The "Twitter detectives" who track down trolls and killers online

In 2010 Jamie Starbuck murdered his wife Debbie eight days after they married, and embarked on a round-the-world trip paid for by her savings.

For two years he masqueraded as his dead wife, sending texts and emails to friends and family.

When they became suspicious, police turned to a very specialised team of detectives - who find clues to identify criminals in online chat, emails and 140-letter Twitter messages.

Sherlock Holmes could tell from the briefest glance at a handwritten letter whether the author was a criminal or an aristocrat - but the Centre for Forensic Linguistics at Aston University in Birmingham can identify people without paper or ink.

People’s use of capitals, their choice of words and sentence structure can be as distinctive as any fingerprint - and the technology could be used to track anyone from killers to Twitter trolls who abuse people anonymously.


                               [Samsung "smart watch" kicks off this year's big tech trend]

When police became suspicious that Starbuck’s messages might be a smokescreen, they turned to Dr Tim Grant, director of the centre, and his colleague Dr Jack Grieve, to analyse Debbie’s messages.

‘Jack and I looked at material we knew to be hers and material we knew was Jamie’s. We built up a list of differences and distinctive features, then looked at the disputed messages. We could pinpoint a date when the writing style changed that turned out to match the date on which police suspected murder.’

Starbuck was sentenced to at least 30 years in prison in May this year.

Their work has been used in dozens of criminal trials to decipher conspiracies and identify suspects using merely their texts, emails or online postings.

The number of identifying features in ones writing can be surprising.

Dr Grant and Dr Grieve found as many as 15 factors that showed Jamie Starbuck was impersonating his dead wife.

‘Among them was the fact that she commonly had a more complicated sentence structure than he did. And in the opening and closing of emails, they had very different styles,’ said Dr Grant.

In another case, Dr Grant was asked to help police when a part-time teacher in his 40s was found to have indecent pictures of children on his computer at the secondary school where he worked.

The man, who cannot be named, claimed that students at the school had been using his computer and posing as him to get him in trouble.

By comparing his chat logs and forum posts with messages purporting to show teenagers and young men conversing, Dr Grant was able to show that he had in fact been role-playing as teenage girls or men in their early twenties to attract young girls.

‘There are several things about how we write that we don’t have much control over,’ explained Dr Grant. ‘Things like your average word length, sentence length, capitalisation and the way you use essential vocabulary like personal pronouns – he, she, that sort of thing – can help us recognise your distinctive style.’

Dr Grant has also been instrumental in jailing a terrorist plotter. In 2004, Dhiren Barot, an Al Qaeda chief working in the UK was arrested on suspicion of masterminding a plot to hide gas cylinders inside limousines and detonate them in underground car parks. ‘The police were very interested in Barot, but they only had intelligence on him, rather than evidence.’

The case hinged on proving that Barot was the author of a document recovered on a laptop outlining the plans. ‘He was British born and bred, so his language wasn’t that distinctive. But whenever he was communicating with non-native English speakers, he would provide ‘glosses’, or definitions of certain words in brackets…he’d write “this stuff is poisonous (toxic)”, in emails, and in the subtitles accompanying videos of his targets. That was how we proved it was him, by matching that style to the work on the laptop.’

Recent outbursts of online abuse have seen several tweeters arrested for sending threats to prominent feminist campaigners. A common tactic for twitter ‘trolls’ is to open multiple accounts in a short space of time, to avoid being blocked.

Dr Grant warns that this may not be as foolproof as it seems, and could leave them open to identification from their language alone. ‘If the abuse is distinctive, and it often is, you could tie multiple accounts together quite easily.’

Similarly, claiming your account was ‘hacked’ would be no protection for trolls if it could be shown that the style of the abusive tweets matched that of your normal writing.

Dr Krystof Kredens, also from Aston University and a member of the International Association of Forensic Linguists, adds that your tweets might reveal more than you intend: ‘You can tweet from desktop or a phone – and the modes of speech are different. The mode affects the message – you get spelling variations from autocomplete apps, and people choose different abbreviations.’ It can also be possible to tell whether a message was written on a phone with a QWERTY keyboard, like a Blackberry, a touchscreen or old-fashioned numeric keypad.

‘Twenty years ago, what you wrote was your own. Now you leave traces on Facebook, on Twitter and other social networks. Your emails are private but with things like [the NSA’s surveillance program] PRISM out there, you’re entering an ethically loaded area.’

It’s not just the police that call on the services of forensic linguists. ‘We’ve worked with prosecutors, corporations, private individuals - and some institutions that we’d rather not talk about’ said Dr Kredens.

 ‘A common example is someone sending letters warning people off dealing with a certain company, who turns out to be an employee of that company. They may claim it wasn’t them, but we can usually prove otherwise. Then you’re looking at an employment tribunal.’

So, be careful the next time you go online to rant about your boss...