What is -gua!spi

Keith Ericsson asked me to put together a short description of -gua!spi. In
writing about 20,000 words of Old Loglan prose, I validated the strengths of
Loglan but also encountered a number of severe useability problems (many of
which have been addressed in Lojban).  The goal of Lojban, gloriously
accomplished, has been to finish Loglan and get it launched.  But I felt
that the next step in evolution should be taken immediately.  Here is a
technical description of -gua!spi and a small sample of machine-translated
text.  

--Design Summary: Morphology.  There are 11 V phonemes: aeiouylmnrw (y as in
knIt, w as in stroNG).  There are 14 regular C's: bcdfgjkpqstvxz (c as in
CHew, q as in SHoe, x as in breZHnev, no h).  A word (including foreign
words) is one or more C's and one or more V's.  Each word also has a tone in
the Chinese manner, which cues grammatical structure.  How simple this is! 
One of the British logli, long ago, asked why we needed CCVCV words when we
had affixes, and I took his suggestion to heart.  

--Grammar.  There is only one kind of phrase consisting of prefixes
(e.g. conversions), pre-subphrases (usually arguments), the predicate
(possibly compound, i.e. more than one word), and post-subphrases.  The
various roles of these phrases are not distinguished by the grammar -- a big
simplification.  Rather, the phrases are classified in a subsequent
organization step.  Falling tone (Mandarin #4, written !) cues the start of
a subphrase; rising tone (#2, /) returns from it; and high-even tone (#1, -)
connects words at the same level.  Up-down (^) starts another subphrase at
the same level; down-up (#3, |) marks a subordinate clause; and low-even (=)
is for a transitive argument.  Multi-level upjumps are possible but are
needed rarely.  A non-extensive experiment shows that naive non-Chinese
listeners can hear the tones.  

--Organization.  Caselinks and conversions are removed and arguments are
assigned to cases (X1, X2...)  Infinitives (X1 is an event of [sentence])
start out either with the infinitive marker -vo as a prefix or as a compound
word (i.e. -can-xyn = changes to be inside = enters); the subsentence is
pulled out as a separate subphrase and -vo becomes the main phrase
predicate.  Extensive defaults are provided, of which -vo on X2 of -can is
an obvious example.  The definition of -can says that its X1 is replicated
as X1 of its infinitive X2 so the actor which changes is also the thing that
is inside.  (Loglan never had this, and Lojban could use it.)  There are
similar services for parallel compounds like -bil-fli = fly from below =
ascend.  The extensive defaults and transformations make -gua!spi quite a
bit more compact than Loglan; experience shows that they are almost always
needed, but when not, they can be evaded easily.  

Any predicate in a subordinate clause is acting as a modal operator. 
Extending pc's compound tense idea, there is a stack for every modal case;
dialog and relative tenses are handled this way.  The stack holds the
antecedents of modal (personal) pronouns.  There is a complete set of
phrase-relative pronouns but they are used less than in Loglan because, at
the semantic level, an argument with -xe (the) and one predicate can
reliably reach a prior argument using that predicate.  Names, including
variables, are recognized during organization and are processed like
pronouns.  

--Semantics.  As in Loglan, predicates represent relations, that is,
lists of thus-related objects (e.g. for "eat", my rat plus my cheese, and
numerous other pairs of eater and food).  Following pc's lead, I define a
sentence to call the listener's attention to certain members of the main
predicate's referent set, designated by the arguments.  

--Vocabulary.  The word list of Lojban has been swallowed almost in toto,
and I have extended the scientific vocabulary in mathematics, chemistry,
zoology, botany and agriculture.  All your favorites are here: verbs of
motion, causal connectives, directional properties, and so on.  Nonetheless
there are still only 1400 words, not counting structure words and letterals.
I didn't get much out of the relation of Loglan words to natural language,
and I had neither the skill nor the time to locate equivalents in eight or
so languages (which the Lojban team did).  Hence as word creation fodder I
used only English, Chinese (Mandarin, from a native speaker) and Latin (as a
surrogate for European languages).  A trial with random assignments was
unacceptably ugly.  

In Loglan and Lojban there are many compound words with idiosyncratic
definitions, and (most) multi-word predicates are interpreted through
metaphor.  In -gua!spi there are no special definitions for compound words;
they are interpreted as multi-word predicates, and they are then transformed
to infinitives, parallels and transitives by specific rules, and are then
interpreted by the definitions of the individual primitives.  Metaphor is
available but is much less prominent than in Loglan.  

