“How do I look?”
Without the benefit of other clues, this conversation could have gone any number of ways. Perhaps Ed Miliband was consulting Justine on his latest portrait with a bacon sandwich. And even then, who knows, maybe Miliband’s wife really loves his bacon buttie face. The English language is so delightfully, confusingly rich in meaning that “great” can mean anything from “wonderful” to “mediocre” to “awful”.
Which brings me to a great piece of news. The US Secret Service is looking to commission a Twitter sarcasm detector to improve its online social media surveillance. It is inviting analytics firms to bid for a five-year contract to monitor and analyse online trends and sentiment, and one of its requirements is the ability to “detect sarcasm and false positives” – presumably because when scanning the web for potential threats to national security, you don’t want to deploy police to the home of the tweeter who could “totally kill for a bacon sandwich right now.” (We know who you are.) And you need to be able to distinguish between “I love Al Qaeda” and “I love Al Qaeda”.
Humans are not actually very good at detecting sentiment in written language. Consider, for instance, that for over 500 years, scholars have tried to improve the way in which irony is expressed on paper – including by developing several irony marks, from backwards question marks to squiggly exclamation marks (more on which here).
Because tweets and text messages are too short to give much context, there is an even greater potential for misunderstanding. The emoticon might have helped a little, and yet over the years, a billion sentiments have been furnished with a winky face.
According to one study published in 2005 in the Journal of Personality and Social Psychology, respondents had a 50/50 chance of correctly judging the tone of an email – although they thought they were right 90 per cent of the time.
Still, the Secret Service can take heart. A number of attempts to develop computerised sarcasm detectors appear to have slightly better odds of being correct than the humans in the above study. In 2010, scientists at the Hebrew University in Jerusalem reported they had developed an algorithm to judge sarcasm that had a 77 per cent success rate at identifying snark in Amazon reviews. The French company Spotter claims to have an 80 per cent success rate at identifying sentiment correctly, and can work in 29 different languages. Its clients include the EU Commission, the Home Office and the Dubai Courts. Perhaps they will be offering their skills to the US as we speak.
If this new information on government surveillance gives you the heebie-jeebies, there is some cause for optimism. The US Secret Services computer technology might not be as advanced as you feared – as the BBC points out, the Secret Service requests that the software be compatible with Internet Explorer 8, a web browser released over five years ago. Good luck with it all, guys!