From ben@goertzel.org Mon Mar 28 10:06:28 2005 Return-Path: X-Sender: ben@goertzel.org X-Apparently-To: lojban@yahoogroups.com Received: (qmail 44109 invoked from network); 28 Mar 2005 18:06:27 -0000 Received: from unknown (66.218.66.172) by m2.grp.scd.yahoo.com with QMQP; 28 Mar 2005 18:06:26 -0000 Received: from unknown (HELO intelligenesiscorp.com) (208.234.8.229) by mta4.grp.scd.yahoo.com with SMTP; 28 Mar 2005 18:06:26 -0000 Received: from zombiethustra (pcp06586041pcs.nrockv01.md.comcast.net [69.140.24.121]) by intelligenesiscorp.com (8.12.10/8.12.10) with SMTP id j2SI6GYl014978; Mon, 28 Mar 2005 13:06:22 -0500 To: , Date: Mon, 28 Mar 2005 13:06:25 -0500 Message-ID: MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_NextPart_000_05D5_01C53396.F2346C40" X-Priority: 3 (Normal) X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook IMO, Build 9.0.2416 (9.0.2911.0) In-reply-to: <20050328145742.68117.qmail@web81303.mail.yahoo.com> X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.2180 Importance: Normal X-Originating-IP: 208.234.8.229 X-eGroups-Msg-Info: 1:12:0 From: "Ben Goertzel" Subject: RE: [lojban] Re: (No Content-Type: text/plain; charset=ISO-8859-1 X-Yahoo-Group-Post: member; u=1278257 X-Yahoo-Profile: bgoertzel X-Yahoo-Message-Num: 24047 ------=_NextPart_000_05D5_01C53396.F2346C40 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Well, machine translation technology has advanced a long way since the 1960's, however, it's still very far from adequate. I believe I know how to make it adequate, via combining modern computational linguistics methodologies with Lojban as an interlingua . But there is a significant amount of work involved here, and no one seems to want to fund it. For example, DARPA has a current BAA for machine translation, but they have a specific policy of only funding improvements in already-working translation programs --- a pretty sure way to avoid funding dramatically different new lines of research! -- Ben -----Original Message----- From: lojban-list-bounce@lojban.org [mailto:lojban-list-bounce@lojban.org]On Behalf Of John E Clifford Sent: Monday, March 28, 2005 9:58 AM To: lojban@yahoogroups.com Subject: [lojban] Re: (No Content-Type: text/plain; charset=ISO-8859-1 This is all very depressing: aside from technological improvements (moving from large-room -sized computers to vest pocket ones with vast increases in processing speed and size of handy memory so that "interactive" involves less than a two-day turnaround) it sounds as though "machine translation" hasn't gotten much beyond where it was in 1962, when big improvements were "just around the corner." To be fair, I can see that the linguistics is better, with functioning parsers of reasonable accuracy (and based on much improved analyses, I suppose -- though whether less arbitrary is not clear). I assume that the distributional semantics tables have been extended and improved and their use better integrated into the process. And, of course, Lojban is much better than the Loglan proposed as an intermediate language back then -- on the basis of the 1960 article only (JCB was not answering mail that decade). Still, it appears that the amount of human time involved in a particular translation has not decreased significantly (and, of course, the ratio of human to machine time has gotten much worse). That is, the advantage of machine work over human translator from scratch is still barely perceptible. --- Ben Goertzel wrote: > > > > > >To make a long story short, > > > > >-- translating Lojban into weird, awkward, > sometimes ungrammatical but > > >basically comprehensible English is > definitely a solvable problem, > > >though requiring a moderate rather than > trivial amount of work > > > > This is really only useful, IMHO, as a step > toward the next one: > > Not really true -- if you had an AI system that > was deriving novel > conclusions and could express these conclusions > in Lojban, then having a > crude English dump of its conclusions could be > quite valuable... > > > >-- translating Lojban into fully grammatical > and elegant English would > > >be quite hard > > > > However, if we are able to accomplish this, > it should be repeatable with > > a somewhat smaller but still substantial > amount of workload for other > > target languages. This would be quite useful, > IMO. > > Correct > > > >-- translating English into Lojban would be > extremely difficult given > > >the current (somewhat pitiful) state of > computational linguistics and > > >AI technology. The right approach would be > to create an interactive > > >tool > > > > Is it even certain that this is possible? > > I'm almost certain it's possible to do an > *almost* complete job of this, but > also certain that it's hard. There are tricky > issues of pragmatics that may > be unresolvable without "strong AI", but I'm > pretty sure you can do a 99% > job without solving these... > > > I see this as a valuable - and possibly > essential > > - step in conquering the problem of good > machine translation. > > Agreed. > > Basically this is the same idea as in the > Interlingua machine translation > project > > http://www.isi.edu/natural-language/mt/interlingua.html > > but using Lojban instead of their graph-based > interlingua (which makes sense > because Lojban is more complete and more > thoroughly battle-tested...) > > -- Ben G > > > > > > > ------------------------ Yahoo! Groups Sponsor > > > To unsubscribe, send mail to > lojban-unsubscribe@onelist.com > Yahoo! Groups Links > > > lojban-unsubscribe@yahoogroups.com > > > > > > > > > To unsubscribe, send mail to lojban-unsubscribe@onelist.com Yahoo! Groups Sponsor ADVERTISEMENT ---------------------------------------------------------------------------- -- Yahoo! Groups Links a.. To visit your group on the web, go to: http://groups.yahoo.com/group/lojban/ b.. To unsubscribe from this group, send an email to: lojban-unsubscribe@yahoogroups.com c.. Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service. ------=_NextPart_000_05D5_01C53396.F2346C40 Content-Type: text/html; charset="US-ASCII" Content-Transfer-Encoding: quoted-printable
 
Well,=20 machine translation technology has advanced a long way since the 1960's,=20 however, it's still very far from adequate.
 
