From pycyn@aol.com Tue Aug 07 13:35:36 2001 Return-Path: X-Sender: Pycyn@aol.com X-Apparently-To: lojban@yahoogroups.com Received: (EGP: mail-7_2_0); 7 Aug 2001 20:35:35 -0000 Received: (qmail 33135 invoked from network); 7 Aug 2001 20:34:59 -0000 Received: from unknown (10.1.10.142) by l9.egroups.com with QMQP; 7 Aug 2001 20:34:59 -0000 Received: from unknown (HELO imo-m04.mx.aol.com) (64.12.136.7) by mta3 with SMTP; 7 Aug 2001 20:34:56 -0000 Received: from Pycyn@aol.com by imo-m04.mx.aol.com (mail_out_v31.9.) id r.c1.120bdaf0 (4403) for ; Tue, 7 Aug 2001 16:34:42 -0400 (EDT) Message-ID: Date: Tue, 7 Aug 2001 16:34:41 EDT Subject: First cutting for a new record To: lojban@yahoogroups.com MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="part1_c1.120bdaf0.28a1aae1_boundary" X-Mailer: AOL 6.0 for Windows US sub 10531 From: pycyn@aol.com X-Yahoo-Message-Num: 9303 --part1_c1.120bdaf0.28a1aae1_boundary Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Anaphora in Lojban, as in natural languages, is less than perfect. There is not always a short expression to recall some earlier expression and, even when there is, it may not be unequivocal or it may be too complicated to use or have understood in the speech flow. The situation is further complicated by some conflicts among various presentations of the anaphoric systems and by the apppearance of some casess that do not seem to have been considered or dealt with. In an ideal system, every sumti -- and every bridi for Lojban's system -- would come with a label attached that could be call up at any later time to recall that item. And there would be a device for combining labels into new ones as well. Formal logic, when presented in a language which allows no terms but bound variables, comes close to this for sumti, though it cannot combine labels easily and is no help for bridi. But that formal logic is entirely written and often confuses even those most familiar with it. For a language that is ever spoken, such a system runs through the available memory too rapidly to be practical (and, in the spoken language, there is no external record to return to to check or refresh). A variation that might seem somewhat more useful is a structural approach: items are tagged not in order but by their position in sentences and subsentences. It may be easier to remember what was the subject of the second sentence back than to remember what was sumti 27 or the one 15 sumti ago. But, of course, this uses up memory as fast as the other -- or even faster, though a back-counting version (as suggested by two of the previous examples) could work better, by dropping sentences after a certain time (isn't it 6+/- 2?) And a sentence of any serious complexity (subordinate clauses of various sorts with their own subordinates) might exhaust the limits even with a single sentence. As a practical matter, then, we need something to fit the memory wiring of the human brain that will still geet us a good ways toward dealing with common situations: 1) anaphora within a single sentence (loosely, reflexivity and reciprocity issues, but others as well), 2) generally, picking up terms from earlier (and, Lojban being Lojban, later) sentences, within some reasonable limits and 3) combining references into new terms efficiently. 1. The Book says that {vo'V} refer to the terms occupying the corresponding places (a=1, e=2,...) of "the present bridi"; the cmavo list says of "the main bridi of the sentence." While these *could* be taken as meaning the same thing (most clearly the latter case), they seem naturally to conflict when a sentence contains subordinate bridi in abstractions or relative clauses (or quotations? or...? -- this is apparently a further unresolved unclarity). This basic problem got resolved -- with some hesitations --in favor of the second interpretation ("main bridi") on the grounds that it covered the most cases and that upper level sumti were more likely to be repeated in lower levels. What then about cases of reciprocity and reflexivity on lower levels? One answer was the back-counting RI series and the other usual techniques, imprecise as they might be. The second was the introduction of {nei}, a repetition of 'this bridi' (with all the possibilities for paradox that that allows) to be used in descriptions {le nei}, {le se nei} etc. for the various places -- replacing the first interpretation of {vo'V}. Paralleling that came {no'a} with subscripts for the supervening bridi in embedded structures: {le no'a} is the first sumti of the immediately supervening bridi, {le no'axire} for the first of the bridi above that and so on. {le no'axiro} always comes down just to {vo'a}, the first sumti of the topmost (main) bridi. This reduction allows that the default subscript on {no'a} is {pa} rather than {ro}, as was also suggested. This technique only works for linear superordination and does not deal with (nor provide ready expansion to) parallel cases: picking up a term from another bridi subordinate to one to which the present bridi is also subordinate. These cases fall back upon the general 'solutions' (though one can imagine a devise for going up to a bridi picking out a sumti in