Comment Difficulty with non-standard orthography (Score 2) 91
I work on Welsh-English machine translation, and have looked at doing this in the browser with a bookmarklet so that comments in Welsh can be read by non-Welsh speakers. The trouble is that people tend to use non-standard spelling etc in informal postings such as facebook, whereas the majority of parallel text available to train a statistical machine translation model tends to be formal language (government documents, press releases from business, etc).
Possibly this could be solved with two stages of translation, i.e. (Lang1 with informal spelling -> Lang1 with standard spelling -> Lang2). If this mapping is relatively straightforward (e.g. common spelling substitutions such as 'ough' -> 'u' or 'uff') then a statistical model might work quite well, if you could first split words into syllables with some rule-based algorithm.
Just a totally trivial, obvious thought about the application of statistical machine translation (sorry if I've pissed on some patent troll's livelihood, heh)