When you need a quick translation, you turn to Google Translate. But it doesn’t always work the way it should. But there’s an alternative that TechCrunch claims is “as quick as the outsized competition, but more accurate and nuanced” than any they’d tried. It’s called DeepL.
What is DeepL?
DeepL Translator is a translation service that uses neural machine translation (NMT). It launched in August 2017, is privately owned, and based in Cologne, Germany.
Ever since the launch of DeepL Translator in August 2017, DeepL has been setting records for machine translation quality.
DeepL Pro, released in March of 2018, allows subscribers to unlock the full capacity of DeepL’s AI translation technology. DeepL Pro subscribers can benefit from an optimized web translator, integrate our algorithms into translation software, or build new apps and services with the DeepL API.
Let’s test an example. I’ve taken an original poem called Círculo Vicioso by Machado de Assis:
Bailando no ar, gemia inquieto vaga-lume:
– Quem me dera que fosse aquela loura estrela,
que arde no eterno azul, como uma eterna vela!
Mas a estrela, fitando a lua, com ciúme:
– Pudesse eu copiar o transparente lume,
que, da grega coluna á gótica janela,
contemplou, suspirosa, a fronte amada e bela !
Mas a lua, fitando o sol, com azedume:
– Misera ! tivesse eu aquela enorme, aquela
claridade imortal, que toda a luz resume !
Mas o sol, inclinando a rutila capela:
– Pesa-me esta brilhante aureola de nume…
Enfara-me esta azul e desmedida umbela…
Porque não nasci eu um simples vaga-lume?
| Google Translate | DeepL Translator | English translation |
|---|---|---|
| Dancing in the air, an uneasy firefly moaned: I wish it were that blond star, that burns in the eternal blue, like an eternal candle! But the star, looking at the moon, with jealousy: I could copy the transparent fire, that from the greek column to the gothic window, she gazed, sighingly, at the beloved and beautiful forehead! But the moon, staring at the sun, with bitterness: Misera! had I that huge, that immortal clarity, which all light sums up! But the sun, tilting the rutile chapel: This brilliant number aureole weighs me… Put on this blue and immeasurable umbilicus … Why am I not born a simple firefly? | Dancing in the air, a restless firefly moaned: I wish it were that blond star That burns in the eternal blue, like an eternal candle! But the star, staring at the moon, jealous If I could copy the transparent light, That from the Greek column to the Gothic window from the Greek column to the Gothic window, gazed, sighing, upon the beloved and beautiful brow! But the moon, staring at the sun, with bitterness: Misera! had I that enormous, that immortal immortal clarity, which all light sums up! But the sun, inclining the ruddy chapel: This bright aureole of nume weighs me down… Pities me this blue and unmeasured umbrella? Why was I not born a simple firefly? | Dancing in the air, the firefly moaned restlessly: – How I wish I was that blonde star, That burns in the eternal blue, live an infinite candle! But the star, gazing at the moon jealously: – Who am I to copy the transparent light, That from the Greek column to the gothic window, Contemplated, sighing, forehead beloved and beautiful! But the moon, gazing at the sun, sourly: – Misery! Had I that huge, that Immortal brightness, in which all live is summarized! But the sun, tilting its shining chapel: – This bright halo weighs down upon me… This blue and unmeasurable umbrella sickens me… Why wasn’t I born a simple firefly? |
While neither translator could emulate the original English translation (which likely had its own artistic flair), they both did a good job. DeepL picked up “restless” in the first verse compared to Google Translate’s “uneasy” and in the word fitando, DeepL goes for “staring”, which is closer to the original “gazing”, while Google Translate chooses “looking” which doesn’t have the same feeling.
That said, there were some glitches with DeepL, for example duplicating the line “from the Greek column to the Gothic window”, and neither picked up the word misera translated to “misery” but that’s not the traditional translation so that makes sense.
Maybe a poem wasn’t the best example. Here’s a paragraph from a news story from Le Monde:
L’heure de la réconciliation entre la France et le Rwanda est-elle arrivée ? Vingt-sept ans après le génocide des Tutsi, qui a fait près de 1 million de morts entre avril et juillet 1994, on veut le croire au sommet de l’Etat français. Emmanuel Macron reçoit son homologue rwandais, Paul Kagame, lundi 17 mai, à Paris, en marge d’une conférence sur le Soudan et à la veille d’un sommet sur le financement des économies africaines. Les deux hommes « ont pris l’habitude d’échanger en anglais dans un climat de confiance, de sincérité et de respect »,précise un membre de l’entourage du chef de l’Etat. Au cours de sa visite, Paul Kagame doit aussi s’entretenir avec d’anciens officiers de l’armée française, dont le général Jean Varret, responsable de la mission de coopération de 1990 à 1993, qui s’était opposé à sa hiérarchie lors de la tragédie.
| Google Translate | DeepL Translator |
|---|---|
| Has the time come for reconciliation between France and Rwanda? Twenty-seven years after the genocide of the Tutsi, which killed nearly 1 million people between April and July 1994, we believe him at the top of the French state. Emmanuel Macron receives his Rwandan counterpart, Paul Kagame, on Monday, May 17, in Paris, on the sidelines of a conference on Sudan and on the eve of a summit on the financing of African economies. The two men “have become accustomed to speaking in English in an atmosphere of trust, sincerity and respect,” said a member of the head of state’s entourage. During his visit, Paul Kagame must also meet with former officers of the French army, including General Jean Varret, responsible for the cooperation mission from 1990 to 1993, who opposed his hierarchy during of tragedy. | Has the time for reconciliation between France and Rwanda arrived? Twenty-seven years after the Tutsi genocide, which killed nearly one million people between April and July 1994, people at the top of the French government want to believe so. Emmanuel Macron received his Rwandan counterpart, Paul Kagame, on Monday 17 May in Paris, on the sidelines of a conference on Sudan and on the eve of a summit on the financing of African economies. The two men “have become accustomed to exchanging in English in a climate of trust, sincerity and respect,” said a member of the entourage of the head of state. During his visit, Paul Kagame is also expected to meet with former French army officers, including General Jean Varret, head of the cooperation mission from 1990 to 1993, who had opposed his superiors during the tragedy. |
For me, it’s a tie on that one. Google Translate’s first sentence felt more natural while DeepL’s “Tutsi genocide” vs. GT’s “genocide of the Tutsi” sounded better. DeepL’s “climate of trust, sincerity and respect” was better than GT’s “atmosphere of trust, sincerity and respect” as well as “opposed his superiors during the tragedy” vs. “opposed his hierarchy during of tragedy.”
Ultimately, this both use machine learning based on data that’s already out there and as language continues to change and evolve, it’ll be almost impossible to get things exactly right. But, for me, DeepL offers more nuance and less literal translations for words which is what you want as a human being.