Gua\spi, an Artifical Natural Language

James F. Carter

15 September 1991

Abstract: Gua\spi is an artificial language suited to both humans and machines. It can express real human conversation. Yet the vocabulary and the grammar are two and three orders of magnitude simpler than English. Word and phrase meanings are defined through predicate calculus and hence can be represented and manipulated efficiently and unambiguously by programs (and people).

Gua\spi is a language artifact. It can do everything that can be done by a natural language like English, though it is wholly artificial. But since it is so much simpler, studies of the phenomenon of language can also be simpler, more conclusive, and less costly in the investigator's time, and software which functions by imitating human language behavior can be written more easily.

The demonstration that gua\spi is fully functional as a natural language involves a difficult process: the reader must think of some language behavior, it is translated into gua\spi, and the reader then judges whether the translation is adequate, using his or her skill in gua\spi to understand the result. This paper hasn't enough length to be a complete users' guide to gua\spi, but I will present many of the basic concepts so the interested reader can make an independent judgement of whether simple phrases are adequately translated. Then I will give a short sample of live discourse, machine-translated from gua\spi to pseudo-English as an example of the kind of topics that can be represented and how much of the meaning survives mechanistic parsing. I hope the reader may be lead to wonder how various advanced topics are handled in gua\spi and that, on getting more complete information, he or she will find the treatment satisfactory.

Salient characteristics of gua\spi that particularly suit it to use by humans and artificial intelligences together are:

The language artifact Loglan, developed by James Cooke Brown [L1], was the inspiration for gua\spi. Brown realized that a very small set of content words could form a basis of a language, and produced such a set. By successfully writing large amounts of prose in Loglan while creating almost no additional words, I validated his insight. Loglan is almost 100 times simpler than English, and I have simplified the deep structures laid bare in Loglan by almost 100 times more to give gua\spi.

The description of gua\spi that follows is necessarily incomplete. I hope that the reader will be led to wonder how gua\spi handles this or that problem --- and that, on getting more information, he will find that the problem is handled adequately. A fuller description of gua\spi is available, and a dictionary and teaching materials are in preparation [Ga].

Morphology --- What is a Word

The phonemes (sounds) are divided in two classes, C's and V's, or kona and vumu in gua\spi. All C's are consonants in English and those English vowels used in gua\spi are all in the V class, hence the names. In addition each word has a tone (dinu), a frequency modulation of the V's of each word in the Chinese manner. A word is written as a tone (see Table 3 [Tones]), one or more C's and one or more V's. What could be simpler?

C/V Length Sound Labial Dental Palatal Velar Glottal
C Plosive unvoiced p t c* k ---
C Plosive voiced b d j g :*
C Spirant unvoiced f s q* --- ---
C Spirant voiced v z x* --- #*
V Vowels u o y i,e a
V Nasal etc. m n l w* r
Table 1 [Phonemes]. Gua\spi phonemes, arranged by tongue position front to back (reading across) and sound type (reading down). Letters marked `*' differ from European standard usage.

Gua\spi English Examples of Pronunciation
c ch CHew, Ciao (Italian)
q sh SHoe
x zh aZure, breZHnev (Russian)
: (pause) the:apple, hawai:i (glottal stop)
# uh thE, Among (schwa)
u u, oo flUte, bOOt
o o, oa bOne, bOAt
y i knIt
i i, ee grEEn machIne (not eye)
e e bEd
a a fAther (not cAt)
mnlr mnlr LeMoN RiNd (no silent R)
w ng stroNG
Table 2 [Pronunciation]. How to pronounce gua\spi phonemes. Nonstandard C's are shown; C's without examples are as in English. Standard radio broadcast accent is close to correct for the V's; Spanish is closer. Pronounce the vowels as one sound, not a glide between two sounds as in ``eye''.

Table 1 [Phonemes] shows the phonemes, categorized by tongue position and sound type. Some phonemes are represented confusingly in English, e.g. `sh' which sounds like neither `s' nor `h'. So in gua\spi they are assigned individual letters which differ from English usage --- `q' for `sh'. Table 2 [Pronunciation] gives examples of these, and all the vowels. There is a 1-1 relation between written and spoken gua\spi. Written blanks have no sound, and are optional. There is no distinction between upper and lower case.

All words must begin with a C and end with a V so foreign words must be modified to fit. They appear in gua\spi as possibly several CnVn syllables. A prefix, most commonly ``qo'' for a foreign name, signals that the syllables are not to be looked up in the dictionary and are to be kept together when other compound words are split up during organization.

Grammar by Tones --- How Words Join

The grammar is stated in Backus-Naur form in the Appendix. The job of grammar is to stick words together into phrases (zdua). Gua\spi grammar produces a strict tree-structure parse for each sentence wherein one phrase is at the root forming the main sentence, and each phrase has an ordered list of zero or more sub-phrases. The grammar does not support meaning of any kind --- no tenses, no possessives, no nouns, no verbs. These ideas are handled at the organizational and semantic levels, using the grammar as a foundation. Like its morphology, the grammar of gua\spi is nearly minimal.

For grammatical purposes there is only one kind of phrase (though distinctions are made at the organizational level), but words have five categories: the two grammar words ``fu'' and ``fi'', sentence start prefixes, other prefixes, and everything else. The main part of a phrase is a sequence of one or more words collectively called the ``phrase predicate''; any prefixes in this must come first. During grammatical analysis there may be several content words in the predicate, though later transformations split up all compound words into separate phrases. After any of the prefixes or after the whole predicate the sub-phrases are interspersed where convenient; they attach to the current phrase at the next higher level.

