Secret Service Stumped By Sarcasm

If you want to fool the Secret Service, it appears sarcasm might be the way to go. The agency is looking for software that can figure out when social media users don’t mean exactly what they say.

In a procurement document, spotted by the BBC, the agency is asking for a commercial firm to provide a social media software analytics tool. It wants something to automate the process of monitoring social media posts and then putting together reports on patterns of activity. Twitter is the only service specifically mentioned in the brief.

Such analysis can be productive. Researchers have suggested that, to some extent at least, it’s possible to predict an increase in civil unrest by looking for an increase in use of politically-charged terms, particularly when the two sides of a political dispute have different words for the same thing.

While most of the requirements listed in the procurement document are about purely technical measures, it does call for the task of sentiment analysis. That’s an attempt to go beyond the literal words of social media posts and instead try to distinguish patterns of tone.

However, the brief acknowledges one limitation in this tactic and specifically asks for the “ability to detect sarcasm and false positives.”

If such a technology is possible, it would certainly save wasted time and effort. It sounds like something of a challenge as it would require a degree of empathy with the writer’s motivations, something not traditionally associated with algorithms.

In fact judging tone can be a particularly difficult task for some humans when law enforcement and officialdom gets involved. That was demonstrated by a case in which it took more than two years for the British legal system to clear a man originally convicted of sending a message of a “menacing character” after posting a vow to blow up an airport that many readers believed was clearly not a sincere threat.