To really make a good translation you need an absurdly huge database of objects, it's meaning, properties, relationships and labels in each language. For exampl

To really make a good translation you need an absurdly huge database of objects, it's meaning, properties, relationships and labels in each language. For example, in japanese, you must use different prepositions if the object in question is round or flat. How do you know if the object being talked is round or flat? You just need an absurdly huge database with every little property possible, for every object in existance. No one is anywhere near a good complete database yet, like we do in our memories. That's why machine translation currently suck. But Google is building such a database, which they call the "Knowledge Graph". Which, among many other things, will allow for much better translation.