Is manual translating still needed today?

When I see a text in a foreign language and want to know its content, I just copy it into a form available online or click the existing link "translate" in my application, and the text appears in my language. Where is the problem?

What does the result of a machine translation look like?

Translation engines can be better or worse. The worse ones usually return just a bunch of words from which you can guess the meaning of the original text if you're lucky. The good ones usually deliver grammatically and often even stylistically correct results. But you cannot bank on their correspondence to the meaning of the original. They often omit or replace units of meaning. The result can even mean the opposite of what the original said.

However, quality depends not only on the software used. It also depends a lot on the language pair and on the topic. The more (good) translations there are, the better the machine translation gets. In addition, texts containing a lot of special terms return better results. Such a term usually can and should be translated the same way throughout a technical text. This can also be seen by the way different people and translation engines deliver very similar or even identical translations: "Reihenmotoren haben meist zwei Nockenwellen" – "In-line engines usually have two camshafts""Рядные двигатели обычно имеют два распределительных вала".

But now let's see a couple of wrong machine translations!

Comparing your salary

A translation from Russian:

"Хоть я и зарабатываю втрое меньше вас, но без вас мы обойдемся." English transcription: Khot' ya i zarabatyvayu vtroe men'she vas, no bez vas my oboydyomsya.

("Although I earn only one-third as much as you, we could do without you.")

DeepL, an online translation engine with a good reputation, returns:

"Even though I earn three times less than you, we can do without you."

"Three times less" is not a correct way to say it, but at least in English you understand what it means. The translation to German is much worse, it says:

"Auch wenn ich dreimal so viel verdiene wie Sie, kommen wir ohne Sie aus."

which means "... I earn three times as much as you, ..." rather than the other way round.

However, adding some more context "а без меня никак" ("a bez menya nikak"), which means "but without me—no way", we get:

"Auch wenn ich dreimal weniger verdiene als Sie, können wir auf Sie verzichten, aber nicht auf mich."

So now the engine suddenly "got it". The translation is now very similar to the English one, with an incorrect way of saying the right thing.

But of course the engine has not really "got" anything, it just uses different model texts. But as we shall see now, you can't bank on more context leading to a more precise translation.

Kisses in the middle of the chatroom?

"Чмоки всем в этом чатике!" English transcription: Chmoki vsem v etom chatike!

(approximately: "Smacks/kisses to everyone in this chat!")

"Чмоки" ("chmoki") is an onomatopoeic word imitating the smacking sound of a kiss or an air kiss. "Чатик" ("chatik") is just the English word "chat" with a diminutive suffix. The diminutive does not necessarily mean the chat is small, but rather that it is nice or at least that you like it somehow. But this is not enough enthusiasm to really greet everyone with "kisses".

For "чмоки всем" DeepL returns

"xoxo to all" XOXO: hugs and kisses. The X means a kiss, the O a pair of arms hugging someone.

Wow, this is exactly what we need! Just what people really write in chats. This piece of software is a genius! The German translation also uses the word "xoxo" here.

But when I add the rest of the sentence (still with no capital letter in the beginning and no exclamation mark, the English translation remains acceptable ("smoochies for everyone in this chat room")—by the way, "chat room" is a good idea—but the German says just

"mitten in diesem Chatroom".

which means "in the middle of this chat room". The entire greeting gets lost. We get it back in the form of "smoochie" by putting the exclamation mark in place, but "smoochie" does not make any sense in German as far as I know.

In English the exclamation mark makes the phrase change to:

"smokey to everyone in this chat room!".

What, I beg your pardon, is "smokey"? Do people write that in chats nowadays? As a greeting?

Last of all I change the first letter to a capital letter. This does not change anything in the English version, but the German one changes to:

"Mwah! Mwah! Mwah! Mwah! Mwah! Mwah! Mwah!".

There you are! No comment ...

What is artificial intellect able to do?

Generally speaking—two things:

  • following rules and
  • imitating the behaviour of people.

Processing algorithms, that is, following fixed rules, is really the basic way any computer software works. So it seems strange that translation engines are able to use correct grammar only if the second strategy, analysing and using existing texts, is implemented really well. Obviously the rules of a language are too complex to be introduced as a fixed set of rules.

But this—following rules and imitating others—is exactly what we also do when translating and generally when speaking! What, then, is it an artificial intellect cannot do?

A piece of software cannot understand the meaning.

As a rule, correct grammar and good style is opposed to the reproduction of every single word. A human being who translates well also will omit a concept sometimes—but knowing that it isn't an important concept in the given context. You have probably heard that "good afternoon" is translated into many languages (including German and Russian) as "good day". The information that it is afternoon is not crucial and can be omitted.

A human being can imagine or even draw a picture of the meaning of a sentence without using words. And a human being knows what parts of the picture or moving scene are more or less important. So instead of saying "In-line engines usually have two camshafts" you might as well say "An in-line engine usually has two camshafts". The plural is not meaningful in this case; you imagine one engine representing many. In the same way, "good afternoon" makes you imagine a person greeting politely, while you know as background information that the time is between 12 and 5 p.m. Well, and "chmoki..." makes you see a young, probably female person coming into a favourite chat-room and warmly greeting the friends gathered there.

"But people also make mistakes!"

That is true. People even generate many typos and volatility errors a piece of software would never make. In addition to that, people with little translating experience tend to bend the grammar of the target language (even if it is their mother tongue!) in order to stay closer to the original (see "three times less"). In a hurry you might also fail to read the original text properly and mess up "one third" and "three times as much".

But a professional translator has a linguistic education and is sensitive to those sources of errors, reads the translated texts multiple times to spot different kinds of errors and—last but not least—also knows how to use the electronical helpers—from simple spellcheckers to special software for translators that helps you to overview the technical terms, always see original text and target text at the same time, and more. And this is also where machine translations enter the game as sources of good ideas (see "xoxo").

Back to top Back to top