String splitting logic could be flawed and needs review

I received several issue reports by email from RaulMuroc, one of which is this:

> I noticed that the logic in breaking the string
> into the shorter segments had a flaw: it relied on sentences being ended with
> ". ", while in my text fragment I had a lot of "!" and "?" signs. That caused
> segments with broken sentences to be translated, which is not something
> Google API handles very well. So I fixed that logic too.

To my mind the module *should* be ok, in so far as it looks for the last sentence that ends with a ". " and works back from there, using it as a marker. In theory it shouldn't care about anything else, but Raul says there's a problem. Unfortunately he didn't provide a patch of code snip, so I'm hoping he will provide a patch for review here. =)