Dialog with Bob McIvor on Loglan - part 1 of 2 (long)

[Logical Language Group <cbmvax!uunet!GREBYN.COM!pucc.PRINCETON.EDU!lojbab>]

70674.1215@compuserve.com Wed Oct 30 11:59:08 1991:
>John Ross has sent me copies of comments by Bob LeChevalier on issues
>discussed in Lognet(TM), which were submitted to the conlang forum.  I
>would like to enlarge on some of his comments, which do not, in my
>opinion, correctly express the views of The Loglan Institute.

Thank you for responding, RAM.  I much prefer dialog.  I do question
whether TLI has monolithic 'views', unless it is by definition that of
JCB.  When I comment, I do so on the basis of my understanding of his
views.

The following is extremely critical of your comments in the referenced
message.  I think you are taking an insider's view as to what is
appearing in print, and missing the obvious implications of what is
being said.  I would prefer that my comments be taken constructively,
and not as an attack on you or TLI per se (to the extent I may trash
TLI's current design of the language, you may take my statements as
being what I feel are reasoned criticisms, each of which we have
resolved for Lojban.)

I note in advance that there is an apology embedded before for a couple
of mistakes I made that you legitimately criticized me for.  Thank you
for being gentle in correcting me.  I alas, am not as skilled in gentile
criticism.  But I'm trying.

In spite of its great length and occasionally intricate detail, I am
posting the whole thing to conlang a) to respond completely and
analytically to your points and b) because while the details may be
irrelevant to those not interested in Loglan in particular, the
discussion clearly points out many issues that a conlang inventor had
better take into consideration, as well as the range of philosophies
that can exist between two conlang workers that think they are working
on the same language, and at my last understanding, respect each other
highly.

I am also copying this to the Lojban List, for the edification of many
there not on conlang.  This is a far more reasoned discussion than the
flame exchange with Steve Rice in June.

______

>As is apparently the situation with Lojban, case tags are being actively
>studied at the moment, with a view to clarifying their role in the
>language.  None of the opinions expressed in Lognet should therefore be
>taken as gospel, and I welcome any insights on this problem which might
>be contributed by the Lojban group.

There has been recent debate case tags in conlangs in general, and some
debate vis a vis Lojban, but we rejected case tags a couple of years ago
after pc did an indepth study.  The conclusion:  linguists cannot agree
on any set of cases as being necessary or sufficient for language, and
one group of researchers actively working with cases is gaining some
success with the effective assumption that there is no finite set of
cases.

We generalized from that and realized that ANY set of cases if treated
as cases per se, imposes a metaphysical bias - a no no according to
JCB's assumptions.  The argument:  if two places of the same or a
different predicate have the same case tag, that common case tag implies
the existence a second predicate relation between the two equivalently
tagged places, specifically that they have the same (or essentially
similar) role in their respective relations.  It seems inherently likely
therefore, that someone will come to identify all 'actors' as having
something inherently in common, likewise all 'sources', etc.  (I
recognize that this commonalty may be poorly identified by the 'name' of
the tag, but strongly suspect that it would color the interpretation of
that common predicate.

It is all well and good to overtly claim that a 'child' is a product of
either parental relationship, but to equate this with the 'product' of
'madzo' make and other 'products' imposes an equivalence that may be
desired by a speaker, but not necessarily by all speakers.  The same
thing results in standard prepositional languages that share the same
preposition.  An English speaker, I argue without proof, subconsciously
assumes some similarity among all objects which are linked to their verb
by the preposition 'to', likewise for 'of' (in the latter case,
possession).  But what if you wish to claim the relationship 'matma'
WITHOUT implying that the child is in some way a 'product', and the
mother a 'source' (personally I would consider the egg and sperm equally
'the source' of a child, and hence ultimately both parents, and feel it
is sexist to imply otherwise).

If you argue that leaving the case tags off removes this implication, I
respond that they are inherently being included as part of the
definition of the words.

The result is that predicates can be semantically classified by the
cases that they relate.  All predicates with the same cases on their
arguments are arguably the same in some semantic sense.  You have thus
subdivided predicates into semantic categories, and I argue that that is
incompatible with a predicate language, where 'predicate' is defined in
its logical sense.

The alternative is hinted at by James Jennings in the new Lognet, where
he proposes that all 1 place predicates have the B case on their single
case, using the argument of the universal grammatical role of the
predicate.  Alas, he does not carry this to its logical conclusion:
that all 2 place predicates should have the same case tag for their x1
place and the same case tag for their x2 place, and the same for three
place predicates, etc.  Any other classification violates the 'universal
grammatical role' assumption.

The only remaining argument is whether the case tag for x1 of a 1 place
predicate should be the same as the x1 place of a 2 place predicate;
i.e. whether the sets of case tags should overlap those of
other-number-of-argument predicates.  If that were not so, each separate
set would be a different selma'o (grameme*), because it would never be
valid to put a 1-place case tag on a 2-place predicate.  Now JCB has in
the past rejected a grammatical division of predicates by numbers of
arguments, and I would be surprised if such a division has crept back in
during the trade secret period.

The result is that case tags are only needed to number the places, and
the Hixson-Bonewitz numbering tags (FA series for Lojbanists) are the
necessary and sufficient set.
_____

*I urge you to start the ball rolling within the TLI community to
correct the mis-terminology of JCB's use of 'lexeme', as we are
hurriedly trying to do.  All of Loglan is hurt by it.  Suffix '-eme'
refers to the smallest significant distinction of its prefix morpheme.
'Lexeme' is thus a word unit as seen by your parser (le, po, and lepo
are TLI version lexemes, not LE, PO, and LEPO).  Your 'lexer' thus
categorizes 'lexemes' into 'gramemes'.  (In YACC parlance, I believe a
'lexer' is something that breaks down the input stream into 'tokens',
which are the word-units or lexemes.  Loglan parsers do the extra step
of classifying 'gramemes' that is not applicable to most computer
languages, since their grammar rules include the final step of
enumerating all the words that are equivalent for each rule, making
'lexeme' and 'grameme' effectively identical.  This is NOT true for
Loglan, and virtually every professional linguist has called us on it.

_____

Instead of case tags, Lojban has generalized the other set of semantic
tags that JCB calls relative modal operators (Lojban BAI), and
incorporated the causals in this set (but keeps them separate from the
rest of the tenses, for which we have a full and detailed grammar).  To
incorporate the causals, we made it so that all relative modal operators
can be 'converted' in the same way that the TLI causals can with NU and
NOI (Lojban SE and NAI).  As the relative modal operators are loosely
associated with certain prims (gismu) that suggest their meaning, so are
Lojban BAI words.  But you use the appropriate conversion to access the
place that you want.  Thus Lojban pilno (x1 uses x2 for x3) gives rise to
"pi'o",  and then "sepi'o" gives semantic reference to the x2 place of
"pilno", thus roughly equating to the preposition "using ...", or the
instrumental/tool case.  The linkage in meaning is somewhat metaphorical,
as are the relative modal operators.

The linkage to prims and their place structures makes a much larger set
memorizable.  We have some 60 BAI tags, but with conversion, that is
probably over 150 meaningful 'cases'.  But we also provide a mechanism
for converting any predicate into a BAI equivalent, so our set of
potential 'cases' is indeed infinite to match the most liberal case
theory.

Unlike JCB, we do not ban these operators from the numbe