This seems to be one specific category of lying detection (new training data with no prior training). This article doesn't seem to discuss machine learning tec

This seems to be one specific category of lying detection (new training data with no prior training). This article doesn't seem to discuss machine learning techniques. For example what if we could have someone talk about "nothing" for x hours for a training set, then interrogate them. That's essentially what lie detection machines do when they ask a series of baseline questions. But heart rate is the only input feature they use. We could use word frequency, spoken word speed, sentence patterns, and all the rich feature sets of ML. For those who have watched the popularity of deception games like Among Us, Secret Hitler, Werewolf, etc, there are verbal and non-verbal cues we all use. And playing with the same players iteratively yields a lot of meta sussing. For example: "You sound guilty because you're usually calm when you're guilty and accused of being imposter. Right now you're excitedly defending yourself"