You just answered a question I pondered earlier today as I watched the MacOS X demo I linked to above somewhere. And that is to the question if a native speaker would recognize the synthetic speech as unnatural. As the video progresses it lands on international languages such as Arabic and Chinese among other things. To my ear and my best memory of how those languages sound the synthetic versions came out pretty convincing. It made me think about a concept I've had about a test to language recognition: if a native speaker were to speak gibberish in their native tongue would an outsider to the language recognize it as gibberish or assume it was vocabulary in that language.