10 Reasons Why Machine Translation Engines Will Hardly Ever Replace Human Translators

In 2017, Google announced the launch of a neural machine translation system (GNMT) and declared the following: “human and Google Neural Machine translations are nearly indistinguishable”.

However, ex-post, Google Translate still cannot provide an intelligible translation of texts that are a bit more complex than everyday speech.

Copy and paste a part of a foreign literary text or a marketing copy and translate it into your native language using Google Translate and you will see what I mean.

On the other hand, rumors are in the air that machines will replace human translators soon (actually, programmers promised to eliminate language barriers over 60 years ago when the first machine translation process was invented).

Some people truly believe that this future is near at hand. No wonder, large corporations can make you believe in anything — this is an irresistible power of marketing.

In my opinion, the main problem of the modern translation “industry” is a misunderstanding of the role of machine translation or deliberate false representation of MT as a technological panacea to cure the consequences of the fall of the Tower of Babel.

In fact, computer technologies, including machine translation software, are nothing but tools designed to help translators with routine tasks, while some individuals and companies try to sell the output of machine translation as a ready-to-use product.

However, it is a topic for another article. Let’s get back to machine translation and why it will hardly ever replace professional human translators.

Machines translate words, humans translate ideas

Machines are programmed to follow certain algorithms. In case of machine translation, they stick to inter-language algorithms developed by humans to convert a line of words from one language into another using dictionaries and grammar rules. However, translation is not a simple conversion of words of one language into another. Translation is a more complex process. Therefore, it does not matter how good the algorithm is, machines have one big disadvantage — they cannot grasp the idea behind the text.

Machines cannot identify and rectify mistakes in translation

To err is human, but what happens when a machine translation engine comes across an error in the source text? I can tell you — MT transfers this error into the target text, as it cannot tell right from wrong (semantic mistakes, wrong terms, erroneous and contradictory statements, etc.). However, some MTs can handle the most frequent typos. On the other hand, human translators can identify and correct such mistakes during translation.

Machines cannot convey emotions

Emotionally colored messages (like marketing and advertising materials) require specific approach called “transcreation”. Translators have to find an adequate equivalent in the target language and sometimes even rebuild the entire sentence, paragraph or text from scratch using different words than in the original text to get the same or similar response from the target audience. In this case, translators have to find and use the words and structures inherent to certain language and culture. To do that, linguists must have a deep understanding of both languages (source and target) and cultures. As you already guessed, machine translation engines cannot and probably will never be able to do that.

Even statistical methods applied by MT solutions with the increasing number of relatively good publically available bilingual texts will not help to resolve this issue as the number of possible contextual settings is literally endless, while statistical models applied to languages will predict only a miserable part of such cases.

Machines cannot use translation transformations

I don’t want to go deeper into the translation theory here, but there are a number of specific transformations used by linguists during translation (the number varies depending on the linguistic school and theories). They cover various aspects starting from grammar and syntax, ending with lexis and semantics. Such transformations are used in correlation with context and their proper application is a result of intellectual effort and in-depth text analysis by a translator. Artificial intelligence in its current form is incapable of such transformations as complex interlingual processes with numerous variables cannot be expressed through programmed algorithms.

Machines can hardly see the difference between homonyms

As you know, homonyms are the words that look and sound like another word with a different meaning.

Here are a few examples of homonym pairs:

• address (to speak to)/address (location)
• bark (a tree’s outer layer)/bark (the sound a dog makes)
• current (up to date)/current (flow of water)
• die (to cease living)/die (a cube marked with numbers one through six)

As a rule, we understand the meaning contextually, or even intuitively with our linguistic feeling. MTs can handle some simple homonyms. However, the error probability increases significantly due to cross-language homonyms — “false translator’s friends”. This term relates to interlingual homophones in particular (words that sound similar but have different meanings in various languages).

Such false friends may lead to funny or even offensive translations as with the word “preservative” in English, which means a “preserving food additive” and “презерватив” (reads the same as “preservative”) that means a “condom” in Russian!

As you can see, MT can easily mess up the translation, however, human translators can identify such disputable points and provide an accurate translation.

Machines cannot translate linguistic lacunae

Lacunae are the words that exist in some languages but have no direct analogues in other languages. This term closely relates to the cultural identity and local specifics. Lacunae shall be translated either by means of transliteration with the explanation in a footnote or through a combination of words or phrases that can be perceived by the target audience without distorting the meaning of the original. You can find several good examples of lacunae in this post: 38 Wonderful Words With No English Equivalent.

Here is one from this list:

Lagom (Swedish)

This word is hard to define, but it means something like “Not too much, and not too little, but just right.”

Machines cannot coin new terms

These days, new technologies are invented almost every month. New technologies often coin new terms to describe new processes that did not exist before. “Blockchain” is a good example of such technology producing new terms.  Due to globalization, content related to such technologies have to be translated into multiple languages. It is a complex and very important task to translate new terms the right way taking into account all industry-specific features of such new technology.

Machines will hardly be able to translate new words, as this work requires a thorough analysis of new technology, researching reference materials and finally reinventing a new term in another language often using a creative approach even for technical concepts.

Machines cannot understand subtle interconnection between words in large linguistics units

Sometimes the meaning of a single word can only be understood through the context. It might be explained in the next paragraph, in several sentences scattered across the text or even in another document used as a reference. Machines cannot see the connection between such words and the entire text, at least now.

Machines cannot convey the true beauty of literary works

Ability to use the expressive means of a language, combine words and invent beautiful metaphors to evoke strong emotions is a rare gift and in certain cases is a result of a hard creative work of a writer. Only the best authors can use the true power of words to light up the mind of their readers and only the best translators can recreate the same effect in another language.  Even if machines have the potential to replace humans in certain spheres, literary translation will remain an exclusive privilege of the best human translators for decades or even centuries to come.

Machines will replace only those “translators” who translate like machines

Development of machine translation technologies and the increasing number of people who can speak several languages resulted in a common misconception that translation is not a profession and every bilingual person can be a translator.

In fact, if you have a cool photo camera, it does not make you a professional photographer. On the analogy, knowing two languages or using machine translation engine does not make you a translator.

I think it becomes obvious from the above points that translation requires specific knowledge and set of skills that can only be acquired through learning (higher education in particular) and experience.

Nevertheless, with the growing demand for translation services, a temptation is high to bite a piece of the “translation pie”. It results in a large number of amateurs claiming to be translators after translating a couple of articles for WikiHow and offering “high-quality translation services” for 5 bucks on Fiverr, UpWork and similar bidding portals.

However, I know that some business people already understand the importance of quality translation and hire professional translators who have knowledge and skills to help them bring their businesses to the new markets. And I’m sure that the number of entrepreneurs who take translation seriously will be increasing.

But, in order to work with such clients, a translator should grow professionally and always keep in mind that those translators who produce the same low-quality translations like machines, will be inevitably replaced by machines.

Like this post? We have some more for you: