WikiDiscuss

WikiDiscuss


BPFK Section: Anaphoric Pro-sumti

posts: 2388

While the systems for anaphoric prosumti in
Lojban are not as bad as possible (we could for
example pick references by counting the number of
syllables since the beginning of the last
same-referring expression), they are pretty
dreadful — certainly for spoken conversation but
even for written. The problem areas are at least
two-fold, reference identification and word
individuation.

To take the latter first, prosumti tend to come
in clumps: {ko'V}, {dV}, {Cy} and so on,
differing one from another by typically one
phoneme, often a final vowel. It does not take a
very noisy channel to reduce a clump of these to
an indistinguishable mumble, at least in theory.
Considering that some effort was put into making
the nodes in gismu space as widely separate as
possible (under certain other constraints), the
cavalier dumping of prosumti is hard to figure
out (except, of course, for the bad examples set
by other constructed languages, especially logic
itself). Ease of learning is also offered as a
reason, but hardly counts for much of one, since
the gismu are meant to be learned and are far
less systematic than the prosumti. Using the
same stock of words but mixing them in different
ways, so that there are two — even three --
points of difference between two items in the
same category would have been more practical in
the end. (Of course, we then get similarlities
across categories, but that seems to be less of a
problem, since context ought often to tell what
kind of prosumti is due).

When you come across an anaphoric prosumti, how
do you know what earlier reference it repeats? In
Lojban the clues are all (with one largely
useless exception) superficial features of the
utterance stream: a count back of sumti, a
grammatical analysis of sentences with references
by placement of nodes, the initial letter of the
main word of the original referring expression,
and so on. The exception is the system of
forethought anaphorics, {ko'V}, {fo'V}, which
require foreseeing the course of a conversation
-- even a monolog — beyond what is normally
possible. And, of coure, remembering what is
assigned to what.

If we look at natural languages, we find that
anaphoric pronouns are categorized not by
featurees of the *expression* they repeat, but by
features of the *referent* of that expression:
gender and number (and randomly a few other
things) in English and most SAE languages, other
categories and number in other languages
(references to a book or books are ki- or vi- in
Swahili, following "ki/vitabu," "book/s.") Of
course, Lojban "metaphysical neutrality" (snort)
means that there are not many classes of things
overtly recognized — and no singular or plural.
Roughly, Lojban recognizes individuals, groups,
sets and maybe abstraction (of various sort).
And any given conversation is likely to be heavy
on one sort or antoher (usually thing). Dropping
back to "natural" categories (as English has to
do) would help a bit — if we decided what the
natural categories were. The SAE M/F/N helps
somewhat, but rapidly loses the accuracy which
the Lojban system — when it works — gives.

Of course, none of this grousing helps here at
all; we cannot overhaul the system or get it to
work as it stands, we can only explain how it
would work if it did (OK, does work when it
does). The rest goes into notes for LoCCan 3.