --- say, Spot (z). Spot has to have a breed to count as a dog; let's say it's a Saint Bernard (w). Now, the White House counts as a "gerku zdani" if there is any relation at all (v) between the White House and Spot. The sky is the limit for (v); it can be as complicated as "The other day, *Spot* chased Socks, who is owned by Bill Clinton, who lives in the *White House*". (If no such v can be found, well, you take another dog, and keep going until you run out of dogs. This may look a bit unfeasible, but then, this is philosophy of meaning.)

After all that, a Lojban restatement of the definition of "gerku zdani" is due. In Lojban, (1.7c) looks like this:

1.7d) lu'i ro da co'o ro de vu'o poi su'o di su'o daxi1 su'o daxi2 zo'u:

       ge da zdani de
       gige di gerku daxi1
       gi da ckini di daxi2

And this is the point we've been leading up to. The gismu "gerku" and "zdani" each have two arguments. But, although the tanru "gerku zdani" is still a relationship between two entities (describing as it does a type of "zdani",) no less than five arguments are involved in its definition, and are used to determine its denotation: the house, the house dweller, the dog, the dog breed (everywhere a dog goes in Lojban, a dog breed follows), and the relation between the house and the dog. Since tanru are explicitly ambiguous in Lojban, the "daxi2" here cannot be expressed within a tanru (if it could, it wouldn't be a tanru anymore!) All the other places, however, can be expressed --- thus, (1.8):

1.8) la blabi zdani cu gerku befa la spot. bei la sankt.bernard. be'o

       zdani la bil.klinton.
       The White House is a dog (namely Spot, who is a Saint Bernard) type of
       house for Bill Clinton.

Not the most elegant sentence ever written in either Lojban or English. Yet if there is any relation at all between Spot and the White House, (1.8) is true. If we concentrate on just one type of relation in interpreting the tanru "gerku zdani", then the denotation of "gerku zdani", and its meaning, change. So if "gerku zdani" only meant "dog house", the White House would no longer be a "gerku zdani" with respect to Spot, given the scenario where Spot chased Socks. (Indeed, it probably wouldn't be a "gerku zdani" even if Spot lived in the White House; a point I'll come back to later.)

So what does a tanru mean? A tanru denotes a kind of head gismu, related somehow to a modifier gismu. The denotation of the head gismu is determined by all of its arguments. The denotation of the modifier gismu is determined by all of its arguments. Therefore, the denotation of the tanru is determined by the combination of all the head gismu arguments, all the tail gismu arguments, and the presence of some relation between what the head and the modifier denote. In short, every sumti in sight gets roped in in the attempt to work out what the tanru can denote.

1.3. What lujvo mean.

This is a fairly long way to go to try and work out how to say "dog-house"! The reader can take heart; we're nearly there. Recall that one of the arguments involved in fixing the denotation of a tanru --- the argument left deliberately vague --- is the precise relation between the head and the modifier. Indeed, fixing this relation is tantamount to giving an interpretation to the ambiguous tanru.

Lujvo are defined as single, disambiguated instances of tanru. That is to say, when we try to work out the denotation of a lujvo, we don't need to add to our list of "there exists"'s to look for, the relation between the head and the modifier. We already know what kind of relation we're looking for; it's a given. Thus, recall (1.7c)

1.7c) {<x, y> | \E z \E w: zdani(x,y) & gerku(z,w) & \E v: ckini(x,z,v)}

We're talking about all the things (x), for which there is, out there: something these things house (z --- making x a house), plus at least one animal (z) and a breed associated with that animal (w --- making that animal a dog), plus some sort of relation (v) between the house (x) and the dog (z).

Let us also, for the moment, sidestep issues of explicit place structure, by using the denotation of the sumti "lo gerku zdani":

1.7d) {x | \E y \E z \E w: zdani(x,y) & gerku(z,w) & \E v: ckini(x,z,v)}

In working out what "gerku zdani" denoted, we had to find a housed entity, a dog, a dog breed, and a relation between the dog and the house. For the corresponding lujvo "gerzda", on the other hand, we already know what the relation we're looking for is. Let's say for now that the relation we're looking for is that the house houses the dog. Then (1.7d) turns into (1.9), the denotation of "gerzda":

1.9a) {x | \E y \E z \E w: zdani(x,y) & gerku(z,w) & ckini(x,z,(zdani(x,z))}

We're talking about all the things (x), for which there is, out there: something these things house (z --- making x a house), plus at least one animal (z) and a breed associated with that animal (w --- making that animal a dog), where the house (x) and the dog (z) are related in that the house houses the dog.

And in Lojban:

1.9b) lu'i ro da poi su'o de su'o di su'o daxi1 zo'u:

       ge da zdani de
       gige di gerku daxi1
       gi da ckini di leka da zdani di

So we've eliminated one place required to work out the denotation of the tanru: the "daxi2" associated with the precise nature of the head-modifier relation.

The insight driving the rest of the paper is this. While the relation expressed by a tanru can be very distant (e.g. Spot chasing Socks, above), the relationship singled out for disambiguation in a lujvo will be quite close. This is because lujvo, parallelling natural language compounding, pick out the more salient relation between head and modifier to be expressed in a single word. The relationship of "dog chases cat owned by person living in house" is too distant, and too incidental, to be likely to be expressed as a single word; the relationship of "dog lives in house" is not.

In fact, the relationship will almost always be so close, that the selbri of the te ckini expression --- the predicate expressing the relation between head and modifier --- will be either the head or the modifier predicates themselves. Thus in the definition of "gerzda" in (1.9b), the head of the lujvo (the veljvo; the modifier is the terjvo) is "zdani", and the relationship between dog and house is expressed as "da *zdani* di".

This should come as no surprise, given that a word like "zdani" in Lojban is a predicate. Predicates express relations; so when you're looking for a relation to tie together "zdani" and "gerku", the most obvious relation to pick is the very relation named by the veljvo, "zdani": the relation between a home and its dweller. (There are some exceptions to this rule; they will be addressed much further down.)

So, given that the te ckini, the relation, will use the veljvo or the terjvo as its selbri, let's look at (1.9b) a bit more closely. The definition tells us that we're looking for at least one entity ("de") that is housed by "da". But we know from the te ckini that there *is* an entity, the dog "di", that is housed by "da". So we've already found our "de", our house dweller: it is the dog, "di". So we lose nothing by eliminating "de", and saying:

1.9c) lu'i ro da poi su'o di su'o daxi1 zo'u:

       ge da zdani di
       gige di gerku daxi1
       gi da ckini di leka da zdani di

But (1.9c) merely says "X is a house for Z (a dog of breed W), *and* X is related to Z in that X is a house for Z." But whenever X is a house for Z, it's pretty obvious that X is also going to be related to Z, precisely in that X is a house for Z. So the statement of relationship tells us nothing we don't already know; and we can readily recast (1.9c) as:

1.9d) lu'i ro da poi su'o di su'o daxi1 zo'u:

       ge da zdani di
       gi di gerku daxi1

Which leads to a conclusion, and a corollary. The conclusion is that, since the relationship between the terjvo and veljvo is expressed by the terjvo or verljvo selbri, at least one of the places of the veljvo is always going to be identical to a place of the terjvo --- and is thus redundant, and can be dropped from the list of sumti required to fix the lujvo denotation. The corollary is that the precise relation between the terjvo and veljvo can be made implicit by finding a place of the veljvo and the terjvo to overlap in this way.

So what is the place structure of "gerzda"? According to (1.9d), we're left with three variables, since the dweller ("se zdani") turned out to be identical to the dog ("gerku"). Ignoring for now the question of whether the dog breed matters or not, we can proceed as follows:

Let's use the notation xn to denote the nth place of the gismu starting with the letter x. Thus, z1 is the first place of "zdani", and g2 is the second place of "gerku".

The place structure of "zdani" is: z1 z2.

The place structure of "gerku" is: g1 g2.

The maximum possible set of arguments for "gerzda" is: z1 z2 g1 g2.

(Two notes. First, these are for the moment unordered sets; we'll come to the question of ordering later. Second, we're consistently assuming that lujvo can't have more arguments than its terjvo and veljvo put together; i.e. that the gismu "gerku" and "zdani" give us all the information we need to work out what a "gerzda" is, and we won't require extra places, like "the builder" or "the postman". More on why this might not always be the case, later.)

But z2=g1.

Therefore, the possible set of arguments for "gerzda" now becomes: z1 g1 g2

By stating that z2=g1, we have declared what the relation between the terjvo and veljvo is. Assuming for now that g2 is included in the place structure, and that the places are ordered as given, we can state the denotation of "lo gerzda" as follows:

1.10a) lu'i ro da poi su'o di su'o daxi1 zo'u:

       da noi ke'a zdani
       cu gerzda di noi ke'a gerku ku'o
       daxi1 noi ke'a se gerku

And, finally, for the selbri itself, we will (for now) take all the remaining arguments out of the prenex, and consider the lujvo "gerku" a relation between z1, g1 and g2:

1.10b) lu'i ro da ce'o ro di ce'o ro daxi1 vu'o poi

       da noi ke'a zdani
       cu gerzda di noi ke'a gerku ku'o
       daxi1 noi ke'a se gerku

Our task is not yet done: we still need to decide whether any of the remaining places should also be eliminated, and what order the lujvo places should appear in. These concerns will be addressed in the remainder of the paper; but we are now equipped with the terminology for those discussions. In the remainder of this introduction, I discuss some more general issues that won't be dealt with in the remainder of the paper.

1.4. What are lujvo for.

The job lujvo have is to extend the Lojban vocabulary. To see how they do this, let us look at how we might try to express the notion "taller" in Lojban. There is no single gismu in Lojban capable of expressing the concept. Therefore, we have four alternatives for expressing the concept compositionally, as a combination of gismu. We can express "taller" as a fully expanded phrase, as a sumti tcita phrase, as a tanru, or as a lujvo. Each of these possibilities is illustrated below:

1.11a) .i (mi) zmadu (do) [leka (mi) clani (lo'e remna) kei]

       (lo mitre be li pimu)

1.11b) .i (mi) semau (do) vemau (lo mitre be li pimu) cu clani

1.11c) .i (mi) clani be (lo'e remna) be'o zmadu (do) (zo'e)

       (lo mitre be li pimu)}

1.11d) .i (mi) clamau (do) (lo mitre be li pimu)

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

Version (1.11a) is a fully explicit expression of a particular instance of "taller", using only gismu and cmavo. It is as unambiguous an expression as is possible in Lojban, and can be considered a valid disambiguation of a more vague expression of the same concept. In fact, Lojbanists tend to think of an expression like this as "underlying" more ambiguous forms; it is somehow the 'real' way of expressing a concept. This is because tanru are inherently ambiguous; and any unambiguous definition of lujvo within Lojban, in order not to be circular, needs to be made in terms of units whose definition is already known. But the only units whose definition have been fixed from the outset are gismu and cmavo. For that reason, I refer to such an expression as a "gismu deep structure" (GDS) in this paper --- by analogy with the 'Deep Structure' used in early transformational syntax. (Note that this is not intended to make any generative-transformationalist claims about the language.)

Note, however, that while (1.11a) expresses the notion of "taller" as required, the Lojban vocabulary has not been extended. In particular, if a gismu were being designed from scratch to express the notion "taller", it would necessarily include the following places: the entity which is taller; the entity which is shorter; and the difference in height. (Thus "_I_'m _half a meter_ taller than _you_".) Neither of the two gismu in (1.11a) have such a place structure. Instead, they express an equivalent notion by sharing out the required places, as well sense information like 'length' and 'more', between the places of "zmadu" and "clani". Since there is no new selbri, the Lojban vocabulary has not been extended.

Example (1.11b) comes closer to expressing the notion "taller" using a single selbri. The only selbri in the entire phrase is "clani"; and the phrase is just as unambiguous an expression of "taller" as is (1.11a). The way this has been achieved, however, is by tacking on the required places for "taller" onto the place structure of "clani" by using sumti tcita. In Lojban thinking, sumti tcita do not provide essential information for the definition of a selbri. All arguments necessary for the proper definition of a selbri are already meant to have been included in the place structure defined for it. This is the reason why, while Lojban has the means to supplant numbered sumti places with sumti tcita places, giving a case grammar language (for example, "gau mi catra vebe'i le puljyralju" instead of "mi catra le puljyralju" for "I shot the sherrif"), this is not normally done.

So sumti tcita actually play a second-class role in the specification of arguments of a bridi; and in fact (1.11b) would not be regarded as a gismu deep structure. It would rather be regarded itself as having something like (1.11a) as a GDS. So while the use of sumti tcita might give a 'quick fix' to the problem of extending Lojban vocabulary, it does not go to the heart of the matter. Nor, it should be noted, can it eliminate any of the numbered places of the bridi the speaker may feel are irrelevant to the notion she wants to express. If, for instance, she wants to eliminate the "se clani" place as irrelevant to "taller", she can't: she's stuck with it.

Example (1.11c) goes a lot further towards increasing Lojban vocabulary: the tanru "clani zmadu" is a unit that can correspond directly, as a selbri, to the notion of "taller" --- which the gismu "zmadu" and 'clani" in isolation could not. But tanru have the problems already discussed: tanru are inherently ambiguous; they cannot be entered into anything like a dictionary because of the on-the-fly way in which they are created; and they do not express a straightforward syntactic relation between their arguments. As discussed, we need three arguments for our notion of "taller" --- but they are scattered as arguments #1,

  1. 2 and #4 of "zmadu", while argument #3 of "zmadu" has ended up totally

superfluous, and argument #2 of "clani" may end up being undesirable. In other words, a new concept has been created semantically; but it has brought with it a host of alternative interpretations, and the creation is not a single lexical unit as far as sumti distribution is concerned. A new place structure has been created; but as discussed already, it is too diffuse to be of much use. So tanru do not extend Lojban vocabulary in the way we require.

Which brings us to (1.11d), which uses a lujvo, and which --- no surprise --- is exactly what is required: it is a new, single selbri, with a well-defined sense and denotation, and a predefined relation between its terjvo and veljvo. It has its own place structure, which corresponds to the place structure intuitively required for "taller", without introducing extraneous places (like "te zmadu") or syntactic convolution (like the tanru). The lujvo "clamau" is a new Lojban word.

As stated, a lujvo unambiguously expresses a specific relation between its arguments, out of the many its ambiguous seljvo (base 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'. (Dictionary writers could override someone's nonce usage if they see a more obvious or useful interpretation for a lujvo.) The purpose of this paper is to give some principled parameters within which the different possible interpretations of the seljvo can be constrained, and the new place structure built up.

1.5. Place structures and Sense.

The guidelines outline in this paper discuss how to derive the place structure for a lujvo. Is that sufficient to define the new meaning of a lujvo?

Surprisingly, possibly, in view of the extensive discussion on denotation above, the answer is no. As John Cowan once said, "lujvo [...] have meanings that are constrained by the place structures of the underlying gismu, but not fully determined by them.")

Take for instance the lujvo we've been discussing so far, "gerzda". We have assumed that the GDS of "gerzda" is basically "lo zdani be lo gerku", and have described its place structure accordingly. However, the same shades of grey that came up in discussing the denotation of gismu (remember that carpet) turn up for "gerzda"; if anything, they multiply in number, given the semantic complexity of the notion. What happens, for example, if a tortoise moves into Fido's doghouse? It's all very well to say "le zdani cu gerzda la faidon.", because we know that "le se gerzda" is a "gerku". But can we say "le resprtestudine. cu se gerzda"? The doghouse is still a doghouse, after all, and it is housing the tortoise. In this particular case, I would argue no: a doghouse is a "gerzda" in that it houses dogs; but with respect to anything else taking up lodging inside it, it is just a "zdani". So I would argue that the phrase "le resprtestudine. cu se gerzda" is wrong (I'm loath to say 'false', for reasons outside the domain of this already wide-reaching paper), because "gerzda" carries with it a presupposition that the "se zdani" is a dog.

Or, to take another difficulty, is it sufficient that the "gerzda" houses a dog? Or should it be a construction explicitly designed for the purpose of housing a dog? (Which is what a kennel is.) If a dog takes up residence in the city park, is the city park a "gerzda"? What about Uncle Bob's spacious country villa, one of whose rooms is Rex's own domain: is the villa a "gerzda"? The room? It might be true of the villa that "ri zdani lo gerku"; but, if we choose to restrict "gerzda" to a meaning like "dog-kennel", it might not be true of the villa that "ri gerzda".

These are complicated issues, that are not directly connected with the determination of a place structure for "gerzda", which might have the same place structure whatever they decide. Arguably, however, each of these alternative interpretations has a different GDS (one that might involve a few more ZAhO cmavo, or a few more seltanru than are in the seljvo); so they will be further discussed below, in the section on Choice of seljvo.

The guidelines outlined here should be regarded as precisely that: guidelines. If the issues to do with the sense of a lujvo become so complicated that the derived place structure needs to be fiddled with, the result is certainly not ungrammatical or unacceptable in any formal way. Indeed, these guidelines were not included in the definition of the language, but were formulated after the fact, and independently of the main language-engineering effort. For this reason, this paper does not have the same prescriptive force as other papers published by the LLG, and does not represent official LLG policy.

But the guidelines still matter. If a desired place structure does not match that obtained by these guidelines, it will be difficult to interpret and to anticipate. It may indicate that the seljvo chosen may not be the best for the job, and may not actually express what is desired. In general (although this cannot be ruled out) there should be no need to introduce places external to the component gismu, or to radically reorder the lujvo places.

2. lujvo place selection.

The approach taken in this paper to the question of what places a lujvo has in its place structure, as already discussed, is that the lujvo places are a subset of the set of places of its component gismu. So the places of a lujvo are derived from the set of places of the component gismu by a process of eliminating places, until just enough places remain to give a proper denotation to the lujvo.

It is also possible to design the place structure of a lujvo from scratch, treating it as if it was a gismu, and working out what arguments contribute to the notion to be expressed by the lujvo. Although this is the procedure implicit in the official LLG position on lujvo, there are two reasons arguing against it, and in favour of the procedure detailed in this paper.

The first is that it might be very difficult for a hearer or reader, who has no preconceived idea of what concept the lujvo is intended to convey, to work out what the place structure actually is. Instead, she will have to make use of a lujvo dictionary every time a lujvo is encountered, to work out what a "se jbopli" or a "te klagau" is. But this means that, rather than having to memorise just the 1300-odd gismu place structures, a Lojbanist will also have to memorise thousands of lujvo place structures with no apparent pattern or regularity to them. No natural language forces such a task on its speakers: their predicates rarely have more than three places, and when they do, adpositions (prepositions and postpositions) are used to disambiguate what the places mean. But in Lojban, as we have seen, sumti tcita (which correspond to adpositions) are regarded as a secondary, non-intrinsic way to form place structures; so we're left again, again, with thousands of lujvo place structures, with no regularity to them which can be exploited by the language learner. The purpose of these guidelines is to apply and enforce such regularity.

The second reason is related to the first: if the seljvo of the lujvo has not been properly selected, and the places for the lujvo are formulated from scratch, then there is a risk that some of the places formulated may not correspond to any of the seljvo gismu places. If that is the case --- that is to say, if the lujvo places are not a subset of the seljvo gismu places --- then it will be very difficult for the hearer or reader to understand what the particular place means, and what it is doing in that particular lujvo. This is a topic that will be further discussed below.

Where second-guessing the place structure of the lujvo is useful is in guiding the process of subsequently eliminating places from the seljvo. If the Lojbanist has an idea of what the final place structure should look like, she should be able to pick an appropriate seljvo, to begin with, to express the idea, and then to decide which places are relevant or not relevant to expressing that idea.

Now, as discussed, there are two criteria for eliminating places from the component gismu of a lujvo:

  • The place conveys redundant information.
  • The place conveys irrelevant information.

As argued, if the te ckini between the terjvo and veljvo is expressed by either the terjvo or veljvo selbri, then at least one of the places in the veljvo will be redundant, as it will be copied in the terjvo, and can therefore be eliminated. Depending on whether the te ckini is the the terjvo or veljvo, there are thus two classes of lujvo, which I now outline.

2.1. Eliminating redundant places: lujvo classes.

2.1.1. lujvo classifications: "be"-lujvo and "gi'e"-lujvo.

Consider again the tanru {gerku zdani}. Let us attempt to use gismu deep structures, as defined above, to disambiguate the tanru within Lojban. Note that we will follow all the restrictions applicable to gismu deep structures; in particular, we cannot use sumti tcita. The possible meanings of "gerku zdani" include:

2.1) "a house that houses dogs; a dog-house"

       lo zdani be lo gerku

2.2) "a house which is also a dog; a dog-cum-house"

       lo zdani poi ke'a gerku
       Or, alternatively,
       da poi ke'a zdani gi'e gerku

(We could also say "lo zdani je gerku", but "je" forms tanru, so this would not be a GDS.)

2.3) "a house shaped like a dog"

       lo zdani poi ke'a se tarmi lo gerku

2.4) "a house named after a dog"

       lo zdani poi ke'a se cmene lo cmene be lo gerku

2.5) "a house whose inhabitant's cat is chased by a dog"

       lo zdani be da be'o poi lo mlatu poi ke'a se ponse da cu se jersa
       lo gerku

Now, let us attempt to state the te ckini of these candidate interpretations for the lujvo "gerzda". Let's take the last three interpretations first:

2.3a) lu'i ro da ce'o ro de ce'o ro di ce'o ro daxi1 vu'o poi

       ge da zdani de
       gige di gerku daxi1
       gi da di ckini
               leka su'o daxi2 zo'u:
               di da tarmi daxi2

2.4a) lu'i ro da ce'o ro de ce'o ro di ce'o ro daxi1 vu'o poi

       ge da zdani de
       gige di gerku daxi1
       gi da di ckini
               leka su'o daxi2 su'o daxi3 su'o daxi4 zo'u:
               ge daxi2 cmene di daxi3
               gi daxi2 cmene da daxi4

2.5a) lu'i ro da ce'o ro de ce'o ro di ce'o ro daxi1 vu'o poi

       ge da zdani de
       gige di gerku daxi1
       gi da di ckini
               leka su'o daxi2 su'o daxi3 su'o daxi4 zo'u:
               ge da zdani de
               gige de ponse daxi2 daxi3
               gige daxi3 mlatu daxi4
               gi di jersi daxi3

What's worth noting in these examples is that none of the te ckini involve the terjvo or veljvo alone. The predicates they involve are, instead, "tarmi", "cmene", and, for (2.5), "zdani", "ponse", "mlatu" and "jersi". This is a reflection of the fact that the terjvo-veljvo relation is rather distant. Indeed, as will be argued later, to properly express these notions, a lujvo ought to include in its seljvo those very te ckini selbri. Thus, (2.3) would be better expressed as "gertaizda" ("dog shape house", "dog-shaped house"), (2.4) as "gercmezda" ("dog name house", "dog-named house"), and (2.5) --- if one really must --- as "se gerselje'ilatpo'eselzda" (from "se ke gerku se jersi mlatu ponse se zdani").

Note also that, as a result of using selbri other than the terjvo and veljvo, new places required for the definition of the te ckini have been introduced into the prenex of the te ckini; for example, "daxi2" in the examples above. This further reflects that the relation described by these lujvo interpretations is broader than an interpretation involving just the terjvo and veljvo.

For the two first interpretations, on the other hand, the denotation is as follows:

2.1a) lu za'e gerzda li'u poi se ciksi lu zdani gi'e gerku li'u cu sinxa

       lu'i ro da ce'o ro de ce'o ro di vu'o poi
               ge da zdani de
               gige da gerku di
               gi da ckini da leka
                       ge da zdani de
                       gi da gerku di

2.2a) lu za'e gerzda li'u poi se ciksi lu zdani be lo gerku li'u cu sinxa

       lu'i ro da ce'o ro de ce'o ro di vu'o poi
               ge da zdani de
               gige de gerku di
               gi da ckini de leka da zdani de

In interpretation (2.2), which we've already looked at at length, the veljvo is the te ckini; in other words, the terjvo describes one of the non-x1 arguments of the veljvo (i.e. "gerku" describes the x2 of "zdani"). The disambiguating GDS for this interpretation involves the word "be" ("zdani be lo gerku"); so this type of terjvo-veljvo interpretation gives rise to what we will call "be"-lujvo. Note that, although it is usually the x2 that is described in this way, that is not always the case. For example, due to reasons of plausibility to do with what type of argument typically fills what place, the lujvo "karcykla" is interpreted as "klama befu lo karce", "car travel", and not "klama be lo karce", "going up to cars".

In interpretation (2.1), on the other hand, the relation lies in that both the veljvo and the terjvo are describing the same entity --- the x1 of both predicates. The GDS for this interpretation (or at least one version of it) involves the word "gi'e" ("zdani gi'e gerku"); so this type of terjvo-veljvo interpretation gives rise to what we will call "gi'e"-lujvo. As an illustration of this, consider the lujvo "balsoi": it is interpreted, not as "sonci be lo banli", "soldier of a great [army]", but "sonci gi'e banli" "both a soldier and and a great one" --- i.e. "great soldier", which is the interpretation we would tend to give its seljvo, "banli sonci".

In this case the x1 of "sonci" is redundant, since it is identical to the x1 of "banli". Therefore the place structure of "balsoi" cannot include places for both the x1 of "sonci" and the x1 of "banli", which refer to the same thing. So the place structure set of "balsoi" is at most {b1=s1 s2 b2 b3}.

Which interpretation is selected for any given lujvo depends on the semantics of its component gismu, and what speakers regard as plausible and implausible in the world, as well as what concept a speaker intended to convey. For instance, while it is true that many insects and micro-organisms live on or in dogs, we do not tend to associate the host-parasite relation with our (prototypical) notion of "housing"/"ka zdani", since our mental model of "house" is not usually a living being. Therefore, we would reject the "gi'e"-interpretation of "gerzda" as implausible. Similarly, the interpretation of "karcykla" as "going by car", rather than "going up to cars", accords with our mental model of what types of sumti we would expect to fill the various places of the predicate "klama"/"go": cars are frequently means of locomotion, but are not often thought of as destinations in themselves.

On the other hand, terjvo which correspond to adjectives and adjectival verbs in natural languages tend to be construed as adjectival in Lojban as well. (It would be interesting to ponder whether this is "malrarbaukai" --- the natural language equivalent of "malglico" --- or not.) But for a terjvo to function adjectivally means that it is describing the same entity its head, the veljvo, is describing. Therefore, 'adjectival' gismu like "banli", when used as terjvo, will tend to give rise to "gi'e"-lujvo. Hence "balsoi" is interpreted as a "gi'e"-lujvo, giving "great soldier".

2.1.2. lujvo classes: "belenu".

As a special case of "be"-lujvo, consider (2.6a) as a candidate GDS for the lujvo "nunctikezgau":

2.6a) mi gasnu lenu le gerku cu citka loi guzme

       "I make the dog eat melons."

giving the seljvo:

2.6b) mi nu lo gerku cu citka loi guzme kei gasnu

       which simplifies to
       mi nuncitka be fa lo gerku bei loi guzme be'o gasnu

and the lujvo:

2.6c) mi nunctikezgau le gerku loi guzme

       "I feed the dog melons."

Note, for the time being, that the place structure of "nunctikezgau" given in (2.6c) corresponds to that of the English verb "feed". Both of them have a place for a feeder, someone fed, and the food used. The interpretation of "nunctikezgau" is that of a "be"-lujvo; thus, "le nunctikezgau" can be paraphrased as "le gasnu be le nuncti".

Both "feed" and "nunctikezgau" are examples of causatives, predicates in which an agent brings about an action described by the modifier. Thus "feed" is the causative corresponding to "to eat"; the former can be derived from the latter by paraphrasing "I feed the dog melons" as "I bring it about that the dog eats melons". Similarly, the transitive verb "sink" (eg. "I'm sinking boats in the ocean") is the causative of the intransitive "sink" (eg. "I'm sinking in the ocean"). It is thus possible to paraphrase the causative version of "sink" in terms of the intransitive: "I'm sinking boats in the ocean" means the same as "I'm causing boats to sink in the ocean".

While all languages have causatives, they use 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 read" is the causative for "to read"). The derivation of causatives in Turkish and Esperanto, on the other hand, is quite regular. "To eat" in Esperanto is "manghi", and "to feed" is "manghigi"; "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-word causatives, like those in Turkish and Esperanto, would be useful in a language like Lojban. The alternative, using a fully expanded GDS as in (2.6a) every time the idea of "feeding" needs to be expressed, is longwinded and counterintuitive. The phrase "mi gasnu lenu da citka de" involves sumti nesting and the complexity of place abstraction; and it conceals the fact that we perceive feeding as a direct relation between three entities (the feeder, the fed, and the food). The lujvo "nunctikezgau" has no such disadvantage: as (2.6c) shows, the three entities are related directly by just one selbri.

Causatives have already 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" corresponds not 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". (The place structure ordering may seem somewhat odd; see the next section for an explanation.)

Now, the selbri "nunctikezgau", by which we have translated the verb "feed", expresses a relation between an actor and an event of eating ("nu citka"). As noted, this makes it a "be"-lujvo. In this lujvo, we would like a place for the actor (g1), for the eater (c1), and for the food (c2). The seljvo of "nunctikezgau" is "nu citka kei gasnu"; since we would prefer to work with a single word for both the terjvo and the veljvo, let's rephrase this as "nuncitka gasnu". The lujvo "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.

So now we have to decide on the place structure of "nunctikezgau". The candidate places are: n1 c1 c2 g1 g2. The place g2 (what is brought about) obviously denotes the same thing as n1 (the event of eating: "le se gasnu cu nuncitka"). So we can eliminate g2 as redundant, leaving us with n1=g2 c1 c2 g1.

But it is also possible to omit the n1 place itself. The n1 place describes the event brought about; an event in Lojban is described as a bridi, by a selbri and its sumti; the selbri is already known (it's the terjvo), and the sumti are also already known (they're in the lujvo place structure set. So n1 would not give us any information we didn't already know. To demonstrate the validity of this, compare the following two place structures:

2.7) mi za'e nunctikezgau lenu gerku cu citka loi guzme kei

       le gerku loi guzme

2.6c) mi nunctikezgau le gerku loi guzme

       "I feed the dog melons."

Example (2.7) leaves in the n1 place, which promptly repeats all the information already given by the terjvo ("nuncti" = "nu citka"), the x3 place ("le gerku") and the x4 place ("loi guzme"). Therefore, the n1 place doesn't actually seem to be doing much constructive; and in (2.6c), it is eliminated, giving the place structure we have already discussed as desirable.

