From the world of machine translation

This area has been established to allow you, our visitors and contributors, to get to know one another a bit better, or to discuss subjects of general interest, without feeling obliged to restrict your postings to language-related topics. But we draw the line at floccinaucinihilivilification.
Post Reply

From the world of machine translation

Post by Phil White » Mon Jul 18, 2005 7:44 pm

Most of the witty examples you will see on the Web are probably apocryphal ("out of sight, out of mind" = "blind and mad"), but machine translation is often good for a laugh if you don't have to try to use it.

I was loosely involved in one of the major projects of the eighties (which has since been dropped). There were many gems which were occasionally circulated by the team. One of my favourites displayed a level of self-referential humour which endeared the machine to me ever after. The German "leicht verständliche Dokumentation" was rendered as "slightly understandable documentation". It should have been "easily understandable documentation".
ACCESS_POST_ACTIONS
Signature: Phil White
Non sum felix lepus

From the world of machine translation

Post by Miia » Thu Jul 28, 2005 1:22 pm

Machine translation is rarely working properly simply because machines cannot consider the sentence as a whole, but translate only one word at a time. One of the most confusing sites I have found was the site of hotel us (http://www.hotelus.com) that has a Finnish version of the page. The so called Finnish language has no sense in it.

One example could be condos, translated for some reason to mean Declared Confiscated! The program has not even understood the difference between post- and prepositions, which makes it kinda hard to read (for example, talossa (in a house) is written -ssa talo).

On the other hand, this is very encouriging. I think there will still be use of us human translators for a long time :) BTW, I'd like to know have the other language versions done any better? The Swedish was totally confusing, and for the French version, my language skills are not good enough...
ACCESS_POST_ACTIONS

From the world of machine translation

Post by Phil White » Thu Jul 28, 2005 5:57 pm

The translations were almost certainly done by Babelfish (the Google translation engine), although much of the German (if not all) of it was clearly done by a flesh-and-blood translator. Some bits are still very mucky, though.

Although there are some products that make a fair stab at clearly, simply written texts, Babelfish is not one of them, and the disjointed nature of a website is not ideal for MT. Most of the more advanced products (i.e. not Babelfish) do attempt to take the context of the entire sentence, and often entire paragraphs and complete texts into account when attempting to render a text in a different language. They do not simply translate word-by-word.

My experience is that they fall down for two reasons
1) Despite the efforts of Chomsky and his disciples over the past decades, nobody has found an adequate (machine-compatible) means of describing the structure of a single language, let alone one that applies theoretically to all languages. Most of the engines that drive the machine translation programs base heavily on Chomsky's work on transformational grammar, dating back at least in part to the fifties and sixties, which he has now rejected in favour of a fundamentally different theory (government-and-binding). Until systematic methods of describing multiple languages are available, MT will never be a seriously effective tool.

2. It is very rare for any single word/phrase in one language to occupy an identical semantic space to that occupied by a single word/phrase in a different language (with the possible exceptions of names, proper names and similar designations). This makes it impossible to simply "map" one word to another without a clear understanding of the connotations of the word in context, and it is this "understanding" that is missing at present - and will remain missing until the scientists have worked out how humans map knowledge in the brain and then build machines to mimic this. (I'm not talking about the simple concept of neural networks, but about the fundamental and little-understood structures of human memory.)

No, we translators are safe for a very long time yet.
ACCESS_POST_ACTIONS
Signature: Phil White
Non sum felix lepus

ACCESS_END_OF_TOPIC
Post Reply