I=20 believe I know how to make it adequate, via combining modern computational= =20 linguistics methodologies with Lojban as an interlingua .  But there i= s a=20 significant amount of work involved here, and no one seems to want to fund= =20 it. 
 
For=20 example, DARPA has a current BAA for machine translation, but they have a=20 specific policy of only funding improvements in already-working translation= =20 programs --- a pretty sure way to avoid funding dramatically different new = lines=20 of research!
 
--=20 Ben
 
-----Original Message-----
From:=20 lojban-list-bounce@lojban.org [mailto:lojban-list-bounce@lojban.org]On= =20 Behalf Of John E Clifford
Sent: Monday, March 28, 2005 9:58= =20 AM
To: lojban@yahoogroups.com
Subject: [lojban] Re: (= No=20 Content-Type: text/plain; charset=3DISO-8859-1

Th= is is=20 all very depressing: aside from
technological improvements (moving=20 from
large-room -sized computers to vest pocket ones
with vast incr= eases=20 in processing speed and size
of handy memory so that "interactive"=20 involves
less than a two-day turnaround) it sounds as
though "machi= ne=20 translation" hasn't gotten much
beyond where it was in 1962, when=20 big
improvements were "just around the corner." 
To be fair, = I can=20 see that the linguistics is
better, with functioning parsers of=20 reasonable
accuracy (and based on much improved analyses,=20 I
suppose  -- though whether less arbitrary is not
clear). I a= ssume=20 that the distributional
semantics tables have been extended and=20 improved
and their use better integrated into the process.
And, of= =20 course, Lojban is much better than the
Loglan proposed as an intermedi= ate=20 language back
then -- on the basis of the 1960 article only
(JCB wa= s not=20 answering mail that decade).  Still,
it appears that the amount o= f=20 human time involved
in a particular translation has not=20 decreased
significantly (and, of course, the ratio of human
to mach= ine=20 time has gotten much worse).  That is,
the advantage of machine w= ork=20 over human
translator from scratch is still=20 barely
perceptible.  

--- Ben Goertzel=20 <ben@goertzel.org> wrote:
>
>
>
> > &= gt;To=20 make a long story short,
> >
> > >-- translating Loj= ban=20 into weird, awkward,
> sometimes ungrammatical but
> >=20 >basically comprehensible English is
> definitely a solvable=20 problem,
> > >though requiring a moderate rather than
>= =20 trivial amount of work
> >
> > This is really only usef= ul,=20 IMHO, as a step
> toward the next one:
>
> Not really = true=20 -- if you had an AI system that
> was deriving novel
> conclu= sions=20 and could express these conclusions
> in Lojban, then having a
&= gt;=20 crude English dump of its conclusions could be
> quite=20 valuable...
>
> > >-- translating Lojban into fully=20 grammatical
> and elegant English would
> > >be quite=20 hard
> >
> > However, if we are able to accomplish=20 this,
> it should be repeatable with
> > a somewhat smalle= r but=20 still substantial
> amount of workload for other
> > targe= t=20 languages. This would be quite useful,
> IMO.
>
>=20 Correct
>
> > >-- translating English into Lojban woul= d=20 be
> extremely difficult given
> > >the current (somewh= at=20 pitiful) state of
> computational linguistics and
> > >= AI=20 technology.  The right approach would be
> to create an=20 interactive
> > >tool
> >
> > Is it even ce= rtain=20 that this is possible?
>
> I'm almost certain it's possible = to do=20 an
> *almost* complete job of this, but
> also certain that i= t's=20 hard.  There are tricky
> issues of pragmatics that may
>= ; be=20 unresolvable without "strong AI", but I'm
> pretty sure you can do = a=20 99%
> job without solving these...
>
> > I see this= as a=20 valuable - and possibly
> essential
> > - step in conqueri= ng=20 the problem of good
> machine translation.
>
>=20 Agreed.
>
> Basically this is the same idea as in the
>= ;=20 Interlingua machine translation
> project
>
>
http://w= ww.isi.edu/natural-language/mt/interlingua.html
>=20
> but using Lojban instead of their graph-based
> interlingu= a=20 (which makes sense
> because Lojban is more complete and more
&g= t;=20 thoroughly battle-tested...)
>
> -- Ben G
>
>=20
>
>
>
>
> ------------------------ Yah= oo!=20 Groups Sponsor
> --------------------~-->
> In low income= =20 neighborhoods, 84% do not own
> computers.
> At Network for G= ood,=20 help bridge the Digital
> Divide!
>
http://= us.click.yahoo.com/EA3HyD/3MnJAA/79vVAA/GSaulB/TM
>
----------= ----------------------------------------------------------~->
>=20
>
> To unsubscribe, send mail to
>=20 lojban-unsubscribe@onelist.com
> Yahoo! Groups Links
>
&= gt;=20
>     lojban-unsubscribe@yahoogroups.com
&g= t;=20
> 
>
>
>
>
>
>
= >=20






To unsubscribe, send mail to=20 lojban-unsubscribe@onelist.com


------=_NextPart_000_05D5_01C53396.F2346C40--