[lojban] Re: Loglish: A Modest Proposal

[John E Clifford <lojban-out@lojban.org>]

It seems we really agree about both the
practicalities and the theory; the rest is just
details.

--- Ben Goertzel <ben@goertzel.org> wrote:

> 
> 
> > Thgough not, I think, as much less as the
> numbers
> > suggest.  Notice, also, that all this is
> largely
> > predicated on continuing to learn Lojbna in
> the
> > worse known way: as isolated pieces of
> vocabulary
> > and syntax.  Hopefully, someone will soon
> (after
> > a mere 50 years) get around to an immersion
> > teaching system for Lojban (steal and modify
> any
> > of the dozens already out there for Uzbeki
> and
> > Tlon) and reduce the learning cost to the
> point
> > where it will largely (though never
> completely)
> > overcome this objection.  In theory, Lojban
> ought
> > to be completely learnable in a day and total
> > facility in a werek. 
> 
> This last sentence seems completely unrealistic
> to me...

Well, I said it was in theory.  The figure is
based on a vocabulary of 4000 words and a
"standard-size grammar" including conjugations
and declensions of words.  Lojban as it stands
has a smaller vocabulary and an
orders-of-magnitude smaller grammar, so I
arbitrarily chopped the time in half for
"learning" but left the time for facility (I
can't find the reference for this formula and it
was a long time ago). Realistically, I think a
month of good training would produce someone
better than any one now (or almost anyone)even
starting with less than ideal users: someone able
to carry on a conversation on most topics at
English conversational speed.
  
> I am a pretty intelligent person, but I don't
> have
> a photographic memory, and I'm sure there's no
> way
> I could achieve a full working understanding of
> Lojban
> (including all that vocabulary) in a week...
> 
> I'd say that 2-3 months would be more realistic
> for
> learning Lojban in a total immersion
> environment --
> which as you point out does not exist.  By
> "learning"
> I include learning enough lujvo to hold an
> actual
> interesting conversation at something
> approaching, say,
> half of English conversational speed...
> 
> I'd say that 2-3 weeks for learning *Loglish*
> might be
> possible, for individuals very familiar with
> both English
> and predicate logic.

For reasons already given, I find this as
optimistic as my revised figures.But you could
have something that sounded right, but contained
many actual  errors, in probably even less time
(starting with relatively intelligent logicians
who were native speakers of English).
 
> > but
> > learning restrictions is often harder than
> > learning new words altogether (a similar
> remark
> > applies to grammar, so Loglish may have more
> > problems ther as well).
> 
> Again, we disagree on this point, and the only
> way I know to resolve the disagreement is
> empirical...

There is data on this one.
 
> I understand that in some other cases
> maintaining a
> familiar vocabulary with an unfamiliar grammar
> has proved
> difficult, but I have a strong feeling this
> won't be the 
> case with Loglish -- based on (among other
> thigns) my 
> recent preliminary experiments trying
> to speak a (flawed version of) Loglish with my
> wife...

Which have been, of course, uncontrolled so that
you can choose to accept whatever you say as
being OK. Several experiments of this intuitive
sort with other modified Englishes have come a
cropper  on this problem.
   
> > I would think that the ultimate aim was to
> get
> > away from English altogether.
> 
> OK, this is a fine *ultimate* aim, but in order
> to achieve it,
> one may need to adopt other aims for the short
> and medium
> term
> 
> I want my AI systems to be able to read world
> literature,
> science, and so forth -- and the only way all
> that stuff
> is going to translated into Lojban, IMO, is if
> an AI does
> it...
We have several problems. 1) A system to read
world literature and get it into digestable form
2)prior to that a system for storing all of this
in accessible format 3) a system for querying
this stored data.  English plays a necessary role
(for English material) only for 1.  It may have
some heuristic value for 2 and 3 but is
inessential for them -- and, indeed, positively
disruptive in at least some parts of each.  The
use of a modified form of English for 3 (I can't
imagine it seriously considered for 2 unless it
were assumed that only English literature was
worth including and only English speakers were
going to query the data) is a concession to the
market, although it may prove adequate for the
task (eventually augmented by similar
modifications of other querying languages (not
too difficult in theory if Loglish works).  

> I do think Lojban is superior to English (and
> to Loglish,
> in principle, once Lojban's vocabulary is more
> fully built
> out), but that doesn't mean I think Lojban is
> going to 
> obsolete English anytime soon....  
> 
> OTOH, I do think that Lojban or Loglish could
> obsolete 
> English in the short/medium term for the
> particular task 
> of communicating with 
> semi-intelligent computer programs...
> 
> And then, once there are AI's that know both
> English and
> Lojban and Loglish, the translation of science,
> literature
> and so forth into Lojban and/or Loglish becomes
> a possibility --
> and the use of Lojban and/or Loglish as a
> primary language 
> becomes a possibility...

