Huge productivity gains… entire industries transformed… fears of plagiarism and deepfakes… There’s a revolution going on and it can be hard to make sense of it. Generative artificial intelligence (AI) is upending our world! So what’s the scoop with translation?
Some background
People have been trying to get machines to translate for decades. After all, if a machine can do it in seconds, why ask a human who needs hours or days? A collaborative effort in 1954 by IBM and Georgetown University was one of the first forays into machine translation, with a project to translate Russian into English.
Early attempts were rules based. The thinking was that if you gave the machine the right set of rules, it could just follow them. But coming up with the rules was another matter altogether—every time they taught the machine a new one, humans created a new exception.
If at first you don’t succeed…
Researchers then hit on the idea of using statistics—teach the computer to predict the most likely translation. New, more advanced processing power made it possible for computers to analyze vast corpora of professionally translated texts and discern complex word patterns. Instead of following a rule, the computer guessed, much like your phone guesses which word you want to type next. The results were better but still inconclusive.
…try, try again!
Artificial intelligence has now broken the logjam. Large language models like ChatGPT with their billions of parameters have learned to mimic human language by mining vast storehouses of textual data. They are still guessing, but with a predictive power multiplied many times over. The results range from unusable in some fields to impressive in others.
What this means for your business
Make no mistake about it: neural machine translation (NMT) is a boon—to business and to humanity.
First, the good news:
- It has never been easier to translate: NMT puts translation resources at your fingertips that you could only dream of before. For a fraction of the effort it once took, you can now connect with the world in all the major languages, not just English and French, but German, Spanish, Chinese, and others. And as time goes by, AI’s large language models are expanding into lesser-used languages.
- It has never been faster to translate: Translation has always been labour intensive. Some bigger jobs could take weeks. With NMT, you can now generate an editing-ready draft in minutes, if not seconds. And that’s important when you’re in a race to be first to market.
- It has never cost less to translate: Makes sense, doesn’t it? Easier, faster, less costly. Just don’t make the mistake of assuming this means it costs nothing—it’s risky to use computer-generated output without any type of human editing.
- The world can translate more: The world has billions of words to translate, but not the millions of translators needed to do it. Now, NMT is putting translated materials into the hands of more people than ever before. The quality is not first rate, but it’s a start.
- NMT takes the tedium out of translation: Truth be told, most documents are rather tedious to translate—product lists, technical descriptions, administrative notices, and more. Many of them are repetitive too. These kinds of text are ideally suited to automation, leaving translators free to work on challenging jobs that machines struggle with.
What you need to remember about translation
This is all very exciting, but a few words of caution are in order:
- There are data security issues: Once your materials find their way onto a free online tool, they’re used to train AI tools and populate public databases. Make sure you’re operating in a secure environment that respects your business’s confidentiality standards.
- Your output needs to be edited: You can get into serious trouble by using straight computer output. Language has too many nuances of meaning. At TRSB, we can help you sort through your materials and decide on a quality control strategy for each type of document, depending on audience, technical difficulty, reputational risk, and intended purpose.
- Your output may require localization: Computers simply cannot know whether your text is suitable for the target market. Are there cultural references to change? Examples to update? Additional explanations required? This is a job for a professional communicator.
- Machines can actually hinder productivity on some jobs: This is particularly the case in marketing and public-facing documents. Your text needs to be reviewed by a professional translator with advanced writing skills and subject-matter expertise. TRSB will pair you with a specialized team.
- Your translation engine needs to be trained: DeepL, ChatGPT, and other translation and writing tools have broad access to language, but not to your in-house terminology, the jargon used in your field, or your style guide. You’ll need the help of a reputable firm like TRSB to refine your parameters for optimum output.
Neural machine translation is a huge step forward with enormous potential. We can help you take advantage of it. With nearly 40 years in the business, we’ve grown to be Canada’s number one provider of translation and language technology services.
