“A Little Practical Arithmetic”: Natural Language Processing and the Sport of Translation
by John Zuern

Paper delivered at the symposium Translation: Theory, Practice, Trope at the
University of Hawai‘i at M?noa in January 2007

In “Translation,” an animated electronic poem by the London-based poet John Cayley, passages from the works of Walter Benjamin and Marcel Proust are displayed against a black background, where they shift gradually, letter by letter, among the languages German, French, and English. Each change in a letter is accompanied by a musical tone, a corresponding “letter” from the computer-generated sonic “alphabet” devised by the composer Giles Perring. From time to time, selections from Benjamin’s “On Language as Such and on the Language of Man” and Proust’s In Search of Lost Time float to the surface of this roiling current of letterforms and are momentarily legible in one of the three languages, only to slip back into the eddies of cycling letters. “If a passage drowns in one language,” Cayley explains in his commentary on the piece, “it may surface in another.”

Cayley, who also works as a literary critic and a translator from Chinese, has attempted to capture in “Translation” something of the dicey night-sea voyage from one language into another that all texts must endure when they are translated. This journey is never simply a straightforward conveyance from source to target; it is always a transformation that entails not only the proverbial loss of “original” meaning but also the emergence of unexpected, unwarranted, parasitical, even anagrammatic meanings that flow back upstream toward the source, muddying the waters of its pristine originality, making it the “target” of its own translation.1 In assigning the task of translation to an algorithm—“Translation’s” letter-switching is governed by a computer program that scans the text and sequentially replaces characters uk essay writing services with others it deems visually or phonetically “similar”—Cayley also takes on the issue of “machine translation,” obliquely raising the question of what it means to outsource to computers the painstaking labor of translating texts written by human beings. What surfaces and drowns, not only in terms of an original meaning but also in terms of all the philosophical quandaries that translation entails, including questions of the ethic of translation, when the machine takes the helm?

I’ll be saying a little more about Cayley’s work later, but I’m going to turn first to another poem, one written by the Greek poet Yannis Ritsos sixty years ago in 1947 during the Greek Civil War, at the end of which, in 1949, Ritsos was imprisoned for four years for his participation in the Communist insurgency. The poem is called “Understanding” [Katanoisi]; it appears in the facing-page volume Ritsos in Parentheses, rendered into English by the eminent translator of Modern Greek poetry, Edmund Keeley. I’d like this poem to be in our minds as I make my few comments about translation and computation.

The poem begins with a concrete description of a Sunday morning street scene, rich in details: people greet each other, a bus pulls away, a worker plays a mouth organ, and “buttons on jackets gleam/Like scattered laughter.” The second stanza shifts to a much more abstract meditation on human connections and the importance of seeing oneself as part of a collectivity, an idea summed up in the phrases “[a] little addition/A little practical arithmetic, easily grasped” (25). In his specific designation of the “worker” and his appeal to collective understanding we can detect inflections of Ritsos’s political philosophy in the poem.

It happens that in 1947, the same year Ritsos completed “Understanding,” Warren Weaver, who during the Second World War had headed up the Applied Mathematics Panel of the United States government Office of Scientific Research and Development, sent a letter to his fellow mathematician Norbert Wiener at MIT proposing that computational processes might be used to translate between human languages. Weaver had been turned on to this idea by his panel’s wartime work in cryptography, enemy code-breaking, and in a now infamous analogy, he suggests to Wiener that we might think of “a book written in Chinese [as] simply a book written in English which was coded into the ‘Chinese code.’”3

Weaver had reason to expect Norbert Wiener to be exited about this proposal. Wiener’s own work on artillery guidance systems during the war had led to the foundation of a new field of applied mathematics, to which Wiener himself had given the name “cybernetics.”4 Wiener, however, was an accomplished student of languages, and actually knew some Chinese, and didn’t find the notion of an “English book in Chinese code” terribly convincing: “as to the problem of mechanical translation,” he writes back to Weaver, “I frankly am afraid the boundaries of words in different languages are too vague and the emotional and international connotations are too extensive to make any quasimechanical translation scheme very hopeful.” Though Wiener certainly recognized that languages could be described as systems of discrete variables governed by more or less regular rules, that is, something like “games,” for him human languages—which would soon come to be called “natural” languages to distinguish them from the “artificial” languages of computer programming—are in important respects, and these are Wiener’s words, “the reverse of mechanical.”5

