The goal of translation is to produce text that reads as if it had been written in the target language in the first place. In other words, readers should not be left wondering about the original language. Bilingual editions were pretty much science fiction before machine translation, but the technology that makes them possible today comes with both risks and perks.
Is This Catalan?
Ricard Fité is the linguistic coordinator for the Catalan edition of El Periódico, which is published daily in Spanish and in Catalan. He explained to me that, in his experience, human reviewers get the same feeling when offered a machine translation: “This is Catalan, sounds like it, the words are in the dictionary, but I would have never said it like that.” And, he is not referring to incorrect translations, but to something subtler: the order of the sentence, the choice of words, and all the intangibles that are the collective heritage of the speakers. A more natural translation requires precise knowledge of what the machine can or cannot do. Above all, insists Fité, it requires human intervention.
A Machine with Style
The assets a machine translator should flaunt in order to impress a newspaper editor are impeccable orthography and stylistic regularity, but it would never occur to a machine to alternate morphological variations. Spanish and Catalan happen to be Latin languages, and are favourably similar to each other for this type of work. However, the ultimate reason to hire a machine translator to translate journalism is that literal translation is a good match for standard language, and it is the kind of text that does not offer much room for interpretation.
When human translators intervene in the configuration of the machine, they have to make choices that unavoidably cause one problem in order to solve another. They may choose not to translate capitalized words, so that Antonio Banderas in Spanish never become Antonio Flags in an English text, but then “Banderas” at the beginning of a sentence won’t be translated either. They can do it the other way around, and suffer the opposite problem. The software is also unable to truly grasp the intended meaning of a polysomic word or detect the presence of a fragment from the target language. A machine translation from Spanish into English would convert something as simple as “soy sauce” into the weird realization that “I am a willow tree.”
The system that I am describing is an update made in 2005 to the software El Periódico began to use in 1997, and it has the ability to attach some grammatical attributes to words and sequences. The chance to configure the software is an advantage that not all translation tools offer, but translators have to be quite ingenuous in order to resolve their difficulties. In the type of situation described above, it is possible to make the machine translate the same word differently, depending on the use or absence of capital letters. But again, a polysomic word could expose the trick, and this takes us back to Ricard Fité and his views on machine translation: “Translation software is a great help, but we need to remember that machines cannot access the knowledge of the world contained in the human mind.” Or, as he put it, “the rookie in the newsroom has flawless orthography but needs supervision because, you know, I would have never said it like that.”