The Machine Translation (MT) is the type of translation performed by a computer programme (software, website, etc.), which analyses a source text and translates it into another language (target language) without human intervention.
Machine translation is not to be confused, however, with Computer Assisted Translation – CAT, that is when a person performs the translation task and uses CAT tools in doing so.
The first approach for machine translation ever created is the rule-based approach. This first model is based on grammar rules and bilingual dictionaries for each source-target language pair.
The system, equipped with syntactic rules and vocabulary, analyses the sentences of the source text, breaks them down by recognising individual lemmas and replaces them with the corresponding ones from the target language.
Since rule-based machine translation is totally devoid of any interpretative capacity of the text, the quality of the final translation depends on the reliability of dictionaries and the correctness of syntactic, semantic, and morphological rules.
Although newer software has higher performance, the final translation is still a rough product, useful only to understand the general sense of a text.
Statistical-based translations “learn” to translate by analysing large amounts of data for each language combination.
This type of approach collects numerous texts in digital format (the corpora) written in a source language, their translations into a target language and texts belonging to the same domain written in the target language. Statistical systems are not equipped with linguistic rules, but can be “improved” by adding more and more domain-specific texts.
The quality of statistical-based translations is not very predictable, especially since terminological consistency is not always ensured. Keywords in texts on the same subject may vary depending on the references in the system during the translation process, which may lead to the use of many variants for the same term and thus confuse the final reader.
Many companies use “closed” systems for their statistical-based translation processes. Their system is “trained” with a corpus of their own texts and constantly supplemented with new translations: in this way, greater uniformity and consistency of terminology can be maintained in the translations.
Neural translation systems (Neural Machine Translation) allow to translate entire sentences taking the context into account.
Neural translation systems are based on artificial neural networks similar to those in the human brain, whose functional units receive, process and transmit information to other “neurons”, extrapolate patterns by observing huge amounts of data and generate new representations that also correct previous information in a repeatable self-learning mechanism.
Neural Machine Translation can adapt to the entire text because it is able to take into account the context of the subject matter and the purposes and intentions contained therein. The end result is a more fluent and natural text, the translation is consistent, accurate and faithful to the original, but still needs to be verified by a human translator.
It is fast.
A large amount of text can be translated in a short time.
It is cheap.
Through subscription access to the most suitable machine translation software.
It is the quickest choice if you have to translate non-creative texts, such as manuals, technical documentation, etc., thanks to the creation of a translation memory.
It is not creative.
Machine Translation is unable to grasp the nuances of meaning and cannot “read between the lines”.
It may commit some errors of terminological imprecision.
Localisation for the target audience.
It is not able to completely adapt a text to a specific target audience.
Despite its advantages, machine translation has not yet replaced the human translator. Why? Because the technology is not capable of performing certain tasks.
For example, in localisation tasks, it is necessary for the linguist to also translate the culture. Localising a text means not only translating the text, but also translating idioms and cultural references, and positioning the target text for an audience with a completely different history and culture from the source audience.
Another example that shows that machine translation is not enough is the field of transcreation. In marketing and advertising, it is not enough to translate the words of the source text, but it is necessary to know how to convey the message. To do this, it is necessary to follow the path from sense to words, taking into account the context, the tone of voice, the means of communication, the culture and the values of the target audience.
One aspect, however, in which the human translator must necessarily associate his work with the machine translation is the post-editing. Post-editing is the process by which the translator, or linguist, revises a translation generated by a machine translator to correct errors and make it fluent and comprehensible.
When you can rely only on machine translation? When the objective is only to understand a text for internal use, then the machine translation is the best choice.
For all other texts, especially if they are addressed to an external audience, machine translation is not sufficient. The public image of a company or professional also depends on the way it communicates; an ineffective translation would risk compromising public trust.
For this type of texts, the best solution is to rely on a professional translator, who can also use CAT tools and review and verify the output of the machine translation.
Do you want to make perfect a text translated with the Machine Translation?