It could be argued that the n1 place might be necessary as a place to put sumti tcita places, involved in the specification of the terjvo event, but not included in the terjvo place structure. For example, the motivation of the terjvo and veljvo events might differ: thus,

2.8a) mi gasnu lenu le gerku cu citka mu'i lenu ri co'u xagji kei kei mu'i

       lenu mi na'e se fanza lenu gy. krixa
       "I brought it about that the dog ate to end its hunger, so that it
       wouldn't bother me with its barking",

where satiation is the motivation for the dog eating (the terjvo event), while quiet is the motivation for me feeding it (the veljvo event.) We would counterargue that this is not sufficient motivation to prefer (2.7) over (2.6c) --- particularly since, if we do want to express the notion in (2.8a), we don't need to use the lujvo "nunctikezgau", as in (2.8b):

2.8b) mi za'e nunctikezgau lenu le gerku cu citka mu'i lenu ri co'u

       xagji kei kei le gerku mu'i lenu mi na'e se fanza lenu gy. krixa

Example (2.8b) simply duplicates the places of "citka" both inside and outside its x2, without reducing the syntactic or semantic complexity of (2.8a) in the slightest. In fact, (2.8a) is much more effective in expressing what it sets out to than is (2.8b). The function of lujvo, and their natural language counterparts, is to encode direct relations between a small number of arguments, just as is the case for gismu. Much more complicated relations, such as that in (2.8a), occur much less frequently, and a simple lujvo place structure would only obscure them.

So now we have arrived at a place structure for "nunctikezgau". There is one further step that can be taken. As we have already seen with "balsoi", the interpretation of lujvo is constrained by the semantics of gismu, and of their sumti places. Now, any "be"-lujvo with "gasnu" as a veljvo will involve an abstract event, since that is how the x2 of "gasnu" is defined in the gismu list. Therefore, if we assume that "nu" is the type of abstraction one would expect to be a "se gasnu", then the rafsi "nun" and "kez" in "nunctikezgau" are only telling us what we would already have guessed. If we drop them out, and use instead the lujvo "ctigau", rejecting its "gi'e"-interpretation ("someone who both does and eats"; "an eating doer"), we could still work out that the terjvo refers to an event. (You can't "do an eater"/"gasnu lo citka", with the meaning of "do" as "bring about an event"; so the terjvo must refer to an event, "nu citka". The meaning in English slang of "do someone" as "socialise with" or "have sex with" is not related to "gasnu".) So we can say simply:

2.9) "ctigau": agent x1 causes x2 to eat x3. (x1 feeds x3 to x2.)

Here the place g2, the action performed ("se gasnu"), is equivalent to an abstraction composed of all the places of "citka". Rather than having a "be"-lujvo, where a single place of the terjvo is equivalent to a place in the veljvo, we now substitute a number of terjvo places for that single veljvo place --- all the places, in fact, of the terjvo. This particular interpretation, in which the terjvo describes an abstraction which is a place of the veljvo, will be called a "belenu"-lujvo (since for example, "le ctigau" is equivalent to the GDS "le gasnu be lenu citka"). This lujvo interpretation and place structure turns up not only for causatives ("rinka" and "gasnu" veljvo), but also for most veljvo with an abstract sumti in x2 or x3. Thus for example, "limdji" would be probably interpreted as "want to swim", rather than "want a swimmer" or "both want and swim".

Remember that "belenu"-lujvo are really an abbreviation of "be"-lujvo based on abstract sumti. The lujvo "ctigau" is really an abbreviation for "nunctikezgau" (a point already discussed in the Abstraction paper), and the conditions for such elision --- namely, that it does not lead to excessive ambiguity --- are no different to those for other elidable rafsi, discussed in section 4. There are in fact only a few veljvo for which the rafsi "nun" would be clearly implicit in the terjvo; for other veljvo with abstract places, an ambiguity results between "be"- and "belenu"-interpretations.

For example, the x2 of "nelci" can be either an event or a simple sumti. The lujvo "sonynei", in the "be"-interpretation, refers to someone who likes a soldier ("nelci be lo sonci"). In the "belenu"-interpretation, it is someone who likes being a soldier ("nelci be lenu le no'a cu sonci"), or perhaps someone who likes for someone else to be a soldier ("nelci be lenu zo'e sonci"). In these cases, it is safer to say "nunsonynei" for the latter case, and "sonynei" for the former (since only the "belenu"-interpretation is available for "nunsonynei", while there is no other lujvo which can express the "be"-interpretation of "sonynei"). In general, Lojbanists should 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, "ctuca" or "nelci".

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

2.1.3. lujvo classes: Summary.

  • In "gi'e"-lujvo, the x1 of the veljvo is semantically "doubled up" by

(denotes an object identical to, or is also described by) the x1 of the terjvo. Therefore it should not be necessary for both these places to appear in the lujvo place structure.

  • In "be"-lujvo, the x2 of the veljvo (typically; sometimes, the x3) is

doubled up by the x1 of the terjvo. Again, the two places should not both appear in the lujvo place structure.

  • In "belenu"-lujvo, the x2 (typically) of the veljvo is an event

abstraction, whose selbri is the terjvo, and whose internal sumti are the sumti of the terjvo. This place in the veljvo is thus replaced by the places of the terjvo selbri.

The claim implicit in presenting these classes as ways of interpreting lujvo, and identifying redundant places in their place structure, is that they are sufficient to analyse any lujvo; and when they do not apply to a lujvo, it is because a selbri has been left out of the seljvo. Lojbanists should take this claim with a grain of salt, and make sure they are careful with how they define any lujvo they create.

It should be noted that, while the types of redundancy covered by these three types are the most frequent occuring in Lojban lujvo, they are not the only redundancies possible. It is possible that other terjvo and veljvo places have the same referent, and can thus also be eliminated.

Take the lujvo "ninpe'i", "to newly meet", which has been used to translate "to meet someone for the first time". (Note that this cannot translate "to be introduced to", since there is nowhere in the seljvo an introducer place could come from. The verb "to introduce" would probably be conveyed by "ninpe'igau", "to make someone newly meet someone".) We can analyse "ninpe'i" as a "be"-lujvo: its GDS paraphrase is "penmi be lo cnino". The fully elaborated 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: out of the five places involved in the seltanru, two are redundant. Therefore, the place structure of "ninpe'i" can be given by "da ninpe'i de di": x1 meets x2 (who is new to x1) for the first time at x3.

There seems to be little motivation to eliminate places from je-lujvo, but much motivation to merge them: if the x1 of both the terjvo and veljvo refer to the same entity, it is likely that other arguments of the two selbri will have the same referent as well. Thus in "zugyxu'a", "to guilty-assert, to assert guiltily, to confess", not only do we merge the asserter (x1) with the person feeling guilt (z1), as normal, but we also merge the assertion (x2) with what guilt is felt about (z2). The GDS of "zugyxu'a" is thus "da zugni de gi'e xusra de": someone both feels remorse about a misdeed, and states the misdeed.

2.2. Eliminating irrelevant places: lujvo definition.

2.2.1. Place elimination in terjvo.

[Note to reviewers: I need more examples. Please give generously.]

Let us go back to considering how we determined the (maximal) place structure of a lujvo, using "gerzda". We started with the sumti "lo gerku zdani":

1.7d) {x | \E y \E z \E w: zdani(x,y) & gerku(z,w) & \E v: ckini(x,z,v)}

We then gave the lujvo "gerzda", with seljvo "gerku zdani", a "be"-interpretation. As a result, "lo gerzda" has the interpretation:

1.9a) {x | \E y \E z \E w: zdani(x,y) & gerku(z,w) & ckini(x,z,(zdani(x,z))}

Or, stated otherwise,

(2.10a) {x | \E z \E w: zdani(x,z) & gerku(z,w)}

This leaves us with three places involved in the definition of "lo gerzda". We then decided that, since these three places are involved in the definition, "gerzda" must be a relation between all three. We thus gave the selbri "gerzda" the denotation in (1.9b), restated in logic as (2.10b), where the z and w places have been lifted out of the prenex of the relative clause and enstated as places of the predicate:

2.10b) {<x, z, w> | zdani(x,z) & gerku(z,w)}

Consider, however, the case where we do not lift both arguments out of the prenex, but leave w, the species of the dog, inside the prenex, giving the following denotations:

2.10c) {<x, z> | zdani(x,z) & \E w: gerku(z,w)}

2.10d) lu'i ro da ce'o ro di vu'o poi su'o daxi1 zo'u:

       ge da zdani di
       gi di gerku daxi1

What is the difference between (2.10b) and (2.10d)?

The difference is in the number of arguments each predicate has: in (2.10b), "gerzda" has three places, while in (2.10d) it has only two. Further, the place eliminated in (2.10d) is not 'doubled up' by any other place: it is distinct information about the relation, which has been discarded. And as pointed out above, changing the place structure of a predicate changes its denotation; so the two versions of "gerzda" refer to different sets of things.

So why would anyone want to eliminate such places? The intuitive reason invoked is that they do not make any difference to the definition. Thus, it would be argued, 'Mon Repos' is Spot's doghouse, whether Fido is a Saint Bernard or a beagle. In other words, we have an intuitive notion of what relation the lujvo should express. If the terjvo and veljvo introduce arguments which are irrelevant to that definition, they should be eliminated.

This intuitive approach is very appealing, in that it gives the lujvo creater flexibility to express what they're talking about; the problem it gives rise to is that what might be obviously irrelevant to one person might not be so obviously irrelevant to another. Therefore it would be desirable for another, more objective rationale to be used.

In most of the instances of place elimination we will consider now, there is such a rationale: the place eliminated does not essentially make any difference to the denotation of the predicate.

So look at "gerzda". We have defined the denotation of selbri as ordered sets of arguments --- pairs for the two-argument case --- for which the selbri holds. Thus, for "zdani", we have <Bill Clinton, White House>, <Socks, White House>, <Spot, Mon Repos>, <Rover, box in ACME Car Yard>. For "gerku", we have <Spot, Saint Bernard>, <Rover, Doberman>, <Fido, beagle>. In the three-argument version of "gerzda", we have <Mon Repos, Spot, Saint Bernard>, <box in ACME Car Yard, Rover, Doberman>, and so on.

Now, if we change our house, we also change our doghouse relation: "la mon.rePOS. gerzda la spat." is something different to "le tanxe pe vi le la .akmis. karcyfoi cu gerzda la spat.". Likewise if we change our dog, we change the housing relation: if Rover moves in with Spot, then Mon Repos houses two dogs distinctly. So "la mon.rePOS gerzda la spat." is distinct from "la mon.rePOS. gerzda la rover."

If you change the dog's breed... well, the problem is, you can't. Dogs don't change their breeds. But, if each dog is associated with just one dog breed, then the dog breed argument doesn't serve to narrow down the denotation of "gerzda" at all. Say there are 3,000,001 doghousing relations in the world: <Mon Repos, Spot, Saint Bernard>, <box in ACME Car Yard, Rover, Doberman>, <Mon Repos, Rover, Doberman>, and so forth. Now, we already know (from the denotation of "gerku") that Rover is a Doberman and Spot a Saint Bernard. So if we remove the third argument from the relation, we'll be left with 3,000,001 doghousing relations again: <Mon Repos, Spot>, <box in ACME Car Yard, Rover>, <Mon Repos, Rover>... The third argument hasn't affected the denotation: it didn't give us any information we didn't already have from the second place.

The reason this has happened --- and it happens a lot with terjvo places --- is that the third place was describing not the doghouse, but the dog. The phrase "la mon. rePOS. gerzda la spat." really means "la mon. rePOS. gerzda la spat noi gerku", since that is the interpretation we have given "gerzda". But that in turn means "la mon. repOS. gerzda la spat noi ke'a gerku zo'e", namely "la mon. rePOS. gerzda la spat. noi ke'a gerku la sankt.bernard." And in that case, it makes little sense to say "la mon. rePOS. gerzda la spat. noi ke'a gerku la sankt.bernard ku'o la sankt.bernard": the dog breed is redundantly repeated, and (to return to intuitive argumentation) is repeated in the wrong place, since the dog breed is supplementary information about the dog, and not about the doghouse.

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 Melbourne. Does that mean that he is bellowing as far as New Yorkers are concerned, but not as far as Melburnians are concerned? Surely not! 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. Since the "se cladu" will be "zu'i" in the meaning of "laurba'u" intended, it can be ommited.

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 still 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 relative qualities.

This kind of elimination even works with "belenu"-lujvo. If we define "depstu" as a waiting-place, to include bus-stops and waiting rooms, we appeal to the Deep Structure "da stuzi lenu de denpa di daxi1 daxi2". But why include a place for "le denpa"? Whether Estragon or Robin Leech are waiting under the tree, the tree is still the same waiting-place. That is to say, the definition of "depstu" with the "denpa" place left in is "the place where someone waits for something", etc.; the definition with it left out is "the place where anyone waits for something", etc. So the particulars of who is waiting at the bus-stop at any given moment, say, are irrelevant: there is no essential difference between "this is where John Smith waits for the 217 bus" and "this is where Mary Black waits for the 217 bus". In particular, the denotation of "depstu" is not affected: the 217 bus bus-stop is the same entity, no matter who waits at it, and is defined as "the place where anyone waits for the 217 bus".

On the other hand, if the "se denpa" changes from "lenu la goDOS. cu klama" ("Godot's arrival") to "lenu le katnymikce cu bredi" ("the surgeon's becoming ready") does make it a different kind of waiting-place.

2.2.2. Place elimination in veljvo.

With "laurba'u", we eliminated place information from the terjvo; this is not problematical. Just like with tanru, lujvo describe a subclass of what their head, their veljvo describe. So far, no places from the veljvo have been eliminated (other than through redundancy), so this has not been challenged.

But it is possible to eliminate places from the veljvo; and this is much more disruptive than any of the previous changes, because it eliminates information which is not either doubled up by another argument, or specific to a place in the terjvo.

Consider for example the translation of "handle" as "terjai". The gismu "jgari" has three arguments: the grasper, the graspee, and the place of grasping. Obviously, the place of grabbing corresponds to the handle, and the thing grabbed is what the handle is a handle of. This corresponds to the intuitive place structure we would give "handle" from scratch: x1 is a handle of x2.

But what happens to the grasper? One could argue it does not affect the denotation of "handle": a handle is a handle no matter who grabs it. Thus, a saucepan handle does not become a different handle whether it is grabbed by John Smith or Mary Black --- whereas it does become a different handle if it is stuck to a different appliance. Similarly, "text" could be conveyed by "seltcidu", but is not contingent for its definition on anyone actually reading it; "beverage" could be conyeved by "selpinxe", but is not contingent for its definition on anyone actually drinking it; and so on. A beverage is not something drunk by someone, but something drunk by anyone.

But, in this respect, compare the bridi "cidja" and "se citka". It is not true that "ro cidja cu se citka" (although it is true that "ro cidja ka'e se citka"). Display food, for example, is "cidja", even if it never becomes "se citka". The expression "se citka" refers to something actually eaten: it means "se citka be zo'e". So it is contingent for its definition on someone actually eating it.

For generic food, this is not the case. That is to say, they can't be considered necessarily "se citka" as long as a "citka" is involved. It is possible, however, in Lojban, to explicitly turn off places of selbri --- to say that, as we have been saying with lujvo, the places are irrelevant for the definition of the selbri. This is done with the dummy-sumti "zi'o"; and in fact, it is true that "ro cidja cu se citka zi'o". This corresponds, not to "all food is eaten", or "for all food, someone eats it", but "all food is the kind of thing that is eaten, but we're not interested in whether anyone actually does eat it." And in effect, "lo se citka be zi'o" corresponds to "lo cidja": they have the same denotation.

So: we've translated "text" as "thing read", "beverage" as "thing drunk", and "handle" as "place grabbed from". The seljvo of these lujvo --- "se tcidu", "se pinxe", "te jgari", mean explicitly "thing read by somebody", "think drunk by somebody", "place grabbed from by somebody". (The place corresponding to "somebody" still exists in these tanru, and is filled by "zo'e".) If we want to say, not "thing drunk by somebody", but just "think drunk", we need to turn the place off, and change "se pinxe (be zo'e)" to "se pinxe be zi'o".