Fig. 1 [Parsetree] shows a parse tree for a simple sentence with two levels of sub-phrases. The tones (see Table 3 [tones]) show the level of each word relative to the one before it.

^:i !tara /vme -crw !kseo ^vu -tum !kfor ^fe -fnauThe rat devours (violently eats) the cheese with a fork and knife
   ^:i       /vme-crw
  (Start)    -1 devours
     |           |
            |         |           |
          !tara     !kseo      ^vu-tum
          +1 rat   +1 cheese   +0 using
                                     |              | 
                                  !kfor         ^fe-fnau
				  +1 fork       +0 knife

Figure 1 [Parsetree]. A Sample Parse Tree. Each phrase is part of the most recent one at the next higher level. Each phrase may have sub-phrases. Tone symbols (see Table 3 [Tones]) show jumps into and out of sub-phrases; they show the difference in level between adjacent words.

The tone, or frequency modulation, of a gua\spi word represents its parse tree level relative to the word just before it. Table 3 [Tones] shows the sounds and interpretations of the tones. The tones are the most terrifying aspect of gua\spi for speakers of European languages. Informal experimentation shows that naive non-Asian listeners can hear the tones reliably. Please remember that over a billion people in China and Southeast Asia speak tonal languages. If they can do it, so can you.

Number Sound Level Type Symbols
1 High-even Predicate Compound - -
2 Rising One higher Phrase / /
3 Down-up One lower Clause | *
4 Falling One lower Argument \ !
5 Up-down No change Phrase ^ @
6 Low-even Predicate Transitive = %
Table 3 [Tones]. Sounds and interpretations of the tones. ``Level'' refers to the parse tree level of the word with that tone, relative to the structure before it. ``Type'' indicates the organizational type of that word or phrase. The first set of symbols shown, ascii characters, is preferred but the second set can substitute on a manual typewriter. In this paper, `!' is used instead of `\' for convenience in typesetting.

Tones `-' and `=' join adjacent words of a compound phrase predicate. Tones `!' and `|' start a sub-phrase of the current phrase. Tone `^' closes the current sub-phrase and starts a new one, part of the same containing phrase. Tone `/' closes a sub-phrase and resumes the predicate of the containing phrase, if among its prefixes, or otherwise starts a new phrase at the higher level. Distinctions within these tone classes are important later but do not affect the grammar.

A sentence start prefix such as ``^:i'' is automatically at root level, and ``/fi'' jumps to root level without ending the sentence. Other multi-level upjumps are available with ``fu'' but are needed rarely.

Cases --- Members of a Relation

The next layer of gua\spi syntax is the organizational level, but to understand the reason for some organizations we have to make a detour into semantics to find out about cases (skam). We also get to see some examples of gua\spi sentences. To minimize vocabulary we will use variations on this one:

^:i !tara /crw !kseo The rat eats the cheese (sentence: rat eat cheese)

Please pronounce it correctly: `c' as English `ch' and `i' as `ee'. Mind the tones, lest you change it into ``the eat rats the cheese'' or some such. (Chinese is worse: you could change ``mother'' into ``horse'' with a wrong tone. But Chinese people survive nicely.) The notation ``tara-rat'' in examples means that ``tara'' is the example word, and it means ``rat'' in English. Isolated words or phrases like this are written without a tone because it depends on the context where the word is used.

Natural languages generally distinguish between ``things'' and ``actions'', where an ``action'' is a relation between ``things''. The formal term for such a relation is a ``predicate'' (gna). Take for example:

^:i !tara /crw !kseo The rat eats the cheese

``crw-eats'', called a ``predicate word'' (qury), is a symbol for the predicate by which the rat and the cheese are related. The predicate is like a function whose arguments are things that might be related; the value of the function is true or false (or fuzzily in between) depending on whether or not they actually are thus related: in this sentence, whether the first actual parameter eats the second.

The formal parameters of a predicate, regarded as a function, are referred to as ``cases''. English has ``nominative'' and ``accusative'' cases (the rat occupies the nominative case and the cheese occupies the accusative case), and Latin has in addition ``genitive'', ``ablative'' and others, but gua\spi simply numbers the cases. Some gua\spi words have as many as five numbered cases. In our example, ``tara-rat'' fills the first case of ``crw-eats'' and ``kseo-cheese'' fills the second. Natural languages and -gua!spi have obvious regularities in how particular cases are used, but it is not possible, at least in -gua!spi, to make a universal theory of what cases mean. Users should attend closely to case patterns in related words, but each category must be learned individually.

The words denoting the actual parameters of a predicate are called ``arguments''; being sub-phrases, they have their own predicate words. Here, ``!tara'' and ``!kseo'' are the arguments. The ``thing'' represented by an argument, which is the actual parameter of the sentence predicate, is something that can fill the first case of the argument's predicate. It is referred to as the ``referent'' of the argument. For example,

crw !kseo Eater of cheese

is an argument phrase; the first case is left open, and our rat (which we have seen previously in the first case of this relation) is a candidate for the argument's referent. Not every first case occupant is a referent of the argument. The rules for forming the referent subset are presented later.

While English partially segregates nouns and verbs, gua\spi uses the same predicate words (qury) in both argument and sentence phrases.

What Definitions Mean

In a dictionary words are defined in one or two sentences, but for gua\spi these sentences are considered to be merely a learning aid. The effective definition is a set of lists of thus-related referents. For example, the referent set of ``eats'' includes a member list with our example rat in first case and our example cheese in second, as well as numerous other members containing other rats, foods, and so on ad (almost literally) infinitum. Other predicates like ``cu-pair'' have referent sets that are actually infinite.

