AI Translation: good or bad?

By Carlos de Paula 

I have studied so-called machine translation from the very beginning, since the 90s. To tell you the truth, it was very bad back then.

 Thirty years down the line translation software has improved. However, most AI translations still sound like AI text, devoid of style and life, incapable of understanding figures of speech, irony, humor, nuances, and the proverbial text between the lines.

 I figured that sooner or later it would be widely adopted as a tool, so in the last couple of decades I continued to study several software titles in myriad languages.

 As it stands, AI produced translation continues to have  a major problem: it reads context very poorly. This can breed trouble in several languages, for the same word can have different meanings in the same language, for instance, “recurso” has several meanings in Portuguese. This results in all types of nasty havoc, embarrassment for starters. Legal issues are more serious, for the most popular AI translation tools continue to inexplicably make positive negative, and negative positive, with disturbing frequency. This in a contract can lead to litigation, great expense and loss of face.

 Not only that, AI frequently picks up the wrong translation for a given term, often leading to unflattering renderings. In other words, translation software fails to connect the dots where the dots are often very important.

 Another problem is that AI translation works reasonably with well written texts. As writing skills are in short order these days, AI is often used to make sense of the senseless. A badly written text will sound wacky, bizarre, after being put through translation software.

 I always say that commercial planes are flown by automatic pilot for the longest part of a trip, but qualified pilots have to take-off and land the darn things. It is no different with translations.

 Let us face it, DOGE is not the only one seeking to cut costs. Businesses penny-pinch as much as possible when it comes to translation work, it has always been so, for it is often seen as nuisance.  Now that it is available a few clicks away, for free, the perception is that paying for translation work is wasteful. 

Thanks, Silicon Valley.

I saw the writing on the wall and specialized in editing AI produced translations, for it is the future of the written translation industry. I have been able to turn atrocities and inaccuracies into good and precise text,  even making some of them enjoyable. When a client comes to me with “a translation he did”, I already know what that means. I only draw the line on certified document translations: I do not accept AI done translations prepared by clients, after all, I have to certify that I did it. Those are done from scratch.

 Whether AI will ever reach perfection is open to heated debate. Brazilians, for one, like to be witty, and AI fails to handle wittiness all that well, so that a culturally competent editor will always be necessary. In other words, rather than making it my number one enemy, I decided to coexist with it. We translators have no choice, I, for one, am not that charismatic to become an influencer.

 

Carlos de Paula is one of the top Brazilian Portuguese translators in the USA since 1982. And now a top Portuguese AI Translation editor as well.