How linguistic software helps companies catch crooks

Mind your language

Corporate fraud

A wise thief tries not to draw attention to himself

IN THE film “Superman 3”, a lowly computer programmer (played by Richard Pryor, pictured) embezzles a fat wad of money from his employer. The boss laments that it will be hard to catch the thief, because “he won’t do a thing to call attention to himself.

The traditional way to snare them is to hire an accountant to scrutinise accounts for anomalies. But this is like looking for a contact lens in a snowdrift. So firms are turning to linguistic software to narrow the search.

Rip-offs tend to occur in what gumshoes call the “fraud triangle”: where incentive, rationalisation and opportunity meet. To spot staff with the incentive to steal (over and above the obvious fact that money is quite useful), anti-fraud software scans e-mails for evidence of money troubles. Phrases like “under the gun” and “make sales quota” can indicate that an employee is desperate for a bit extra.

Spotting rationalisation is harder. One technique is to identify those who seem unhappy about their jobs, since some may rationalise wrongdoing by telling themselves that their employer is an evil corporation that deserves to be ripped off.

Ernst & Young (E&Y), a consultancy, offers software that purports to show an employee’s emotional state over time: spikes in trend-lines reading “confused”, “secretive” or “angry” help investigators know whose e-mail to check, and when. Other software can help firms find potential malefactors moronic enough to gripe online, says Jean-François Legault of Deloitte, another consultancy.

To work well, linguistic software must adjust to the way different people talk.

Dick Oehrle, the chief linguist on the project, explains how it works. First, the algorithm digests a big bundle of e-mails to get used to employees’ language. Then human lawyers code the same e-mails, sorting things as irrelevant, relevant or serious. The human feedback and the computers’ results are then reconciled, so the system gets smarter. Mr Oehrle says the lawyers also learn from the computers (presumably such things as empathy and the difference between right and wrong).

To find employees with the opportunity to steal, the software looks for what snoops call “out of band” events: messages such as “call my mobile” or “come by my office” suggest a desire to talk without being overheard. E-mails between an employee and an outsider that contain the words “beer”, “Facebook” or “evening” can suggest a personal relationship.