Language users are not expected to be familiar with every object list that was, is now or ever shall be thus related. A big part of language behavior consists of the listener adding to his knowledge of which items are thus related, which information the speaker sends to him or her. Each person has his own limited experience of the world, but we speak of ``the referent set'' of a word independent of a person because words are supposed to mean the same thing to each person, and language users really do agree on meanings most of the time.

Humans are very good at generalizing from a few referent set members so as to recognize novel referents, and they are not satisfied with a word until they can do such a general recognition algorithm and usually come out with the same answers their neighbors do. But mechanical users of gua\spi cannot be expected to show such skill, and neither can beginning human users such as infants. They must build up a referent set for a word by exhaustively hearing referent set members. If an advanced human, or advanced software, can transcend the official definition of gua\spi words, that's fine --- a common (but risky) strategy for humans will be to use their native language as a guide to gua\spi meanings. However, gua\spi words are still defined officially in terms of referent sets simply because this definition is known to be tractable both for theory and for practical implementation. A gua\spi referent set is perfectly suited to be represented as a Prolog database, if truncated to a practical size.

The Interpretation of Language Behavior

When you speak an argument in a nonsentence you call the listener's attention to its referents. For example,

^:i |va -jiw /vn -sper -jiolHey, a crocodile!

When you speak a sentence or a subordinate assertion you do the same thing: you call the listener's attention to the members of its referent set. (Thanks to John Parks-Clifford, editor of The Loglanist, for this insight [TL43].) Thus in:

^:i |qnu !qo -jan /tara /jun !kseo |zey !juJohn, the rat is after your cheese!

your knowledge of the referent set of ``jun-hunt'' includes a member which John will want to append to the ones he knows, before the cheese is stolen. This is the ultimate meaning of the gua\spi sentence.

The Organizational Syntax Level

Now that we have an unambiguous parse tree made up of phrases, what shall we do with it? Modern theories of parsing are very good at describing the transition from input tokens to a parse tree, but they leave the subsequent use of that structure pretty much to ad-hoc patchwork, and gua\spi is no better. However, the use of the parse tree can be divided profitably into two phases, organization and semantics. Semantics in gua\spi consists mainly of computing and updating referent sets, whereas organization refers to a collection of preparatory transformations including assigning sub-phrases to cases, handling imbedded sentences, replacing pronouns by their antecedents, and transforming compound words into sub-phrases. One way to look at the organizational level is that these surface structures of -gua!spi are transformed into a single deep structure, the predicate with its arguments, which is a uniform and simple interface into the semantic level.

Which Words Go in Which Cases

The tones of grammar deliver to the organizational syntax level, for each phrase, an ordered list of attached sub-phrases, which are the arguments of the phrase predicate. For example in ``!tara /crw !kseo'', ``tara-rat'' and ``kseo-cheese'' are attached to ``crw-eats'' as sub-phrases and therefore are its arguments. In the simplest and most common variation the arguments fill a sentence predicate's cases in order by number, much like English and Chinese, so ``tara-rat'' fills the first case of ``crw-eats'' and ``kseo-cheese'' fills the second. In arguments the first case is skipped over, being filled by an invisible placeholder for the referent. This organizational syntax can be so simple because the grammar delivers unambiguous lists of arguments, whereas in English or Latin a combined syntax has to deal with both getting the arguments on the right predicates and getting them into the right cases, and so is a lot more complicated.

The root phrase is assumed, in the absence of special cue words like ``vn'', to be a sentence; thus its first sub-phrase fills its first case. All sub-phrases are assumed to be arguments with empty first cases, except if they have tones or prefixed cue words that make them subordinate or infinitive clauses.

Should it be inconvenient to have cases filled in order, gua\spi has ways to change the order. First, certain prefixes signify that the relation word is ``converted'': one case is moved in front of the others. This is most useful for arguments, and lets one gua\spi word do the job of as many as five English words. For example in ``zu -crw'' the second case comes first, and the referent of such an argument would be something occupying the second case of ``eats'' before conversion: the meaning is ``food''. The original first case, the eater, comes afterward: ``zu -crw !xo -tara'' means ``rat food''. The English ``passive voice'' is a conversion in a sentence.

Second, an argument can be directed to a specific case by a ``caselink'' prefix analogous to the case endings of Latin. For example, take

^:i !tara /fer !se -dowu The rat carries (something) from the house

``fer'' means ``X1 carries X2 to X3 from X4 via X5''. Its arguments are ``!tara'' in the first case, but ``se'' links the next argument, ``dowu-house'', to the fourth case: the start point.

A predicate word can act as a caselink too, linking a ``modal case'' by means of an imbedded sentence. Such sentences are covered in the next two sections.

The motion words have complicated definitions, so all the definitions have been made similar: ``X1 (moves) to X2 from X3 via X4'' or ``X1 makes X2 (move) to X3 from X4 via X5''. Many other word categories have uniform definitions too.

Sentences as Arguments --- Infinitives

A gua\spi sentence or argument expresses a relation between specific referents, and this specific referent set member is called an ``event''. (Frequently the sentence represents several similar events.) It is common for several cases of the predicate to be vacant: in the previous example the thing carried, the destination and the route were not specified. Nonetheless there must have been a thing carried, a destination and a route, and the sentence asserts a relation between all five arguments. The next organizational elements we will look at are linking words that attach sentence predicates (with their arguments). The linked sentences represent lists of specific events with specific argument referents and with all cases filled even if not specified by words.

