Are Machine Translators going to steal my job? No!

Please stop asking professional translators whether machine translation is going to make our skills redundant. It isn’t. At least not in the foreseeable future.

I don’t know whether people do it to intentionally rile us up, but this seems a popular question to ask translators. It could be that people really are unaware of what we do and the limitations of machine translations. Regardless, the following is my response to this question.

Would you trust a machine to write your content? Particularly if that content is for publication, on a specific technical subject or intended to fulfil a specific purpose. I hope the answer is no.

Why then would you trust a machine to translate it?

I think a lot of the hype around machine translation stems from a fundamental misconception of what a translator does. As argued by the Troublesome Terps in their first podcast, translation is not simply pattern recognition. Our ability to understand one another, to interpret language, is not simply pattern recognition. It requires creativity, abstract thinking, an understanding of the context, of how language works and develops, of who is speaking and what their purpose or agenda might be.

While all these skills are required to use and understand one language, translation takes this a step further. Translators need to apply these skills to two languages. They need to understand the source language fully, and convert the underlying message, purpose, style and tone into their target language – their native language.

If every language worked in the same way, used the same style and constructions for the same purpose, and had the same cultural influences, then yes, this would simply be pattern recognition and machines could manage it. But that is not how language works.

Every language develops independently of every other language. Of course, languages do borrow from each other and are influenced by each other and certain shared experiences, but no two languages develop in exactly the same way. Translators develop the skills to deal with this uniqueness, and reproduce, as far as possible and as naturally as possible, the message, style and tone of the source text in the target language.

Is true equivalence even possible for professional human translators?

Given the above, I would say no. Most translators would agree that there is more than one way to translate any given text ‘correctly’. Translation is a puzzle with several possible solutions rather than an equation with a single correct result.

Idioms, slang, jargon and neologisms can all be interpreted by professional translators creatively. Translating this type of language is rarely a case of find an exact equivalent in the target language. Creativity is the key to communicating the underlying message effectively.

Does the effectiveness of machine translation depend on the type of text?

To some extent it does. Simple, technical texts written in highly standardized language can be translated more effectively by machines than, for example, reports on social issues or novels.

This could in part be explained by the purpose of a given text. If a text is simply written to inform the user of simple facts, in simple standardized language, machine translation could do an adequate job. But texts that have persuasive, creative or cultural elements, that are particularly complex or that require knowledge of a specialist field or lexicon are usually rendered poorly by machines. Perhaps one day machines will become artificially intelligent enough to decipher these nuances, but not in the foreseeable future.

But translators make mistakes too!

Of course, translators make mistakes. We are only human! That is why quality controls are so important. Firstly, it is imperative that the right translator works on the right job. If I attempted a literary translation, with inadequate experience of studying literature in my source languages and no experience of writing literature in my target, I imagine I would make a pig’s ear of it. I know that, so I would not accept a literary translation project. A machine would not, as far as I am aware, reject a translation task because it knows it is not up to it. Professional translators, in line with the ITI’s Code of Professional Conduct, will not accept a project that they are not qualified to undertake to a high standard.

This is why working with professional translators (rather than simply amateurs that speak multiple languages) is important. We have studied the quirks and nuances of multiple languages, our specialist areas and how these fit together. We work to continuously improve our skills and keep up to date with the latest developments. We know how a text can best be converted from one language to another to produce a text that is more than fit-for-purpose.

Quality controls should be embedded in the translation process. Machines can help, with segmentation, spelling and grammar checkers, etc. Yet even these tools are limited in their usefulness. You certainly should never rely on a spellchecker to catch every error. Moreover, even if technically correct, sometimes a phrase will ‘just not sound right’. The ability to recognize this is, I would argue, uniquely human (for now).

Another advantage that professional translators have over machines is their ability to spot mistakes in the source and extrapolate the intended meaning. Of course, a professional translator would highlight any errors to their client, improving the source text through the translation process as well.

In the interests of quality, ideally more than one professional translator will work on a text - one to translate and one to revise. Sometimes a text will then be edited further by a target language editor. If you are producing a report, a novel or marketing material in any language, should it not be checked, and rechecked, by multiple pairs of eyes? The same applies to translation.

Machines are emotionless.

Few texts lack emotion entirely. The documents that I usually translate (annual reports, corporate governance statements, funding requests and documents produced by international organizations), might not be considered emotive in the traditional sense. Nevertheless, understanding and interpreting the nuance of purpose and tone is as important as the content of the messages. If my translations do not evoke the same response as the source texts and if they do not emphasise the same points, I have failed in my task.

Until machines can understand emotions and how texts are intended to serve a particular purpose, they will not be able to produce translations that are fit-for-purpose.

Can machine translation engines help human translators?  

Sometimes.

Machine translation post-editing is when a human translator corrects the product of machine translation. It can be more efficient for a human translator to edit higher-quality machine translation, usually of simple, technical or standardized texts, than to translate a text from scratch; however, this is not always the case. For texts that are more complex, or that require more creativity to be fit for purpose, a human translator can often produce a better translation in the target language faster when working from scratch.

Moreover, machine translation relies on pre-existing translations (by humans). A machine translation engine cannot distinguish between a good translation and a poor translation, an appropriate translation and an inappropriate one, in the corpora used to train it, so poor or inappropriate translations are easily replicated. The quality of the corpora used to train machine translators therefore impacts the quality of the machines' output. This means that human translators need to check the quality of the input and ensure that the right corpora are used for the right projects.

Many words have more than one meaning and these multiple meanings are not the same across different languages. I recently edited a machine translation and ‘main courante’ had been translated as ‘handrail’ instead of ‘app’ and ‘pointeaux’ as ‘needle’ instead of ‘checkpoint’. Professional translators are unlikely to make this type of mistake and are far more likely to produce a text that that sounds like it was actually written in the target language.

When is machine translation alone appropriate?

When you want to understand what a text is about, rather than specifics.

When accuracy doesn’t matter.

When the language is highly standardized and designed for machine translation and the product will only be used internally. Even here I would recommend machine translation post-editing by a professional translator to check for significant errors.

Do you want a reliable rendering of the source in the target language that is easy to read and understand? If yes, use a professional translator.

Do you need to publish your text? If yes, use a professional translator, a reviser and even an editor.

Do you want an unreliable, difficult to read text that potentially contains significant errors? If yes, use a machine translator (or a cheap amateur).

It is as simple as that.