Marie Wilcox raced to save a language before its last native memory closed — the Library of Congress kept every public tweet and found that keeping everything, without choosing, can strand the kept thing just as surely
R813’s danger was scarcity: one person held a language, and the work was to get it out of her before it was gone. Here the danger runs the other way. In April 2010 the Library of Congress agreed with Twitter to receive the public tweets from the platform’s start in 2006 onward, and then the whole ongoing stream, as a record of a new form of public speech that was beginning to stand in for letters and journals. The Library normally collects selectively; for tweets, while social media was still new and its direction unknown, it made a deliberate exception and chose not to choose.
The exception filled up fast. The Library’s own January 2013 update put the archive at about 170 billion tweets, with the daily intake having risen from about 140 million a day in early 2011 to nearly half a billion a day by late 2012. Its stated first aims were modest and were met: acquire the 2006–2010 backfile, set up a secure daily intake, and organise the whole by date. What it could not do was let anyone into it. By that same update about 400 researchers had already asked to use the archive, for work ranging from citizen journalism to tracking vaccinations, and none had been given access: the tools for searching a pile that large, the Library said plainly, lagged behind the tools for making and sending the pile. Chronology had been supplied; reading had not.
Notice the exact shape of the drowning, because it is not simple bulk. First, kept is not the same as accessible: the collection sat embargoed, in 2013 and still in the Library’s December 2017 update, until access could be resolved in a way it called cost-effective and sustainable. Second, it was complete in one dimension and thinning in another: the Library received text only, while tweets were becoming more visual — images, video, and linked material it never held — so the longer the total capture ran, the larger the share of the record whose context was missing. Third, the remedy the Library finally adopted was not more storage or a better index retrofitted onto everything. Announced on 26 December 2017, effective 1 January 2018, it ended comprehensive collection and went back to acquiring tweets selectively, on a thematic and event-based basis, the way it collects websites. The twelve years already kept, 2006–2017, were to be preserved as they were. Choosing, postponed in 2010 as a precaution, turned out to have been part of the keeping all along.
One day, unchosen. A schematic toy, with its terms stated plainly. Each press adds one day of the stream at the late-2012 rate of about 500 million public tweets, none of them read, ranked, or thrown away. The reader works at one tweet a second, without sleep. There is no score; the only question is how far behind keeping, without choosing, puts a single future reader.
Wilcox’s keeping failed if she waited; this one faltered because it refused to. A dictionary made by one woman in a hurry can be opened by a stranger; an archive of at least 170 billion short messages, organised by date and opened to no one, asks its future knower to begin by drowning politely. Neither scarcity nor abundance is the lesson on its own. The lesson is the middle act both cases expose: someone, at some point, has to choose, rank, and discard — and a keeping that skips that act has not finished, however many shelves, or servers, it has filled.