The question now is: should we allow "selpinxe" to mean, not "se pinxe be zo'e", but "se pinxe be zi'o"? In other words, is it proper to allow elimination of veljvo places?

My position in compiling lujvo place structures has been conservative: I have not yet considered this distinction valid. The deletion of places advocated is unpredictable and unsystematic; it reads in much more information in building the lujvo than a naive reader would have access to. But I am also aware that this process may turn up in future research, if the language heads in that direction. For now, lujvo like "selpinxe" have not been listed separately in the jvoste: "selpinxe" is considered fully equivalent to "se pinxe".

The danger in general with lujvo with veljvo place deletion is that the veljvo selected is simply wrong. Different place structures imply different concepts, and the lujvo maker may be trying to shoehorn the wrong concept into the place structure of their choosing. This is obvious when someone tries to shoe-horn a "klama" tertanru into a "litru" or "cliva" concept, for example: these gismu differ in their number of arguments, and turning of "klama" places in a lujvo doesn't make any sense if the resulting place modified place structure is that of "litru" or "cliva".

It is not as obvious, however, for a "seltcidu" veljvo, where no other selbri seems to supply the concept of "text". A "se tcidu" is the closest we can come to "text", but in being read by some reader, it is still not really a "text". There is a solution involving "zi'o" in lujvo building, which is detailed in Section 5. But playing games with "zi'o" is a convoluted intellectual exercise which will probably not come intuitively to people creating lujvo on the fly. So for now, we will have to bear in mind that such shoe-horning will happen.

2.2.3. Lean lujvo.

Another important issue is: how much information should be contained in a lujvo which could just as conveniently be given by a gismu? Consider "brulu'i", "to sweep". Its component gismu have the following place structures:

burcu: x1 is a brush for purpose x2 (event) with bristles x3

lumci: x1 (agent) washes/cleanses x2 in/with cleaning material x3

We eliminate the obvious redundancies: the "se burcu" is "lenu lumci" and can be omitted altogether; the "te lumci" is the "burcu". We can also omit the "te burcu" information as specific to the broom rather than the sweeping, like we did with "gerzda" earlier. We are left with the place structure: l1 l2 l3=b1.

But what information do we gain in saying "mi brulu'i le loldi le xekri burcu", that isn't already given by "mi lumci le loldi le xekri burcu"? According to a school of thought prevalent amongst Lojbanists, known as Lean lujvo ("si'o toltiljvo"), we often make lujvo to blindly match the lexicon of our native languages, where a gismu by itself (given the appropriate context) is sufficient.

For example, we can talk of a waiter as a "djabeipre", a food-carrying-person. But once it's obvious we are talking about a worker in a restaurant, Lean lujvo would prefer that we refer to her just as a "bevri", instead of blindly translating "waiter" wherever it occurs by the redundant "djabeipre". This introduces a pragmatic approach to denotation: the class of "carriers" ("bevri") in the world is much greater than that of people who specifically carry food ("djabeipre"); but once context comes into play, "le bevri" ("the carrier --- you know what I mean") has the same effective denotation as "le djabeipre".

This way of thought affects place structures significantly. According to it, "mi brulu'i fi le xekri burcu" says as much as "mi lumci fi le xekri burcu". So if we want to relate "mi" and "le xekri burcu", it seems pointless to use "brulu'i" as the bridi, when "lumci" will do the job just as well and without any redundancy. But in that case, do we need the "burcu" place of "brulu'i" at all? If whenever the place is mentioned, it would be better mentioned in the context of the gismu "lumci", then perhaps only the x1 and x2 places need be left in.

This issue is pervasive with "be"-lujvo. Typically, the x2 of the veljvo is described by the terjvo. One can argue, in some cases, that the x2 is so well described by it, it needn't appear in the final place structure at all; including it would be redundant. A good example of this is one of the most widely-known lujvo: "lobypli", "Lojbanist". It seems ridiculous to include a place for "le se pilno", when we know perfectly well that what is being used is Lojban. "mi lobypli lenu cilre fi loi gerna", "I use-Lojban/am-a-Lojbanist to learn about grammar", makes much more sense than "mi lobypli la Lojban. lenu cilre fi loi gerna", "I use-Lojban/am-a-Lojbanist using Lojban to learn about grammar".

But as alluded above, a case can be made for retaining the terjvo place, by interpreting x2 to be a generic, rather than specific, entity. Take "gerzda": if we interpret the place structure as: x1 is-a-house-for-a-dog for a dog (x2), we risk redundancy, because we already know x2 is a dog from the seltanru. If we make (x2) a specific dog, then our predicate has the same place structure as "zdani" itself. We need then to show that "gerzda" conveys significantly more information, nonetheless, than "zdani", and thus has its own raison d'etre.

An underhand way of doing so would be to make x2 the dog breed, rather than the specific dog. We would then have two rather distinct concepts expressed by the two distinct selbri: "la monrePOS. cu gerzda lai sankt. bernard gi'e zdani la spat.": "Mon Repos is a doghouse for Saint Bernards, and is Spot's house". Something like this has actually happened at times in the jvoste. For example, for "rulzda", "flower bed", it doesn't make sense to say "this is a flower bed for flowers

  1. 53276, #62347 and #723489" (which is what we'd get for z2=x1); but it

does make sense to say "this is a flower bed for chrysanthema", "ti rulzda loi xrulrxrisantemo." (which has z2=x2). Obviously, this can only happen when x2 of the terjvo gives the class or category of entities in x1; and it only makes sense when the lujvo relation concerns the veljvo and that whole class of terjvo, not just an individual. If there is a real difference between Saint Bernards doghouses and Doberman doghouses, it makes sense to say "la monrePOS. gerzda lai sankt. bernard". If not, we might as well just say "la monrePOS. cu gerzda gi'e zdani la spat.", and leave it like that.

Body part lujvo support the Lean lujvo idea. Take the lujvo for "skull", "sedbo'u". There are four places involved: the bone (b1), the system of which the bone is a part (b2), the head (s1), and the system of which the head is part (s2). What the bone is a part of, b2, can either be a person or a body part. If it is a person, then s2=b2 (since s2 is obviously a person): thus, if this is the skull of Yorick, then it is a bone of Yorick, and is somehow associated with the head of Yorick. (Note that this is not a "be"-interpretation! The equation s2=b2 does express a relation between terjvo and veljvo, but it is not of the type "le bongu be le stedu".) If, on the other hand, the bone is considered part of a body part, then s1=b2. (This does give the "be"-interpretation of the lujvo: "this is the skull of the head of Yorick"/"ti bongu be le stedu be la .iorik.")

In either case, though, do we really need a separate place for the head? Does it makes sense to say x1 is the skull of the head x2 of person x3 ("this is the skull of the head of Yorick")? Obviously not: we know that the bone is a head-bone, so we don't need a place in the place structure telling us so. The final structure can only be: x1 ("le bongu") is the skull of x2 ("le se stedu") --- which is the place structure desired from the start ("this is the skull of Yorick"). This case is comparable with "rulzda" above (flower-bed for chrysanthemums): x2 of the terjvo gives the information that really matters in describing what is related to what; x1 is obvious from context or not particularly salient, and is ignored.

Or take "jbogerna", "Lojban grammar", and "gicygerna", "English grammar". "gerna" has three places: grammar g1, language g2, and text g3. For "jbogerna", we already know the language to be Lojban; so, as with "lobypli", we delete the place for language. But things are not as clear-cut for "gicygerna". It is reasonable to claim there are different Englishes (dialectical variants), and place ge2 might be kept to denote the particular form of English specified by the grammar. (This is another instance where an otherwise redundant place is kept in the place structure by undergoing a meaning shift.) We would probably decide, though, that such a nicety is best served by a form like "nort,mbria. zei gicygerna", which will still have just two sumti (unless you start concentrating on idiolects of Northumbrian for ge2, but that is probably going too far). "gicygerna" itself, in the interest of compact place structures, will probably also have just two arguments: grammar and text.

There are far-reaching implications here as to what precision we wish to invest our lujvo with. No categorical verdict would be helpful at this stage of development of the language. To help keep place structures as small and manageable as possible, I have cautiously supported Lean lujvo in my proposed place structure list. But the judgement typically needs to be made independently for each individual lujvo.

3. lujvo place ordering.

So far, we have concentrated on selecting the places to go into the place structure of a lujvo. However, this is only half the story. In using selbri in Lojban, it is important to remember the right order of the sumti --- particularly since Lojban does not use a case grammar in the same way most natural languages do: there is a lot of difference between "fi" and "fu"! With lujvo, the need to attend to the order of sumti becomes critical: the set of places selected should be ordered in such a way that a reader, unfamiliar with the lujvo, should be able to tell which place is which. The ordering of places should somehow be reproducible, and follow a consistent pattern.

If we aim to make understandable lujvo, then, we should make the order of places in the place structure follow some conventions. If this does not occur, very real ambiguities can turn up. Take for example the lujvo "jdaselsku", "prayer". In the phrase "di'e jdaselsku la dong.", is Dong the person making the prayer ("this is a prayer by Dong"), or the entity prayed to ("this is a prayer to Dong")? We could resolve such problems on a case-by-case basis for each lujvo. But this makes the task of learning lujvo place structures unmanageable. People need consistent patterns, to make sense of what they learn. Such patterns can be found across gismu place structures, and are even more necessary in lujvo place structures. Case-by-case consideration is still necessary; lujvo creation is a subtle art, after all. But it is helpful to take advantage of any available regularities.

The place structures of gismu tend to be ordered according to some notion of psychological saliency, or importance. There is an implication within the place structure of "klama", for example, that "lo klama" will be talked about more often, and is thus more important, than "lo se klama", which is in turn more important than "lo xe klama". A similar tendency may be observed in lujvo; but this criterion is too subjective and context-dependent to use by itself as the primary ordering criterion.

Instead, we propose the following guidelines:

3.1. gismu place ordering consistency.

The places of the component gismu appearing in the final lujvo should not be arbitrarily re-ordered. Rather, they should appear in the order they have in the original gismu, allowing for interleaving with places from other gismu.

To illustrate this, let us consider the lujvo "jdaselsku" again. The places of "jdaselsku" deemed relevant to its definition are: "le cusku", "le se cusku", "le te cusku", "le ve cusku", and "le lijda". We presume "le cusku" is equivalent to "le se lijda". Now the component gismu "selsku" has its arguments appear in the following order:

3.1) lo se cusku cu selsku lo cusku lo te cusku lo ve cusku

The gismu Deep Structure of "jdaselsku" will be something like:

3.2) lo se cusku cu jdaselsku lo se lijda be lo lijda be'o po'u lo cusku lo te cusku lo ve cusku "A prayer is something expressed by someone belonging to a religion, who is also the expressor, to an addressee through a medium."

Based on this, in the final lujvo, "le cusku" (the person making the prayer) should precede "le te cusku" (the entity to which the prayer is addressed), paralleling the place structure of "selsku" itself. Accordingly the place structure proposed is: c2 c1=l2 c3 c4 l1. (We will see why l1 is tagged on at the end shortly.)

Interestingly, the lujvo was first proposed in ju'i lobypli in the phrase "le jdaselsku be la jegvon." ("The Lord's Prayer"). That is, the "te cusku" was the second, not the third argument of the lujvo --- which would presumably have the place structure c2 c3 c1 c4 l1. Such a structure may make some sense: "a prayer to X by Y" may order its arguments in a way more familiar than "a prayer by X to Y". But I would contend such rearrangement from the Deep Structure place order is usually unneccesary and confusing, with no substantial advantage over the order proposed, which follows the place order of "selsku", and is thus more predictable.

The ordering principle should be maintained even if places from other gismu are interleaved with the gismu being considered.

Thus in "jditadji" "policy", it was decided the relevant places were: t1, t3, j1, j3. The Deep Structure is: "lo tadji cu tadji lonu lo jdice cu jdice zo'e lo te jdice kei lo te tadji". To reflect this place structure, and to have consistent ordering, the places should appear in the order: t1 j1 j3 t3. t1 precedes t3, and j1 precedes j3.

There are cases where the component gismu have many places in common, and each gismu orders these places differently. To be consistent with the Deep Structure, we choose the tertanru of the veljvo (by default, the rightmost gismu) as the selbri whose place structure is used. This is because the place structure of a tanru is based on that of its tertanru: since the lujvo describes a type of veljvo, the place structure of the veljvo should take precedence over that of the terjvo.

For example, in "ctucku", "textbook", we have the following places: ct1=cu3, ct2=cu4, ct3, ct4=cu2, ct5, cu1, cu5. We could order the places based on the place structure of "ctuca", as given above, or based on "cukta", as follows: cu1, cu2=ct4, cu3=ct1, cu4=ct2, cu5, ct3, ct5. Since a "ctucku", a "ctuca cukta", is a kind of "cukta", the Deep Structure will look like this:

3.3) lo cukta cu te ctuca be lo ve cukta bei lo te ctuca bei lo se

       cukta bei lo xe ctuca be'o gi'e cukta lo se cukta lo te cukta lo ve
       cukta lo xe cukta

The corresponding tanru, however, has the following place structure:

3.4) lo cukta cu ctuca be di bei lo te ctuca bei da bei lo xe ctuca be'o

       cukta da de di daxi1

So it makes sense to use the latter place structure, based on "cukta".

3.2. "belenu"-lujvo place structure ordering.

In a "belenu"-lujvo, place structure ordering is simple, and reflects that in the Deep Structure: the terjvo places appear nested amongst the tertanru places, in the place of the veljvo abstraction place they together describe.

For example, "posydji", "to want to have something", has places d1=p1, d3, p2. d2 is "lenu ponse", and is thus redundant. p3, the law of ownership, is an irrelevant detail in an expression of desire ("I want to have it, and I don't care about the red tape"). Also, "posydji" should express wanting something for oneself --- which is the default interpretation of its Deep Structure, "djica lenu ponse". Wanting something for someone else (namely, "le ponse" being different from "le djica") is a less frequent and more complex concept, and can be relegated it to a longer expression.

(This, by the way, occurs frequently with lujvo: leaving in a place, as already discussed, makes the concept more general, while leaving out a place, e.g. by overlap as above, makes the concept more specific. Since we usually want to express this more specific concept, we feel it deserves its own "word" more than the more general concept. "To want to have something for oneself", for example, does have its own word in English: "to want X". "To want someone else to have something", on the other hand, does not get compressed in that way. So we say "ko'a posydji le solji", instead of "ko'a posydji ko'a le solji", for "she wants the gold [for herself]", and "ko'a djica lenu ko'e ponse le solji", instead of "ko'a posydji ko'e le solji", for "she wants him to have the gold, she wants the gold for him".)

As outlined, the places of "ponse" should be nested in the d2 place of "djica". Thus, the Deep Structure of "posydji" is:

3.5) lo djica cu djica lonu lo djica cu ponse lo se ponse zo'e kei lo te djica

and it makes sense to make the place structure of "posydji" correspond to this template, by putting the "ponse" places in place of the redundant "lenu" place in "djica":

3.6) lo djica cu posydji lo se ponse lo te djica

which matches what we'd expect in English (X wants to own Y for purpose Z).

