For the translator in the twenty-first century, one of the most fervently repeated collocations of the times has to be machine and translation, those two suggestive siblings that have gone on to establish their reputations as the acrobats in a traveling circus of technology. As members of the audience, both captivated and unsettled by the performance in the ring and the apparently high-flying stunts of our little siblings, translators have front-row seats that make them the perfect observers of the madness and implicate them in the occasional fatalities.
Once the lights have been extinguished in the audience, we might be dazzled, disturbed, cowering, maybe even fearful that one of the death-defying siblings will fall during a performance and pulverize a translator or two upon impact. Sometimes fear can become reality. And sometimes, machine and translation can fall very hard indeed. And it’s not a pretty sight.
The following is an example of one of these potential catastrophes, often attributable to an overestimation of the capabilities of machine translation. It is one, moreover, that the technical translator might be familiar with: the context is safety data sheets, where the highly formulaic language would seem to make machine translation an ideal supplement for the human translator. But in this case, the word-for-word precision of a machine translator is less effective, and ultimately less accurate, than a comparison of the documents and a patient search for equivalent terminology. The translator should be looking for contextual and field-specific equivalence rather than for the definitions of words found in a dictionary.
It is the safety data sheet of a Spanish chemical supplier that in theory could be looking for an English-language translation at a reasonable cost. Thinking that a machine translation would be economical and efficient, if only just as a start, it puts the safety data sheet through the grinder.
But the results spit out by a machine translator, as the company soon realizes, are less than stellar. All this formulaic, technical language from the safety data sheet, or ficha de datos de seguridad, turns into something monstrous and unintelligible in the target language, and it’s almost too horrific to describe. The English-language target text seems like an absurd jumble of nonsense, and the machine-translation output was, in this case, misleading and ineffective.
What the human translator can make, and what machine translation tends to overlook entirely, is a comparison of what are called in the translator’s practice parallel texts, which are supposed to exemplify the terminology and context of the source language in its target-language equivalent. By comparing source- and target-text documents that are equivalent in a given field, context, or industry, the translator can assess the differences and determine how much translation is actually necessary, and how much would be reinventing the lexical wheel.
And while it’s certainly correct that some bad machine translation might not be representative of the capabilities of modern-day technology, it’s also true that an effective and studied comparison of the specialized language and context of these kinds of documents would be more suited to a human being than to a machine, and for the following reason: machines can’t yet account for all the contextual subtleties and implications of a text. They have significant semantic (or we could say connotative) limitations that are the bread and butter of the human translator and, in particular, of any professional translator worth his salt.
In many cases, machine translation is completely crude and inadequate; in others, merely sufficient without the more studied approach of an at-length comparison; and in a number of others, impressively accurate, especially in those where language is highly repetitive and there is no need for extensive cross-checking or contextual inferences like those to be found, for example, in literature, commerce, public relations, marketing, subtitling, or any other creative field where the context and connotative variety of a text brings out, to put it one way, the deficient humanity in the machine translator.
And in view of this article, another category to be included on occasion among these creative fields and text types—and more often than one would expect—would be technical data sheets. May the technical translator steer clear of all egregious pitfalls.
