Opinion – Marcelo Viana: How have automatic translators improved so much?

by

The first experiments to automate text translation are almost as old as electronic computers. In 1954, researchers at IBM and Georgetown University wrote a program to translate phrases from Russian into English.

At the time, rule-based algorithms were used: programmers tried to include the grammars of the two languages ​​in computer instructions, explicitly describing how to translate each type of sentence.

It took a lot of work and a lot of time and resources. An IBM 701 computer translated 60 sentences from Russian, proving that it was possible, but the results were poor. By the end of the decade, virtually no one was investing in this effort.

In the 1980s, Japanese researchers developed the first example-based algorithms. They used examples of translations already made (“The book is on the table”, The book is on the table) and adapted them to similar situations (“The pillow is on the sofa”, The pillow is on the sofa). For the first time, algorithms could evolve as they “teached” them more examples.

In 1990, IBM introduced the first statistical translation methods. Such methods use text equivalents in different languages ​​to develop statistical models, from which the computer can identify translation patterns. Translation just got better and faster. It was also the era of the universalization of the internet, which contributed a lot to popularize online translation services.

The most famous, Google Translate, was launched in 2006 with statistical translation technology. Despite relying on the colossal amount of data Google has access to to improve its performance, we all remembered that translations were still pretty much more or less, requiring human review.

And then, around 2015, the quality dramatically improved. It was the result of the migration from Google Translate to the technology of neural networks, which mimic how the human brain works.

Neural networks can be trained from existing translations, using so-called deep learning, in order to produce better and better results. They do not yet compete with professional translators, but their translations are already adequate for most practical purposes, putting foreign languages ​​within everyone’s reach.

.

You May Also Like

Recommended for you