"belenu"-lujvo place structure ordering can be applied to "be"-lujvo as well. Typically, however, so few terjvo places remain after place selection (thanks to the Lean lujvo notion discussed above), that the effect is imperceptible. Thus "ninpe'i", "to be introduced to, to meet for the first time", is a "be"-lujvo: "lo penmi po'u lo se cnino cu penmi lo cnino po'u lo se penmi lo te penmi". The places of the lujvo are: p1=c2, p2=c1, p3. "belenu"-lujvo ordering would place "lo cnino" 'in place' of "le se penmi". But leaving the places in the order they have in the veljvo, "penmi", has exactly the same effect: p1=c2, p2=c1, p3.

An interesting example of a be-lujvo is "xandegycalku", "fingernail". The seltanru of this lujvo is "xandegji", "finger". The relevant places here are d1 and x2=d2 (following the same reasoning as was applied to "sedbo'u" above.) Now "calku" has three places: the shell itself, what the shell encases, and what the shell consists of. Treating "xandegycalku" as a "be"-lujvo, the "se calku" is the "xandegji", and its places substitute c2 in the final lujvo. So the place structure is: c1 d1 x2 c3. But the same reasoning which led us to discard "le xance" from the place structure of "xandegji" makes us drop "lo degji" from the place structure of "xandegycalku", leaving the place structure:

x1 is a fingernail of entity x2, made of x3,

where thanks to "belenu"-lujvo type interleaving of places, the place for "lo se xance" has substituted the place for "lo se calku".

3.3. "gi'e"-lujvo place structure ordering.

Most lujvo in our list of proposed place structures have their place structures ordered by "gi'e"-lujvo ordering. In this scheme, the selected places of the veljvo are simply given in order, and then followed by the selected places of the terjvo. This ordering makes sense if we consider the terjvo and veljvo in the Deep Structure linked by "gi'e": each bridi-tail simply lists the arguments of each selbri in order. In the lujvo, the veljvo sumti come before the terjvo sumti, following the principle that veljvo arguments have priority.

Besides being useful for "gi'e"-lujvo themselves, this type of ordering is also natural for those lujvo which do not fall into a well-defined terjvo-veljvo relationship. If the relationship isn't like "be", then the terjvo places can't be placed in any veljvo place slot, and it is safer to dump them at the end, 'out of harm's way'.

As an example of a "gi'e"-lujvo, consider "tinju'i", "to listen" ("to hear attentively, to hear and pay attention"). Its places are: j1=t1, j2=t2, t3. The places of the veljvo appear first, and the remaining places of the terjvo follow: "lo jundi cu tinju'i lo se jundi lo te tirna".

As an example of a lujvo with an obscure terjvo-veljvo relationship, take "jdaselsku", which we considered earlier. For this selbri, the thing expressed is in some relationship with a religion. This relationship cannot be expressed by either "be" or "je"; the Deep Structure looks something like:

3.8) le selsku cu selsku le cusku gi'e steci le lijda

(As will be argued below, this is because the seljvo is, in some respects, elliptical, and does not fully express the components involved in the terjvo-veljvo relation.) The easiest way to deal with this is simply to dump the terjvo places after the veljvo places: "lo selsku cu jdaselsku lo cusku lo te cusku lo ve cusku lo lijda".

      • be-lujvo can also have je-lujvo ordering. more explicit examples***

4. lujvo structuring.

4.1. lujvo with more than two parts.

The theory we have outlined above is an account of lujvo with two parts: a terjvo and a veljvo. But often lujvo are made containing more than two parts. An example is "bavlamdei", "tomorrow": it is composed of the rafsi for "future", "adjacent", and "day". How does the account we have given apply to lujvo like this?

The best way to approach such lujvo is to still classify them as based on binary tanru, the only difference being that the terjvo and/or veljvo is itself a lujvo. So it is easiest to make sense of "bavlamdei" as having two components: "bavlamji", "next", and "djedi". If we know (or invent) the lujvo place structure for the component selbri, we can compose the new lujvo place structure as normal.

In this case, "bavlamji" is taken as having the place structure: b1=l1 b2=l2 ("x1 is next after x2"). We combine this with "djedi", which has the place structure: duration x1 is x2 days in duration (default 1) by standard x3. The place d2 should have its default value of 1 here, and doesn't provide us with new information; so it is omitted. While for "gi'e"-lujvo we would normally put any trailing veljvo places before any terjvo places, the day standard is a much less important concept than the day the tomorrow follows, in the definition of "bavlamdei". (This occurs a few times in our "gi'e"-lujvo. We can say that "gi'e"-lujvo ordering, having a less clear Deep Structure motivation than "belenu"-ordering, needn't be as rigidly adhered to. It can be violated, as happens here, to rank places in order of relative salience. Alternatively, one might contend the concept really intended to be conveyed is "djebavlamji".) The resulting place structure is: d1=b1=l1 b2=l2 d3 --- x1 is the tomorrow of/day after x2, under the day standard x3.

Not all lujvo with more than two parts lend themselves to a two-part analysis; the relation between all of them can be imprecise. Even if a relationship is identifiable, "ke"-elision (which we will consider shortly) can mean that the relation is quite different from what one would have guessed. So it is useful to have a catch-all place ordering principle for such lujvo, just as we had the "gi'e"-lujvo ordering principle as a catch-all for all two-part lujvo with an unclear interrelationship. The principle we propose is just an extension of the "gi'e"-lujvo principle: have the places of the rightmost gismu first, then those of the second rightmost, and so forth up to the leftmost gismu. This is consistent with applying "gi'e"-lujvo ordering if you already know the binary structure of the veljvo.

So in analysing "cladakyxai", "long sword", we select the following places as relevant: x1=c1=d1, d3, c3, x2, x3. If we dump the places in the order we just suggested, we get: x1=c1=d1 x2 x3 d3 c3. But if we were to analyse "cladakyxai" as a two-part lujvo ("cladakfu xarci"), and apply "gi'e"-lujvo ordering to it, we would get the same result. The place structure of "cladakfu" would be: d1=c1 d2 d3 c3 (as "cladakfu" is itself a "gi'e"-lujvo). In forming a lujvo out of "cladakfu" and "xarci", we note that d2 is the purpose for which the knife is used --- which is obvious if it is a weapon --- so it can be left out. Dumping the places of the terjvo "cladakfu" after those of the veljvo "xarci" gives: x1=c1=d1 x2 x3 d3 c3 --- which is exactly what we obtained above.

4.2. Eliding SE rafsi from lujvo.

We have stated that, in theory, all lujvo should be expressable as either "gi'e"-, "be"- or "belenu"-lujvo. It is often the case, however, that while the terjvo and veljvo do share arguments, they do not share them in the way predicted by these lujvo schemas. On the other hand, though, simply inserting a SE word into the seljvo would bring the lujvo in line with one of these patterns. In that case, we treat the lujvo as having the tanru including the SE word as an underlying form, and consider the SE word to have been elided fron the seljvo.

Of course, SE words are not the only words which might be absent from the seljvo in such a way as to make its structure more opaque. It is also frequently the case that other structural cmavo, such as "ke", "ke'e", "nu" and "kei" are 'missing', in that sense --- namely, for the lujvo to make sense following the guidelines outlined here, such a cmavo would need to be inserted into the seljvo. It may even be the case that a BAI cmavo --- or, following the restrictions on BAI cmavo of gismu deep structure, an entire selbri --- is absent from the seljvo. We consider all these cases below.

It should be pointed out that, in all cases, the lujvo with 'elided' cmavo is not, in and of itself, 'wrong' (with the possible exception of lujvo eliding SE words from their veljvo). There is no official restriction of the terjvo-veljvo relation to being either "be", "belenu" or "gi'e"; and indeed, it cannot be denied that the terjvo and veljvo are related in these gismu. However, it is also the case that the seljvo of these lujvo does not have as its seljvo the most explicit possible representation in tanru form of the terjvo-veljvo relation. In addition, because such lujvo add a new interpretation for the seljvo to the three interpretations already considered, they increase the ambiguity of the seljvo.

This is not itself bad, as pointed out. Indeed, there are usually good reasons for such 'elisions' in the first place: the terjvo-veljvo relationship is typically conceived of as a relation between two gismu, without any intervening cmavo machinery. Indeed, in a real sense, 'elided' lujvo come more naturally to the language-user than more explicit, disambiguted lujvo. Because of the risk of ambiguity, however, Lojbanists should be aware of the complications raised by such lujvo. In the case of ambiguous readings resulting from 'elisions' in lujvo, the following extract from the morphology paper in this grammar may serve as a useful guideline:

"If the lujvo made by a shorter form of tanru is in use, or is likely to be useful for another meaning, the decider then retains one or more of the cmavo, preferably ones that set this meaning apart from the shorter form meaning that is used or anticipated. Generally, therefore, the shorter lujvo will be used for a less complicated concept, possibly even over a more frequent word. If both words are needed, the simpler one should be shorter. It is easier to add a cmavo to clarify the meaning of the more complex term than it is to find a good alternate tanru for the simpler term."

So, when possible, the simpler concept --- in terms of this discussion, the simpler, canonical interpretation, as a "be"- or "gi'e"-lujvo --- should be preferred over more complex interpretations --- that is, interpretations requiring the insertion of extra cmavo, be it NU ("belenu"), SE (SE elision), or entire new selbri (more distant relations). This is because it is easy to add such a cmavo on to the seljvo, but impossible to take it away if it is not formally present. So for example, according to this guideline, it is preferrable to interpret "soidji" as a "be"-lujvo, rather than a "belenu"-lujvo. This is because, if "soidji" means "djica be lo sonci", it is easy to get a lujvo for the "belenu"-reading by adding "nu" to the seljvo ("nunsoidji"). If, on the other hand, "soidji" has a "belenu"-interpretation to begin with, it will prove impossible to create an analogous lujvo with a "be"-interpretation: there is no "nun" rafsi to take away. If, however, the formally simpler interpretation is implausible, and the formally more complex interpretation is required, then it should be allowed; examples of this can be seen below.

So let us now turn our attention to the first case of 'elision', the omission of SE words from the seljvo. There are two ways to elide SE rafsi from a lujvo. They can be dropped off the terjvo, or the veljvo. The implications of each are quite different.

4.2.1. Eliding SE rafsi from terjvo.

SE rafsi elision in the terjvo is extremely commonplace in Lojban; so much so, that it requires an effort to notice that it has even happened. Consider as an example one of the most-used lujvo in the language: "le'avla", "loan word". This lujvo expresses some relation @@@between a borrower ("lo lebna") and a word ("lo valsi"); but what precisely this relation is is unclear. It certainly doesn't fit into any of the schemes we have set up: it's neither a word meaning "borrower" ("valsi be loi lebna"), nor a word which is a borrower ("valsi je lebna"). But if we prefix the terjvo with "se", everything becomes clear. The lujvo "selyle'avla" has the "gi'e"-lujvo interpretation "valsi gi'e se lebna", which is exactly what a le'avla is: not a borrower word nor a borrow word, but a borrowed word. Similarly for "ti'ifla", "bill, proposed law", the relation between "lo stidi" and "lo flalu" is obscure. But "ti'ifla" can be considered an abbreviation of "selti'ifla", a straightforward "gi'e"-lujvo, since "lo se stidi" (that suggested) is "lo flalu".

As described above, there is nothing intrinsically wrong with lujvo like "ti'ifla" and "le'avla": they are more intuitive than "selti'ifla" and "selyle'avla", since they don't require the coiner to sit down and work out the precise relation involved in the seljvo: they can just rattle off a gismu pair. (It is, however, the case that nowadays the "gi'e"-lujvo "fu'ivla" is preferred to the older "le'avla".) But should the lujvo get to the stage where a place structure needs to be worked out, then the precise relation does need to be specified. And in that case, such reduced lujvo form a trap in lujvo place ordering, since they obscure the most straightforward relation between the terjvo and veljvo. To give our lujvo-making rules as wide an application as possible, and to encourage analysing the terjvo-veljvo relation in lujvo, lujvo like "le'avla" are given the place structure they would have with the right SE added to the terjvo.

Note that, with these lujvo, an interpretation requiring SE insertion is safe, because the alternatives are implausible: words do not borrow, 'word meaning "borrow"' is not likely to provide fertile ground for word-coining, and laws do not of themselves suggest. (Laws can be made about suggestions, so a "be"-interpretation is possible for "ti'ifla", but the "bill" interpretation would probably be considered more deserving of the shorter word.) This may not always be the case, and Lojbanists should be aware of the risk of ambiguity.

4.2.2. Eliding SE rafsi from veljvo.

Eliding SE rafsi from veljvo gets us into much more trouble. To understand why, recall that lujvo, following tanru, describe a type of veljvo. Thus, "posydji" describes a type of "djica", "gerzda" describes a type of "zdani", and so on. What is certain is that "gerzda" does not describe a "se zdani" --- it is not a word that could be used to describe a dog, say.

Now consider how we would translate the word "two-sided". Our first impulse might be to translate each element literally, and come up with "relmla". But try using this lujvo. Can we say "the board is two-sided" as "le tanbo cu relmla"? The place structure of "mlana" has: x1 is a side of x2. A board is not a side, a "mlana"; it is something that has sides, a "se mlana". The one thing a naive reader should be sure of, coming across the lujvo "relmla", is that it is a kind of "mlana"; that's what the veljvo ("remei mlana") says. To have "relmla" turn out to be a "ba'e se mlana" is something noone could guess without a dictionary; and even then, they'd scarcely believe it. If the lujvo has nothing to do with its veljvo, one of the two is wrong.

All is not lost, of course; all we need do is insert the cmavo "se": "le tanbo cu se relmla". While we can get away with this here, however, consider another example: "dark-skinned". Let's translate this word as "xekskapi" (Note that "manku" means something in the dark, not something dark-coloured). As we have seen, we cannot say "la djak. cu xekskapi", because Jack is not skin, "skapi", but someone with skin, "se skapi". So we say "la djak. cu se xekskapi". But look now at the place structure of "xekskapi": it is a "gi'e"-lujvo, so the place structure is: x1 is the black skin of x2. But this gives us two problems. First, the place we are interested in talking about most is x2, not x1. Indeed, x2 was the meaning we were trying to render in the first place.

Second, not only is the x1 place less important; it's hard to think of a circumstance in which we'd want to speak of it at all. The jufra "la djak. cu se xekskapi leri skapi" looks downright silly: we know that the skin Jack has will be his own skin, so why give it a place in the place structure? This seems to be an instance where Lean lujvo could be applied, eliminating the irrelevant skin place. But doing so leaves us in an embarrassing situation: we are left with a selbri with no x1! This doesn't make any sense at all.

What is happening here is that we are translating the veljvo wrongly, under the influence of English. The suffix "-sided" does not refer to a "side", but something with sides, "selmla"; similarly, "-skinned" does not mean "skin", but someone with skin, "selskapi". Because we've got the wrong veljvo (eliding a "se" that really should be there), any attempt to accomodate the resulting lujvo place structure into the mould of our desired concept is fitting a square peg in a round hole. Since they can be so misleading, lujvo with SE rafsi elided from the tertanru are frowned upon in the jvoste. The two concepts mentioned appear in the jvoste with more appropriate tertanru: "relselmla" and "xekselskapi".

In fact, while what we have just described is the most Lojbanic way to approach the problem, it is not the most general description of what is happening with the English forms being translated. The English term "two-sided" is an adjective formed from the phrase "two sides". Adjectives have the property of specifying qualities of their referents. Thus, the very fact that a word is an adjective makes it correspond to the Lojban gismu "ckaji", "has the quality". Thus, "two-sided" means "the referent has the quality of having two sides", which corresponds to the gismu deep structure "da ckaji be lenu da se mlana lo remei", and the lujvo "relmlakai", where the rafsi "kai" ("ckaji") corresponds to the English suffix "-ed". Since this correspondence is not obvious to beginners, "ckaji" is a veljvo often left out in lujvo-formation; we return to this point in Section 5.

