Conference 7.286::digital

    	I haven't worked out yet how to use a Web browser, and with French
    as only a second language I would be a poor critic.
    
    	I would suggest posting your note also in ULYSSE::VALBONNE, and
    possibly in local notes files in Munich and Paris dedicated to their
    languages. I can find references if you don't have them. 
    Valbonne had (and probably still has) techies of every European 
    nationality, and a few others as well,
    but more than 100 will be leaving the company within the next couple of
    weeks. I don't know of a notes file dedicated to the Italian language,
    but I could give you electronic mail addresses of individual Italians.
    
    	I get the impression that most of JOYOFLEX readers (with Denis
    Maillard as a notable exception) are native English speakers.

             Tnx, Brian -- I just crossposted into ULYSSE::VALBONNE.

    Does this wonder s/w have the ability to translate between American and
    English too?  There are three aspects: the differnet spellings being
    the largest part; the unique words and phrases; and the idioms.
    
    My favourite mis-translation was something rendered into French as 'un
    boeuf couchant'.  It was a non-technical person's attempt at
    'bulldozer'.

    	I doubt if there will be in the near future anything that will
    replace manual translation.
    
    	On the other hand, a machine translation can at the least remind a
    human translator of word meanings, saving him having to check in a
    dictionary, save him typing, and reduce his job to debugging the
    classical machine translation errors of which there are examples in
    JOYOFLEX.

    ... and back again... e.g., my favorite examples from the classical
    AI/MT literature:
    
    "The spirit is willing but the flesh is weak" translated into some
    other language, and thence back to English... ==>
    
                    "The vodka's OK but the meat is rotten"
    
    and my alltime fave
    
    "Out of sight, out of mind" getting the same treatment: 
    
                               "Invisible idiot"
    
    So yeah, there's still a large r�le for people to play in USING
    mechanical translation tools... :-)
    
    Dan
    

    I had a quick look at the German translation. German is only my second
    language, but I couldn't help laughing... 
    
    Seriously, there are a few sentences which are almost correct, at least
    understandable, but when Al Gore is translated into 'AL Bahn' (means
    roughly AL Railroad), the meaning of the sentence suffers a bit...
    but there are quite a few plain errors.
    
    All in all, if I didn't have the original, I probably would have
    understood no more than half of it.

    ...we had the best laugh in our department here in Munich when reading
    the German translation: never heard of "ranch-clusters" before.
    
    I am really glad that this in internal only...
    
    /Henning
    

    Same remark as .6 but for the French version : without the english
    text, you hardly understand 50% of what's in there.
    
    Nice errors : 
    
    	"Digital Equipment Corporation" 
    	 becomes 
    	"La soci�t� d'�quipement num�rique"
    
    	Someone named "Dan Cross" becomes "La croix Dan"
        Another named "Al Gore" becomes "le godet d'Al"
    
    	"...one of the systems--the one running DEC OSF/1..."
        becomes
    	"... le OSF � un dirigeant DEC/1..."
    
    	and a lot more ...
    
    If the purpose of this is to be funny, then the goal is reached.
    Otherwise, there is still work to be done ...

    Yep, even I (who am only semi-"fluent" in French and "can get by" in
    German, Spanish, & Italian) saw the humor in the translations,
    especially around the proper names -- _vide_ my basenote.
    
    However, please also note the leverage exemplified by the concept of
    the "jargon file."  Anyone who's ever used a spelling corrector that
    could load such a file, or which can build one interactively with you,
    knows the irritation and humor than can spring from its reactions to
    unrecognized words, and even better the laffs that you can get when a
    benighted spelling-corrector suggests nonsensical alternatives to
    unrecognized tokens.  
    
    You will also remember the feeling of relief & power you get when you
    tell your "jargon file constructor" that you really DO mean it when you
    type in "Frammisization" because the name of your new company is the
    "International Amalgamated Association for the Frammisization of
    Transport."  Once you tell it that Frammisization is OK, and that it
    needn't bother to suggest that what you REALLY might have meant was
    "Framing nationalization", you get happier real quick.
    
    Now, in the basenote, I ask(ed) you to imagine that the obvious tokens
    in this area of discourse (Digital Equipment Corporation, OSF/1,
    Internet, hypertext, etc.) are OK just as is, or are translated THUS
    into French, German, etc., ... and that Dan Cross, Al Gore, Gail Grant
    etc. are names as common as Bill Gates and Bob Palmer and are OK as
    they stand in this jargon file, how many of the tokens and laffs we
    were sharing go away?
    
    I can recall the word "burgeoning" not being in the generic translation
    table.  
    
    Which others need adding ...?
    
    Given these observations, I am *STILL not* claiming that the accepta-
    bility rating is "fine as is."  
    
