SETimes

» see setimes.com as it lived, on the wayback machine

~ the eulogy ~

SETimes told you who was paying for it. The About page, last revised in September 2009, carried the line plainly: "The Southeast European Times is sponsored by the US European Command, the joint military command responsible for US operations in 52 countries." No digging was required. The sponsorship sat in the footer navigation next to the podcast link.

The site went up in October 1999 under the name Balkan-Info. It became Balkan Times in May 2001 and SETimes in October 2002, which is the name it kept for the next twelve years. The final version of the front page listed thirteen countries across its top navigation, a span that reached from Albania at one end to Ukraine at the other and took in Turkey on the way. Stories were filed by paid correspondents rather than wire copy, and the topic menu sorted them under headings like Extremism and Integration.

What that produced was a newspaper with a visible center of gravity. The last homepage ran six stories about military cooperation and NATO accession before it got to anything else, including one about Montenegro trying to secure membership before the end of 2015 and another about Kosovo's security force becoming an army. Underneath, the URLs gave away the plumbing: every page sat under a path beginning with /cocoon/, after the Apache publishing framework the thing was built on.

It stopped quietly. The Internet Archive holds sixteen working captures of the homepage between February and the end of March 2015, all of them around 10 KB, and the last one is dated 29 March. After 31 March 2015 there are no captures of the site at all. By then the front page was already stale, still leading with an audio story about a Moldovan election held the previous November.

~ the dossier ~

~ ten languages, one story ~

The translation operation was the expensive part and the reason anyone still cites the site. A single article was published across the region's languages rather than summarized in them. Over its life the roster covered Albanian, Bosnian, Bulgarian, Croatian, English, Greek, Macedonian, Romanian, Serbian and Turkish, with Russian and Ukrainian added later for the Moldova and Ukraine desks. The 2009 About page counted eight languages; the 2015 navigation bar offered six toggles on the front page alone.

For a region where several of those languages had almost no professionally translated text available in bulk, that was an unusual amount of matched output, and a military command produced it continuously for a decade and a half without once intending to.

~ what the machines kept ~

Francis Tyers and Murat Serdar Alperen scraped it. Their SETimes corpus pairs roughly 200,000 sentences across eight Balkan languages plus English, and because every article existed in every language, the alignment came close to free. They used it to build 72 machine translation systems, one for each direction between the language pairs.

The corpus went into OPUS as SETIMES2 and stayed there. It turns up in the training data of neural machine translation papers years after the newsroom closed, and for several of those languages it was the standard benchmark for a long stretch, because there was nothing else of that size and quality to use.

Turkey sat on that masthead for sixteen years and I never once opened the site. I was reading Turkish news in Turkish from Turkish outlets, which is what you do when you live there, and a US military site covering my country as part of a Balkan regional package was never going to be on that list. I came at this from the other end entirely. I work in AI, and the SETimes corpus is a name you encounter in the literature long before you learn there was a newspaper attached to it.

~ the last front page ~

The final capture is an ordinary Sunday. The compass-rose logo sits at the top, saved out of Pixelmator in January 2009 according to the metadata still inside the JPEG. Below it are the six language toggles and six stories about NATO. Nothing on the page says goodbye or mentions that funding had run out. Someone simply stopped putting new articles into the Cocoon pipeline, and a few days later the server stopped answering.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »