©2023 Diana Rhudick
Full disclosure up front: a human, professional translator wrote this article, not a bot. The article is intended for anyone who needs documents translated into another language and is wondering whether they should use a tool like ChatGPT for the translation. You might expect me, as a human translator, to recommend that you always use a fellow human. Instead, I would like to lay out some points to consider before using an artificial intelligence language model as a universal translator, and offer ways to get the best results.
What are my options for using an AI model?
The best-known version is likely ChatGPT from the company OpenAI. You can access it free on their website, https://chat.openai.com. The model it is based on is GPT-3.5. If you pay for a monthly subscription, you get the GPT-4 model. This one is faster and will still be available when demand is high. It is the version that has been incorporated into the Microsoft Bing browser, bing.com/chat. Use is also free with the browser. It works only with Microsoft Edge, and also has a phone app. Google’s entry into the field is Bard, at Bard.google.com, which currently has a wait list for access. Less well-known is Phind.com, a site favored by developers. OpenAI has an API for programmers who want to incorporate ChatGPT into their own software to customize its use.
Will my data be secure?
One very important point to make about these bots is that none of them protect your data. In fact, they use it to train their models. So if you don’t want anyone outside your company to have access to the information being translated, don’t use a chatbot. It’s that simple.
What else should I know?
Consider that if you haven’t integrated the AI language model into your own software, or acquired an interface for uploading documents, you will be copying and pasting the source text manually. Then you will have to copy and paste the resulting translation into another file, and reformat it.
If you need 100% accuracy in your translation, chatbots are not for you. Sometimes they just make stuff up; this phenomenon has the amusing name of hallucinations. The accuracy decreases even further for translations into or out of a language of limited diffusion, because there is less data for the bot to train on. Particularly data-starved are Indigenous and lesser-known African languages, and regional dialects. AI models cannot recognize and produce all diacritical (accent) marks, nor older or less commonly used scripts. The earlier OpenAI module (GPT-3.5), has almost no information after the year 2021.
Another consideration is that you have to know the best way to prompt the language module to get the results you want. To specify a language register, for instance, you might tell it to “Write this for a high-school-level audience,” or “Use language that a lawyer would understand.” You may also have preferences regarding the tone of the translation, or which regional variety of a language to use: British English? Canadian French? If you have a list of preferred terms or terms to avoid, you will have to type those in individually.
What are the benefits?
Despite all these caveats, I have to acknowledge that the quality of machine translation has improved vastly since the early days of Google Translate. In many cases, most of the text translated by a language model will sound fluid and natural, if at times dull. That itself can pose a problem. The translation may sound very good, but since these AI models invent things unpredictably, how will you know if the results are accurate?
Two more benefits of automatic translations are that they are produced astronomically more quickly than with a human, and the price can’t be beat. To harness the advantages of this type of machine translation and avoid the pitfalls inherent in it, the best way forward is to work with a machine in combination with a human. You can get the translation quality you need if a professional translator reads and edits the automated output, applying his knowledge of both languages, as well as his ability to understand meaning and nuance. Many translators have been working with machine translation for several years. They know what types of errors to look for, when the output is false, and how to improve automated text.
The advent of AI language models has created a brave new world for many industries. A language specialist can guide you through the uncharted territory. You are more likely to succeed in your quest for high-quality, automated translation using a professional translator as your adviser and editor.
Image created by DALL-E
"Weak and fragile is the realm which is based on a single language or a single set of customs."-St. Stephen of Hungary 