The translation of texts between different languages is a known complicated problem. The difficulty starts with the fact that the meaning of each word depends on the context. My high school English teacher liked to remember that “play” can be either a piece of equipment (“piece” in English) or a theatrical performance (“play”). To determine the meaning of each word, a good translator needs to pay attention to the entire sentence and possibly the surrounding sentences.
Then there are the grammar rules, which vary from one language to another. Unlike in Portuguese, in English the adjectives usually go before the nouns: “a big man” is not “a great man”. And in German verbs often go at the end of the sentence.
A joke that circulates among me speaks of an algebra book written by a German professor in two volumes: the theorems in the first and the verbs in the second. Have you thought about how the translation would be?
Not to mention idioms like “the cow went to the swamp”, Millôr Fernandes’ playful translation of “the cow went to the swamp”. A good translator also needs to know the culture of the society that generated the text.
For these and other reasons, the first attempts at computer translations produced mediocre, if not ridiculous, results. And yet today there are excellent quality translator algorithms that rival human ones.
The most famous and used is Google Translator, which covers more than 100 languages and 99% of the texts on the internet. DeepL Translate, developed by a small German company, claims to be the best, and my friends in scientific computing agree. I decided to test it with a trial by fire: translating the first stanza of the Brazilian national anthem into English, with its baroque exaggeration of stylistic artifices. The result of the translation was practically perfect, showing that DeepL understood the hymn text much better than I did the first time I read it!
How did we get to this point?
The first machine translations are almost as old as the first computers, dating back to the 1950s, but the technology has undergone several revolutions in these seven decades. The most recent —translation by neural networks— took place just half a dozen years ago, around 2015, and had a notable impact on the quality of the results. I’ll comment next week.
.
