(Original) Auch bei längeren Texten hat uns die Qualität der Übersetzung überrascht. Lediglich bei komplizierten Nebensatzkonstruktionen liegt das Übersetzungsportal nicht immer richtig. Der Anwender erhält dann lediglich eine Aneinanderreihung der Wörter und muss sich mühsam aus den verschiedenen Bedeutungen die richtige heraussuchen. Kurze, einfache Sätze wurden dagegen meist verständlich übersetzt, wenn auch nicht immer idiomatisch korrekt.
(webtranslate version) The quality of the translation has surprised us also at longer texts. The translation portal isn’t always correctly located merely for complicated subordinate clause constructions. The user gets then merely a stringing together of the words and must the right with difficulty find themselves from the different meanings. Short, simple sentences were usually translated against this understandably if also not always idiomatically correct. There are more sophisticated versions for sale.
Babelfish / Google translate produces this:
(Babelfish/Google translate) Even with longer texts the quality of the translation surprised us. Only with complicated subordinate clause constructions the translation portal does not lie always correctly. The user receives then only a lining up of the words and must laboriously from the different meanings the correct pick itself out. Short, simple sentences were usually understandably translated against it if also not always idiomatisch correct.
The article emphasizes that webtranslate handles complex sentences well and has a good dictionary. The samples above show that Babelfish did not have idiomatisch in its dictionary. However, the results of the two programs are both OK, and Babelfish doesn’t have the 500-character limit.
We often make fun of machine translation. Of course machine translation is not bad – up to a point. For instance, it is useful for skimming texts in the Internet. It can be improved if its dictionary is enlarged or if, say, you define some terms as legal or economic and tell the program it is translating a legal or economic text – then it will translate Bank as bank, not bench (which will nearly always be right) and bar as Anwaltsstand, not Bar. Conversely, if you run a hotel and have your menu machine translated for your website, you are almost certain to fail. If you translate 2,000 menus, you may be able to automate the translation if you feed the right material in (it’s not easy to tell what is the right translation of a menu in a foreign language). But if you want to use MT in a firm, you will need to consider if it doesn’t cost you more to have for-publication or for-serious-understanding texts revised by human translators who would have been faster starting from scratch.