Machines translate. Humans write.



Six reasons why machines can never replace human translators



Humans read between and behind the lines.

 (What is the author’s intent?)

“What did you think of our report?”

“It’s OK, I guess.”

Depending on the human source and target cultures, this reply can be construed to have very different meanings. If it comes from a British person, “OK, I guess” is actually less than lukewarm. It could really mean “It’s not at all OK, but I don’t want to hurt your feelings so I’m being diplomatic. We need to talk.”

Likewise, a Japanese business email may drag on for twenty lines of pleasantries in which the actual message lies buried, whereas a hard-headed business executive from Singapore will want the email to get straight to the point and would find such an email tiresome and impractical. A human translator can understand this and adapt the message accordingly.



MT is digital; humans are analogue.

(Does the text “flow”?)

One area in which it still seems universally accepted that humans do best is in post-editing, one of the last inspections of the text before it reaches its intended readers. A simple example of why this is so can be found in the matter of translation software segments.

Let’s say a paragraph of six sentences has been neatly translated sentence-by-sentence and thus divided into six such neat chunks now known as “translation segments”, easily handled by a machine. On the final review, the human translator may decide to use a different sentence order and completely re-arrange the paragraph to make more logical sense in the target readers’ culture. Or they may decide to chop up one or two gigantic, meandering sentences into smaller, more digestible phrases. On the other hand, they may prefer to join two or three short, stuttering sentences into a longer, flowing one. Such stylistic decisions are still the domain of the human translator and will be for a long time to come.



Machines don’t get culture shock.

(Will this bikini ad work in Saudi Arabia?)

As all diplomats know, sometimes in human communication the best thing to say is nothing at all. Languages are born out of cultures with all of their historical human baggage: humour, gender roles, work ethic, religion, concepts of justice, family roles, political and judicial systems, manners, taboos, etc. A human translator will ring the alarm bell if they see an erotic ad or hear a sexist joke intended for an audience with whom it will go down like a lead balloon. Machines just translate.

Nevertheless, it should be said that perhaps in this case the machine is at least more honest. In a perfect world, maybe the translation should be allowed to go ahead with all of its disastrous consequences, so the reader truly understands the author. But in the real world the customer is not always right, which is why they need a human translator to correct them.



Human writers make mistakes; computers don’t.

(And that’s why human translators are more reliable.)

A common shortcoming of translation courses is that they invariably use perfect source texts for students to practise translating. In the real world, texts that are yet to be translated and published are hardly ever free of mistakes of all kinds; not just punctuation and spelling errors, but dodgy factual information, wrong vocabulary usage or simply tired writers’ lapses. There may even be intentionally poor or ambiguous wording from a wily lawyer or else texts by authors who are simply not good at writing because it’s not their job, such as engineers or hurried doctors with their infamous shorthand. Human translators—and particularly those who have worked in professional fields other than translation—understand this and spot such human mistakes that the computer doesn’t notice, precisely because it isn’t human.

Furthermore, the fact that human translators can also tactfully point out mistakes to the author and offer alternatives is another service the computer doesn’t provide. This leads us to yet another advantage we have over machines…



Humans ask human questions.

(What’s the text for?)

Is the text for children or the elderly? Specialist doctors or their patients? Mexicans or Spaniards? Jews, Muslims or atheists? Hilarious or deadly serious? Is it advertising a product to sell it or simply providing practical public information about it? Will that ad campaign work as well in China as it does in the USA? If not, perhaps it needs re-thinking and re-writing from scratch (“transcreation”), in which case your translator has just saved your company some heavy losses simply by pointing that out.



Humans write the message and humans read the message.

(So who should be in the middle? Well, duh.)

Texts are written by humans to be read by humans. But if the writers and their intended readers work in different languages, the text has to be translated. So will a machine or a human best understand and translate a text written with human intent for human effect? Well, duh. The true test of artificial intelligence is not whether a machine can “fake” human interaction, but whether it can actually “do” it. It is one thing to translate Macbeth; quite another to write it.

Human translators write.


© Gary Smith, 2017. Image: Kaiyodo action figure.


(Related links:




Gary Smith

Professional translator, proofreader and content writer. ES, CA > EN.

This Post Has One Comment

Leave a Reply