Returning to organization, the first sentence link word is ``vo'', which acts to change a sentence into a one-argument predicate, referred to as an ``infinitive'', which means that the occupant of its first case is an instance of the sentence relation. Though the infinitive can itself be a sentence predicate it is much more commonly used in arguments, like this:

^:i !tara /vyu !vo -crw !tara The rat likes for the rat to eat

``vyu'' means ``X1 enjoys doing (vo) X2'', where the second case is some kind of activity --- a natural place to fill with an infinitive. The sentence linked by ``vo'' is ``!tara /crw'' = ``the rat eats'', and an instance of that relation, an event, is the referent of the argument ``vo -crw !tara''. (The argument ``!tara'' may come before or after the sentence predicate ``crw'', wherever convenient.)

``vyu'' includes the prefix ``vo'' on its second case by default, as do all words which commonly have infinitive arguments. Also, there are various patterns in which main sentence arguments are replicated into infinitives, and by far the most common is for the argument just before the infinitive to be replicated, if the infinitive has none --- ``!tara'' here. So you could just say

^:i !tara /vyu !crw The rat likes to eat

Subordinate Clauses

A subordinate clause is a sentence within a sentence. Its predicate relates one (or more) of its internal arguments to the phrase it modifies; the internal argument is called a ``modal case'' of the modified phrase. A subordinate clause can specify a tense, location, possession (genitive case), listener (vocative case), speaker in dialogue, gender, plural number, repeated action, and numerous miscellaneous cases as in the examples below. Its purpose is signalled by a linking word:

A supplementary comment, giving additional information about the modified phrase, typically adding a modal case.
A restrictive clause, which events of the modified phrase must satisfy, or they are thrown out of the referent set.
A discursive comment, a helpful assertion by the speaker of the relation between the modified phrase and the previous sentence.
A supplementary assertion often stating the speaker's relation to the modified phrase.

Subordinate clauses are more common in gua\spi than in English, and also can be complicated, so two special rules are provided to make them simpler:

Here is a subordinate clause restricting an argument, illustrating ``vu'':

^:i !tara /crw !xi -kseo |vu -xel The rat eats smelly cheese

Not all cheeses but only those which smell (|xel) are eaten by the rat. The restricting sentence is ``X1 smells like X2'', and argument referents (cheeses) are automatically brought into X1 through the placeholder. When in English we use adjectives and adverbs, in gua\spi we usually use subordinate clauses like this one. Here are examples of the other subordinate linking words:

^:i !tara /crw !kseo ^ve -tum !vden !xgnoThe rat eats the cheese with its teeth
This subordinate clause adds a modal case. The clause is ``!vden !xdro !fu -tum !vo (event)'' = ``its teeth are the tool for doing (event)'', and the asserted relation ``!tara /crw !kseo'' = ``The rat eats the cheese'' also satisfies this sub-sentence. Because of the clause we know that the rat does not gum the cheese. The effect is as if an additional case were added to ``crw-eat'' for the cutting tool.
^:i |vi -bwy ^tara |va -cul !zu -crw /fi -go -crw !kseoBut the rat, which was full of food, didn't eat the cheese

``|vi -bwy'' appears to have no modal argument, but for this discursive category a pronoun is provided by default to represent the previous sentence. Thus the subordinate clause says ``this sentence differs from the previous one''. The other clause beginning with ``va'' is a subordinate assertion, which is similar to a main sentence, but the reader can understand it better when it is imbedded.

^:i |va -tan /pur -far !tara |zey !jiDamn, my rat ran away

In ``tan-annoy'' who is annoyed? ``ji-me'' is provided by default in the first case (after conversion) of supplementary assertions, main phrases and infinitives that otherwise lack one (provided certain conditions are met). Forceful or emotional speech seems more free and expressive with this feature. The other clause ``|zey !ji'' is a possessive phrase; most languages have special grammar just for possessives, but gua\spi uses the general clause mechanism.

Pronouns Represent Words, Not Things

The next organizational issue is the pronoun. English pronouns have referents just like any other argument. But gua\spi pronouns represent words, not the referent of words. In computer terms, they are like functions that are expanded in-line rather than being called. The represented words are called the ``antecedent'' of the pronoun, and the sentence is analysed as if each pronoun were taken out and replaced by its antecedent. The antecedents, not the pronouns, have referents. In this way the organizational syntax level can be kept free of meaning, and the semantic level has to deal with only one class of words, predicates.

For example, a document typically will have a signature line saying in effect ``this text is the output of Jim Carter''. (Spoken discourse is analogous though identification is by sight or voice tone.) Then when there appears the pronoun ``ji'' (``me'' in English) the effect is as if the words ``Jim Carter'' had been written in its place. That is, ``A rat ate my cheese'' and ``A rat ate Jim Carter's cheese'' mean exactly the same thing.

As illustrated below, various kinds of context are carried into a phrase by a pronoun-like mechanism. When an antecedent is replicated to replace a pronoun the context is replicated with it, so the antecedent will have the same referent in both places despite intervening context changes. And when the antecedent is copied any pronouns originally in it have already been replaced by their own antecedents.

Gua\spi includes question pronouns, phrase-relative pronouns, names and modal pronouns. For question pronouns the listener is supposed to say the antecedent; in other words, the speaker provides a sentence and the listener is to fill in the blanks. There are question pronouns corresponding to English ``who'', ``how'', ``how many'', ``which'' and ``isn't it''.

