[96233] in tlhIngan-Hol
Re: [Tlhingan-hol] Bingonese Translator
daemon@ATHENA.MIT.EDU (Rohan Fenwick)
Tue May 14 19:35:19 2013
From: Rohan Fenwick <qeslagh@hotmail.com>
To: "tlhingan-hol@kli.org" <tlhingan-hol@kli.org>
Date: Wed, 15 May 2013 09:34:56 +1000
In-Reply-To: <1368570492.17419.YahooMailNeo@web172605.mail.ir2.yahoo.com>
Errors-To: tlhingan-hol-bounces@stodi.digitalkingdom.org
--===============3967461171878601382==
Content-Type: multipart/alternative;
boundary="_1dba7cdd-cfd3-40f6-9d80-2245538beb67_"
--_1dba7cdd-cfd3-40f6-9d80-2245538beb67_
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
ghItlhpu' Alex=2C jatlh:
> The Bing Klingon translator went online today. I've tried a few sample
> sentences=2C rudimentary stuff anyone could do with a cursory examination
> of TKD.
> It needs work.
net Sovchu'.
What you have to understand is like most machine translators these days it =
depends less upon a predetermined set of rules and more upon a database of =
considered-to-be-grammatical text from which it extrapolates the rules for =
itself. With languages like French it has titanic quantities of textual evi=
dence from which to do so. Sometimes it still gets it wrong even in a major=
language like French. Compared to French=2C Klingon as a language is only =
a step or two up from having no textual backing at all=2C and while most of=
the largest texts are in the Bing engine=2C many are perhaps not the most =
ideal for machine translation (Hamlet=2C while a wonderful text=2C is a rel=
atively free retelling and so sentence-by-sentence comparisons are often no=
t fruitful - get Bing to translate "Qo'noS" and you'll see what I mean).
In order to remedy this=2C I and several other Grammarians have spent the l=
ast little while actively participating in the testing and training. We've =
been madly feeding it specimen sentences as well as monolingual and paralle=
l texts better suited to machine comparisons. Most of the published Klingon=
texts have been fed into the engine=2C as well as a wide range of unpublis=
hed others. And please don't get the idea that the training has completed j=
ust because it's gone live. In response to your "let's hope they get improv=
ing"=2C work on such improvement is ongoing and KLI people are about as dee=
ply involved in the process as it's possible to get. All of the issues you =
raise have already been raised with the folks at Microsoft=2C and the Gramm=
arians are doing all we can to help them to make the system as good as it c=
an be.
> Till then=2C *sigh* expect journalists to come up with "Hov trek 'oH tlhe=
gh ngo'." for "Star Trek is money for old rope."
Any journalist who relies on a Bing translation for any language=2C and Kli=
ngon is no exception=2C deserves Rura Penthe.
QeS
=
--_1dba7cdd-cfd3-40f6-9d80-2245538beb67_
Content-Type: text/html; charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
<html>
<head>
<style><!--
.hmmessage P
{
margin:0px=3B
padding:0px
}
body.hmmessage
{
font-size: 12pt=3B
font-family:Calibri
}
--></style></head>
<body class=3D'hmmessage'><div dir=3D'ltr'>ghItlhpu' Alex=2C jatlh:<br>>=
=3B The Bing Klingon translator went online today. I've tried a few sample<=
br>>=3B sentences=2C rudimentary stuff anyone could do with a cursory exa=
mination<br>>=3B of TKD.<br>>=3B It needs work.<br><br>net Sovchu'.<br>=
<br>What you have to understand is like most machine translators these days=
it depends less upon a predetermined set of rules and more upon a database=
of considered-to-be-grammatical text from which it extrapolates the rules =
for itself. With languages like French it has titanic quantities of textual=
evidence from which to do so. Sometimes it still gets it wrong even in a m=
ajor language like French. Compared to French=2C Klingon as a language is o=
nly a step or two up from having no textual backing at all=2C and while mos=
t of the largest texts are in the Bing engine=2C many are perhaps not the m=
ost ideal for machine translation (Hamlet=2C while a wonderful text=2C is a=
relatively free retelling and so sentence-by-sentence comparisons are ofte=
n not fruitful - get Bing to translate "Qo'noS" and you'll see what I mean)=
.<br><br>In order to remedy this=2C I and several other Grammarians have sp=
ent the last little while actively participating in the testing and trainin=
g. We've been madly feeding it specimen sentences as well as monolingual an=
d parallel texts better suited to machine comparisons. Most of the publishe=
d Klingon texts have been fed into the engine=2C as well as a wide range of=
unpublished others. And please don't get the idea that the training has co=
mpleted just because it's gone live. In response to your "let's hope they g=
et improving"=2C work on such improvement is ongoing and KLI people are abo=
ut as deeply involved in the process as it's possible to get. All of the is=
sues you raise have already been raised with the folks at Microsoft=2C and =
the Grammarians are doing all we can to help them to make the system as goo=
d as it can be.<br><br>>=3B Till then=2C *sigh* expect journalists to com=
e up with "Hov trek 'oH tlhegh ngo'." for "Star Trek is money for old rope.=
"<br><br>Any journalist who relies on a Bing translation for any language=
=2C and Klingon is no exception=2C deserves Rura Penthe.<br><br>QeS<br> =
</div></body>
</html>=
--_1dba7cdd-cfd3-40f6-9d80-2245538beb67_--
--===============3967461171878601382==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
_______________________________________________
Tlhingan-hol mailing list
Tlhingan-hol@stodi.digitalkingdom.org
http://stodi.digitalkingdom.org/mailman/listinfo/tlhingan-hol
--===============3967461171878601382==--