    Perhaps it IS, for a very limited area of the marketplace, to be
    created later, perhaps...
    
    But I would ask you to imagine that it's a few months down the
    timeline, and the jargon-files are tracking about 98% of the
    specific-to-DIGITAL stuff... as the frontier of DECjargon moves
    inexorably forward...  And imagine that you have the job of producing
    polished final copy of all of DIGITAL's PR blurbs...  
    
    Do you want to do them all _de novo_ or would you want LOGOS's product?  
    
    Would you want to use someone else's similar product (which one, why?)?  
    
    If you were an end-user, would you be willing to pay for a
    business-level RAW translation, assuming that the jargon file were in
    better shape than the NONEXISTENT one we're using here?  
    
    How much more would you be willing to pay for a human-polished version? 
    What's the tradeoff between time and accuracy?  Etc., etc. ... :-)
    
    I'd be interested in comments along those lines...  I will of course
    stipulate that the raw stuff is pretty amusing!!  :-)
    
    Dan                      
    

I speak and write French and German fluently and have done translations
into and from both languages. After reviewing the samples, it is obvious
to me that this software is great for translation offices, but should not
be used standalone. If I was a translator, the software would provide
a great starting point, but that is all. Humans need to stay in the loop.

Marc

    I didn't want to go to too much detail - but there are quite a few
    places where I think a 'jargon' file just doesn't cut it. I'm certainly
    not an expert in automatic translation, but the problem simply seems to
    be that the program is not able to decide in what context various
    homonyms are used, and often chooses the wrong translation.
    
    A good example is 'in echange for this service' which is translated as
    "in B�rse f�r diesen Dienst" in German (the other languages are
    similar). Yes, 'B�rse' is exhange when we're talking about a stock
    exchange, but in this context, the translation simply doesn't make any
    sense (it's not even funny - just gibberish). The translation of "Put
    wheels on it..." is total nonsense. These are just a couple of
    examples.
    
    I wouldn't want to make any of these public without previous human
    intervention.
    

    Personally, I'd have to doubt the competency of a translator (and
    the company that produces it) if it can't recognize a "standard"
    name (eg "Joe Smith") and know enough not to translate it.  How 
    often do you see a string of the form: Capitalized-common-first-name
    followed by Capitalized-whatever, in a spot in a sentence where a
    noun is expected, and NOT have it be a persons name?

Dan, could you perhaps use this as Web-entertainment for
the masses?    The idea is, we'd feed in names of our
colleagues, and be amused with the inapproriateness
of the translation into some intermediate language
and then back to the language of the submittor.

  :-)

    The following is a raw machine translation, from Japanese to English. 
    I added a few carriage returns and heads for uh, clarity. 
    
    
                   Yuji no Monogatari  ->  The Tale of Yuji

    
In which Yuji endures summer camp:
----------------------------------

Same grade life, 40 persons joined at the time of a first-year boy, and is 10 
persons at the time of 3th grade. Yuji fell because of the year when Yuji
joined, summer lodging together of hell. It was carried by an ambulance.

Half of a new admission into a club member resigned by the summer lodging 
together. But, Yuji was not resigned. We thought that it was lacking in spirit
resign by such a thing. 


In which Yuji and Shinsuke join the rugby team:
-----------------------------------------------

The Shinsuke which came over at rugby in a new admission into a club member
inside with Yuji from a junior high school was. A position of Shinsuke was
stand OFF. A line was already produced on BAKKS and Shinsuke did not turn a
ball in BAKKS and even a condition applicable in open attack sometimes did a
kick.

But, when doing GAIN by a kick within an opponent camp, there was, and when not
doing it, there was. Rugby is skillful as for Shinsuke. But, a teammate was not
trusted. Promote a game alone it was ignorant it was done. Confidence from a
teammate has not been also secured. The position where horsepower of a forward
is decided about lock. A linear out of ZUSHI KOUKOU is not FUKKA and franc car 
throws into a ball from a touch line. Franc car and Yuji were unfriendly. 

But, Yuji and a rule rate of a ball of a linear out of combination of franc
car were 100\%. 


In which Yuji and his fellow students have a food fight:
--------------------------------------------------------

Showa 62 years, YUJI-TACHI amounted to 2th grade. They finished and retired at
3 years ginger last game in January of the year when they came insight.

Following it combined with a retiring expression, Yuji's teammates mixed a
fermented soybean and Indian ink with a senior student by and was about to show
a heart of senior student affection. YUJI's teammates postponed but to use a
fermented soybean by self-control of a lower-grade boy of a fermented soybean.

