Another challenge is that there is very little corpus data available for ASL (for one thing, there's no natural orthography so notation systems can vary wildly between studies and have only started coalescing recently). So you really can't build a statistical MT system the likes of Google Translate because there simply isn't the raw data to train it on.