Return-Path: <lojbab>
Message-Id: <m0jkYqW-0001fNC@snark.thyrsus.com>
Date: Tue, 4 Jun 91 06:42 EDT
From: lojbab (Bob LeChevalier)
To: lojban-list
Subject: response to nick on PU/BAI
Status: RO
X-From-Space-Date: Tue Jun  4 06:43:49 1991
X-From-Space-Address: lojbab

Nick (cbmvax!uunet!ee.mu.OZ.AU!nsn) writes:

>the tense case tags, pu and ba, are used analogously to BAI. Then I would
>interpret

... [discussion showing that they are in fact used oppositely to BAI in
     semantic interpretation relative to the associated gismu]

>which I think is alien to lojban, where, it seems to me, the semantics
>of a case tag has some independence from the main bridi.  And yet in the
>lojban I've seen to date, the NL way of timing is used, and the case tag
>is used as an offset of the main bridi from the abstracted subordinate
>bridi:  before I die, I am born:  "I am born" is previous (before) to "I
>die".
>
>Should we keep it like this?  I don't know any Chinese, but isn't that
>language taken with 'main claims' rather than baroque links, and
>wouldn't the Chinese (and the lojbanis) consider it more logical to
>offset things like I said?

History lesson time, again!  These discussion are useful, in that in
some cases they are the first time I've put in writing the rationale for
some of our early decisions.  Indeed, this discussion will explain
another inconsistency we've been called on, that of "ci'a".  So please
bear with me for the sake of Posterity; and it might even be relevant to
the discussion.  I will use "lexeme" instead of "selma'o" since it is
the historical term, and it is hard to write this without it.

You point out an inconsistency that is historical to Loglan.  I will
briefly say that logic is here defeated by history - there is too much
historical usage the other way to allow this drastic change (funny
arguing that Loglan/Lojban has too much historical usage for any thing).
I certainly would have severe relearning trouble.  In any case, there is
a second requirement that the tense system be self-consistent, that
overrides the requirement for consistency with lexeme BAI, which has
evolved separately and converged to being grammatically near-equivalent
to PU.  The Lojban tense system is difficult enough (note that we've
never yet written it up - >I< don't know it well enough yet), without
imposing a foolish consistency with an external system that makes it
more difficult.

Lexeme BAI derives from what Jim Brown called 'modal relative phrases',
which is why I call them 'modals' (these words may at one time have been
found only in relative phrases, i.e, those attached to sumti with "pe"
and "ne", but their use in Lojban has far outgrown these limits).

An important point is that the members of lexeme BAI are recognized as
being malglico - heavily English biased.  The Lojban list was made by
taking Brown's list, then by going through our gismu list, and adding a
BAI member for anything that suggested itself as useful for an adverbial
or prepositional phrase in English.  The reasoning was that Lojban
should be able to translate any English expression.  As we further
developed the language, we hoped to get perspectives from other
languages to expand BAI to meet non-English uses.  FIhO was seen as an
'emergency measure' to pull in anything we had omitted and had no cmavo
space for.  The combination of BAI and FIhO was therefore neutral since
it COULD cover any desired tag.  The key word in the above is "suggest"
- no one thought of modals as equating to their gismu - we chose the
cmavo to give us memory hooks - sometimes malglico ones.

In 1985 Brown proposed and in 1987 adopted by fiat a case tag system
that in Institute Loglan operates >SEPARATELY AND APART FROM< and in
some ways contradictory to his set of modals.  In that version of
Loglan, case tags apply to places that ARE part of defined place
structures, and modals in effect allow addition of places that are NOT
part of the standard set.  This latter set is very limited - only about
a dozen modals.  There are some case tags that could/should be modals,
but Brown did not accept this.  In some cases, therefore, he has a case
tag and a modal with the same meaning; in others you cannot add the
place.  Brown's version has no "fi'o" construct that allows formation of
new modalities at will; I believe I stole that from jimc's ideas.

