These kinds of analytical methods (i.e. word counting) work well in controlled experiments when you know you're going to be lied to by a subject, and also told

These kinds of analytical methods (i.e. word counting) work well in controlled experiments when you know you're going to be lied to by a subject, and also told the truth so you have a point of reference. In the real world, you don't know if you're being lied to. That's the entire point! You don't have any normalized data that says this is the number of words a person uses when they're telling the truth and when not, so the detection technique will be as inaccurate as this missing calibration info. And since there are wide variances between individuals in both general memory, recall, and "talkativeness" you have to develop a sample corpus per person. Also you need to interview the person on multiple days and randomize when you ask the questions so as not to introduce bias and to protect from ordering. It's easy to conduct a study like this and proclaim an obvious conclusion, but this means almost nothing for real-world application of the technique.