Though a lower-grade boy did self-control, the senior student who discovered a
fermented soybean does not do self-control. The senior student who was hung
Indian ink went out for a counterattack by a fermented soybean. It was Yuji to
be performed. A fermented soybean was poured upon on a head. The entire staff
saw and laughed at Yuji about both a senior student and a lower-grade boy.
Hanging a fermented soybean from a head, Yuji also laughed. 

They do not laugh. 

When a man meets a tragic thing, or why they laugh. Fermented soybean KUSA
{smell} and tenacity and NURU-NURU {scrubbing} did not readily take it from a
head of Yuji. There was only one piece about a fermented soybean. The thing
that it was covered with a fermented soybean from a head concerning history of
a ZUSHI KOUKOU rugby section is not merely Yuji.


In which Yuji becomes CAPTAIN of the team:
------------------------------------------

Doing it 3 years ginger retirement, they must choose new CAPTAIN from 2th
grade, and became. A retirement did it, and utterance power mingle the 3th
grade which was maintained still, and there was CAPTAIN election conference.
Anyone thought that Shinsuke would be CAPTAIN about a lower-grade boy. 

But, it was not Shinsuke to be nominated by CAPTAIN -- it was Yuji!

Yuji must command practice as CAPTAIN, and became. There was "they thought a
practice menu" in terms of Shinsuke to that place, and to say sometimes. 

Yuji was hard to perform it. 


An injury to Shinsuke, after eating noodles:
--------------------------------------------

Shinsuke was injured on a leg by a traffic accident in July when they refrained
from summer lodging together in imminence. They happened an intersection for
Chinese noodles roofs from a school. 

After practice ended, they went to Chinese noodles roofs. It was the tolerable
luxury which had been partly changed habit. The entire staff goes be able to
not the pocket money that a high school student was fixed every day. 

While having GAYA-GAYA unnecessary a talk, they walk, and has turned to the
left in the road a shopping center ends a shopping center in front of a station
of Zushi, and which continues to a house town. While bending, a road was
crossed. 

The car which came from the back was about to do it a left time. A building
becomes and does not look like a blind from an operational person about a
figure of Yuji and his teammates. An operational person turned a handle in the
left, and entered left time action. It is done a left time a new road. Yuji and
his teammates was before eyes. They stepped on a sudden brake. Yuji and his
teammates ran away. A front wheel of the car which was moved by power of
inertia stepped on shell of a right leg of Shinsuke. 

About Shinsuke practice for 2 months. 


They reach the playoffs:
------------------------

And Kanagawa prefectural great meeting of Showa 63 years. It is a great meeting
at the end of the Yuji and his teammates which amounted to 3th grade. They win
the championship in this great meeting, and in order to do rugby in a flower
garden, Yuji and his teammates has been practiced. In order to win the
championship at everything time, all energy in this great meeting for 3 years,
it has been dedicated. Yuji and his team was burning. They won through at a
battle once. 

And a battle twice!

If prefectural Zushi high school of this year overcame an opponent of a battle
twice, the ability which could win through to a final contest by a tournament
was had. YUJI-TACHI was fighting bravely. It left it of latter half, and was
overtaken on a tie by 10. It has turned into time sharpness draw. In any case,
a score is wanted. Prefectural ZUSHI-KOUKOU got a penal - kick. And, Shinsuke
did a kick from a position of 20 meters within an opponent camp. 

Loading a lower-grade boy of reserve, expectation of OB, a ball went away from
the ground at thoughts being of prefectural Zushi high school five. A ball
became disconnected from a goal mailbox. 

After all, a game was parted a tie. A winner depends on a lottery in conformity
with history of rugby and a tradition. Even though they say with a lottery, if
they say to an existence condition, it is KUJIBIKI. The direction where the
direction there are "O" and a label of "X",  and where it was pulled with "O"
subtracted "X" by winning is defeated. It is Yuji of CAPTAIN to pull Kuji as
Zushi representative.

Yuji pulled Kuji. A label of "X" was grasped on a hand. Yuji grasped a label of
"X" and saw a teammate. Yuji was laughing. All things did and came to life
again in a head of the Yuji which begin to walk in an one moment for 3 years.
It recurred in a head it was various and they were vivid SOU and a start made
Yuji confused. Address Yuji to a teammate it was ignorant it was done. 

But, it could not be said anything. A tear appeared instead of a word. 

When returning to a place of a teammate, Yuji was crying. And, rugby has ended
for 3 years of Yuji. 

Newpage game contents and psychological conditions of the time detailed.

=============================   THE END   ===================================

    Neil (.14) clearly demonstrates that .10 is wrong - machine translation
    is ideal!  Why would human intervention be needed?  ;-)  --S

    So, wuz it good for you, too???
    