In short, Brown has never recognized the predicate relation nature of
lexeme BAI.  We have come to that position only after considerable
evolution in thought.  BAI originally did not have the converters -
resemblences to gismu were just learning aids.  This is why "ci'a" still
means "written by" when what is actually meant is the relation "ciska
cusku" or "ciska finti" or even "cusku finti" (the exact meaning is a
topic open to all of you to comment on - which is most useful?).  At
that point "pi'o" (his "tie") meant what "sepi'o" now means.

The causals that are now "mu'i", "ki'u", "ni'i" and "ri'a" (JCB's "moi",
"rau", "soa", and "kou", respectively) were distinct from the modals,
and were not originally part of BAI; they have always had the "se" (his
"nu") and "nai" (his "noi") conversions and negations that the modals
did not have.  They had a much more elaborate grammar, including causal
connection between sentences that exists today as ".isemu'ibo".

The set that is now lexeme BAI grew during mid-1988 (as I note below, it
was still part of lexeme PU at the time), as we first started using
Lojban, and realized how many modals could be useful if available, and
noted that we had lots of cmavo (little word) space given the revised
phonology with the apostrophe and the removal of the lerfu letterals
from regular cmavo space.  This was also when Jeff Taylor and I were
redeveloping the Lojban grammar from scratch to prevent more arguments
about copyright from Jim Brown - who even today wrongly believes that we
used the Institute's LYCES program to develop our grammar.

The Institute Loglan 'tense' grammar then (and still now) consisted of
little more than lexemes PU, VI, and ZI, occurring in strings.  The
'useful' strings had historical dictionary-defined meanings, like "puba"
= "was going to", and were not inherently analyzable as Lojban's are
now.  These tenses were evaluated in effect as a single-word unit:  the
existence of an internal grammar of these strings was ignored; they were
always written as one word, and the existence of a tense word at the
head meant that the entire word was a 'tense'.  Brown's machine grammar
would let his cmavo substitution equivalent of "pufana.uimamo" be a
legal 'tense'.  Brown hand-waved past such nonsense by saying that the
machine grammar defined allowable language structures, but that certain
things permitted by those structures were forbidden as "bad usage".  (We
consider 'bad usage' an invalid excuse - the grammar defines what are
allowable strings of words, and how they group.  A string that is hard
to analyze is merely one that we haven't figured out how to use - yet.)

With the aid of "bad usage", Brown compressed tenses, modals, causals,
and a couple of other things (including what is now our CU) into one
lexeme 'PA', that could be used as a sumti tcita (argument tag) or as a
tense 'inflection'.  Causal connection between sentences, and associated
with logical connectives was handled by writing them as one word with
the 'head-word'.  Thus his equivalent of ".i se mu'i bo", which I in
fact write as one word ".isemu'ibo" WAS one word - part of lexeme I. He
did not need the "bo" in his grammar since his equivalent of
".isemu'i(bo) mi blanu" (/"Inumoi mi blanu"/) could not be grammatically
confused with ".i semu'i mi [KU] blanu" (/"I numoi mi, blanu"/), because
writing it as one word meant that the parser did not internally analyze
it.

The latter was unacceptable to us - Brown's language was, and still is,
in violation of audiovisual isomorphism, because these two sentences
that have the same pronunciation have different grammars depending on
how they are written (all of these features are to my knowledge still
true of Institute Loglan - readers who know otherwise are welcome to
correct me).  When we redeveloped the grammar from scratch, we required
that all cmavo (little word) compounds be broken into their components
before parsing and that the grammar completely describe all rules for
compounding.  At the time we started Lojban, we had no formal grammar
of our own, and Brown's lexeme PA became our lexeme PU - then containing
all of PU, BAI, VI, and a few others.  The 6/88 cmavo list reflects this;
by 10/88, these cmavo were all in separate lexemes.