4.3. Eliding KE and KEhE rafsi from lujvo.

People constructing lujvo usually want them to be as short as possible. To that end, they will discard any cmavo they regard as niceties. The first such cmavo to get thrown out are usually "ke" and "ke'e", the cmavo used to structure and group tanru. We can usually get away with this, because the interpretation of the veljvo with "ke" and "ke'e" missing is less plausible than that with the cmavo inserted, or because the distinction isn't really important. For example, in "cladakyxa'i", the veljvo is "clani ke dakfu xarci", "long knife-weapon, long dagger". But it doesn't seem to make much difference whether the veljvo is this, or "ke clani dakfu ke'e xarci", "long-knife weapon". The lujvo "zernerkla", "to sneak @@@in", has the veljvo "zekri ke nenri klama"; we can guess this is the veljvo intended, because the alternative, "ke zekri nenri ke'e klama", "to crime-inside go", doesn't make much sense. (To go to the inside of a crime? To go into a place where it is criminal to be inside --- an interpretation almost identical with "zekri ke nenri klama" anyway?)

There are cases, however, where omitting a KE or KEhE rafsi can lead to misunderstanding, particularly if the lujvo contains a SE or NAhE rafsi. An example of this is "selxagmaugau", which was intended to mean "improved": this would give it the veljvo "se ke xamgu zmadu gasnu" (from "xagmaugau", "improve" --- make more good). If we interpret the lujvo with default tanru bracketing, we come up with "ke ke se xamgu ke'e zmadu ke'e gasnu", which means "acting ("gasnu") so that something is more ("zmadu") of a beneficiary ("se xamgu")". This looks like meaning "making someone benefit more from something", and is not at all an implausible reading of the veljvo. It can describe, for example, what I am doing for you in building a better oil well for you on your site: I am making you benefit more from your site's resources.

Such misinterpretation is more likely than not in a veljvo starting with "se ke", "na'e ke" or "to'e ke": the scope of the cmavo will likeliest be presumed to be as narrow as possible, following the guideline from the Morphology paper given above. If "selxagmaugau" meant "se ke xamgu zmadu gasnu", there would be no possible lujvo to express "se xamgu zmadu gasnu". For that reason, if we want to modify a lujvo by putting "se", "na'e" or "to'e" before it, it's better to leave the result as two words, or insert "ke", than just stick the SE/NAhE rafsi on: "se xagmaugau" or "selkemxagmaugau", but not "selxagmaugau".

Note that, if the lujvo we want to modify with SE has a terjvo already starting with a SE rafsi, we can take a shortcut. "gekmau", for instance, means "happier than...", while "selgekmau" means "making people happier than..., more enjoyable than..." ("zmadu leka dakau se gleki"). If something is less enjoyable than something else, we can say it is "se selgekmau". But we can also say it is "selselgekmau". Since two "se" in a row cancel each other ("se se gleki" means the same as just "gleki"), there would be no good reason to have "selsel" in a lujvo with that meaning. So we can feel free to interpret "selsel-" as "selkemsel-". The rafsi combinations "terter-", "velvel-" and "xelxel-" work in the same way. Other SE combinations like "selter-" needn't work like this, because "se te" does make sense in a veljvo. Usually, thouugh, there won't be much motivation to say something like "se te"; so it remains to be seen whether our interpretation of "selsel-" might be extended to combinations like "selter-" too.

4.4. Eliding NU and KEI rafsi from lujvo.

Eliding "nu" rafsi, as we have seen, gives rise to the distinct "belenu" category of lujvo. As we also said earlier, whether the "nu" is elided or not depends on how confident one is that the lujvo will be understood properly. This is why both "zvaju'o" and "nunzvaju'o", for example, are listed in the jvoste as renderings of "to know something is there, to be aware of something". The latter lujvo is an expansion of the former, and is less ambiguous.

It does, however, introduce a second ambiguity, resulting from the elision of "kei" from the seljvo. Thus, "nunzvaju'o" is really an abbreviation of "nunzvakezju'o", but it could also be interpreted as "nu zvaju'o". The issues are the same as with the elision of KEhE, considered above. The jvoste contains entries with both possible interpretations of the veljvo: "nunclapi'e" means "nu clapi'e" (long jump), whereas "nunmrostu" means "nunmro stuzi" (place of death). As before, factors of plausibility and succinctness enter into the equation. In a case like "nunclapi'e", the interpretation "nuncla plipe" (length jump) is much less plausible than the interpretation given; however, all other things being equal, the default interpretation should probably be narrower scope for "nu", to be consistent with other cmavo. Note however, that, unlike the case for "ke'e", there are disambiguiating longer forms available for both interpretations. Thus, "nu morsi kei stuzi" can be rendered as "nunmrokezystu", while "nu ke morsi stizu" can be rendered as "nunkemymrostu". For consistency, though, the former interpretation would be preferred, since the latter can be expressed quite well as "nu mrostu".

It might be asked whether there are any circumstances under which "nu" can be omitted from the veljvo; for example, is "solcanci" ('sun-vanish', cf. English 'sundown') an acceptable rendering of 'sunset', or should the eventhood of the selbri be made explicit by leaving in "nu", as in "solnuncanci" or "nu solcanci"? In our opinion, while context can be relied upon for sorting out terjvo, veljvo should be left alone, since they are critical to establishing what a lujvo is talking about. So to be consistent with the approach to eliding SE cmavo from veljvo taken above, we would discourage such forms as "solcanci" meaning "solnuncanci".

4.5. Choice of seljvo.

As already mentioned, this paper is driven by the idealisation that all lujvo express a relationship paraphrasable in a gismu deep structure by "be" or "gi'e". While this is a useful generalisation which captures most lujvo coinages, for the reasons we have outlined (namely, that the closest, most salient relation between terjvo and veljvo is itself the veljvo or terjvo bridi), it does not universally apply. There are many cases in which there is a clear relationship, because the terjvo and veljvo share places, but where the place sharing does not share either the "gi'e"-, the "be"- or the "belenu"-pattern.

An example of this is "jdaselsku", 'prayer', considered above. Recall the deep structure proposed for the lujvo:

3.2) lo se cusku cu jdaselsku lo se lijda be lo lijda be'o po'u lo cusku lo te cusku lo ve cusku "A prayer is something expressed by someone belonging to a religion, who is also the expressor, to an addressee through a medium."

This is not a "se cusku be lo lijda", nor a "se cusku gi'e lijda", so it does not fall into the established patterns. The real meaning of "jdaselsku" is "se cusku pe lo lijda"; however, "pe" is itself intrinsically ambiguous. But note that the terjvo and veljvo do share arguments --- in particular, l2 and c1. This means that, while "jdaselsku" does not fall into the established patterns, a selbri concentrating on a different argument --- namely, a selbri whose x1 contained the doubled-up sumti --- would fall into place, as follows:

4.1) lo cusku po'u lo se lijda cu cusku lo se cusku lo te cusku lo ve cusku gi'e se lijda lo lijda

This gives the "gi'e"-lujvo "seljdasku" ('expresses and is in a religion'), with place structure c1=l2 c2 c3 c4 l1. It is probably not the case that this is a more 'correct' lujvo than "jdaselsku", but it makes the relation between terjvo and veljvo is made somewhat more explicit, and thus it might be preferred by some Lojbanists. Note that "se seljdasku" gives the same place structure as "jdaselsku": c2 c1=l2 c3 c4 l1.

However, there is a further problem with "jdaselsku", not resolved by "seljdasku". This is that no seljvo involving just the two gismu "lijda" and "cusku" fully express the relationship implicit in prayer. A prayer is not just anything said by any adherent of a religion; nor is it even anything said by them as adherents of that religion. Rather, it is what they say under the authority of that religion, or using the religion as a medium, or following the rules associated with the religion. So the seljvo is somewhat elliptical.

If some of the disambiguations look like BAI-relations, there is a good reason for that: the relation between the terjvo and veljvo may only be captured by a "be" + BAI relation, which in strict gismu deep structure (which reduces BAI to gismu), means adding a further gismu to the seljvo. For example, consider "xancyminde", 'to order by hand, to beckon'. The relation between the terjvo and veljvo is close enough for there to be an overlap: x2=m1. But "xancyminde" is not a "be"-lujvo, and the "gi'e"-lujvo "selxancyminde" sort of misses the point: the real relation expressed by the lujvo is "minde sepi'o le xance", 'to order using the hand', which can variously be expressed (depending on how strictly you want to constrain the seljvo) as "xancypliminde", "mindyxancypli" or "mindykemxancypli" (built up out of "xancypli lenu minde": "xance" fills the x2 of "pilno", leaving x3 open in "xancypli"; "minde" partly fills the x3 of "pilno" as a "belenu"-lujvo, leaving open the x2 and x3 of "minde". Thus "mi mindykemxancypli la bab. lenu klama": 'I beckoned to Bob to come over'.)

Does this make "xancyminde" wrong? By no means. But it does mean that there is a latent component to the meaning of "xancyminde", the sumti tcita "pi'o" (and, eventually the gismu "pilno"), which is not explicit in the seljvo. And it also means that, for a place structure derivation that actually makes sense, rather than being ad-hoc, the Lojbanist should probably go through a derivation like that of "mindykemxancypli" above, even if she decides to stick with a shorter, more convenient form like "xancyminde". Plus, of course, the possibilities of "be" + BAI interpretation of lujvo increase their potential ambiguity exponentially --- an unavoidable fact which should be borne in mind.

Note lastly that, if a Lojbanist requires a place from a lujvo not given to her by the terjvo or veljvo --- if, in other words, she feels tempted to add an extraneous place to the lujvo, other than those normally derived --- then the seljvo probably doesn't fully describe the relation involved, and a selbri describing the required place should be added somewhere. This is a different case to "xancyminde", where the places involved are all present, but the relation is not explicit. In the case of extraneous places, not all the places are present in the seljvo, and therefore the extraneous place will not be accessible by anyone trying to decode the lujvo without prior information. Adding to the seljvo gives the reader the ability to access that place.

Consider as an example the concept "meeting, workshop", which involves, at the least, people meeting (or workshopping), and a topic for the meeting (e.g. "Workshop on lujvo place structures". 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? No: there is nothing in the place structure of "nu" or "penmi" that could give us a topic of discussion. Lacking such a place, all "nunpenmi" can properly mean is "an act of meeting someone, an encounter". For a meeting with a topic of discussion to be expressed, any seljvo must include a selbri referring to topics of discussion. Such a selbri, of course, is "casnu", 'to discuss'. And indeed, "snununpenmi" expresses the desired meaning much better than "nunpenmi".

4.6. Relative ordering of terjvo and veljvo.

As we will see below, the way to analyse lujvo such as "catrysimxu", 'are reciprocal in killing, kill each other' is with "simxu" as the veljvo in a "belenu"-relation, paraphrasable as "da ce de simxu lenu catra". However, it is much more usual for "simxu" to be used as a terjvo than as a veljvo, in lujvo which have exactly the same meaning either way. Thus, the form "simcatra", which means the same as "catrysimxu", is found much more frequently.

The reason why "simxu" usually occurs as a terjvo, when formal analysis says it should be a veljvo, is that natural languages do not treat the word "reciprocally" as a head, but as a modifier. Before condemning this as muddy thinking, we should remember that, in some ways, the Lojbanic way of expressing reciprocality in Deep Structure ("da ce de simxu lenu brode") is counter-intuitive: the natural language equivalents of "simxu" ("each other", "inter-", reflexives, passives, reciprocal verb forms etc.) are mere grammatical machinery, while the complement of "simxu" still conveys the gist of the meaning. While we can't form an explicit Deep Structure with "catra" as the tertanru, we know intuitively that "simcatra" is a kind of "catra". We don't care that it is also a kind of "simxu", since "simxu" is a fairly abstract relation.

So it isn't that Lojban is uncovering something natural languages are missing, in its Deep Structure analyses. Rather, it is merely imposing its own sense of order in an environment that typically runs along quite different rules; lujvo place structures do reflect this conflict between a logical head and a 'psychological' head. This issue of logical veljvo giving Surface Structure terjvo occurs for quite a few gismu similar to "simxu": "simlu", "simsa", "mabla", "zabna", "mintu", "dukse", "carmi", "milxe", "fadni", "cnano" --- and, it may be argued, even commonplace gismu we would not consider just as grammatical machinery, such as "banli", "barda", "cmalu", "purci", and so forth. All these gismu correspond to natural language expressions which are typically modifiers rather than heads: adverbs ("seemingly", "somewhat"), adjectives ("damned", "great"), affixes (diminutives, "pre-", "ex-"), and so on.

Indeed, it seems to have become something of a shibboleth for Lojbanists to use 'logical' veljvo as either the veljvo or the terjvo . If they do the latter (e.g. "tepcai" instead of "camterpa", "intense in fear" instead of "intensely fear", for 'frightened'), it's a fair guess they are more commited to a methodical logical analysis of speech; if the former, they are indifferent, or make a point of keeping their Lojban from straying too extravagantly from natural languages.

5. Some common lujvo patterns.

Many of the lujvo we have collected are based on a small number of terjvo or veljvo. These lujvo fall into natural patterns, so it is makes sense to use these regularities as much as possible, to make their place structures consistent. In lujvo-making in general, the specific meaning and context of use of a lujvo may alter its place structure from the patterns we encourage. These patterns however, often corresponding to other languages' affixed rather than compound words, are so prevalent, that there is high motivation for consistency.

We list the most common such patterns below.

5.1. cmavo-based lujvo.

5.1.1. NU-based lujvo.

All lujvo based on a NU rafsi and a gismu require regular place structures, since they have unconstrained productivity ("nu" makes sense in front of any selbri whatsoever), and since there so little information in the veljvo to help decide the place structure on any other basis. Such a regular place structure is already outlined in the Abstraction paper in this grammar, whose results we summarise here:

"nunbroda": n1 b1 b2 b3 b4 b5 "dumbroda": d1 b1 b2 b3 b4 b5 d2 "jezbroda": j1 b1 b2 b3 b4 b5 j2 "kambroda": k1 b1 b2 b3 b4 b5 "lizbroda": l1 b1 b2 b3 b4 b5 l2 "mufbroda": m1 b1 b2 b3 b4 b5 "nilbroda": n1 b1 b2 b3 b4 b5 n2 "puvbroda": p1 b1 b2 b3 b4 b5 p2 "sizbroda": s1 b1 b2 b3 b4 b5 s2 "suvbroda": s1 b1 b2 b3 b4 b5 s2 "zazbroda": z1 b1 b2 b3 b4 b5 "zumbroda": z1 b1 b2 b3 b4 b5 z2

(For the definition of d2, j2 etc. see the place structures of NU cmavo as defined in the Abstraction paper.)

5.1.2. "jai"-based lujvo.

The cmavo "jai" is discussed in the Abstraction paper and the Place Structures paper; its function is to extract a place out of the abstract place in the x1 of the selbri, making it the new x1 place; the old x1 place is relegated to the 'extradimensional' "fai" place of the bridi. This is a function known as 'raising' in linguistics. Thus, the following two sentences are equivalent:

5.1) lenu do tavla fo la tlingan. cu cfipu mi

       Your talking in Klingon confused me.

5.2) do jai cfipu mi fai lenu do tavla fo la tlingan.

       You confused me by talking in Klingon.

