In our last post we addressed translators as professionals and how important they are to the modern society. In the same line of thought, today we would like to address a topical and highly important issue in the translation industry, i.e. machine translation. We all know how online machine translation software, such as Google Translate, are readily available nowadays. Besides being seemingly intuitive and easy to use, this type of software generally offers a more or less accurate translation of the content and is constantly improving striving towards offering the user an increasingly precise and reliable service. A new free online translator based on an artificial intelligence system exploiting neural networks capable of mapping the natural language, DeepL represents a tangible example of this type of software. With its 26-language combination constantly growing DeepL is becoming an increasingly competitive and advanced tool in the industry.
Striving to improve its current machine translation services, DeepL is based on machine learning technology (automated learning basically) through which the computer memorises the most appropriate language mechanisms, progressively applying them in a continuous translation improvement and boosting process. Use of neural networks enables picking up various language nuances in a text and transferring them in the translation, adapting expressions to the context.
Google translate applies these smart technology solutions with the aim of providing “quality translations” too. Driven by curiosity, we tested these two major machine translation software solutions with the aim of understanding which of the two offers better translations and, above all, whether DeepL could actually replace a human translator. In our test we firstly took a previously translated text (our website’s “About us” page) and uploaded it on the two platforms. Then we went ahead to compare the three versions, i.e. the Google Translate version, the DeepL version and our version. In our assessment we tried to be as objective as possible, only highlighting objectively glaring mistakes. Below are some excerpts from the three versions:
Source: “WE BRIDGE LANGUAGE BARRIERS”
DeepL: “SUPERIAMO LE BARRIERE LINGUISTICHE”
Google: “WE BRIDGE BARRIERE LINGUISTICHE”
NENO LS: “ABBATTIAMO LE BARRIERE LINGUISTICHE”
On closer look at “WE BRIDGE”, we see that DeepL used the Italian verb “superare” (i.e. Overcome). An interesting choice of lexicon which however falls short of fully encompassing the “bridging” concept in this context. On the other hand, Google opted to leave the English expression as it is. On our part, we opted for the verb “abbattere” which basically means “breaking” or “bringing down” barriers. To us, this was the most appropriate option given that we are metaphorically talking about “barriers”.
Source: “..i.e. rendering written texts into audio files.”
DeepL: “..ovvero di rendering di testi scritti in file audio.”
Google: “..ovvero il rendering di testi scritti in file audio.”
NENO LS: “..vale a dire la trasformazione di documenti scritti in formato audio o video.”
Curiously, both DeepL and Google Translate opted to leave the English expression “rendering” as it is. Though correct from a grammatical standpoint, translating the English expression offers a more exhaustive meaning to the service we offer our customers as observable from the segment appearing on our page. Despite use of English words in Italian being an increasingly common practice, translating the expression adds clarity and professionalism to the text in this case.
Source: “Our history, as Neno Language Services (NLS), dates back to 2007…”
DeepL: “La nostra storia, come Neno Language Services (NLS) risale al 2007..”
Google: “La nostra storia, come Neno Language Services (NLS) risale al 2007..”
NENO LS: “Neno Language Services (NLS) nasce nel 2007..”
In this case, translating “as” with “come” is too literal. The rest of translations confirm that despite being correct, DeepL and Google’s translations are literal and schematic. On the other hand, our translation flows more easily whereas the use of a conversational style boosts the reader’s willingness to continue reading.
In conclusion, for the sake of a more exhaustive comparison of style, here is the entire sentence:
| Since then, we – as NLS – have come a long way expanding our radius of action increasing the number of specialised in-house and external translation collaborators. The expansion of our language combination and fields of specialisation was obviously a natural consequence. | Da allora, come NLS, abbiamo fatto molta strada per ampliare il nostro raggio d’azione, aumentando il numero di collaboratori specializzati interni ed esterni nel settore della traduzione. L’ampliamento della nostra combinazione linguistica e dei nostri campi di specializzazione è stata ovviamente una conseguenza naturale. | Da allora, come NLS, abbiamo fatto molta strada espandendo il nostro raggio d’azione aumentando il numero di collaboratori di traduzione interni ed esterni specializzati. L’espansione della combinazione linguistica e dei campi di specializzazione è stata ovviamente una conseguenza naturale. | Tuttavia, col passare del tempo, NLS cresce e allarga il proprio raggio d’azione aumentando il numero di collaboratori specializzati in traduzioni e, di conseguenza, le combinazioni linguistiche e i campi di specializzazione. |
The latter example clearly shows that despite often translating correctly, machine translation is all about just transposing words from one language to another. This without considering factors such as repetitions in the same sentence or reformulating the text to make it more pleasant and flowing for the reader.
Basically, the three examples show that translations carried out by professional human translators offer an exhaustive understanding of the source text based on an appropriate execution from a grammatical and style standpoint. However, full understanding of context proved to be the true game changer. Something that can be fully perceived by a human only being as opposed to a machine translation software.
In conclusion, we believe in the importance of being able to rely on machine translation software as useful tools for a quick look-up. Any other use must always be reviewed by human translators. Last but not least, the relationship between the customer and human translator cannot be replaced. As a matter of fact, translators sometimes have to consult the author or the source text or specialists in the industry (engineers, technicians, doctors, biologists etc.) to clarify some points. Something unthinkable for a machine.