Phrase-relative pronouns are for copying neighboring phrases --- arguments or entire sentences. One of the more common phrase-relative pronouns is vgry, the whole phrase that the current listener last said, which typically is the question the speaker is filling blanks in.

In gua\spi a name is a pronoun. A name consists of a predicate prefixed by ``qu'', or ``qo'' for foreign names, which disconnects the usual meaning of the predicate and substitutes the pronoun behavior. People are assigned permanent names at birth through a performative (ritual) statement like this:

^:i !qo -ben /ga -zu-xim !jw |cilBen is (declared to be) the name of this child

From then on, ``!jw |cil'' (``this child'', with context so the listeners remember which one) is the antecedent of the name ``qo -ben''.

The six variables da, de, di, do, du, dy are names which you can assign to important concepts in nonfiction or characters in fiction. In mathematics it is also common to use letter words as pronouns for mathematical expressions.

Modal pronouns are like ``ji-me'' and ``jn-now''. A modal pronoun's antecedent is set by a modal phrase with a special prefix, saying to save the modal phrase on a kind of stack, separate for each modal predicate, from which it can be retrieved. The previous antecedent can be replaced, but of more interest, it can be saved and later restored.

The modal stack is used for more than modal pronouns, though. For each kind of modal case, e.g. tense or speaker, every sentence that lacks a modal phrase for that case gets the stacked phrase automatically. Here is an example of stacked speaker cases, in story dialog:

^:i |qe -jai !qo -kira /py /zu-zni !cyr -far !juSays Kira, ``Why do you flee?'' (default saved, set)
^:i -po -sfa -daw -can -siw -dan !ju``Don't you want to be rescued?'' (default inserted automatically)
^:i |va -pli ^vi -zu -gre ^jo /kuo !ji``Please, at least talk to me!'' (default inserted automatically)
^:i |qa -jai ^qo -kira /jun !suyKira pursues the swimmer. (prior default = narrator)

Kira said the first three sentences. In the first, ``|qe'' indicates that the current speaker, the narrator, should be saved while Kira speaks. The words ``|jai !qo -kira'' are added to the second and third sentences automatically. Finally, ``qa'' restores the narrator as speaker and his modal phrase is put on automatically.

Tenses are handled the same way. If you put ``|qe -cnu !X'' on the opening descriptive sentence (where X is an event identifying when the sentence happens) then it will be propagated to subsequent sentences automatically --- unlike in English where a syntactically complicated and less precise tense has to be used on every sentence. John Parks-Clifford, then with the Loglan Institute, originally developed this concept of tense defaults [TL43].

Compound Predicates

A key organizational element of gua\spi is the compound predicate, a sequence of words heading a phrase. The motivation to make compounds is threefold. First, you can use a single argument list to say what amounts to two sentences, which when compounded are much easier for the listener to interpret. Second, just as we use Latin prefixes in English to make many words from one, e.g. ``ob-ject'', ``pro-ject'', ``in-ject'', ``ab-ject'', most meanings in gua\spi are achieved by combining a much broader range of predicates. A beginner can learn the primitive words (qury), about 1400, and then stick them together in self-created compounds which he can expect any listener to understand, while to achieve the same range of expression in natural languages the speaker and the listener must master a huge vocabulary in which most of the words are rarely used. Third, the compound words are deconstructed into phrases headed by one qury word each, and semantically processing these phrases is much easier than in natural languages because there are so few qury that must be kept track of.

There are three main patterns to the compounds. First, if the main word has a case with a default linker of ``vo'' or ``bi'' --- that is, a case for an infinitive --- a word compounded with high even tone `-' is the predicate of that infinitive, and the main word case before the infinitive (before conversion) becomes the infinitive's first case. (Exceptions are noted in the dictionary.)

^:i !qo -kira /can -xna !fyniKira takes hold of the oar
canX1 changes so (vo) X2 becomes true
xnaX1 holds X2 with (body part) X3
!qo -kira /xna !fyniKira holds the oar (X2 of /can)

Second, the words may share an argument list. The effect is as if you had made two sentences with the arguments copied into each. This pattern is cued by tone `-' when the infinitive argument pattern does not apply, or by a conjunction ``-fe'' when it does.

^:i !do /suy -pne -qmy !kqua It swims down through the water
!do /suy !kqua + !do /pne !kqua + !do /qmy !kqua It swims to water; it penetrates that water; it is above that water.

A third pattern is found in which a transitive main word is followed by its object as a compound. It is cued by the tone `='.

^:i -spo !bri =kqua |bir ^dri =fliMaybe the pilot already drowned
bri X1 breathes X2
kqua X1 is a serving/portion of water
bri =kqua X1 drowns
dri X1 drives X2 to X3 . . . (transitive motion word)
fli X1 flies to X2 . . . (motion word)
dri =fli X1 pilots the flyer (airplane) to X2 . . .

Though humans like to think of compound predicates as separate words analogous to the primitive words, compounds are actually defined through these transformations, so that each primitive word heads a separate phrase. For example in the third type of compound the compounded object is to be taken off and put in its proper case as a sub-phrase. Thus one can easily and reliably interpret a compound word that one has never heard before, as long as one knows all the primitive words.

That is how gua\spi is organized. Let us now turn to the semantics of arguments.

Argument Semantics --- Referent Sets

As stated earlier, a predicate word expresses a relation between the occupants of its cases, and is defined by a referent set consisting of lists of case occupants that are thus related.

