AI Translation, A Long Way From Perfect

October 25, 2017


Not to be a buzz kill—everyone has their preferred translation methods—but artificial intelligence (AI) translation has a long way to go in terms of reliability. Why? For starters, until it stops finding ass where there isn’t one, I can’t use it.

Every year, I ghostwrite 10-12 books for CEOs. After literally months of interviews, edits, rewrites and design, myself the CEOs—with a team of designers, PR people and publishers—produce 250-page masterpieces about making billions, creating strong corporate cultures, selling businesses and creating social media strategies. While I’ve met very few of these CEOs face-to-face, we get to know each other intimately. They tell me about their families, their failures both professionally and personally, their fears and their weaknesses. They are forthcoming with me in ways they can’t be with journalists and business associates. They trust that because I’m acting in their best interest, I will help them represent themselves as they are while also safeguarding them against sticking their foot in their mouth.

So imagine if I turned over a chapter draft that had one of these CEOs calling someone an ass or speaking gibberish? Usually, I spend many dollars for humans to translate the dozens of interviews I conduct with each CEO however, a few weeks ago I thought, “Why not give AI a try?” I hired an AI translation company that shall remain nameless to translate a two-hour interview. Before uploading my MP3, I had to answer a few simple questions that ran along the lines of:

Does the interview include strong accents?

If so are they easy to understand?

Is there any background noise that could interfere with translation quality?

No accents, not even Southern US. No background noise, we were on a secure line. My interviewee even spoke slowly. As far as transcription goes, that interview should have been a dream. 

I paid for the service, uploaded my doc and in less than five minutes, I had my interview. Excited, I downloaded it, opened it and looked the content over for about four seconds before jumping on another call. What a mistake. Three days later when I sat down to write the chapter, which was due the next day, I found myself staring at words I didn’t recognize neither linguistically or contextually.

Case in point, ass.

My AI translation crew thought my interviewee responded to one of my questions by saying: “Ass. OK well this is this is where we get into our.”

Imagine if following the ass insult, I wrote the following paragraph verbatim from the AI translation?

“Certainly we're dealing with with regard to Flav's they still have to be certain you know competency the gold standard that became that foundation for integration. So I left Canada with some dirty Churchkey in 1989 and the word got into the system security. We. Weren't quite sure about the lack of integration."

Ooo. Flav and a dirty Churchkey. Come on Danielle Steele! But seriously, what a bunch of mildly entertaining gibberish.

While AI’s potential is almost limitless, AI translations have a long way to go. Of the 24-page interview, a total of four full pages were legible. Until AI improves, I’m returning to far more accurate transcription services. Sorry AI. You may be intelligent but when it comes to translation, you’re in third grade and I need a PhD.