Of the examples everyone
has provided, Katherine’s are perhaps the most hopeful, like the story of the
World War II codebreaker deciphering a message in Turkish even though he had no
idea what language it was in. The idea that every language can be reduced to
atomic elements and then reconstituted reminds me not only of John Wilkins’s
analytical language (which, it should be mentioned, no one ever learned to
speak) and digital technologies that reduce everything (including this text that
I will soon e-mail to my editor) to ones and zeros before reconstituting it. It
also reminds me of a low-tech solution that long preceded the digital: the
Persian Empire pragmatically adopted Aramaic as the intermediary language of
communication across its expanse. Messages were translated into Aramaic at
point A and out of Aramaic at point B, whatever languages were spoken at A and
Anthony maintains this hopeful note, at least partially, noting that human input into a number-crunching system like Google’s, with tremendous processing power applied to huge amounts of data, is finally producing results, after fifty or so frustrating years. I confess: after researching an article about eight years ago, I concluded ruefully that computers may never be worth a damn at translation. I was wrong. We erred in trying to teach computers syntax and real-world understanding, like for example that the “pen” in the sentence “The pen is in the box” is different from the “pen” in “The box is in the pen.” Instead, Google taught them to be dumb but massively inductive, learning from vast amounts of human-translated text. No human will translate “The box is in the pen” into Spanish as “La caja está en el bolígrafo” (i.e., a writing pen). And so Google, unlike so many of its predecessors, won’t make that mistake either. Computer translation is getting so good that Nicholas Ostler, author of The Last Lingua Franca: English Until the Return of Babel, predicts that no language will ever replace English because none will need to. Computers will suffice, so that learning English will be for future generations like learning calligraphy: a graceful skill of the leisure classes, but practically unnecessary.
But Anthony and Biljana both provide cases where translation goes wrong, Anthony with Google’s still-rocky software and Biljana in the case of trained diplomats with high stakes: the famous Resolution 242. But Biljana points in her Hainan example to the flexibility that ambiguity can provide. I’ll offer another. Not only strategic ambiguity but all kinds of trickery is possible with a creative use of multiple understandings. When Liu Xiabo won the Nobel Peace Prize last year, censors in China tried to hide the news. Chinese dissidents cleverly began using Chinese’s extensive homophony to use other similar-sounding Chinese characters to talk about him. Well might they now be doing the same about the imprisoned artist Ai Weiwei.
So translation is getting better, translation will never be perfect, and having something “lost in translation” is essential to our humanity. In our last round, I hope our experts will share their takeaways from this fascinating discussion. What do people understand least about translation that you would want them to know—from the linguistic, practical, and literary perspectives?
Recently I heard speak a chess grandmaster who worked on Deep Blue, the computer program that beat the reigning world champion, Gary Kasparov, in 1997. I was surprised when he announced, “Deep Blue did not play chess.” He explained that chess is a game with historical, cultural, and affective dimensions; oblivious to these, Deep Blue by contrast calculated moves. A similar situation obtains with modern machine translation. As Lane observes, computers nowadays do not translate anything; rather, they search databases for close matches to phrases that humans have already translated. Like chess, translation is an activity entwined with historical, cultural, and affective contexts. Finding equivalent phrases is not so much translation as it is an exercise in rapid search and retrieval operations.
The importance of context is playfully, famously explored in
the Argentine writer Jorge Luis Borges’s short fiction “Pierre Menard, Author
of the Quixote.” The narrator
explains that Pierre Menard is engaged in an ambitious project to re-create Don Quixote, not as Cervantes wrote it
but as it would signify to a twentieth-century writer. Phrases that were
platitudes for Cervantes—“truth, whose mother is history”—become for Menard
iconoclastic thoughts so radical that it takes the utmost effort to conceive of
them. “History, the mother of truth: the idea is astounding,” the
narrator exclaims after comparing Cervantes’s text with an apparently identical
passage from Menard. “Menard, a contemporary of William James, does not define
history as an inquiry into reality but as its origin,” calling such a
conclusion “brazenly pragmatic.”
Given that the human touch is essential to full and accurate translation, the question remains whether intelligent machines can ever understand (not merely process) language. A promising beginning is the nascent field of machine reading. At Carnegie Mellon University, Tom Mitchell and his team of computer-science students have developed NELL (Never-Ending Language Learning), a program that reads text “in the wild” (i.e., unstructured, unrestricted text on the Internet) and draws inferences of the type “X is a type of Y which is a Z.” Based on its reading, the program proposes “candidate facts,” tests them against facts already in its knowledge database, and if they are consistent, promotes them to the level of “beliefs.” Some of the facts sound mundane—“golden_bellied_euphonia is a bird”; “vastus_medialis is a muscle”—until we realize that the operations required to create them include drawing inferences from contexts where the fact itself is not explicitly stated.
