delayed response to jimc on indicators (long)

[cbmvax!snark.thyrsus.com!lojbab]

Much delayed comments on jimc's 06/08/90 discussion of indicators, 'modals',
and 'freemods'.  If you don't still have this message, don't worry - I'll
cover carter's main points in responding.

1. First a note of terminology - always a problem with jimc's comments
what jimc calls 'modals', we call "sumti tcita", the tags found on sumti that
are not part of the place structure.  The problem with calling them modals, is
that the latter term has too many different meanings - etymological it is
derived from 'setting the mood' of the sentence.  Most frequently, the term is
applied to adverbs like 'probably' that modify a sentence from a pure claim
into something else.  Most often, such mood-modifying in Lojban is done using
attitudinals and discursives.
     Now it turns out that I use the term 'modal' erroneously myself.  I call
these sumti tcita 'modals' when they are used in the tense position to modify
the selbri.  They really are 'aspectuals', there, and maybe I'll try to use
that word instead in the future.  Begone, 'modal'!

2. jimc argues that we use only a limited set of these 'modals', when
theoretically any gismu should be usable to express such an additional
relationship.  He further argues that the grammar of adding in these extra
sumti should be identical to the regular sentence grammar, i.e. like a bridi.
     He's half right, and we already do.  Not only any gismu, but any selbri,
can be used to 'define' a sumti tcita relationship, using constructs based on
lexeme FIhO.  The grammar within a FIhO construct is that of a description
(i.e. a complete selbri).  The reason is that a sumti tcita describes the
relationship of ONLY ONE sumti to the rest of the bridi.  A bridi describes
the relationship among several sumti, and you would have no way of knowing
which one was the 'tagged' one.
     However, FIhO constructs are complicated, because selbri can be
complicated; in everyday conversation it would be easy to come up with a
perfectly unambiguous sentence that was just too complicated for someone to
figure out.  We defined the set of lexeme BAI tags to express the most common
relationships as 'abbreviations'.  The only grammar appurtenant to lexeme BAI
is a single pseudo-conversion, using lexeme SE, that reminds the speaker and
listener which place of the gismu that was the source of the lexeme BAI member
is relevant.  This is essential for some. e.g. "ri'a" accesses the first place
of "rinka", a cause; "seri'a" accesses the second place of "rinka", an effect.
     A secondary advantage of lexeme BAI is that they look and act like
natural language prepositions in one of their uses, as 'sumti tcita' on
trailing sumti; they aren't prepositions, though, since they have two other
major uses that are not similar to prepositions:
     a. Tagged or untagged sumti can be used to modify another sumti as a
restrictive or non-restrictive relative phrase.  This is how possessives and
appositives (alternate identities) are expressed, as well as comparatives.  In
Lojban, we prefer to say the equivalent of
   (x, more than y) likes z      or      x likes (z, more than y)

whereas a 'prepositional' tagged sumti use would be the same as English:
                            x likes z more than y
which in Lojban is interpreted as
                           (x likes z), more than y

which is strange when y is a name.
     b. Sumti tags can be used as aspects in tense position before the selbri.
In this case, there is no sumti (it must be elliptically inferred), and the
lexeme BAI words resemble most closely English adverbs.

3. The term 'subordinate clause' isn't too useful in Lojban either.
Effectively it means any bridi-like construct within a sentence other than the
main bridi.  jimc may not realize how many possible ways there are to inject a
subordinate bridi into a Lojban sentence.
     First not that there are two forms for subordinate bridi.  In
abstractions, the subordinate bridi is truly identical in structure to a main
bridi.  As a result, you frequently have to append the terminator "kei" to
keep an abstraction from absorbing any following sumti on the main bridi.
Many subordinate bridi use the grammar of specified descriptions.  You append
trailing sumti onto the selbri of the clause using "be" and "bei".  Only in
very complex situations do you need to close these clauses with a terminator;
the rest of the time the terminator is elided.
     Thus we can use relative clauses (a kind of subordinate bridi using
standard bridi grammar) to modify a sumti restrictively or non-restrictively.
There are also the FIhO constructs mentioned above, which use be/bei syntax.
description sumti and tanru can be composed of pieces which are subordinate
bridi (the equivalent of "the faster-than-light spaceship").

