In what ways would you say its awkward? I'm really curious about its performance in languages that might not have shown up in its training data much. I will say, I've tried using it with a dialect, and it's good when it chooses to use the dialect, but slips into formal/prestige language extremely often.