David Bellos, the author of Is That a Fish in Your Ear? is the director of Princeton University’s program in Translation and Intercultural Communication. Graduates of this program might get work in international legal firms, as interpreters in government, or translating blockbuster novels. The title refers to the Babel-fish of Hitchhiker’s Guide to the Galaxy, which, when stuck in your ear allows you to understand language spoken by anyone.
As fluent speakers of at least one language, each of us is some kind of expert on how language works. This book engagingly delivers all sorts of information that will make you an even bigger expert, and also make you wonder, “Why didn’t I ever think about this before?”
Here’s an example:
In college, I learned that a device like a Babel-fish could only be the stuff of science-fiction. Linguists and mathematicians had already demonstrated how it’s impossible to encode all the rules of two languages so that one can be turned into the other by a machine. The broad rules were easy enough, so at the outset the endeavor seemed promising. But every attempt to manage the exceptions eventually failed.
For instance, a machine would have to know how to render the meaning of the word “pen” depending on whether the sentence to be translated was “The pen is in the box,” or “The box in is the pen.” The machine would also need instructions for distinguishing between the meaning of “penning a chicken” versus “penning a manifesto,” and that the chicken, but not the manifesto, could ultimately be “unpenned.”
In practice, it turns out that the list of exceptions (and exceptions to the exceptions) for any language spoken by humans is impossibly long. Sooner or later—probably sooner—a translating machine makes a blooper that is so ridiculous that the whole translation isn’t credible. Eventually, the whole idea of machine translation, which held great promise in the 60s and 70s, fell by the wayside. You just can’t write a computer program for turning English into Japanese.
Are you thinking perhaps that someone forgot to tell that to Google Translate? Or that Google is so arrogant that they don’t treat theoretical impossibility as a limitation? Could it be that Google technicians are so smart that they did figure out how to stick the full grammar of a language and all its exceptions into a machine? None of the above. Bellos explains that Google Translate works because it relies on what Google does best—searching.
In order to translate, Google mines already-translated internet documents, looking for phrases contained in the original and noting how those phrases have been translated into the target language. This wouldn’t have worked in 1962, when there was no internet to speak of and the science of digital archiving and retrieval was embryonic.
Today, at a bare minimum, Google can access all of the documents of the European Union, which exist in 24 languages. The work of the UN and its agencies is published in at least 6 languages. When Google still can’t locate the phrase it is looking for, it can scan corporate reports, legal proceedings, correspondence, literature, and any other writing that individuals, groups, agencies and organizations have published in more than one language. Processed with Google’s searching and statistical methods, the translations Google makes out of phrase-matchings are better than not-bad.
Sounds straightforward? Asks Bellos. Think harder. What if an Icelandic person wants to read something written in Farsi? It is unlikely that Google is going to be able to locate enough already-translated Farsi/Icelandic documents to find every phrase it needs to make the translation. Not a problem. Enter the pivot language.
A pivot language is a language into which many documents from all over the world are commonly translated. English can serve as a pivot language, although other globally popular pivot languages include German, Arabic, and Russian. To translate a document from Farsi to Icelandic, Google Translate will look for documents in both languages that that have been translated into potential pivot languages. Then Google can rifle through the pivot language documents, looking for phrases that match the original Farsi document and put together a translation that way.
So when a Brazillian student writes to a penpal in China, or a Senegalese manufacturer researches potential markets in Indonesia and Google Translate stands in as the interpreter, the resulting translations could be a patchwork of snippets from a host of sources ranging from international treaties to the novels of Danielle Steele.
Is That a Fish in Your Ear lives up to its subtitle: Translation and the Meaning of Everything. This trove of ideas about language will leave you ruminating for a long time.
(Note: if you ask Google Translate to render the last sentence above into Vietnamese and then translate it back into English, it will tell you, “This thing very many ideas about the language will for your ruminating in a long time.” In other words, humans have yet to be fully removed from the translation process.)