Estimated reading time: 5 mins
In about three years, Google Translate will be able to translate almost anything, even novels. Thus, professional translators won’t be necessary anymore. This is the alarming message that Giodano Vintaloro, translator and English professor at the University of Trieste, sent out last week during an interview with il Mulino magazine.
Will the human mind be ousted by artificial intelligence in such little time?
Automated translation has been around for some time now, but translating isn’t simply a matter of converting each word from source to target language. If you want to be understood, you need to fully comprehend the source text and its context, understand the culture of the target language, and pay attention to details. Although it is fair to admit that Google Translate has improved a lot over the years, the tool still lacks all the above-mentioned characteristics.
Recently, the California-based company declared that one of their most recent projects is the Google Neural Machine Translator (GNMT), a system based on neural network architecture—in other words, they developed a deep learning mechanism and a neural network that imitate the functions of the human brain. Developers are making the system read literature written in different languages in order to help the program to learn words, grammar rules, and the correct method to comprehend texts. Contrary to the previously developed Phrase-Based Machine Translator that translated phrases on a word per word bases, GNMT approaches sentences as a whole. Happy about their progress, developers stated that thanks to this method, automated translation is more precise—almost as good as a professional translator. As a matter of fact, compared to the previous system, the current one makes from 55% to 85% fewer mistakes. During the interview, Vintaloro argued that soon Google could even translate 80% of a novel in an eighth of the total time it usually takes a human being.
To be completely honest, there are so-called CAT Tools (Computer Assisted Translation Tools) that a translator can use to speed up the process and reduce costs. Nonetheless, they are completely different from automated translation systems. Their main feature is translation memory; all translations performed with these software programs are stored in separate files created by the program. Every time a translator starts a new project with the same language combination as previous entries, the program will recollect all identical and similar text segments, thus reducing the amount of text that the translator actually needs to translate.
That said, we can’t rely on language alone. There are a lot of fields of expertise when translating that in order to make sense still need the intervention of the human touch. For example, patents and technical translations require specific terminology. Marketing translations encompass far more than just words—you need to be aware of the final reader and his culture. For this type of translation, transcreation is the only solution, and that can’t be performed by a machine. Similarly, literature translations must convey emotions and accentuate the rhythm and musicality of the language and the style of the author.
Last month, in movie theaters, T2 Trainspotting arrived. It’s the movie inspired by Porno, the second chapter of the Trainspotting saga. This book is written with a strong Scottish accent that people used in Edinburgh in the ‘90s. I am wondering if Google Translate would be able to express in another language the same linguistic peculiarities as the English text. For Italian, the script was translated by Massimo Bocchiola, a literature translation professor at the University of Pavia. When faced the challenges posed by the text, he had to keep in mind the different stylistic choices, choose a speech pattern, and translate the neologisms of the text and slang words.
This type of consideration for Italian is valid also for other languages. For example, Korean would pose other difficulties with its four different levels of politeness depending on the situation in which the communication is taking place and who is speaking.
Google is the first to say that “automated translation is not complete. GNMT can still make significant mistakes, something that a human translator wouldn’t do, […] like translating single sentences without considering the meaning of the paragraph or what is written in the rest of the page.”
In the end, even if Google Translate could translate a patent in just a few minutes or a novel in as little as a month, it is also true that there are stylistic problems and language-cultural competences that go well beyond the literal meaning of the single words and that the machine can’t handle. Furthermore, even if we decided to switch to automated translation, there would still be a need for human translators for reviewing and post-editing.
Written by Marcella Sartore, Marketing & Communication Assistant @ Athena Parthenos