It’s a Human Thing: Translation Nuances Machines Don’t Understand

In recent years, machine translation output has significantly improved in quality. This is a fact that is largely down to the introduction of neural machine translation, an approach that rather than focus on a phrase-based statistical method uses an artificial neural network to predict the likelihood of a sequence of words. Whilst this approach has certainly improved fluency and quality, there are certain translation nuances that machines still don’t understand, nuances that, fundamentally make the difference between a good and a bad translation.

In this blog post, we are going to outline five elements that machines are still struggling with and quite simply require a human touch.

A confused figure to represent translation nuances that machines don't understand.

Playful Language and Humour

The first translation nuance that machines don’t understand is humour. This isn’t surprising given the fact that humans also struggle to understand a joke here and there, however the problem with machines is that not only do they not understand the joke, but they also can’t recognise when something is a joke. For example, in many creative texts you’ll find made-up words, exaggerated language and non-sensical situations all presented for comedic effect. Machines however won’t understand this intention and so won’t catch the subtleties in language, translating rather in a literal manner and failing to convey the humour.


Similar to the use of humour highlighted above, machine translation also fails to pick up on the tone of a text. For example, if you’re writing a letter to a serious client, you may wish to use more sophisticated and formal language, perhaps writing ‘utilise’ instead of ‘use’ or ‘thank you’ rather than ‘thanks’. Whilst a human would be able to identify this formal tone, a machine may struggle to do so and thus may produce a translation with an entirely different tone to what you intended.

Nuanced Vocab

The third element that machines struggle to translate is specific vocabulary. Whether it be a specific medical term or a regional colloquialism, machines struggle to understand the more niche aspects of writing. Whilst humans will be able to identify any unfamiliar terms and conduct detailed research to be able to locate the best translation, machines can’t do this. Whilst the consequence of this may just be a minor loss of meaning for a colloquial term, it could be quite dangerous if the text concerns medical information.

Broader Context

Another aspect that machines can’t process is broader contextual elements. For example, if you’re translating a travel blog you may encounter lots of different national holidays or historical figures such as Bonfire Night or Guy Fawkes. Whilst a human linguist would be able to recognise any elements that would be unknown to a target audience and therefore adjust their translation accordingly – perhaps adding an additional comment to the translation explaining any missing context – a machine wouldn’t be able to do this. Translations produced by a machine therefore might fail to connect with their target reader. After all, if you don’t understand the implicit context behind the text, it can be difficult to follow what’s going on.

Cultural Differences and Localisation

The final element that machines struggle with is localisation. As highlighted above, there are times when translators need to purposefully change the translation in order to make the text relatable for a target audience. Two key examples of this are time and currency. If you’re translating marketing copy outlining the cost of a product or the time an event begins, it’s going to be important that all dates, times and costs are correctly translated. However, whilst a human would be able to recognise that an event starting at 3 pm UK time would actually be starting at 4 pm Spanish time, a machine would likely pass by this piece of information thus leading to confusion! The same goes for any costs in GBP – a human would be able to identify if any currencies need to be converted, but a machine wouldn’t.

Whilst they are getting better at translating content, there are still plenty of translation nuances that machines don’t understand. At the end of the day, if you wish to produce a text that is respectfully localised to your target audience and that accurately conveys humour, tone and nuances, human translation is the best way forward.

If you have any translation needs that require our services, please get in touch today to receive a free quote.