What about a web of associations--like a tag cloud on steroids--for each word and grammar construct that act like priors in a bayesian process that modifies the

What about a web of associations--like a tag cloud on steroids--for each word and grammar construct that act like priors in a bayesian process that modifies the meaning of each atom based on context? It probably wouldn't be more than a few orders of magnitude harder than making a translate.google.com that's as good as the average bilingual.