Arms, Algorithms, and Qualifying Clauses

“A well regulated Militia, being necessary to the security of a free State, the right of the people to keep and bear Arms, shall not be infringed.”

To say this sentence has found itself under the same powerful microscope as some foundational sentences in the Bible ─ “Thou shalt not kill” being the first that springs to mind ─ would not be an example of hyperbole.

Crossfire and Contention

Gun-control activists insist the first clause (“A well regulated Militia”) acts as a qualifier for the second (“the right of the people to keep and bear Arms”). For them the link between the two is obvious ─ the founding fathers wanted to guarantee the freedom of citizens against tyrannical governments by giving the people the right to form a militia, and you cannot form a militia without “the right to keep and bear Arms”. For gun-rights activists, however, the opposite is true: “the right of the people to keep and bear Arms” just happens to find itself in this particular Amendment and is not fundamentally connected to the right to form a militia.

Now while it is true that forming a militia without the right to keep and bear arms would be difficult, taking on drones, nukes and tanks with any sort of militia would be impossible. The advent of these modern weapons in the arsenal of the US military should have destroyed any sentiments about a militia, and along with it “the right of the people to keep and bear Arms”. Or so the gun-control argument goes. But the point of this blog is to describe how algorithms have found their way into the court and why they belong in the toolboxes of translators too, so let’s get to that.

All rise

The last time US jurisprudence got serious about the gun issue was in 2008, when, during the course of DC v. Heller, the link between the “Militia” and “the right to keep and bear Arms” came under scrutiny in the Supreme Court. Oral and written arguments were tossed back and forth, and when the dust settled the link had been weighed and found too light. The late judge Antonin Scalia, who wrote the majority opinion, consulted Samuel Johnson’s dictionary from the 18th century, as well as period prose, to defend the conclusion that “to bear Arms” in the time of Jefferson did not imply a militia.

Algorithm Appeal

Now it’s all good and well to consult prose and dictionaries from a certain epoch, but it is simply beyond the powers of any judge (or panel of judges) to process the entire corpus of text generated in any given epoch. At least not if they plan to reach a verdict before the end of their own epoch (or “prehumously” for that matter). In practical terms they’re always going to end up with selective quotations, and selective quotations employed to prove a point is called cherry picking. Algorithms provide a much more powerful way to search for patterns of word use. Dennis Baron, a linguist at the University of Illinois Urbana-Champaign searched “bear Arms” in the Corpus of Founding Era American English (139 million words from 1760 to 1799) and the Corpus of Early Modern English (1.3 billion words from 1475 to 1800). “To bear Arms” appeared 1 500 times and only in a few cases not in the context of organised armed action.

Set the dials to 1787

In DC v. Heller Scalia’s attempt to “travel” to the 18th century to verify the context of “to bear Arms” failed because he didn’t employ the latest technology, a grave mistake for any time traveller. As translators we should heed the implications of Baron’s research. Any translator worth her salt relies on context to determine the meaning of words and phrases before attempting a translation. In the past translation choices could be defended with the help of dictionaries, prose, acquired knowledge of a culture, and substantiated with well-chosen quotes only. As Google expands its corpora, the time is near when any translation choice not backed up by the superior search skills of algorithms will always be open to the charge of cherry picking, especially where historico-cultural distances between the source and the target are involved. One small consolation is that the mistranslation buck will then stop somewhere between the (wo)man and the machine.

  • This blog is based on an article that was published in The Economist (June 9th -15th 2018).