it and then running down to a term in that embedded bridi). The {nei} and especially {no'a} forms present some problems. With {nei} the question is when is 'the present bridi' there? Can one refer, in the first place, to {le te nei}, before the third place --or indeed the selbri -- has been uttered? Lojbab is for the negative, but that begins to get into counting problems: what if the first sumti of a bridi is a subordinate bridi and we then want to refer back up -- is the immediately super bridi (whose selbri has not yet been uttered) to count as {no'a} or not? It is not yet even {nei} in itself, how can it be {no'a} to the one below. In the case of {ri} the rules held that it could not count a sumti in which it was included in its count. A similar rule here would quite possibly cut the whole system off at the beginning, depending on the structure. On the other hand, it seems odd to say that a bridi exists when there is no selbri (witness the history of the terms, still found in places in the English 'explanations'). With {no'a} the issue is what does it mean used as a bridi alone rather than in a description (one can imagine the same problem for {nei}, but less plausibly). Officially, this question is unanswered, but Lojbab favors taking subordinate {no'a} out as separate sentences and replacing {no'a} by {go'i} then taking that as the correct interpretation of the original. This assumes that the transformation can always be done effectively and also forces a particular interpretation on the results, leaving the question of how to formulate other possibilities as simply. The three cases so far discussed are 1- la djan ba klama lo zarci pu le nu la meris no'a A) John will go to some store before Mary goes to it. B) John will go to some store before Mary goes to one. 2- la djan ba klama ro zarci pu le nu la meris no'a A) John will go to each store before Mary goes to it. B) John will go to every store before Mary goes to any. 3- la djan ba klama le zdani be vo'a pu le nu la meris no'a A) John will go to his home before Mary goes there. B) John will go to his home before Mary goes to hers. The {go'i} solution solves 3 in favor of a, with b being { la djan ba klama le zdani be vo'a pu le nu la meris no'a ro'a}. The other two are still being calculated out -- suggesting that the effective transformation condition may not be met. (2 also has a third suggested reading "John will go to every store before Mary does" i.e. "before Mary goes to every store" "john will complete the store rounds before Mary completes it" but that seems Englishly inspired.) 2. Picking terms from other sentences more accurately that the vague {ri/a/u} or literals suggests the use of {go'i/e/a/u }-- and [yech, ptui] {go'o} -- like the {nei} and {no'a} within a sentence, now looking at bridi in order not in subordination. The {le se go'i} is the second sumti in the preceding sentence and so on. The basic give four sentences, probably as many as can be practically held in mind with any accuracy. And, of course, if the sentences are complex, much more than can be held in mind. The complex sentences raises again the problem of subordinate sumti; the system only gets the topmost sumti, not sumti that are buried in those sumti at some depth. On the other hand, a device for digging up those treasures is likely to be as long as -- and much less clear than -- simply repeating the original. At least the basic descriptions with {go'V} are officially legal. 3. "Jack challenged Bob to a duel. They agreed to fight on Isle Duello. They arrived before dawn" Translate into Lojban, being brief. No can do. Lojban has no immediate device for collapsing two separate terms into a "they" nor for combining two places into one as in "agree." You end up with something like (though surely much better than) {la djek talsa la bab lenu relmemda'a .i ra tugni ru soi vo'e fo lenu damba vi la ildu,elos i. la djek. e la bab. klama tu pu le cermurse}. So far as I can find there is no suggestion on how to deal with these cases -- either one -- within the present framework, except to bite the bullet and spell it all out as above. The possible experimental devices for either of these are not yet clearly nor completely laid out, but hopefully both of the faults will be handled soon, now that the issue is raised (yet again?). And always remember: repetition too is anaphora. --part1_c1.120bdaf0.28a1aae1_boundary Content-Type: text/html; charset="US-ASCII" Content-Transfer-Encoding: 7bit Anaphora in Lojban, as in natural languages, is less than perfect.  There
is not always a short expression to recall  some earlier expression and, even
when there is, it may not be unequivocal or it may be too complicated to use
or have understood in the speech flow.  The situation is further complicated
by some conflicts among various presentations of the anaphoric systems and by
the apppearance of some casess that do not seem to have been considered or
dealt with.
In an ideal system, every sumti -- and every bridi for Lojban's system --
would come with a label attached that could be call up at any later time to
recall that item.  And there would be a device for combining labels into new
