On Measuring the Complexity of Code-Mixing