This reminds me of work done with the Enron Corpus. IIRC, it was used to cross reference statements made in court by the same people. When the emails revealed t

This reminds me of work done with the Enron Corpus. IIRC, it was used to cross reference statements made in court by the same people. When the emails revealed the lie made in court, there were often differences in linguistic structure between other statements. The example that sticks out in my head is that use of the passive voice was a key indicator of the likelihood of a statement being a lie. Of course these insights are statistical in nature, not definitive.