ones as well.  Formal logic, when presented in a language which allows no
terms but bound variables, comes close to this for sumti, though it cannot
combine labels easily and is no help for bridi.  But that formal logic is
entirely written and often confuses even those most familiar with it.  For a
language that is ever spoken, such a system runs through the available memory
too rapidly to be practical (and, in the spoken language, there is no
external record to return to to check or refresh).  A variation that might
seem somewhat more useful is a structural approach: items are tagged not in
order but by their position in sentences and subsentences.  It may be easier
to remember what was the subject of the second sentence back than to remember
what was sumti 27 or the one 15 sumti ago.  But, of course, this uses up
memory as fast as the other -- or even faster, though a back-counting version
(as suggested by two of the previous examples) could work better, by dropping
sentences after a certain time (isn't it 6+/- 2?)  And a sentence of any
serious complexity (subordinate clauses of various sorts with their own
subordinates) might exhaust the limits even with a single sentence.
As a practical matter, then, we need something to fit the memory wiring
of the human brain that will still geet us a good ways toward dealing with
common situations: 1) anaphora within a single sentence (loosely, reflexivity
and reciprocity issues, but others as well), 2) generally, picking up terms
from earlier (and, Lojban being Lojban, later) sentences, within some
reasonable limits and 3) combining references into new terms efficiently.
1.  The Book says that {vo'V} refer to the terms occupying the
corresponding places (a=1, e=2,...) of  "the present bridi"; the cmavo list
says of  "the main bridi of the sentence." While these *could* be taken as
meaning the same thing (most clearly the latter case), they seem naturally to
conflict when a sentence contains subordinate bridi in abstractions or
relative clauses (or quotations? or...? -- this is apparently a further
unresolved unclarity).  This basic problem got resolved  -- with some
hesitations --in favor of the second interpretation ("main bridi") on the
grounds that it covered the most cases and that upper level sumti were more
likely to be repeated in lower levels.
What then about cases of reciprocity and reflexivity on lower levels?  
One answer was the back-counting  RI series and the other usual techniques,
imprecise as they might be.  The second was the introduction of {nei}, a
repetition of 'this bridi' (with all the possibilities for paradox that that
allows) to be used in descriptions {le nei}, {le se nei} etc. for the various
places -- replacing the first interpretation of {vo'V}.  Paralleling that
came {no'a} with subscripts for the supervening bridi in embedded structures:
{le no'a} is the first sumti of the immediately supervening bridi, {le
no'axire} for the first of the bridi above that and so on. {le no'axiro}
always comes down just to {vo'a}, the first sumti of the topmost (main)
bridi.  This reduction allows that the default subscript on {no'a} is {pa}
rather than {ro}, as was also suggested.  This technique only works for
linear superordination and does not deal with (nor provide ready expansion
to) parallel cases: picking up a term from another bridi subordinate to one
to which the present bridi is also subordinate.  These cases fall back upon
the general 'solutions'  (though one can imagine a devise for going up to a
bridi picking out a sumti in it and then running down to a term in that
embedded bridi).
The {nei} and especially {no'a} forms present some problems.  With {nei}
the question is when is 'the present bridi' there?  Can one refer, in the
first place, to {le te nei}, before the third place --or indeed the selbri --
has been uttered?  Lojbab is for the negative, but that begins to get into
counting problems: what if the first sumti of a bridi is a subordinate bridi
and we then want to refer back up -- is the immediately super bridi (whose
selbri has not yet been uttered) to count as {no'a} or not?  It is not yet
even {nei} in itself, how can it be {no'a} to the one below.  In the case of
{ri} the rules held that it could not count a sumti in which it was included
in its count.  A similar rule here would quite possibly cut the whole system
off at the beginning, depending on the structure.  On the other hand, it
seems odd to say that a bridi exists when there is no selbri (witness the
history of the terms, still found in places in the English 'explanations').  
With {no'a} the issue is what does it mean used as a bridi alone rather
than in a description (one can imagine the same problem for {nei}, but less
plausibly).  Officially, this question is unanswered, but Lojbab favors
taking subordinate {no'a} out as separate sentences and replacing {no'a} by
{go'i} then taking that as the correct interpretation of the original.  This
assumes that the transformation can always be done effectively and also
forces a particular interpretation on the results, leaving the question of
how to formulate other possibilities as simply.  The three cases so far
discussed are

