Can machine translation replace professional translators?

Although it may seem like a recent phenomenon, the truth is that machine translation first appeared as early as the 1950s. Over the years, machine translation systems have evolved considerably, to the point where some believe it will eventually replace the work of professional translators.

But to what extent is this true?

To answer this question, we must first understand what machine translation exactly is.

The history of machine translation 

As I mentioned earlier, machine translation systems have been changing and improving over the years. Next, I will tell you about the different types that exist.

Rule-based

This system, known as RBMT, refers to the classical approach used. It gathers linguistic information from dictionaries and grammars of both the target and source languages. 

While this model still has a place in the research world, it is not very profitable due to the amount of time and resources it requires.

Moreover, the system is quite limited for translating linguistic structures not covered in dictionaries and grammars. Not to mention any type of poetic or literary text.

Statistical machine translation

Statistical machine translation serves as an alternative to the costly development processes involved in rule-based machine translation. Its main advantage lies in the fact that it only requires monolingual corpora – as broad and complete as possible – in both working languages.

This system consists of three main components: 

• A language model: for calculating the probability that a sentence is correct in the target language. 
• A translation model: for ensuring fluency in the translation by establishing correspondence between both working languages. It is trained using monolingual corpora.
• A decoder: for finding the most accurate translation among all possible ones. 

This system has improved greatly over time, becoming a great working tool. Regardless of the quality of machine translation, it is always necessary for a professional translator to post-edit and review the content in order to ensure that the translation is of the highest quality.

Neural network

This system is the most recent and is based on a kind of artificial neuron. However, it is no longer so recent. The article published in 1997 by Spanish researchers Mikel Forcada and Ramón Núñez is often considered the precursor of this system. 

At that time, researchers were already proposing the use of neural networks for machine translation processes, which is now possible thanks to supercomputers that train this type of engines. 

These types of systems are composed of neural networks that aim to emulate the functioning of the human brain. This artificial network automatically generates context for each sentence.

Moreover, it also uses a self-learning system whereby the artificial neural network develops its own language, creating equivalences between different languages, thus leading to a more natural, more human conceptual-semantic representation. 

Problems generated by machine translation

As mentioned earlier, while machine translation systems can be very useful in the translation process, there are still problems that prevent artificial intelligence from delivering high-quality results.

There are certain issues that these systems do not know how to solve, and this is precisely where post-editing and professional translation come in. What are its main limitations?

1. Syntactic challenges: The relationships of agreement and hierarchy between terms to form simple or compound sentences. 
2. Cultural barriers: Translating is not just about words and grammar, it also involves a deep cultural understanding. Machines cannot yet grasp cultural nuances, such as the typical jargon or slang of each region, local festivities and customs, cultural references, etc. 
3. Semantic ambiguity: Interpreting the meaning behind symbols, words, or expressions is difficult for a machine to do, especially in cases where very metaphorical language is used. 
4. Idiomatic expressions: Here the idiosyncrasy of each language comes into play. A professional translator will maintain the style and register of the text, whereas a machine will not be able to differentiate it, generating low-quality translations. 
5. Intent and tone: The automatic translator is not capable of deducing the author's intention, so it cannot convey it to the target language. For example, an ironic or sarcastic phrase will not maintain its intention in a machine translation. 

So, will machine translation manage to replace professional translators?

The answer is no. At least for now. At least not to produce a high-quality translation.

As we already mentioned in a previous article about customised machine translation in businesses, machine translation is not the most adequate for every type of text.

The way these systems are currently programmed make them an extremely useful tool for translating texts, such as technical manuals, scientific articles, pharmaceutical products, etc. 

In other words, texts that lack ambiguities and creative language, as this is the main weak point of machine translation engines. 

However, in these cases, it is always essential for a professional translator to review and rework the content to ensure a high-quality translation.

AbroadLink offers this translation service for scientific texts, manuals, or other types of technical documentation. Very useful for companies with a volume exceeding 100,000 words.

Do not hesitate to contact us if you are looking for a good provider of professional translation services. We will analyse your case and propose the best solution for your business.