To interpret an argument, you start with its predicate's referent set. You retain members consistent with any sub-phrases. From each member you extract the first case occupant (after conversion), and out of these you make the ``full referent set'' of the argument. The ``referent subset'', which is the set of actual referents of the argument, is a subset of the full set which depends on a prefix word called an ``article'' (tirl). (More modern terminology might be ``determiner''.)

The most common article is ``xe'', and it is assumed with most predicates when arguments lack an article. Its English translation is ``the''. The referent subset is whichever members the speaker has in mind to talk about, but generally there are prior context cues to show which out of numerous possibilities are intended as the referents. In particular, if a set of referents has been designated before and if it is the only such set that is a subset of the full referent set of the argument, then those are the referents of the argument. For example,

^:i !xo -fkar |xda ^vu -xge /fi -can -tai !qel =fkarAn old, black car emerged from its garage.
^:i !fkar /cyr -vleThe car turned left.

``fkar-car'' appears three times. The first instance designates one referent in detail using ``xo'', described below. The other two instances are typical arguments with ``xe-the'', but the article is unseen, being provided by default. Since the prior referent fits this predicate (and in the second sentence ``its garage'' does not), the prior referent is being redesignated. Because gua\spi words are so short it is just as efficient to redesignate an argument like this as to use a phrase-relative pronoun, so pronouns are less commonly used in gua\spi than in English.

There are also articles that select the entire referent set, ``typical'' members of it, and no members of it (actually making a negative statement about all members). Another important article is ``xo'': From the full referent set one or more members are selected, and it doesn't matter which ones. For example,

^:i |vi -pli ^jo ^sa -ji /gey !xo -kliwPlease give me some nails

All in the box are equivalent and it doesn't matter which you get. ``xo'' is often used for arguments in the ``serving or portion'' category, called ``partitive nouns'' in English.

There are two articles for each meaning; the first unfolds the referent subset so each member is a referent, while the second specifies that the referent is the referent subset as a set. The careful distinction between sets and extensions of their members is characteristic of gua\spi.

Most Indo-European languages distinguish between genders and numbers of arguments. Like Chinese and English, gua\spi has no gender, though you may use a subordinate clause like ``|fmy-female''. Number comes from the referent sets, not the grammar. You may specify the exact number of referents with a numeric predicate, like this:

^:i |vi -pli ^jo ^sa -ji /gey !xo -beol |fmy ^vu -zu -cuPlease give me two nuts (female screws)


A great deal of the machinery of language, which in natural languages is shared between the grammar and the vocabulary, is handled in gua\spi purely by words. Here is a discussion of how the words were created, and four samples from a long list of models of how to say things. Frequently I have thought that some form or meaning required a new primitive word, or even a change in the grammar, but it has turned out that existing words were more than adequate if creatively used.

Word Creation

The words of natural languages appear to be arbitrary symbol strings of tremendous variety of sound. Gua\spi is similar in that its words were generated by a partially random process. However, the words were made to resemble natural language words so they would sound more pleasing and so occasional cognate relations might aid learning. To begin, the word lists of Loglan [L4] and Lojban [Lja] were merged and some additional words were added. An English, Chinese and Latin translation was determined for almost all words. (Latinoid English words were avoided.) Both Loglan and Lojban use many more natural languages as word creation fodder.

Then experimental phonetic data [NB2] was used to assess candidate words for the accuracy with which speakers could distinguish them. For each gua\spi meaning, randomly generated word candidates were evaluated for recognizability, for distance from other gua\spi words, and also for similarity to their natural language equivalents. The final assignments were determined through a process of numerical annealing so as to maximize the summed quality scores.

CV structure words were assigned by hand; related structure words, like articles, have the same C and varying V's. Structure words pertaining to numbered cases have the same V's as the corresponding digits, but contrasting consonants, making learning easier.

A question often asked is, why create new words? Why not use Chinese or English words? First, some attempt has been made to keep gua\spi culturally neutral, and if Chinese words were used it would intimidate English speakers and vice versa. More important, Chinese words are designed for use with Chinese. Many required meanings, like articles, simply do not exist in Chinese, and those meanings that are present are only approximations of the gua\spi meanings. That is why the approach was rejected of simply stealing natural language vocabulary.

How adequate is the word list? Can every required meaning be expressed by infinitive, parallel and transitive compounds? Only extensive literature in gua\spi by a variety of authors can demonstrate adequacy. However, I have written about 20,000 words of prose and fiction in Loglan, and I am satisfied with the coverage of the Loglan word list, especially with the Lojban additions and with my own. Some people are interested to discover just how few basis words we can get by with. While I do not believe that the Lojban word list is minimal, I think it is fairly close. Thus I chose to use existing word lists for gua\spi rather than to try for radical pruning or de novo creation.

Modal Cases

Here are examples of a few modal cases. However, virtually any word can be construed as a modal operator. Be alert for creative opportunities for expression.

birPast tense
^:i !ji /crw !kseo ^vu -bir !jun -vnl !taraI ate the cheese before the rat came hunting
zeyGenitive or possessive case: a relation of pertinence
-fkar |zey !jiMy car (which I lease and which my brother is now driving)
jaiThe speaker
^:i |jai !qo -kira /ju /dwu -csn -zu -jeuSaid Kira, ``You're a monster''
plmSuch as: an example
-xy -pso |psi ^vu -plm !xy -kai |keiBad people such as thieves

Varieties of Negation

In gua\spi negation is not a unitary concept; beside the obvious antonyms there are nine or ten ways to express negative meanings, most of which involve compound words. Here are a few examples.