1- la djan ba klama lo zarci pu le nu la meris no'a

  A) John will go to some store before Mary goes to it.
  B) John will go to some store before Mary goes to one.

2- la djan ba klama ro zarci pu le nu la meris no'a

  A) John will go to each store before Mary goes to it.
  B) John will go to every store before Mary goes to any.

3- la djan ba klama le zdani be vo'a pu le nu la meris no'a

  A) John will go to his home before Mary goes there.
  B) John will go to his home before Mary goes to hers.

The {go'i} solution solves 3 in favor of a, with b being  { la djan ba klama
le zdani be vo'a pu le nu la meris no'a ro'a}.  The other two are still being
calculated out -- suggesting that the effective transformation condition may
not be met. (2 also has a third suggested reading "John will go to every
store before Mary does" i.e. "before Mary goes to every store" "john will
complete the store rounds before Mary completes it" but that seems Englishly
inspired.)

2.  Picking terms from other sentences more accurately that the vague
{ri/a/u} or literals suggests the use of {go'i/e/a/u }-- and [yech, ptui]
{go'o} -- like the {nei} and {no'a} within a sentence, now looking at bridi
in order not in subordination.  The {le se go'i} is the second sumti in the
preceding sentence and so on.  The basic give four sentences, probably as
many as can be practically held in mind with any accuracy.  And, of course,
if the sentences are complex, much more than can be held in mind.  The
complex sentences raises again the problem of subordinate sumti; the system
only gets the topmost sumti, not sumti that are buried in those sumti at some
depth.  On the other hand, a device for digging up those treasures is likely
to be as long as -- and much less clear than -- simply repeating the
original.  At least the basic descriptions with {go'V} are officially legal.

3.  "Jack challenged Bob to a duel. They agreed to fight on Isle Duello.  
They arrived before dawn"  Translate into Lojban, being brief.  No can do.  
Lojban has no immediate device for collapsing two separate terms into a
"they" nor for combining two places into one as in "agree." You end up with
something like (though surely much better than) {la djek talsa la bab lenu
relmemda'a .i ra tugni ru soi vo'e fo lenu damba vi la ildu,elos i. la djek.
e la bab. klama tu pu le cermurse}.  So far as I can find there is no
suggestion on how to deal with these cases -- either one -- within the
present framework, except to bite the bullet and spell it all out as above.  
The possible experimental devices for either of these are not yet clearly nor
completely laid out, but hopefully both of the faults will be handled soon,
now that the issue is raised (yet again?).

And always remember:  repetition too is anaphora. --part1_c1.120bdaf0.28a1aae1_boundary--