    :-)
    

    I know what .14 (English version of Japanese text) reminds me of.
    Madlibs. Remember those few-paragraph stories/articles/excerpts of
    famous text which were missing key words? Underneath the blank for each
    missing word was the part of speech which had been omitted. Noun. Verb.
    Exclamation. etc. One person would look at the madlib, and ask the
    rest of the people to suggest a noun, a verb, an exclamation, then
    write these in the blanks. When read back, the story was supposed to be
    funny. "To be or not to be. That is the armadillo." Ha ha ha.
    
    Well, reading .14, I thought it had been created by the same method.

I tried reading the Italian translation without looking at the English 
original and ended up going nowhere fast; bits were understandable, and it 
might have been a useful starting point for a human translator, but that
was about it.

You could try posting in GVPROD::ITALIANO to get some more opinions.

Antonio

    Great idea; I just took your suggestion, Antonio.  If you'd be so kind,
    would you follow me there & make your same suggestion (reading the
    Italiano first to see if it makes any sense) to the others?  Thanks!
    
    Ciao � Grazie!! :-)
    
    Dan
    

    Dan, 
    
    Just a thought, but have you considered the time it will take to
    actually set up the "jargon list". 
    
    The reason why I am mentioning this is because I was asked to evaluate a 
    PC based translation package (English - French - English). I can't remember
    exactly what it was called, "Power Globe" or something like that. Like 
    you I elicited responses from native French speakers on the results I 
    obtained for a small piece of text.... after the laughter subsided, it was 
    obvious that some tinkering around was necessary if the program was to 
    live up to claims of "98% translation accuracy". 
    
    I investigated a bit further and I found two major stumbling blocks:
    
    1: The program would only accept files written in, 'scuse the language 
       here, a bog-standard text editor, in this case, WRITE! This was not
       acceptable as who uses this editor? I found that the time that would 
       be needed to prepare the source information into this
       "un/acceptable" format would be prohibitive.
              
    2: The "jargon list" itself - to set this up properly you need native 
       language speakers, yes human translators. You might want to consider 
       the time/cost this will take to set this up effectively. In my case, I
       found we would need two native speakers dedicated to this task for a
       solid month if we were to expect anything approaching accuracy from 
       the program and the program setup was not exactly intutive. 
    
    Dan, I am not suggesting that the software you are evaluating will have 
    exactly the same problems as the software I evaluated, but you might
    like to consider these points as it is clear that some work will have 
    to be done before you would dare to release material translated this 
    way externally!
    
    Cheers,
    Elaine.         

    For those who whish to look further at Languages and everything from
    On-Line dictionaries to Commercial translation services may find the
    following page of interest.
    
    http://www.willamette.edu/~tjones/Language-Page.html
    
    Kevin

    I skimmed through the 4 translations listed in .0 and didn't really
    have much trouble figuring out what any were saying except in a couple
    of instances and in the case of most of the names...
    
    As already stated here I think it depends a lot on your objectives. 
    There is hardly a sentence in any of the text which could be left as
    is.  Even the ones which are obviously understandable are awkward. 
    Still a human translator could certainly reformat them a lot faster
    than translate them.  Assuming he is familiar with the subject then it
    should be apparent where the mistranslations have occurred and he can
    either deduce what it should have been or look it up in the original
    text.
    
    hth,
    
    John

    
    >As part of my work as a technocatalyst in Rose Ann Giordano's Internet
    >Business Group (part of Bill Strecker's Advanced Technology Group),
    >I've been doing some informal evaluation of the mechanical translation
    >engine produced by an outfit called Logos Corporation.  
    
    FWIW, International Systems Engineering did indepth evaluation of
    several automated and semiautomated translation (MT) systems back in the
    late 80s, and bought licenses for ALPS, which was the front runner at
    that time.  I suspect that much of this info is still out there
    somewhere.  The most useful data from the studies are readability
    and cost evaluations of cleaning up MT, compared with human
    translation.  
    
    (Basically, what ISE found was that significantly lower cost investment
    in pre-translation editing resulted in much better MT output and was
    far cheaper than having X native language speakers cleaning up the
    output.)
    
    They also started joint project with a Dutch company (BKB?)
    which was intended to deal with the problem of "jargon dictionaries"
    in that the system "learned" jargon.
    
    As for alternatives, Ford Motor Co. uses the Alpnet system with a high
    degree of success in that all their technical ECO information is
    translated and distributed globally.  Alpnet identifies terms/phrases
    that are unique to the client and automatically uses an approved
    translation.  Ford claims that they spend about $250,000 p.a.
    "maintaining" the system.  Systran and Fujitsu have similar
    technologies.  My data is a bit old, but these systems all had similar
    costs back in 1989.
    
    Colin
    
    
    
    

    Thanks for the feedback.  I'd appreciate any reader with knowledge of
    International Systems Engineering giving me a pointer to this info, or
    to someone who might know someone...