11 September 2018 (Hania, Crete) – This past Wednesday The New York Times’s Opinion desk published an Op-Ed by an anonymous senior official in the Trump administration which slammed the U.S. president and has White House staffers, political pundits and journalists racing to try and uncover the writer’s identity. It created … SURPRISE! … a firestorm of political controversy. It brought up issues of sexism, officiating double standards and adverse playing conditions … oops. Sorry. That was the women’s final in the U.S. tennis open.
Back to that Op Ed.
So now we have “the race”: try and uncover the writer’s identity, adding to the search already underway for those who were sources for Washington Post journalist Bob Woodward’s upcoming tell-all book Fear: Trump in the White House. The mystery writer has sparked wild speculation, similar to the frenzy to learn the identity of Woodward and Carl Bernstein’s mysterious Watergate source, Deep Throat, who was later to be revealed to be FBI associate director Mark Felt. Trump himself, according to insiders (do these people ever stop babbling?) thinks the anonymous official works on national security issues or in the Justice Department.
I spent half-a-day on social media and it has been flooded with theories and counter-theories as to the author’s identity, with speculation ranging from Vice-President Mike Pence, Attorney General Jeff Sessions and White House counsellor Kellyanne Conway, to Trump’s own daughter, Ivanka, or his wife, Melania. Political observers have become amateur forensic linguists, as they pore over any possible stylistic clues in the piece that might give away the author’s identity. Just a few bits I pulled off of Facebook, Twitter, etc. (believe me, I could have listed 100+) :
- Newsweek: the senior official references “we” in the context of “Trump appointees,” suggesting the author could be a cabinet member.
- Many have zeroed in on one particular word — lodestar — used in the op-ed to praise the late Sen. John McCain. That word, say some of these new amateur sleuths, is one often used in speeches given by Pence, meaning it could be either the vice-president or possibly his speechwriter.
- Robert Leonard, a linguist at New York’s Hofstra University, who is often retained by defendants and prosecutors in criminal cases involving threats, plagiarism and libel: “A problem with public people is that a lot of their published work is edited, so it’s like mixing fingerprints or DNA. You don’t always know who the real author is.”
- Axiom: we have White House leakers that tell us the person who wrote the article purposely included certain words to cover his or her tracks: “I myself usually pay attention to other staffers’ idioms and use that in my background quotes. That throws the scent off me.”
- Many, many pundits said it must be National Intelligence Director Dan Coats. Key piece of evidence? “He worked very well with Colorado Senator Michael Bennet while in the Senate. That’s the brother of the NY Times opinion editor, James Bennet”.
Enter … technology.
I had to dig out my notes from my artificial intelligence program from last winter when we actually discussed authorship attribution, which is the science of inferring characteristics of the author from the characteristics of documents written by that author. As the instructor noted, this is a problem with a long history and a wide range of application. But the field has perhaps had less uptake and general acceptance than is its due. The discipline is quite successful, even in difficult cases involving small documents in unfamiliar and less studied languages.
And this week I was somewhat fortunate because I had access to the work of Efstahios Stamatotos who is a professor in the Department of Information and Communication Systems Engineering at the University of the Aegean in Samos, Greece. I got a “Master’s Class” in AAA (Automatic Authorship Attribution) and the complicated features of syntactic information, part-of-speech frequencies, use of rewrite rules, etc. and the Natural Language Processing (NLP) tools able to provide accurate measures. The use of NLP tools can also provide useful measures related to the specific procedure followed to analyze the text.
But I did not need to go that far in my contact database. Enter Johannes Scholtes, Chairman and Chief Strategy Officer of ZyLAB.
I have a particular warm spot for Johannes and I have had the opportunity to video interview him several times. In 2011 he introduced me to the keynote speaker at LegalTech, Gonzalo de Cesare, who at the time was the political advisor with the European Union Police Mission in Bosnia & Herzegovina. His keynote was entitled “United Nations War Crimes Investigations – Information Management in High-Volume Legal Proceedings.” He discussed the UN’s process for managing some of the largest and most complex cases in the world with information management and e-discovery management software (all provided by ZyLAB) for the Khmer Rouge trials, the criminal tribunals for Rwanda, and the trial involving Slobadan Milošević and the former Yugoslavia. That introduction led to further contacts at the International Criminal Court and led me to get involved in the e-discovery aspects of war crimes investigations. It also led me to get involved in “extreme e-discovery”: projects in Iran, Libya, and Nigeria (I have a post in progress).
Johannes has a suggestion. Polymath that he is, he wrote a paper in 2011 along with Freek Maes for the Benelux AI conference on Authorship Detection, a paper based on Freek’s MSc graduation project at Maastricht University. In that paper they described a 90% accurate method for authorship detection for texts from unknown authors based on machine learning. It would appear their work is more relevant than ever.
The paper morphed into a patent and you can read it my clicking here.
Don’t be put off by the technical aspects. It is an easy read. Or just scan the summary. But given most of you do this stuff for a living (this post has gone out to my artificial intelligence, cyber, e-discovery, IT tech, and media listservs) you’ll be fine.