Our tense grammar thus moved away from the modal grammar, eventually
becoming the "baroque" (according to jimc) structure that we have today
and cannot yet properly teach.  Our grammar encompassed all of Brown's
idiosyncratic compounds, and systematized them so that they could be
analyzed from their components and so that new ones could be created,
following the pattern.  We have revised this tense grammar completely 3
times, and a recent grammar change proposal makes a 4th fundamental
philosophical change although it amends the actual grammar only
slightly, and in ways that conform to the way people have been already
using tenses.  (I'll let John C. discuss this separately.)

Meanwhile, we still had the causals and the modals.  Sometime in
mid-1988 (possibly at LogFest that year), Nora, pc (John
Parks-Clifford), and I decided that what became lexeme BAI had gotten so
large that it was no longer easy to memorize, just from the memory hook
of the associated gismu, just what relation was implied:  did "pi'o"
mean user, or did it mean instrument?  Since the causals and modals were
grammatically related, it was easy to merge the two completely, allowing
conversion and negation to operate on the lexeme BAI members, and tying
their meanings to the place structures of the source gismu (or tanru in
the case of "ci'a").  As I've stated before, this was not because we saw
an identity between place structures and their associated BAI words, but
rather that using the converters merely gave the BAI words a little
better memory hook.

That this wasn't all that natural, can be seen by looking at the first
published cmavo list - from 10/88 (the 6/88 list was handed out at
LogFest but otherwise not distributed).  Of the 40 lexeme BAI members,
we incorrectly included or excluded the conversion markers on over 1/4
of them.  Even at that time we still thought of lexeme BAI as a closed
set - we had made this set more memorizable by making the form used,
with conversion, matching with the source 'metaphor' - but did not
clearly see that places not on the official BAI list could also be used
(and were grammatically permitted); we would thus have seen Nick's
example of "sebai" as being "bad usage" until someone called us on it.

Actual usage has of course moved the lexeme BAI members into near
equivalence with associated gismu, and we even forgot that some of them
like "ci'a" were actually based on tanru metaphors.  (I may be able to
dig out others from the records - if "ci'a" is not changed, at least the
reference will be to a tanru rather than to the gismu "ciska" in the
next published cmavo list.)

The tenses, after compounding, are still equivalent to the modal/causals
in grammar.  But their evolution to that equivalence was from the other
direction.  Modals started out in relative phrases, and later became
allowed as inflections on selbri (predicates).  Tenses started out as
inflections, but it was later realized that they could serve in relative
phrases.  Indeed, in the 1975 Loglan 1, Section 5.10 is called "modal
relative phrases", while 5.11, on the tenses as tags, is called
"relative modifiers".  They were not then seen as being the same thing.
We now in effect say that PU by itself is virtually the same as a modal,
but this is only a reductio case - if the most used form of tenses as
tags in the language.  A tagged sumti like "puvive'a lenu salci"
(before-at-over a medium area (around where was to be) the celebration"
is not so easily analyzed in terms of place structures.

In contrast, PU is likened to a vector-direction in time.  There are a
bunch of members of FAhA that are corresponding vector-directions in
space, though some are difficult to draw as 'arrows' since they deal
with proximacy situations like "next to" and "inside".  English "upon"
as a preposition is "up from-next to" or "next to-up from" in Lojbanized
English - we haven't defined a semantic difference in the order at this
point (no doubt someone will, though).  Ideally, these would be joined
with "joi" but there are limits in the internal grammar of tenses so that
we have a new kind of tanru.  "pu" is an arrow back in time.

The sumti tag versions are thus originally an expansion of the tense
inflection, while the modal inflection have always been an ellipsis of
the sumti value from a modal tagged sumti.  In other words, we went from
inflection "pu" to "puku" to "pu lenu klama", whereas we went from "mu'i
lenu klama" to "mu'iku" to "mu'i" as an inflection.  The existence of
the story-time convention, as described in draft textbook lesson 4,
shows the vector nature of the time words, and also makes it meaningful
to use the intermediate form "puku", which has a common natural language
translation "earlier" (and which has become even more important after
the 'origin' philosophy change that I'm not describing here).  The
corresponding "mu'iku" has never been used that I know of.

lojbab