Two years after his exchange with Wiener, an undeterred Weaver circulated a memorandum called “Translation” to some 200 members of the scientific community. That memo is now seen as the inaugural document for a number of fields in computer science, linguistics, and engineering that fall into the category of “natural language processing,” including some branches of artificial intelligence research, speech recognition and simulation, and machine translation or “computer-assisted” translation of “natural” languages.6

Anybody who’s played with one of the many free online translation engines will know that the basic problems Wiener pointed out in 1947 have by no means been overcome, and it’s also important to note that the wartime cryptographer’s view of translation as code-breaking is alive and well, for example, in the English-Arabic translation devices the U.S. military has been issuing to troops in the war in Iraq and, for example, in the program launched two years ago by the Defense Advanced Research Projects Agency—DARPA, the folks who brought us the Internet—called, believe it or not, Global Autonomous Language Exploitation (acronym GALE). The aim of GALE (and this is from DARPA’s own mission statement) is “to develop and apply computer software technologies to absorb, translate, analyze, and interpret huge volumes of speech and text in multiple languages, eliminating the need for linguists and analysts, and automatically providing relevant, concise, actionable information to military command and personnel in a timely fashion.”7 DARPA says “multiple languages” but they’re starting with Arabic and Chinese. This conscription into military service and even “weaponizing” of translation is one of the points of departure for Emily Apter’s recent book The Translation Zone: A New Comparative Literature.8

I suspect that many of us in the humanities generally fall in step with Norbert Wiener, Emily Apter, and many other critics who find the attempt to outsource the task of translation to an algorithm unconscionably and even dangerously reductive, for at its current best a translation program can do little more than “string together the most accurate synonyms by the most proximate syntax,” as Gayatri Spivak has put it. It continues to be difficult to come up with algorithms that can tell the difference between “playing by the rules” and “being a good sport,” (or, for that matter, being a bad sport)—a difference that seems crucial to sensitive and responsible translation.

At the same time, while it’s clear that any number of military and commercial ventures into automated translation have been far too quick to deploy shoddy translation algorithms based on impoverished models of human communication, I think we who want to argue against such initiatives can’t afford to be too quick to deplore algorithms as such as a kind of enemy code and to make the tactical error of dismissing them without attempting to decipher and to understand them and to understand the folks who create them.

If we look at what John Cayley is doing with the “ambient poetics” of his “Translation,” the piece seems in some ways to caricature machine translation: the algorithm Cayley developed for this project simply swaps individual letters in samples of texts by Walter Benjamin and Marcel Proust on the basis of more or less arbitrary visual and phonetic similarities from one character to the next—“n” for “r”, for example, “s” for “c.” The program parses and manipulates the original texts in this way so that they gradually come into and out of legibility. Cayley does not claim that this procedure is “actually” translation; his deliberately reductive algorithm is less a true language-processing engine than it is a poetic image—Cayley calls it “a transliteral morph”—that’s meant to emphasize the interdependence of our quite supple and fault-tolerant cognition of linguistic meaning and the comparatively rigid constraints of our inscription technologies.
Cayley’s “Translation” is a pretty simple example of the efforts of number of poets and fiction writers, among them Judd Morrissey and Lori Talley, Stephanie Strickland, and Noah Wardrip-Fruin who are trying to create a much more intimate intersection of computer programming and creative writing than we tend to see with more instrumental applications of natural language processing, So the first suggestion I want to make as I move to conclude is that the kind of hard technical and imaginative work good writers and good translators must do is not utterly remote from the kind of hard technical and imaginative work a good programmer must do, and that we might have much to gain from thinking more carefully about what logics these tasks have in common.

My second suggestion is that sometimes the flat-footed decisions of very stupid algorithms can serve as useful correctives to the elegant but potentially misleading decisions of very smart human translators. I want to go back now to the Ritsos poem “Understanding.” It’s hard to imagine anyone better equipped to translate Modern Greek poetry into English than Edmund Keeley. He spent years of his childhood in Greece, he is himself a creative writer, and he’s written a good deal of criticism reflecting on translation problems. But when we look at several of Keeley’s published statements on translating Yannis Ritsos, it becomes quite evident that Keeley is unable to see Ritsos’s commitment to Communism as anything more than an unfortunate and costly lapse of judgment.9 I want to look at one tiny example of where this lack of sympathy, not to say scorn, for the poet’s politics might be reflected in Keeley’s choices as a translator.