For my first exercise in Loglan I translated a technical article, and I
promptly and painfully felt the lack of mathematical expressions by which to
express dimensioned quantities like "125 meters per second".  In -gua!spi
numbers are classes of equal-count sets, and therefore, with no special
extensions whatever, the grammar supports mathematical expressions with both
Polish and infix notation in the guise of arguments that have arguments (all
post-arguments or mixed pre and post).  

--Progress and Plans.  The most often asked question about -gua!spi is, can
a language simplified to this degree actually support live human language
behavior?  Some time ago I wrote a substantial story in Loglan to
investigate this question, and so far I have translated about 3300 words of
it to -gua!spi.  My correspondents suggest that I have a naive user
translate it to English and see if the meaning comes through, but as only
one user is available so far, this is not yet practical.

However, I have nearly completed a computer program to parse and organize
-gua!spi text: everything above except semantics (which, yes, is planned
eventually).  Output can be a parse tree, or pseudo-English.  A prime design
goal of -gua!spi was that such machine analysis should be feasible, and this
discipline has powerfully shaped the language.  Here is a brief sample of
the program's output, illustrating how much of the meaning can be recovered
and how lively the story remains even after this kind of abuse.  Digits are
case numbers; \\ enclose the predicate of the phrase that a subphrase
modifies; these phrases are in <>; and subsentences are in [].  Pronoun
antecedents are in ().  

{^:i |qe-jai !qo -:mei-ci /vu-qi-qnu !qu-zgiw=xgri /vu-qnu !qu-zgiw=xgri 
/vi-faw ^jw /pei-qmy !ji ^pqua}
{ Mei Chi exclaims: Tigereye, Tigereye, he dunked me! }
{ [this1 push1 i (:mei ci2) some water3] and [(:mei ci1) above2 (water2)]
  <speaker/say/ :mei ci1 \and2\> <listener/emphatic/ you (eye3) \and2\>} %%
{ "Dunk" is best translated "push from above" = -pei-qmy, which is 
  transformed to two sentences in parallel.  English here has the advantage
  of a massive vocabulary of rarely used, specialized words.  "He" (sexist)
  comes out as "this".  Sorry, the program doesn't map "I" to "me".}

\x{^:i |vi-faw ^jo /pny !jw ^zgly }
{ Punish him for that! }{}
{ [imperative (eye1) punish <listener/emphatic/ you (eye3) \punish2\> this2
  (and3) <speaker/say/ :mei ci1 (punish2)>]} %%

\x{^:i |ql ^ju /pny !jw ^vu-gul !jo}
{ Tigereye: You have to punish him yourself. }{}
{ [you (:mei ci1) punish this2 <actor/must/ imperative (:mei ci1) \punish2\>
  <experiencer/attention/ (:mei ci1) (punish2)> <speaker/say/ (eye1) 
  (punish2)>]} %%
{ |ql interchanges speaker and listener modal stack antecedents.  Stack modal
  cases are inserted on each sentence automatically. }

\x{^:i !jo /kuo |fw-xgri ^jw}
{ Nag him. }{}
{ [imperative (:mei ci1) talk <tiger \talk1\> this2 <experiencer/attention/
  (:mei ci1) (talk2)> <speaker/say/ (eye1) (talk2)>]} %%
{ "Tiger talk" is a metaphor for "nag", inspired by Chinese. }

\x{^:e !jo /pei-qmy !jw |va-din /pqua }
{ (Or) dunk him back. }{}
{ and [imperative (:mei ci1) push1 this2 <exchange \this1\> some water3]
  and [(this1) above2 (water2)] <experiencer/attention/ (:mei ci1) (and2)>
  <speaker/say/ (eye1) (and2)>} %%

\x{^:i !jo /voi-zre !ji ^fyo !ju |zu-vem }
{ Don't ask me to fix you up when someone makes trouble for you. }
{ [imperative (:mei ci1) avoid to2 [(:mei ci1) request2 i (eye2) to3 [(eye1)
  repair2 :mei ci2 <victim/trouble/ \:mei ci2\>]] <experiencer/attention/
  (:mei ci1) (avoid2)> <speaker/say/ (eye1) (avoid2)>]} %%
{ Extensive reliance on defaults and replicated tenses to make three nested
  infinitives and one subordinate clause mean something.  Here -gua!spi
  clearly is more compact than English. }

\x{^:i |qa-jai ^qo-:mei-ci /gr-pli !do ^bwu !xgno}
{ Narrator: Mei Chi is not pleased how he helped her. } {}
{ [opposite :mei ci1 please variable~b (eye2) to3 [(eye1) help2
  (:mei ci2)] <author/fiction/ kar tr jym3 (please1)>]} %%
{ The narrator, Kartr Jym, sat on the stack all this time.  !do, a "variable"
  used as a name, was earlier assigned to Tigereye. }