Proposal talk:Preparing a translation machine

From Strategic Planning

Impact?

Multiple machine translations

I think it might be interesting to offer a web service with multiple machine translations. Translators would be able to display the recommended translations for every single sentence of pages hosted by Wikimedia translated with different systems (e.g. Moses, OpenLogos, Apertium). --Fasten 17:03, 9 November 2009 (UTC)[reply]

Parallel corpus

e.g. UN and EU texts. --Fasten 17:27, 9 November 2009 (UTC)[reply]


Might work on an exceedingly limited number of items written in bureaucratese - but as soon as you examine English usage, the number of totally contradictory translations possible (see the OED for the number of potential problems with words having a huge number of meanings like "fast"), the corpus needed to find parallel word usage becomes enormous. The only way we can possibly make a translation engine work is to restrict each language vocabulary to at most 5,000 words (say, a 5th grade vocabulary for each language). Each article to be translated would first be run through a filter to simplify language, and a human would likely have to make decisions about complex usages, idioms and adages. Collect 12:38, 10 April 2010 (UTC)[reply]

Today this Might work on an exceedingly limited number of items written in bureaucratese - but as soon as you examine English usage, the number of totally contradictory translations possible (see the OED for the number of potential problems with words having a huge number of meanings like "fast"), the corpus needed to find parallel word usage becomes enormous. The only way we can possibly make a translation engine work is to restrict each language vocabulary to at most 5,000 words (say, a 5th grade vocabulary for each language). Each article to be translated would first be run through a filter to simplify language, and a human would likely have to make decisions about complex usages, idioms and adages. As the ability of the translation machine improves over the years it should be able to cope with more complex texts. Some human assistance will probably always be needed but the translation machine should be able to ask an english speaker to clarify ambiguous passages.Filceolaire 13:21, 10 April 2010 (UTC)[reply]