Now, this seems pie-in-the-sky and, happily, no
particular part of Lojban planning (the stuff
about translating scientific literature is pretty
much that too, but at least conceivable and I
note you wisely refrain from talking about
translating imaginative literature.)

> > > -- included mappings of each Lojban word
> into
> > > appropriate WordNet senses
> > > 
> > > -- included mappings of each Lojban
> > > argument-position
> > > into an appropriate FrameNet case-role
> > 
> > These seem like likely goals in any case (not
> > necessarily using WordNet and FrameNet, but
> some
> > such schemata).  
> 
> Yes, I agree that WordNet and FrameNet are not
> the only possible
> resources to use in this role ... they're just
> the best-known
> and most fully-fleshed-out examples of
> resources of their kind...

I worry about WordNet because it does not seem to
have a core vocabulary with which to define
everything else (the Platonist in me, I suppose;
but the full carrying through of all these
projects seems to requires some such basis).
 
> > Since the basic Lojban
> > vocabulary is small, unambiguous, and still
> > somewhat under its designers' control, the
> > fundamental part could be done fairly rapidly
> > (does WordNet have a set of basic notions in
> > terms of which all others are defined as
> FrameNet
> > seems to have a set of relations to cover all
> > cases?).  The task of accounting for the
> derived
> > values for lujvo -- and the imported values
> for
> > fuhivla -- will be more complex but still
> > relatively easy compared to dealing with the
> > whole of English -- or even a reasonable
> sample
> > (the classic 10,000 words, say).
> > To be sure, given the nature of the Lojban
> > community, some proposals will be disputed
> and
> > modified but this can be restricted, as has
> been
> > done already for cmavo (indeed, the process
> into
> > which this project might fit is already under
> > way).  We may not have what is needed at the
> > moment, but it is not so far off as you seem
> to
> > think (nor as expensive).
> 
> Well, when these resources are created, I'll be
> very happy
> to think about how to use them in AI systems.
> 
> I believe you are underestimating the amount of
> work required
> to create them, however.

Dictionary writing is a peculiar skill.  If we
get someone good at it -- and there seems to be
no way of predicting this before a few hundred
samples come in -- it will go quite fast;
otherwise it will indeed be a long task with much
backing and filling.  I suppose my optimism is
based on the suspicion that among us there is at
least one person with that skill (though he may
not know it yet) who can be convinced to
undertake the task. The raw material is pretty
much at hand (more so than for English, say) and
is of manageable size.
 
> And, critically, until the derived values for
> lujvo are dealt with,
> such a resource will be of very limited use. 
> Most English words
> will translate into lujvo, obviously.

By the simple fact that there are a lot more
English words than gismu, true.  Although many
may not translate as words at all but merely as
places within other predicates for example.  It
does not seem that we need to aim for a Lojban
expression to cover every English word (and
certainly not every sense of every English word),
merely enough to cover all the topics we want to
cover and the ability to add new ones as needed. 
but you are right that the topic of expanding to
cover lujvo is a difficult one if we have to work
merely with the forms; if we get the
place-structure derivation, it will be much less
a problem -- at worst comparable to figuring out
the meaning of an English word from context. With
occasional error in both cases.

> > Well, I think that if the learning could be
> > improved, a far larger segment of the present
> > community could be brought up to speed and
> that
> > would appraoch at least the critical mass
> needed.
> 
> This may well be correct....
>  
> > Lojban requires less of the AI's, because
> it's
> > much less ambiguous
> > (the only significant ambiguity, so far as I
> can
> > tell, residing in tanru
> > and in reference resolution mechanisms).>>
> 
=== message truncated ===


> 
<<> Yeah, it has actually managed to have a
reference
> system that is less effective than English, a
> fairly remarkable achievement in its own right
> (to be sure, it has, in theory, a foolproof
> reference system, but it has proven unworkable
> even for written communication).

This is something I haven't fully thought through
in a Loglish context.

For Loglish, I may well introduce a completely
different way of doing
referencing -- separate from either English or
Lojban....>>

If you come up with something let us know --
mayhap we can borrow it and save Lojban's bacon.

 
<<> Yes, I agree that selling Loglish would be a
lot
> easier and that, therefore, if either of these
> ideas is to get off the ground, Loglish is the
> one that will make it. My point is only that
this
> is a sad state of affairs, since the
adevantages
> ultimately lie with Lojban (or some improved
> version thereof).

I agree with the above paragraph... 

but I fear that in my old age 
I have become more pragmatic and less idealistic
;-p>>

Gee, as I get nearer to 70 I find myself being
more optimistic about most things that do not
involve politics (which this issue  does I
suppose) and giving up politics as  hopeless
(given the small amount of money at my disposal).


To unsubscribe from this list, send mail to lojban-list-request@lojban.org
with the subject unsubscribe, or go to http://www.lojban.org/lsg2/, or if
you're really stuck, send mail to secretary@lojban.org for help.