For something so important as the copy on your website, leaving translation to chance can be akin to entering a car rally race with one tyre coming lose before you start. In short, it’s a big risk.
Google translate – known to anybody using the internet – is an example of machine translation. Launched in 2007 by Google, to train its engine, Google Translate uses the vast amount of Internet data it indexes in its search engine. It recognises language patterns through sheer quantity, its results improving with the amount of language data published on the Internet. It supports many languages, but not all language combinations yield the same results. Further to this, no appraisal of quality of the target text can be based on the results obtained for a different language pair – linguists are still needed to assess the results and adapt the texts prior to publication, at least for now.
In November 2016, Google released its neural net machine, Google Neural Machine Translation(GNMT) for eight language pairs, a system which uses an artificial neural network to increase fluency and accuracy in Google Translate. This system, one which requires heavy-duty computing power, greatly improved results, especially for languages with lots of available training data. But still, not all languages are equal. This new model uses two networks: an encoder converts each word of an input sentence into a multidimensional vector, taking into account what has happened earlier in the sentence; a decoder then generates a word-for-word translation, taking account of the immediately preceding word. The result? a better text, but still not perfect.
An alternative to Google translate are customised translation engines that use a mixed statistical and neuronal approach, tailored to your specific domain. If well trained they can yield good results, especially for product descriptions or technical texts.
But back to the original question: should anyone rely on Google Translate for their website? Would you solely rely on AI for your system software's code?
There are 7.5 billion people in the world, 80% of whom do not speak English. Translating your website makes perfect sense for many reasons. Considering the ever-growing amounts of content that even small companies need to translate, machine translation could be a smart option to save time and money. But, we caveat this should only be done if you submit all machine pre-translated text to a thorough post-editing process done by linguists specifically trained in your particular domain. Your business has spent a great deal of effort and resource for the development of your site and brand more broadly; a badly translated website could destroy all that hard work.; Your brand could end up in tatters, users might lose confidence in your products or services, and you might miss out on good commercial opportunities.
Simple phrases still produce wrong translations when plugged into machine translators. For instance, in the sentence, "The city councilmen refused the demonstrators a permit because they feared violence” Who exactly feared violence? And chat GPT encounters the same issue, at least in the language pairs tested.
It is still the case that human intervention covering all the bases is needed for good translation. As much as all 4 wheels are essential for the smooth operation of you racing car.
For content purposes, the golden rule still (2024) is to use technology supervised by humans, and recognise the intellectual and cognitive effort needed to deliver correct and precise translation or copy.
If you are interested in a machine translation solution, customized to your domain, we can help you with state of the art neural MT engines, trained on your data, kept securely in regional data centres, enriched with a LLM layer contributing to fluency - supervised by expert linguists with years of domain expertise.
