Machine translation (MT) transfers information from one language to another by a computer without human intervention. NMT is the most advanced method of computer-generated translations. Thanks to autonomous learning based on artificial intelligence.
Neural Machine Translation
Neural Machine Translation (NMT) works on a completely different principle than previous statistics-based or rule-based machine translation methods. The NMT uses a large neural network based on the human brain model with the help of artificial intelligence.
The NMT is the most advanced method of computer-generated translations. Thanks to autonomous learning based on artificial intelligence, Big Data, and Deep Learning, this method has made enormous progress in the last recent years. Nowadays, neural machine translation programs can be used as a basis for professional translations.
Neural algorithms can generate accurate translations as well as learn a language. This technology improves the quality of the translated data continuously. For the NMT to work well, it must be trained to be used. Training means that a large amount of data feeds the algorithms to improve the reliability of the final results.
Human language translation is a tricky business. Languages offer many ways of expressing something. Languages use different letter systems, grammar, and spelling. A lot of pitfalls are on the way from one language to another.
Read about the the most frequent sources of issues with the sections below.
Everyday language (slang)
Everyday language is exposed to rapid social development. Nobody knows which words that are currently on everyone’s lips are yesterday’s news again. In addition, terms change their meaning.
NMT marks the beginning of a new era for artificial intelligence. Neural networks for translation no longer operate according to the rules of grammar. These procedures establish their own rules and even create their own (meta-) language.
NMT algorithms first translate the sentences of a text into their own machine- language. Then the algorithm thinks about how this machine text is translated into an understandable sentence for the human language. Quite similar what a human brain would do.
The next issue comes up by written words in large and lower case. In English, for example, it is often not clear whether there are words the lower case ones are nouns or verbs.
Usually, people recognize it based on their language experience in no time. Computer algorithms do not. They require a lot of calculations to find this out.
The problem with any translation is ambiguity. Human beings notice ambiguities mostly not because they usually recognize the meaning through the relationship. The problems related to ambiguity cannot be solved either by more data. The opposite is the case. Larger amounts of data can even worsen the problem because more ambiguities arise (Paradox of Knowledge).
One study tested Google Translate from 2010 to 2013 with the same text. 2010 was the translation bad. In 2011, the translation got better, and in 2012 it got even better. However, in 2013 the quality of the translation fell and got worse again.
Poems or verse
Artificial intelligence continues to struggle with the enormous complexity of human language, and nowhere is language as complex and meaningful as in literature. In music, novels, poems, and plays, the beauty of words often lies in nuance and subtlety.
A very successful way out of the hell of complexity for the translation of lyrical texts is proven by the algorithms of the NMT (Neural Machine Translation). These systems can learn over time, including the complexity of literary translations.
For a correct translation, it is recommended to use simple language that does not use country-specific expressions (slang). A simple structure of the sentences will help enormously to consider the foreign language’s grammar and spelling.
Write short sentences and especially avoid complicated subordinate clauses. Avoid rare words or words that are only used in certain languages. A simple language will not always be easy, especially when good language is important for you.
Test your text with freely available translators on the internet to check the outcomes. Those first translations will be the base for optimizations on your text for internationalization.
Writing texts for machine translation is a thing of its own. If you run a website international, the effort will be very worthwhile.
Translation with DeepL
DeepL offers one of the best NMT engines worldwide. The translater can be used online or by an application (app) installed on a computer locally. The third option to translate is the cloud-based service offered by the DeepL API to be integrated into custom apps or websites, for example.
J1 integrates the cloud-based translator service for the Free service and professional translations using the Pro system. The current API version V2 supports the translation of more than 25 languages vice versa.
Try the translator with Jekyll One and DeepL on your own!
Check the quality of DeepL’s NMT engine. Translate some of the passages from the sections. See typical machine translation limitations with the following examples.
Slang is a typically spoken language, not written. Many slang words are not known to the algorithms because they are seldom used for written text. Unknown words or proper names are not translated and returned for translation as they are.
Bloke would be the American English equivalent of dude.
Everyday language use quite often common words in a different meaning. A good reason to avoid such expressions.
Not my religion
Typical mistakes students make are bloopers. Translations may become wrongly understood sometimes the same way.
The thing that first caught my eye was a large silver cup that Charles had won for skating on the mantelpiece
Human languages are full of ambiguities. A good example of using them explicitly is a Joke.
A man tells his doctor: Doc, help me. I’m addicted to Twitter! The doctor replies: Sorry, I can't follow you.
Many translations for song lyrics are wrong and often do not give meaning to what the artists express. The reason: using machine translators of bad quality.
Check DeepL using different language scopes!
I guess the time was right for us to say
we’d take our time and live our lives together day by day
We’ll make a wish and send it on a prayer
We know our dreams can all come true with love that we can share
With you I never wonder
will you be there for me?
With you I never wonder
you’re the right one for me
I finally found the love of a lifetime
A love to last my whole life through
I finally found the love of a lifetime
forever in my heart