Example (5.2) is the more natural expression in English; indeed, the place structure of "jai cfipu" as given in (5.2) is so much more natural to English speakers, that it used to be the place structure of "cfipu" itself. This, however, is a logical fallacy: the "jai cfipu" place does not convey any information independent of the "cfipu" place. In other words, the place structure in (5.2) cannot be regarded as logically basic, since "do" does not give any information not already contained in "lenu do tavla fo la tlingan". It has been decided for this gismu that people don't confuse people; events confuse people. Indeed, this is a decision that has been taken for many gismu in the gismu list.

The problem remains that, while redundant, the place structure in (5.2) is still appealing, because humans tend to talk not about abstract events, but concrete agents: it's appealing to say something like "do broda", rather than "lenu do brode cu cfipu" or "tu'a do cfipu". This is why "jai" is extremely useful. Since it is so useful, it makes sense to allow it to appear in lujvo, in order to give a single wod equivalent to the selbri in (5.2) --- which, incidentally, will give us a place structure like "cfipu" used to have before the desirability of unraised place structures was realised. So since "jai" has the rafsi "jax", it is possible to say "do jaxycfipu mi" as well as "tu'a do cfipu mi" or "lenu do tavla fo la tlingan. cfipu mi".

The one outstanding question is what becomes of the "fai" place in such a lujvo --- the "lenu do tavla fo la tlingan." in example (5.2). Should it remain a "fai" place, or should it be appended after the other places of the selbri? In other words, do we say "do mi cfipu fai lenu do tavla fo la tlingan.", or "do mi cfipu lenu do tavla fo la tlingan."? My personal inclination is to have the place structure of "jaxycfipu" match its gismu deep structure, (5.2), as much as possible: in other words, leave the "fai" place as a "fai" place. In the unlikely case that "jai" conversion is applied to a lujvo already containing "jai", subscripting can be used in a manner comparable to "ke'a" subscripting in nested relative clauses (see the Relativisation paper.) Thus, consider the following somewhat stilted example:

5.3a) lenu do se cfipu lenu mi tavla fo la tlingan. cu rinka lenu do fusra

       The fact that you were confused by my talking in Klingon caused you
       to be frustrated.

5.3b) lenu mi tavla fo la tlingan cu jai rinka (or: jaxri'a) lenu do fusra

       kei fai lenu do se cfipu leno'a
       My talking in Klingon caused you to be frustrated by confusing you.

5.3c) mi jai jai rinka (or: jai jaxri'a, or jaxyjaxri'a) lenu do fusra kei

       faixi1 lenu mi tavla fo la tlingan kei faixire lenu do se cfipu 
       lenu tavla
       I was a meta-cause (if you like) of you being frustrated by confusing
       you, by talking in Klingon.

It should hopefully be obvious that this kind of problem will rarely arise, even though "jai" may well become much more prevalent than might be anticipated (because of the appeal of 'raised' place structures like (5.2).)

5.1.3. KOhA-based lujvo.

There exist rafsi allocated to a few cmavo of selma'o KOhA, but they are rarely used. The obvious way to use them is as terjvo, filling in a given place of the veljvo, and giving a "be"-lujvo. Thus "donta'a" 'you-talk' would be interpreted as "tavla be do", and would have the place structure: t1 t3 t4, since t2 (the addressee) is already known to be "do". The lujvo "donma'o" 'you-cmavo', which means "a second person personal pronoun", would be interpreted as "cmavo be zo do", and have the place structure: c1 c4 (since both the se cmavo and te cmavo are obvious from "do".)

An anticipated use of rafsi for cmavo in the "fo'a" series is to express terjvo which can't be expressed in a convenient rafsi form, because they are too long to express, or are formally inconvenient (fu'ivla, cmene, and so forth.) An example would be "fo'a goi le kulnrsu,omi .i lo fo'arselsanga" ("x6 stands for Finland. An x6-song.") Lujvo involving the rafsi for "bu'a" would presumably operate in the same way.

Finally, lujvo involving "zi'o" are also possible, and are fully discussed in the Pro-sumti paper. In brief, the convention is to use the rafsi for "zi'o" as the terjvo, followed by the rafsi for the number of the place to be deleted. Thus, if we consider a beverage as a "se pinxe be zi'o", the lujvo corresponding to this is "zilrelselpinxe" (deleting the second place of "se pinxe"). Presumably, deleting the x1 place in this fashion would move all remaining places up by one. This would mean that "zilpavypinxe" has the same place structure as "zilrelselpinxe", and "lo zilpavypinxe", like "lo zilrelselpinxe", refers to a beverage, and not to a non-existent drinker.

5.1.4. PA-based lujvo.

These lujvo tend to play havoc with our predictions of lujvo meaning. This is because numerals are not selbri, to enter into tanru relations with tertanru in a seljvo. Rather, they quantify their head directly. So in this case, the seljvo is not a tanru. The seljvo of "so'ipre", for example, is "so'i prenu".

Even if we interpose a cmavo of selma'o MOI making the numeral a selbri --- which seems the only intelligible way to proceed --- the choice of cmavo is ambiguous. Thus, "pavjbe", 'one-born', is ambiguous, and in fact only appears in the jvoste as its disambiguated alternatives "pavmemjbe" ("pamei jbena", 'only-begotten') and "pavmomjbe" ("pamoi jbena", 'first born'). Both alternatives have been presumed in the jvoste: thus "relselmla", 'two-sided' ("se mlana lo remei") and "so'ipre", 'crowd' ("so'imei co prenu") presume "mei", whereas "relpru", 'second-last' ("remoi le'i purci leka jibni le se purci") and "cibdei", 'Wednesday' ("cimoi le'i djedi krefu") presume "moi".

Of PA-based lujvo "ro"-based lujvo, in particular, often display eccentric behaviour. For instance "roldei", meaning "daily", properly has a place structure which has little to do with that of "djedi". Its real place structure has more to do with its Deep Structure, "ckaji leka se krefu ca ro djedi". These lujvo are considered as having the implicit veljvo "ckaji", and therefore have a place structure following the conventions for "ckaji"-lujvo (discussed below.)

These lujvo are being constructed to mirror natural language use of numeric prefixes (bi-, mono-, omni-, all-, etc.). They do not fit comfortably with a Lojbanic analysis: in particular, it does not seem approporiate to translate "This newspaper is a daily" as "levi karni cu roldei": this seems tantamount to saying that a newspaper is a totality of days! (See the discussion above on elision of SE cmavo from veljvo for further arguments.) The reason they are tolerated in Lojban is that such lujvo themselves are typically used as tertanru, since they correspond to English adjectives. Thus, 'daily newspaper' may often be rendered as "roldei karni"; this makes sense, because the tanru merely gives a loose relation between 'newspaper' and 'all days'. But as explained, once the tanru is disambiguated, "roldei" doesn't correspond to 'daily', a notion better characterised as "roldeikai", 'characterised by [appearing] every day.' Indeed, a full elaboration of the meaning of 'daily' is given by the lujvo "djerolmemcabrefkai" ('characterised by a recurrence contemporaneous with every day'.) In context, however, a less elaborated lujvo like "roldeikai", and even --- in tertanru --- "roldei" is quite sufficient.

5.1.5. "pe'a"-based lujvo.

These lujvo are discussed in detail in the Attitudinal paper. We note that few such lujvo have been coined. The only ones to have appeared in text to date are: "pevbaknykalci", 'bullshit'; "pevjicla", 'fuss, disturbance'; "pevrisnyjelca", 'heartburn'; and "pevycuvgau", 'purge (political)'. As the attitudinal paper says, the place structure of such coinings is entirely unpredictable; it remains to be seen whether Lojbanists will widely avail themselves of this flexibility. It is also possible (indeed, in natural languages it happens all the time) that the figurative meaning can take over the literal meaning. If this happens, the "-pev-" rafsi may even end up dropped --- although, should that occur, then the lexical transparency of the lujvo will have been lost (it will no longer be related to its seljvo in any consistent way), and the lujvo may as well be a le'avla, or a new gismu.

5.1.6. Other cmavo-based lujvo.

For the same reason as given for NU-based lujvo, SE-based lujvo have the same function in lujvo as they do in tanru, and "selbroda" has the same place structure as "se broda". (See section 4.2.2 for further argumentation on this point.)

In fact, cmavo from all the following selma'o behave just as they do in tanru, and will not be discussed any further in this paper: BO, CAhA, CO (which merely serves to swap the veljvo and terjvo in any place structure determination), FAhA, GOhA (the cmavo "du" and "co'e", which would behave the same as any gismu in a tanru), JA, JOI (which would both consistently follow "gi'e"-patterns of place structure), KE, KEI, KEhE, MOI, MOhI, NA, NAhE, ROI (note that "roi" as a veljvo is not well defined, since PA + "roi" does not form a selbri), VA, VEhA, ZAhO, and ZEhE.

There are also rafsi assigned to "le'e" and "lo'e". These are presumably intended as terjvo, and as a convenient shorthand for selbri corresponding to 'stereotypical' and 'typical'; as such, they would have a "gi'e"-interpretation. (Thus, "lo lempre", 'a stereotypical person' --- disregarding the obvious clash between "lo" and "le'e".) Finally, "bu" has also been given a rafsi --- a resource yet to be explored.

5.2. gismu-based lujvo.

5.2.1. "rinka"- and "gasnu"-based lujvo.

These lujvo have a long history in Lojban. They have been an impetus to lujvo place structure investigation, through their well-defined place structures as "belenu"-lujvo, and their similarity to transitivising and causative affixes in other languages. They are extremely productive, and help make Lojban more speakable, by simplifying the structural representation of complex gismu deep structures.

By looking only at the keywords in the gismu lists, Lojban users may be unaware that English often expresses two distinct concepts with the same verb, where Lojban must use two different bridi. As discussed above, the English verb "to sink" has two meanings. The intransitive meaning, as in "The boat sinks", is that something is lowered. The transitive, as in "The Moors sunk two ships", is that some agent cause something to be lowered.

Lojban gismu usually express intransitive concepts. For example, the related concept of 'immersion' is expressed by "jinru" as: entity x1 is immersed in liquid x2. The related transitive concept is expressed by the Deep Structure "tu'a da cu rinka lenu de jinru di", or "da gasnu lenu de jinru di". As we discussed in section 1, it makes more sense to speakers of many languages to express this transitive concept as a single selbri: "jinryri'a" or "jinrygau".

"broda zei rinka" lujvo have the place structure: r1 b1 b2 b3 b4 b5 r3; "broda zei gasnu" lujvo have the place structure: g1 b1 b2 b3 b4 b5: these are "belenu"-lujvo par excellence. The gismu "rinka" should be used with caution to form lujvo, because the x1 of "rinka" is a cause, not a causer. If one wants to refer to a causer, one should use either "jai" (discussed elsewhere), or "gasnu", whose x1 is always an agent. In general, people prefer to speak of relations between agents ("loi gasnu no'u loi jai rinka") and patients ("loi jai se gasnu no'u loi jai se rinka"), rather than causes ("loi rinka") and effects ("loi se rinka"); so "gasnu" is used much more frequently as a veljvo.

These lujvo do not onlt do the equivalent of transitivising an intransitive; they can also or making an already transitive verb a causative. (E.g. "basti": x1 replaces x2 in circumstances x3 produces "basygau": x1 (agent) replaces x2 with x3 in circumstances x4.). In addition, they can affect what we would consider nouns or adjectives in English. (In Lojban, everything is a predicate, so adjectives, nouns and verbs are all treated in the same way.) This is consistent with similar affixes in other languages.

For example, "glare": x1 is hot by standard x2, can give "glagau", 'to heat': x1 (agent) makes x2 hot by standard x3. Or, "litki": x1 is a liquid of composition x2 under conditions x3, can give "likygau", to 'liquefy': x1 (agent) causes x2 to be a liquid of composition x3 under conditions x4. (This particular case is problematic: x2 seems redundant, and this may indicate that "gasnu" is the wrong veljvo. The veljvo "galfi" is more appropriate in some such cases.)

Note that, particularly with "gasnu", the terjvo need not be in a "belenu"-relation with the veljvo. It may specify the manner of the veljvo instead. The lujvo "kalsygau", for example, may not mean 'to make somethic chaotic, to mess something up)'; it may simply mean 'to act chaotic, to do something chaotically', giving a "gi'e"-interpretation to the lujvo. In such cases, the lujvo-maker may have to augment the lujvo to disambiguate it.

5.2.2. "zmadu"- and "mleca"-based lujvo.

These lujvo also express a frequent construct in languages: comparatives. They express the concept of exceeding, in a way more familiar to speakers of other languages than the corresponding Deep Structure. Compare:

5.4) I am six years younger than you.

       ".i mi citmau do lo nanca be li xa"
       ".i mi zmadu do leni ce'u da citno kei lo nanca be li xa"

The place structure for "citmau" is clearly more intuitive than the complex abstraction given by "zmadu lenu citno". (Note that, since "ce'u da" fills a slot in the abstraction, and is explicitly bound to it, it is not necessary in the lujvo --- where the abstraction does not appear as such --- and so is not included in its place structure. For more discussion on the quantifier "ce'u", see the Abstraction paper.)

These lujvo are extremely productive, "zmadu" much more so than "mleca". Because of their much simpler place structure, they are in fact used much more frequently than "zmadu" and "mleca" themselves as selbri. It is highly unlikely for such lujvo to be construed as anything other than "belenu"-lujvo. But there is another type of ambiguity relevant to these lujvo, and which has to do with the placing of the "ce'u da" expression omitted from the lujvo. For example, does "nelcymau" mean "X likes Y more than she does Z" ("zmadu fi leni ce'u da nelci"), or "X likes Y more than Z does" ("zmadu fi lenu nelci ce'u da")? Does "klamau" mean: "X goes to Y more than to Z" ("zmadu fi lenu klama ce'u da"), "X goes to Y more than Z does" ("fi lenu ce'u da klama"), "X goes to Y from Z more than from W" ("fi lenu klama fi ce'u da"), or what?

We answer this by putting regularity above any considerations of concept usefulness: by convention, we consider the "ce'u da" place to belong to the x1 of the terjvo. In that way, each of the different possible interpretations can be expressed by SE-converting the terjvo, and making the required place its new x1. Thus, the possible interpretations in the previous paragraph can be expressed as "nelcymau", "selnelcymau" ("zmadu fi lenu ce'u da se nelci"), "klamau", "selklamau", and "terklamau" ("zmadi fi lenu ce'u da te klama").

Thus, a "broda zei zmadu" lujvo has the place structure: z1=b1 z2=b1 b2 b3 b4 b5 z4, and a "broda zei mleca" lujvo has the place structure: m1=b1 m2=b1 b2 b3 b4 b5 m4. (But see Section 2.2.1 for an instance where not all the terjvo places need go into the final lujvo.)

The place structure we have given is that of a "belenu"-lujvo, as the Deep Structure above shows. Unfortunately, it displaces the "ve zmadu" place, which expresses by how much one entity exceeds the other, by an number of places different for each lujvo. For example, while "nelcymau" has the measure of comparison as its fourth place, "klamymau" has it as its sixth place. This discrepancy may eventually justify jumbling the places, treating such lujvo as "gi'e"-lujvo, and placing the "ve zmadu" place before the terjvo places. For now, however, I have left the place structure in "belenu"-arrangement.

Note that such lujvo do not presuppose their terjvo. Just as in English, saying someone is younger than someone else doesn't imply that they're young in the first place: an octagenarian, after all, is still younger than a nonagenarian. Rather, the 80-year-old has a greater "ni citno" than the 90-year-old. We could take a fuzzy-logic approach: the 80-year old has a "ni citno" of 1/80, while the 90-year-old has a "ni citno" of 1/90. To be considered young, however, you may have to have a "ni citno" greater than, say, 1/40.

5.2.3. "zenba"- and "jdika"-based lujvo.

