Abstract. Those who want to conceal the content of their communications can do so by replacing words that might trigger attention by other words or locutions that seem more ordinary. We address the problem of discovering such substitutions when the original and substitute words have the same natural frequency. We construct a number of measures, all of which search for local discontinuities in properties such as string and bag-of-words frequency. Each of these measures individually is a weak detector. However, we show that combining them produces a detector that is reasonably effective. 1 Motivation Terrorists and criminals must be aware of the possibility of interception whenever they communicate by phone or email. In particular, terrorists must be aware of systems such as Echelon [3] that examine a very large number of messages and select some for further analysis based on a watchlist of significant words. Given that it may not be possible to evade some examination of their messages, ...
SzeWang Fong, David B. Skillicorn, Dmitri Roussino