^:i -go !ji /kio !tara |zey !ju
I don't have your rat. ``go'' is a mood prefix which means that the asserted sentence is counter to fact.
^:i !jw |kseo /fi -gl -zao
This cheese is flavorless. Some dimensions like ``zao-flavor'' are quantifiable (more or less) but unsigned, so their degree ranges from zero to larger values. Others like ``gal-high'' are signed. In either case ``gl'' modifies a predicate so that its degree is zero or negligible.
^:i !jw |kseo /fi -gr -ksi
This cheese is not fresh. When the dimension ranges from positive to negative values, ``gr'' interchanges positive and negative. For extremes of unfreshness one can use ``fpu-rotten''.
^:i !jw |kseo /fi -vry -can -psl
This cheese is desolidifying. ``vry-reverse'' indicates that the process in its X2 case is occuring in the reverse of the usual order.

Causal Connectives

The root structure of syntax is a discourse, or sequence of sentences. But the sentences need not stand alone; they may be connected by predicates, like this:

^:i -dae !kara ^:o -bal !crw |jro ^tara ^kseoIf the box is open then maybe the rat will eat the cheese.

The speaker may connect sentences with any useful word having suitable cases, such as ``kau-cause'', ``kmo-motivate'' or ``sny-imply''. Like all gua\spi words, the causal connectives can also be useful as arguments and as modal caselinks. In this example ``^:o'' is a ``retroactive downjump'', a special case in organization. A sentence start word, it transforms itself into a pronoun for the previous sentence, which goes into the first case of the following main word, the causal connective in these examples. Human speakers prefer infix causal connectives with a retroactive downjump rather than the obvious form with two explicit infinitives.

Mathematical Expressions

Any discussion sooner or later involves quantitative statements with units of measure. Therefore gua\spi has quite an extensive facility for mathematical expressions, even if the more complicated possibilities are rarely used by non-scientists. First, gua\spi syntax matches perfectly the definition of a ``number'' as an equivalence class of equal-count sets. This concept can be generalized to various extension rings and fields.

!xu -cu -cw -ciThe number 2.5 (the class of all sets of ``count'' 2.5)

Mathematical functions are defined with such classes as formal parameters, and hence have ``xu'' on parameter cases by default --- ``xu'' means the entire referent set of an argument, as a set (or class). The first case of a function is its value, and the function is defined as ``X1 is in the equivalence class that comes from doing (function) on (xu) X2'', possibly with several parameters. For example,

^:i !xa -ca /plw !co ^cu3 is the sum of 1 and 2 (every triplet is in the equivalence class of 1+2)

Specifically, units of measure are defined to multiply a number or other expression by the unit. The resulting equivalence class is considered to contain gua\spi events whose degree or measure are that big; hence the unit expression takes the form of a subordinate clause, and the main sentence predicate tells what dimension is being measured. For example,

^:i !ji /vga |kyam !ku -cyI weigh 70 kilos (I heavy kilo 7 0)

Sample Text

Here is a short passage from a story I am currently translating from Loglan to gua\spi. If a second human knew gua\spi I would have him or her translate it to English, but so far my only colleague is mechanical. Its purpose is to check syntax and organization, and its English is only good enough to substitute for a full parse tree printout. For example, it can't tell an infinitive from a gerund, and it is overly free with possessives. Nonetheless you can get an idea how much meaning a mechanical translator can recover from the text.

Brackets `[ ]' surround sentence-type phrases; angle brackets `<>' mark subordinate clauses; backslashes `\\' repeat the predicate of the phrase a subordinate clause restricts; parentheses `()' enclose the antecedent of any other pronoun; and when a word's meaning as an argument differs from its root meaning, the root comes afterward in slashes `//'. Subscripts give the case number of each argument.

^:a |vi-gza ^vu-qe-kam !sa-cil ^qu -jaiw=tiri |va-ga-xim !do /fi-jiw !su-cana-fer ^vi-gau ¶ Tigereye cries to the children, the barge, look out!
[then the boat1 and2 carry2 surprise < paragraph \surprise1\> < speaker/cry/ tiger's2 eye1 < performative name \eye1\ variable b2> \surprise2\> < listener/warn/ you2 (child) \surprise2\> < time/present/ (something2) \surprise1\>]

``Carry-boat'' is used for ``barge''. ``qe'' sets the speaker and listener modal case antecedents. However, ``gau-warn'' supercedes ``kam-cry'' as the predicate bearing the listener. ``do-variable~b'' is assigned to represent Tigereye. Words for document structure like ``gza-paragraph'' fit naturally into the text, written or spoken.

^:i |faw ^jo /suy !jrSwim over here!
[imperative1 (child) swim place2 (something) < listener/emphatic/ you3\,(child) \swim2\> < speaker/emphatic/ I1 (eye) \swim2\> < time/present/ (something2) \swim1\>]

``jo'' instead of ``ju-you'' makes the sentence imperative. ``faw-emphatic'' has a default ``ji-me'' in the case for its speaker and ``ju-you'' in the case for its listener, thus superceding the default kam otherwise imported from the first sentence.

^:i |qi-qnu !qo-qosefo /ve-faw ^jo /qma -duw !gunu !juJosepho, move your ass!
[imperative1 (qo se fo) make to2 [your2 (qo se fo's) buttock1 move] < listener/emphatic/ you3 (qo se fo) \make2\> < speaker/emphatic/ I1 (eye) \make2\> < time/present/ (something2) \make1\>]

Only a few words like ``fer-carry'' are intrinsically transitive. Normally an infinitive compound with ``qma-make'' makes words transitive.