Also illuminating are the program’s mistakes, which invite speculation about the kinds of texts that could lead the program to this conclusion. My favorite, recently posted at the project’s website (which features an ever-changing list of “recently discovered facts”), is “english is the language of the country japan.” Despite its limitations, the program has the great advantage of reading 24/7. Who knows how smart it may be when, say, it is ten years old, or what a wild, wacky universe it may have inferred from the cacophony of millions of human voices?
Katherine’s image of solitary souls calling out from their towers is so evocative that I can hear their cries, even the silent, Munchian ones! How apt, given the panelists’ focus on technology and the brave new world of translation, that the commentators should draw us back to the sensory nature of language: its distinctive moods, music, and images; its power to modulate one’s sense of self.
Regardless of whether or not we have translation machines in
the basement, we are all calling out to be understood. Every act of
communication is an attempt to share those “illuminations that erupt in the
mind” that Anthony referred to in his first posting. In articulating our
individual take on the world, we affirm our identity, and in sharing those takes
with others, we define ourselves in relation to them. Do we inhabit the same
world? What, as one commentator asks, might I learn from our differences?
These questions of identity and representation arise at a cultural and national level as well. One of the most interesting developments in international relations recently has been the rise of soft power and public diplomacy. Soft power, as defined by scholar Joseph Nye, is the power of attraction, dialogue, and mutual benefit as opposed to the hard power of coercion. Public diplomacy involves the promotion and management of a national reputation in order to gain influence abroad.
What does this have to do with translation? I’m going to focus on choice, since it has been raised in the Forum, and since choice also constitutes my answer to Lane’s question “What do people understand least about translation that you would want them to know?”
In the first place, we are all constantly making choices when deciding what it is we want to translate into words. At a national level, there is a similar concern with what to say, what kind of image to project, what story to tell about oneself, how to make it relevant, attractive, and convincing. As I suggested above, our identities depend on this initial choice.
Second, there are choices to be made when translating: whether to translate what has been said, as in Anthony’s Catalan toast example, or what is understood: the said or the unsaid. The influence of the simultaneous interpreter on political negotiations reflects the central role of choice.
Finally, there are the choices we do not have because the grammar of a language denies them to us. Here we return to Lane’s first posting on the nature of language. I’ll conclude with an example: the term ‘”soft power” has positive connotations in English, by analogy with other compound words such as “soft drink,” “soft landing,” “soft answer,” and “soft sell.” Since no other language has an equivalent array of soft-something compounds, “soft” doesn’t have the same meaning outside English. Quite on the contrary, its connotations err towards weakness. Will we ultimately have no choice but to abandon the term “soft power” because it doesn’t translate well?
When translating, there is a difference between asking, “What does this mean?” and “What do you mean?” One can translate for the text, and thus see translation as an affair of words and languages. Or you can translate for the people—not just the author you construct in your mind but also the users of your translation, similarly constructed, with active stakes in the quality of the communication.
The young French
theorist Arnaud Laygues relates this back to the philosophy of dialogue (Martin
Buber, Gabriel Marcel, Emmanuel Levinas), to the ethics of dialogue with people
rather than the manipulation of things. I would want to extend it to a humanism
of communication in general, with applications to cross-cultural communication
in particular. And I am not too worried about the archaic sense of the term
“humanism”: the Renaissance’s Leonardo Bruni argued, like Laygues, that Plato
should be translated as a person, not as a scholastic textbook. There is
nothing radically new here.
American academics have nevertheless discovered translation belatedly. It seems a useful metaphor, to overcome the way the academy has divided up languages and literatures. But they are drawing a lot on Formalist aesthetics (Borges’s Pierre Menard) and Germanic thought (from Friedrich Schleiermacher to Walter Benjamin). Read those theorists, though, and they’re really talking about relations between languages, and thus relations between cultures. They are not really dealing with communication between people. The Germans attached undue importance to language because it embodied cultural identity, in lieu of a state. And the Formalist play with language was unashamedly antihumanist.
I’ve been suggesting that the words are not so important, that we can translate what is not in the text, that we can and should improve texts. Beyond that, I have little time for the idealization of literary texts. I hope that might upset a few of the academic idées reçues. But here’s what I am really in favor of: I think we have to help people cooperate across cultures, and that translation must be part of that humanistic aim. If not, you’ll never escape from the entanglement of words on words.