4. jimc claims that indicators should be expressed as 'subordinate clauses' in
Lojban.  Again, he is half-right; they can be.  But I'm not sure that even
then they are quite the same thing, pragmatically, when this is done.
     The more serious problem is that jimc doesn't see the difference between
"Damn! I dropped it." and "I am angry that I dropped it."  Say the two
sentences the way you might in the appropriate context, and you'll probably
FEEL the difference.  The first EXPRESSES an emotion; the second TALKS ABOUT
an emotion.  More logically, in the first sentence, the main claim is that of
'dropping'.  In the second, the main claim is that of 'anger'.  They say
different things, and have different logical truth conditions.
     jimc had this misunderstanding several years ago, when he debated the
issue with pc in the early 1980's.  pc said jimc was wrong then and he still
is.

5.  However, jimc CAN express his emotions as logical claims if he wishes.
We've added metalinguistic bridi to Lojban, allowing the speaker to comment on
the main claim of the sentence at a 'different level' without modifying that
claim.  A metalinguistic bridi is a 'subordinate bridi' like jimc requested.
Metalinguistic bridi are marked with "sei", and are "free modifiers" that can
be placed after most gramamtical units.  These clauses have the be/bei grammar
for trailing sumti and seldom require a terminator (se'u), but can also have
any number of sumti before the selbri.  Using this construct changes "Damn! I
dropped it." into "(I'm angry about this!) I dropped it.", where "this" refers
to the main sentence state (in Lojban "la'e dei") and is usually ellipsized
("(I'm angry!) I dropped it."
     But I wouldn't express MY emotions this way.  (I might use metalinguistic
bridi to talk about someone else's feelings though - "(John's angry) He
dropped it.")  For indicators to serve their purpose, they must be spontaneous
and instinctive, not thought out.  They must minimal grammar and interaction
with anything else being spoken.  Sort of talking out of the other side of
your brain.
     Every natural language has some small number of indicators, though none
that I know of have the exhaustive set that Lojban has.  Animals also have
them, too; indicators may be the more basic form of 'language'.  But they
cannot express anything complicated.
     Emotions are NOT logical, and pragmatics requires that we have a way to
express them.  We cannot hobble Lojban by forcing that which is beyond logic
into a logical mold.
     Note that indicators are more free in grammar than 'free modifiers',
which metalinguistic bridi are.  The latter can occur after a couple dozen
'grammatical constructs' that comprise a Lojban sentence.  Indicators and
discursives (lexeme UI) can occur after any WORD, except those of lexeme LA
and DOI, and they cannot occur between a non-indicator word and its trailing
"nai" (in any of these cases, an ambiguity would result - in the former, the
indicator would be assumed part of the name that follows).

6.  A more general disagreement between jimc and Lojban's design shows up
several places in jim's essay.  He favors only one grammatical form supporting
any given type of construct.  We do not.  To do so requires that you impose a
total and rigid classificational scheme covering all constructs within things
that are sayable.  His approach fails if his classificational scheme is wrong,
i.e. two constructs are grouped as being 'the same' when in fact they are
logically different in language.
     I don't think we know enough about language to define an ironclad
classification scheme.  Furthermore, to impose such a scheme is to impose an
unnecessary metaphysical (and possibly cultural) bias on the language.  In
Lojban, we want only to havve the constraint that it be easier to talk using
logical constructs than in natural languages, and we've carrried this to the
extreme of making it easier to talk using logical constructs than non-logical
constructs.
     With regard to jimc's concept, we've taken the exact opposite tack.  If
there are multiple grammatical ways to express something in different natural
languages, the subject to maintain Lojban's unambiguity, we try to emulate ALL
of them (so as not to favor any one).  There is a tradeoff here between logic
and pragmatics; we've favored pragmatics where a construct does not violate
the logical basis of the language.  If there are contradictory forms in
different languages, one form has to be marked differently than the other, and
we've generally chosen to make the simpler metaphysical assumption the less
marked form.  This has led Lojban to a simple elegance, and we've had several
serendipitous discoveries wherein a more complicated construct falls apart
into 1 or 2 simple constructs.
     The prime virtue in simplicity is easy learnability.  But if we simplify
the grammar to an extreme, we make complicated ideas hard to say.  This is a
somewhat different constraint, in terms of a Sapir-Whorf constraint, than
Lojban's.
     A brief example, then I'll end this:  By jimc's argument, a direct quote
should only be expressible in a single chunk:  "... .i ... .i ..." with "he
said" stuck on the front or back of it.  But EVERY natural language allows for
direct quotations to be broken up, even if the two resulting pieces are
ungrammatical by themselves:  ("That", he pointed, "is the open window.") is
much more effective at conveying the integrated speech and action than (He
pointed to the window.  He said, "That is the open window.")

     I appreciate jimc's comments, which in some cases are apropo and lead to
changes (we adopted his suggestion of a 'neutral scalar' for negation, for
example), but overall jimc seems to have different priorities and goals than
we do for Lojban.