There are some concepts in which the "se zmadu" is difficult to specify. Typically, these involve comparisons implicitly made with a former state of affairs, where stating a "se zmadu" explicitly would be problematic. Thus, we can't really translate "I'm going to 78th Street more nowadays" into something involving a "zmadu" selbri, since it is not clear what we would put in the x2 of such a selbri. The best we could do is say "mi zmadu mi pe pu ku lenu klama la 78moi klaji", but "mi pe pu ku" is not a particularly elegant solution to the problem, and is giving an explicit argument where it is not clear it is a necessary part of the relation described.

In such cases, it is best not to leave any comparisons like "zmadu" hanging, but to use instead the gismu "zenba" (and "jdika" for "mleca"). The gismu "zenba" was included in the language precisely in order to capture those notions of increase which "zmadu" can't quite cope with; and we don't have to waste a space in lujvo or tanru on what we'd never fill in with a value anyway. So while "mi ca zmadu fi lenu mi klama la 78moi klaji" means @@@something like "I'm going to 78th Street more than X does", where X is a problematic place to fill, "mi ca zenba lenu mi klama la 78moi klaji" means just what "I'm going to 78th Street more nowadays" does. Accordingly, "klaze'a" would mean exactly the same thing. The phrase "I'm stronger now" would be translated, not as "mi ca tsamau", which implies that I'm stronger than somebody else, but "mi ca tsaze'a", 'I increase in strength'.

The place structure of "broda zei zenba" lujvo is: b1=z1 b2 b3 b4 b5 z3; for "broda zei jdika", it is: b1=j1 b2 b3 b4 b5 j3. These are "belenu"-lujvo, and the same issues concern their use as with "zmadu"- and "mleca"-based lujvo.

5.2.4. "traji"-based lujvo.

Just like "zmadu"-based lujvo are used to build comparatives, {traji}- based lujvo are used to build superlatives. Thus "xagrai" is taken to mean "best". Since the place structure of "xagmau" is x1 z2 x2 x3 z4, we would expect the place structure of "xagrai" to somehow mirror that, given that comparatives and superlatives are comparable concepts: x1 is the best of all x2 as far as x3 is concerned, in quality x4. (The place in "traji" for property is made redundant by the terjvo, and the place for the extreme of "traji" (whether the most, or the least) is presumed by default to be "the most".)

But the set against which the x1 of "traji" is compared is not the x2 (which would make it parallel to "zmadu"), but the x4 of "traji". This decision has been reaffirmed by the gismu list editors, who argue that "traji" is semantically closer to "mutce" than to "zmadu". This makes the resulting lujvo place structure somewhat unwieldy: "belenu"-place ordering, which has the remaining places of "xamgu" replace place t2, calls for the place structure x1 x2 x3 t4: x1 is the best as far as x2 is concerned in quality x3, beating out all x4.

Now this means that "the best of X" is expressed, not as "le xagrai be xy." but as "le xagrai be fo xy." This has the same disadvantage as the "ve zmadu" place in "zmadu"-based lujvo, only compounded. The place t4, the set from which the extreme is selected, is intuitively the most important place of such a lujvo, after t1; we think of the terjvo places, outlining the quality in which t1 is extreme, as subsidiary in meaning. But not only are we forced by "belenu"-ordering to put t4 after other, relatively unimportant places; t4 also ends up occuring in different places for different lujvo, depending on the place structure of the terjvo. Thus "Conan is the strongest of all barbarians" ends up translating as "la konan. cu tsarai fi ro cilcyre'a", while "Judy is the youngest of all Lojbanists" translates as "la djudis. cu citrai fe ro lobypli", and "Einstein was the greatest of all scientists" as "la ajnctain. cu balrai fo ro skegunka". This is obviously unsatisfactory.

What can be done about this is unclear. One option (consistent with Jim Carter's philosophy on lujvo) is to change the gismu place structure to match our lujvo. But this is unlikely, and inconsistent with current thinking on how gismu place structures should be decided. Another is to change the veljvo --- but it is not clear which other veljvo can do the job of 'most'. A third option (in my view the most sensible) is to flex the principles outlined here, and to allow t4 to take the position in the place structure we want it to: namely, to allow x1 t4 x2 x3. This is in fact how these place structures appear in the jvoste, although we acknowledge that this is a violation of the guidelines expressed in this paper.

5.2.5. "ckaji"-based lujvo.

This type of lujvo was mentioned briefly above, in connection with "ro"-based lujvo such as "roldei" and "rolgu'e". These lujvo form the equivalents of adjectives derived from other predicates, when that predicate itself does not apply to what is being described. Thus, it doesn't make much sense to call something "sutrykai", 'characterised by switfness', when once can just as readily say it is just "sutra", 'swift'. On the other hand, it does make sense to say that something like a mountain is "rokcykai", 'rocky', though it might not itself be a rock, "rokci". Since such lujvo are somewhat vague in describing the nature of the characterisation, and since there are often other means available to the Lojbanist for a more accurate description (other places in the same selbri --- e.g. "big-nosed" can be rendered as "se bardynazbi"; other veljvo such as "simsa" and "ponse", and so on), "ckaji"-based lujvo, which would typically be "belenu"-patterened ("ckaji be leka broda"), have not yet been used much in the language; but the trap with "ro"-based lujvo indicated above shows that they should form part of the Lojbanist's inventory.

5.2.6. "simxu"-based lujvo.

The gismu "simxu" occurs in the jvoste as both a terjvo and a veljvo. Both cases seem to have equivalent meaning, and correspond to a Deep Structure giving "simxu" as the veljvo. (More precisely, "simxu" is the selbri of the Deep Structure for a jufra where the lujvo is the selbri.) The reason why "simxu" might be used as either a terjvo or veljvo with the same meaning is discussed extensively above; "simxu" is a straightforward "belenu"-lujvo. Thus "simcatra", 'kill each other', corresponds to: "da ce de simxu lenu da catra de soivo'a di". This gives the place structure: s1 = c1&c2, c3. For example, "Mercutio and Tybalt kill each other with daggers" translates to "la merkucos. ce la tibalt. simcatra tu'a loi dakfu". Note that the x1 of a "simxu"-based lujvo, just like the x1 of "simxu" itself, is a set.

As a terjvo (or veljvo), "simxu" is handy in that it can convert a 2-way relation into an arbitrary n-way relation. For example, "penmi" is something two people do to each other: "la DRAkulys. penmi la godzilys.", 'Dracula meets Godzilla". But what if we want to say that Dracula, Godzilla and Frankenstein meet? The sentence "la DRAkulys. penmi la godzilys. e la FRANkenstain." does not imply that D. meets the other two together; the sentence "la DRAkulys. penmi la godzilys. joi la FRANkenstain." does, but fails to imply that G. and F. meet. However, since we can say "la DRAkulys. ce la godzilys. simpenmi", there is no reason we can't also say "la DRAkulys. ce la godzilys. ce la FRANkenstain. simpenmi", neatly resolving the dilemma.

This is a neat way of extending a 2-way relation if the relation is reciprocal, as is the case with selbri like "pencu", "tavla", "danre", "gletu" and so forth. But if the relation is not reciprocal, "simxu" won't do. An example of this is "vlina", 'logical disjunction': "la'elu .abu .a by. li'u vlina .abu by.", "'A or B' is the disjunction of A and B". Suppose we make 'OR' an operator taking not two arguments, but as many as we like; how could we describe OR(a,b,c)? We might say "ma'o tau .abu .abu ge'a by. cy. vlina .abu by. cy.", sticking an extra place on to "vlina"; but adding in places to gismu at will is risky, and probably won't work for most such gismu (since trailing gismu places might get in the way). The best alternative solution seems to be to using "selkampu" as a veljvo, giving in this instance something along the lines of ".abu ce by. ce cy. vlinyselkampu ma'o tau .abu .abu ge'a by. cy.", "A, B and C (as a set) have in common the disjunction OR(a,b,c)".

5.2.7. "mintu"-based lujvo.

As with "simxu", "mintu" occurs as both terjvo and veljvo in lujvo with no change in meaning, and with a 'logical' Deep Structure where "mintu" is the veljvo in a "belenu"-dependency. So both "broda zei mintu" and "mintu zei broda" are analysed as "da mintu de leka ce'u di broda". The "ce'u di" place indicates that "leka broda" can aply to either "da" or "de", and does not appear in the final lujvo place structure.

Thus the lujvo "vlami'u", 'synonym', can be analysed as follows: v1=m1 v1=m2 m3. The "te mintu" place is left in because two words can be 'the same' in a number of ways: pronunciation, semantics (exactly or approximately so), spelling, etc. We have left out places v2 and v3 (the meaning and language of the words) as information that properly belongs in m3. If the English words "horse" and "steed" are synonyms, part of their "te mintu" is that they mean the same thing in English. On the other hand, the English word "curve" and the Lojban word "curve" are also "vlami'u", but their "te mintu" is that they are spelt the same. It would make no sense to have a language or a meaning as part of the place structure of "vlami'u", since English "curve" and Lojban "curve" have neither in common.

"mintu"-based lujvo are productive in constructing lujvo equivalent to expressions in English using the prefix "con-", or "fellow-". For example: "Mahler was a contemporary of Klimt" can be translated as "la maler. cedrymi'u la klimt."; "Zamenhof appealed to his coreligionists" (or in more modern parlance, "fellow Jews") as "la ZAmenxof. cpedu tu'a lei seljdami'u be ri"; "My fellow Lojbanists!" as "doi jbomi'u be mi".

Incidentally, there is a contrast between "seljdami'u" and "cedrymi'u": Zamenhof was a "seljda", but Mahler was not a "cedra". Mahler and Klimt are in fact "simymi'u ba'e tu'a le cedra". It can be argued that "cedrymi'u" is an abbreviation for "cedrycabzasmi'u", 'the same in that they existed during a given era'. As discussed above, such 'abbreviations' are very frequent in lujvo-making. We also shouldn't be surprised to see SE elision from the terjvo of such lujvo, like "jdami'u" ('same in religion') for "seljdami'u", and "lazmi'u", 'person in the same family as, a relative of', for the more fully descriptive "selylazmi'u": plausibility can be typically relied on in lujvo of this type to help the listener access the desired interpretation.

5.2.8. "cmalu"- and "barda"-based lujvo.

These correspond to augmentatives and diminutives, particularly when they appear as terjvo (which is nearly always), and they are treated as "gi'e"-lujvo. Thus, "cmalu zei broda" has place structure c1=b1 b2 b3 b4 b5 [c2] c3. Sometimes c2, the way in which "le broda" is small (or big), is obvious, and can be left out. Thus "cmalalxu", "small lake", has place structure: l1 l2 c3; it is obvious the lake will be small in area, so the "se cmalu" place would not serve any useful function in the lujvo. Were the lake small in, say, depth, a different lujvo would be used. Similarly, "barda zei broda" has place structure ba1=br1 br2 br3 br4 br5 [ba2] ba3.

Note that, in many natural languages, diminutives and augmentatives are often used to convey some derogatory or affectionate attitude about the word. In Lojban, this function is carried out by attitudinals, and it would be confusing to allow it with these lujvo too: they should only have the literal interpretation. Thus "cmatuple" should mean only "small feet", and more specifically "feet that are small for the entity that has them"; they should not mean "dear little feet", which is conveyed by "tuple .iu".

5.2.9. "stuzi"- and "zdani"-based lujvo.

When "stuzi" is the veljvo, these are either "be"-lujvo or "belenu"-lujvo; in either case, the place structure of "broda zei stuzi" is s1 b1 b2 b3 b4 b5. Thus "depstu", 'waiting-place' (e.g. bus-stop) has the place structure: s1 d1 d2 d3. The same holds for "zdani": eg. "ckuzda", 'library': z1 z2=c1. This kind of lujvo place structure can be extended to analogous gismu, even when their place structure doesn't immediately support it. An example is "kumfa" in lujvo like "jupku'a", 'kitchen': there is no place in "kumfa" corresponding to "se zdani" or "se stuzi", yet "jupku'a" is immediately interpreted as 'room for cooking', after the pattern of such expressions as 'place for cooking' and 'house for cooking'.

5.2.10. "carmi"- and "milxe"-based lujvo.

As with other gismu, these occurs as either terjvo or veljvo without changing the meaning of the lujvo; the Deep Structure has them as the terjvo. The place structure of "broda zei carmi" and "carmi zei broda" is: c1=b1 b2 b3 b4 b5 c3. The place c2, the way in which "le broda" is intense, is considered redundant. Thus "caicta", 'stare', has place structure cat1=car1 cat2 car3, and Deep Structure "da carmi leka da catlu de kei di". "milxe"-based lujvo are constructed similarly, but "milxe" has no observer place. Thus the place structure of "milxe zei broda" is m1=b1 b2 b3 b4 b5.

5.2.11. "mabla"- and "zabna"-based lujvo.

These are among the most productive gismu in lujvo, and indeed occur almost exclusively in lujvo. Their meaning is almost always obvious, which is why they don't feature greatly in the jvoste. The place structure of "mabla zei broda" is m1=b1 b2 b3 b4 b5 m3; m2 is taken as obvious (m1 is "le broda" according to "le te mabla", and m2 is "le broda" objectively). The structure of "zabna zei broda" parallels this. In these lujvo, too, "mabla" and "zabna" are logically the veljvo.

5.2.12. "sevzi"-based lujvo.

These lujvo do the work of reflexives, and indicate that the lujvo action reflects back on the agent. In Lojbanic terms, it means that one of the places in the veljvo (typically x2) is the same as x1, and can be omitted. Thus the place structure of "sezlu'i", 'to wash (intr.), to wash oneself', is: l1=l2 l3. We can explain this from our normal rules as a "be"-lujvo, giving the place structure l1=s2 l3, and omitting l2=s1 as redundant: this gives the same result. So we can say "mi sezlu'i loi zbabu", 'I wash with soap', as well as "mi lumci mi loi zbabu", 'I wash myself with soap'.

Because English often conflates reflexivity with intransitivity, we can find it easy to omit a necessary "sevzi" in the veljvo. Thus "cavlu'i" doesn't mean 'to take a shower', but 'to shower/sprinkle/spray someone/something'". The proper translation for 'to take a shower' is "sezycavlu'i".

5.2.13. "prenu"-based lujvo.

The place structure of these lujvo yield no surprises: their first place refers to a person, and the remaining places are terjvo places describing what the person does. Note, however, that often the veljvo "prenu" is redundant, since the terjvo tends to already imply that we are talking about a person. Thus, as we saw in discussions above, "djabeipre" says nothing much that "djabei" doesn't already say.

6. Worked examples.

[To be supplied]

7. Acknowledgements.

The author (Nick Nicholas) would like to thank the following Lojbanists: Mark Shoulson, Veijo Vilva, Colin Fine, And Rosta, and Iain Alexander for their suggestions and comments; John Cowan, for his extensive comments, his examplary trailblazing of Lojban grammar, and for solving the "manskapi" dilemma for me; Jorge Llambias, for his even more extensive comments, and for forcing me to think more than I was inclined to; Bob LeChevalier, for his skeptical overview of the issue, his encouragement, and for scouring all Lojban text his computer has been burdened with for lujvo; Nora Tansky LeChevalier, for writing the program converting old rafsi text to new rafsi text, and sparing me from embarrassing errors; and Jim Carter, for his dogged persistence in analysing lujvo algorithmically, which inspired this research, and for first identifying the three lujvo classes.

      • I have a note that I should go through further exchanges on the

list, and privately between me and Lojbab, on lujvo --- but I can't currently access my archives. Therefore I rely on my reviewers' judgement! ***