Currently reading Larker & Zakolyukina "Detecting Deceptive Discussions in Conference Calls." Heard about it on NPR Morning Edition. They used a Naive Bayesian classifier to classify the conference call contributions of corporate CEOs and CFOs. Interestingly they used Pennebaker's LWIC word/phrase classifier as a domain specific dictionary builder. I'm only 6 pages in so I don'ut know anything beyond that, but my interested is piqued.