^:i -po !ju /daw -scu-zu-crw-zu-ter !fwa-pei !cana-ferDo you want to be chewed up by the barge's propeller?
[you1 (qo se fo) is it desire to2 [to1 [[(qo se fo2) food1/eat/ by to's2 [push boat1 and2 carry2] device1] and [(device1) tear2 (qo se fo2)] complete] < speaker/cry/ (eye1) \desire2\> < experiencer/attention/ (qo se fo1) \desire2\> < time/present/ (something2) \desire1\>]]

A complicated sentence, just as efficient as English. See how Josepho is replicated into the internal infinitives. ``Push-device'' is used for ``propeller''; ``completely eat-tear'' is ``chew up''. The speaker and time are added automatically.

^:i |qi-koy !do /va-gri ^ji /gul !qou !jw ^vu-cnu !xa-jlDamn, I have to watch them every moment!
[I1 (eye) must to2 [(eye1) watch2 this2 < time/present/ all something2 \watch1\>] < performer/think/ variable b1 (eye) \must2\> < listener/think/ (eye3) \must2\> < actor/angry/ (eye2) I1 (eye) \must3\> < time/present/ (something2) \must1\>]

A subordinate clause gives, literally, ``at all times''.

^:i !pu /kau !qai-kar !jw ^jw /vi-fawWhy can't they take care of themselves?
[what1 cause to2 [this1 fail to2 [(this1) care2 this2]] < listener/emphatic/ you3 (eye) \cause2\> < speaker/emphatic/ I1 (eye) \cause2\> < time/present/ (something2) \cause1\>]

For a reflexive, repeat the argument.

^:i !ji /suy |swa !dman =co I swim for one second...
[I1 (eye) swim < duration all* one's2 second1 \swim2\> < performer/think/ variable b1 (eye) \swim2\> < listener/think/ (eye3) (swim2)> < time/present/ (something2) \swim1\>]

Here is another tense, this time a continuous one. ``One second'' is a simple mathematical expression.

^:o -sno !can-zu-vem !jwAnd that's enough for them to get into trouble.
[conjunction \swim1\ sufficient to2 [to2 [(this2) trouble] change by this1] < performer/think/ variable b1 (eye) (sufficient2)> < listener/think/ (eye3) (sufficient2)> < time/present/ (something2) \sufficient1\>]

Gua\spi is much richer in causal connectives than English with its ambiguous ``because''; thus this gua\spi sentence is half the length of its English translation.

This short sample exercises almost every feature of gua\spi\, even including a mathematical expression. To verify every phrase the reader must know gua\spi fairly well, but one can see easily the simple phrase organization of gua\spi. The lengths of the gua\spi and native English sentences are comparable, showing how efficient gua\spi is. The primitive words cover almost all meanings in this relatively unspecialized text, and the componds for ``barge'' and ``propeller'' are quite understandable. Finally, the sample makes it clear that gua\spi is more than just a dry substitute for SQL; gua\spi can support real life.


What can one use gua\spi for? Here is a brief list:

I hope this brief introduction to gua\spi has whetted your appetite to learn more about it. As you have seen, it expresses typical human sentences easily and efficiently. But the meanings of the words, and particularly the meanings of the phrases and sentences made from them, are defined much more specifically and clearly than in even the best natural languages. Finally, and most significant for artifical intelligences, the resulting meanings are cast in a form that is ideal for modern fifth-generation languages --- which, in fact, those languages were designed to represent. Thus the gap between human and machine languages is closed by gua\spi.

Appendix: Gua!spi Grammar in Backus-Naur Form

``Discourse'' is the root grameme. Grammar for quoted non-gua\spi text is not shown, but foreign predicates and quoted gua\spi are processed by this grammar and are recognized at the organizational syntax level. A procedural definition shows the simplicity of the grammar more clearly.
; Morphology.
C = (choice of letters)
Cseq = (Cseq C) | C
V = (choice of letters)
Vseq = (Vseq V) | V
Word = Cseq Vseq
; Tone categories.
Compound = `-' | `='
Sametone = `^'
Down1 = `!' | `|'
Up1 = `/'
; Grammar. LHS `-' symbol indicates which end has a tone.
Prefix = (subset of Word, e.g. ``vo'' or ``zu'')
Primitive = (subset of Word, e.g. ``tara'' or ``crw'')
Phrase = Prefix Args0 Phrase
| Phrase-w
Phrase-w = Primitive Phrase-w
| Primitive
Phrase0- = Phrase Sametone
| Phrase Down1 Args1
; Args(n) is a list of phrases that jumps up n levels at the end. Args3, 4, . . . are defined similar to Args1 and 2. Some finite bound must be set on n to give a finite grammar.
-Args0- = Compound (Just one tone)
| Down1 Args1
Args1- = Phrase Up1
| Phrase Down1 Args2
| Phrase0 Args1
Args2- = Phrase `!' `fu'
| Phrase Down1 Args3
| Phrase0 Args2
; These add the effect of ``fi''.
Afterargs = Phrase0 Afterargs
| Phrase Down1 Afterargs
| Phrase
After1 | After1 Down1 `fi' `-' After1
| After1 `-' `fi' Down1 After1
| Afterargs
-Preargs- = Down1 After1 `/fi' Preargs
| Down1 After1 Down1 `fi' `/' Args1
| Args0
-Sentstart = `^' (choice of words such as ``:i'')
Sentend = Phrase Down1 After1
| Phrase
-Sentence = Sentstart Preargs Sentend
-Nonsentence= Sentstart Down1 Sentend
-Unit = Sentence | Nonsentence
-Discourse = (Discourse Unit) | Unit