I want to draw your attention to a phrase in the poem, in the third-to-last line, “mia mikri praxi tis praktikis arithmitikis,” which Keeley renders as “a little practical arithmetic.” When I made the experiment of entering this phrase into the online translation engine at WorldLingo.com, the program returned “a small action of practical arithmetic.” The machine translation restores a word that Keeley drops: the word is praxi—praxis, action, act. The word can also carry the sense of an exercise or “problem,” like a “math problem” for schoolchildren. In the original, this noun praxi is echoed in the adjective praktikis, so when Keeley collapses the two he essentially disfigures Ritsos’s rhetorical figure of polyptoton, the deliberate repetition of a word in a different grammatical form. In fact, throughout his translation of the poems in this collection, Keeley appears to be “cleaning up” repetitions in Ritsos’ style—I’ve indicated several places in your copies of “Understanding”— “scattered laughter” in the second line, for example, would be literally “little laughs” [mikra yelia], the “little” echoed later in “a little practical arithmetic.” In doing this in “Understanding,” I think Keeley neutralizes, in subtle ways, the social and political philosophy Ritsos’ poem seems to be trying to work out in adhering to a conspicuous lexical asceticism as it presents simple verbal exchanges, basic mathematical operations, and the harmonics of music—phisarmonika is the Greek for “mouth organ”—to point to the range of symbolic systems that mediate human coexistence and to emphasize that however “natural” it may feel, that mediation entails a practice, an exertion, a learning. I’m in no way saying that WorldLingo.com can out-translate Edmund Keeley, only that in some instances the reductive logic of the dumb algorithm can expose meaningful formalizations in the original that the more expansive intelligence of the translator has obscured.

I’ll end by suggesting that something on the order of Keeley’s collapsing of “practice” into the “practical” in “a little practical arithmetic” occurs again and again when quasimechanical, quasiautonomous natural language processing techniques are enlisted to solve the problems of human understanding that are bound up with the problems of translation. As our co-existence with other human beings comes to be more and more mediated by computational processes, the work of restoring that effortful and always interested practice is likely to demand that we find ways, as researchers, writers, and teachers, to extend the skills of the translator across the divide between the verbal and the quantitative disciplines in the liberal arts. We have to brave that catwalk of logic that runs between dialectic, traditionally in third position in the old trivium after grammar and rhetoric, and arithmetic, traditionally in first position before geometry, astronomy, and music in the quadrivium. We can take from Ritsos’ poem a precise but not reductive formulation for such a translation in the interest of understanding across languages and disciplines: “strange / that you are able to listen and answer.”


1 John Cayley with Giles Perring, “Translation.” 2006. http://www.shadoof.net/in/. Accessed 22 January 2007. In Caley’s poem, between the original source and translated target, a third, virtual, performative text emerges and persists: the text of the act of translation itself, the absent “unmade moves” and unexplored avenues that cluster in the space between source and target. They represent translation’s shaky “aim,” always subject to a tragic hamartia.

2 Yannis Ritsos, Ritsos in Parentheses, trans. Edmund Keeley (Princeton: Princeton UP, 1979), 25.

3 Warren Weaver, “Translation,” in Nirenburg, Sergei, Harold Somers, and Yorick Wilks, Readings in Machine Translation (Cambridge, MA: MIT, 2003), 13-17, 16. In 1949, two years after this exchange, Weaver and Claude Shannon published The Mathematical Theory of Communication (1949). Rita Raley gives a compelling account of the history of machine translation in “Machine Translation and Global English,” The Yale Journal of Criticism 16.2 (2003), 291-313.

4 Wiener is the author of Cybernetics: or, Control and Communication in the Animal and the Machine (Cambridge, MA: MIT P, 1961).

5 Qtd. in Weaver, “Translation.”

6 The first conference on machine translation was convened in 1952.

7 Defense Advanced Research Projects Agency, Global Autonomous Language Exploitation, http://www.darpa.mil/ipto/Programs/gale/index.htm. Accessed 20 January 2007. The Program Manager is Joseph Olive, who in addition to his work in text-to-speech recognition software is a composer of digital music.

8 Apter writes, “Drawing on Carl von Clausewitz’s ever-serviceable dictum ‘War is a mere continuation of policy by other means,’ I would maintain that war is the continuation of extreme mistranslation or disagreement by other means. War is, in other words, a condition of nontranslatability or translation failure at its most violent peak.” The Translation Zone: A New Comparative Literature (Princeton: Princeton UP, 2006), 15-16.

9 See, for example, “With Warren Wallace: on Ritsos,” in Edmund Keeley, On Translation: Reflections and Conversations (Amsterdam: Harwood Academic Publishers, 2000), 69-75. “What I knew of his work initially were some early poems that didn’t much appeal to me—the first of his then eighty-odd volumes was called Tractor and you can imagine what that volume is like” (69).