The evolution of machine translation
October 6, 2020 Specific Translations
Technological advances have been made in the translation market in recent years. Machine translation software is becoming increasingly efficient. As a result, its use is on the rise.
We will define machine translation, and outline the system developed in our translation agency and its advantages.
What is machine translation?
Machine translation is performed by computer using specialised translation software, without human intervention. It is an instant and fast translation method. Several companies, like Google, have created machine translation platforms, some of which are free, although others charge a fee.
Artificial intelligence for the benefit of translation
Artificial intelligence (AI) brings together several systems to imitate human thought and action. These systems are based on creating and implementing algorithms that allow computers to think and act like human beings.
In recent years, we have seen significant developments in translation software. In fact, as a result of the emergence of artificial intelligence, software is increasingly efficient: it therefore delivers better quality translations. That’s why we use this type of software.
When translating technical documents, translation engines provide a quality result. However, when it comes to literary texts, documents with more complex syntax, innuendo, a more formal, creative and thoughtful style, the translations are not as good quality.
Generic machine translation can be used in the legal field.
In marketing, however, it is more difficult for a machine to do a translation, but it is possible to train a machine with data that has a particular style. The machine will then reproduce the style that it has learned.
AI is made up of two disciplines: “Machine Learning” and “Deep Learning”.
They are different in several ways:
- Used for
| ||Machine Learning||Deep Learning|
|Database||Controllable||> 1 million data|
|Training||Must be performed by a human||Autonomous learning system|
|Algorithm||Editable algorithm||Neural network algorithm|
|Used for||Simple routine actions||Complex tasks|
Deep Learning allows machine translation software to record all possible combinations of words from one language to another.
At Intertranslations, we have developed our own system. Our team trains our software to comprehend as many words as possible which have been translated into different languages, and this learning is based on human translations.
The advantages of our in-house method
For about 4 years, the way that we train machines has been evolving; we switched to neural machine translation (NMT). This method, as explained above, was created thanks to the development of Artificial Intelligence.
A machine understands a linguistic combination (source language and target language) as well as specific terminology. One machine is used per language combination, and each machine is structured into terminology categories (legal, technical, etc.)
A machine’s performance depends on how well it is trained and the available data: the more data we have, the better, and the more it is trained, the better it performs. Translation data comes mainly from human translations, the more highly skilled the engineer, the more high-quality translations the machine will produce. The engineer must sort through the data and summarise it. For a small machine, 1 million model sentences are used to train it. For a larger one, it takes about 10 million.
The more our professionals produce human translations, the more they enrich our software with appropriate terminology. As a result, for simple translations with straightforward content, we save time because our in-house software produces them successfully. The result is of high quality thanks to the training carried out by our professionals.
This means that we can deliver high-quality translations, faster and at lower prices, to our regular clients, as well as to other clients – even if you do not have a database with us, you can benefit from the machine translation machines that we have already developed.