Machine Translation Tools can be up to over 90% accurate!

Machine Translation Tools can be up to over 90% accurate!

…however they can also be as low as 55% accurate./their accuracy can be as low as 55%.

Why such a difference?

Simple. Accuracy depends on how much data is available in the source and target languages. This data is then used to train a machine. This means that using a machine to translate between English and Spanish might well be able to hit 90% accuracy, yet translating between Chinese and Zulu is much more inaccurate.

Are the machines taking over?

We are probably a little while away from Skynet taking over. Machine translation (MT) has actually been around for longer than most people realise, beginning way back in the 1930s with Georges Artsrouni filing and receiving the first patent for a mechanical translator in France in 1933, and Petr Troyanskii filing and receiving a similar, albeit more detailed, patent in the USSR in 1935.

Momentum was really gained after the Second World War, predominantly by the USSR and USA up until the 1970s, at which point the US took a bit of a back seat for a while, with Canada, France, Germany, and the UK continuing the research.  Strides were made in the 2000s, and since the 2010s machine translated output has become much more reliable - albeit nowhere near perfect.

Throughout all this time, different approaches to machine translation have been researched and experimented with, each reliant on the technical prowess of the time. From the initial basic rules-based approach, through the approaches of statistical methods, to modern deep learning approach of neural-MT.

Amazing – but just how good is 90% accuracy anyway?

It is often said that when asked about translations, a linguist will always answer ‘it depends on context’. Here is no different.

90% accuracy may be great if you’re on holiday and trying to order lunch or ask directions back to your hotel. The grammar doesn’t need to be perfect. Saying ‘cow’ rather than ‘beef’ doesn’t matter so much as context fills in the gaps and you’re able to enjoy a meal before finding your way safely back to bed.

But.

If you’re trying to translate marketing material and slogans, you need to understand your audience and culture. Machines unfortunately don’t yet understand that.

If you’re translating colloquialisms and acronyms, you need to understand the local culture and context. Machines unfortunately don’t yet get that.

If you’re translating medical documents with complex sentence structures, acronyms, and vocabulary, you need a deep medical and linguistic understanding. Machines unfortunately don’t yet have that.

If you are translating important legal documents, such as contracts and court evidence, you need to ensure accuracy across all points. Is 90% accurate ‘good enough’?

Would you buy from a company that alienates every tenth customer?

Would you read a novel in which every tenth page is complete gobbledygook?

Would you go to a doctor who misdiagnoses illnesses every tenth time?

Would you use a solicitor who drafts contracts incorrectly for every tenth client?

Do you really think 90% right is always acceptable in any context?

A translation that is only 90% accurate could still contain significant errors that would have serious consequences: court evidence that could end one in prison; a contract that offers no security or guarantees; a medical report that doesn’t specify a diagnosis or prognosis…

Machines and AI are indeed developing rapidly, but they have been for decades already. It is human expertise and skill that ultimately produce accurate and reliable translations. Especially in complex and important documents.

At IMD Translation, we pride ourselves on our deep understanding of language, law, and the professional services. Because in legal matters, language matters.

Get in touch at info@imdtranslation.co.uk or 03309121530