Machine translation is no longer a fantasy but a completely familiar technology. With the advent of neural networks, a revolution took place. The first breakthrough resulted when a machine translation system performed translation the same way as a human translator (in a certain scenario, Chinese–English news translation).
It was an exciting breakthrough in machine translation research, but the system that was built for this project was a complex, heavyweight research system incorporating several cutting-edge technologies. Today, there is the availability of the latest generation of neural machine translation models in production.
Translation: From Manual To Machine
Before the advent of computers, the word “translator” was associated exclusively with humans. To translate a book, document, letter, or article, people had to turn to specialists. However, with the advent of computers, computer translation began to develop: first in the interests of special services, and then – science, business, and everyone who had access to a PC.
Real-time website translation tool today demonstrates unprecedented speed and equally unprecedented quality. With the help of modern linguistic solutions, you can get coherent, smooth, and accurate text in another language that does not require human editing in a matter of seconds. This opportunity is given to us by the use of neural networks.
Built, Built … And Finally Built
The first machine translation systems were based on an approach called rule-based translation, or Rule-Based Machine Translation (RBMT). This approach simulated human activity and the system worked based on linguistic information about the source language and the target language.
To create a system based on RBMT technologies, professional linguists and programmers were required, as well as a lot of time to develop rules and bilingual dictionaries. The advantage of the system is that it is easy to add words and phrases to it – since the main RBMT configuration tool is a dictionary.
Neural – A New Word In AI
For machine translation to be applied to specific business tasks, translation of industry documents, business correspondence, user-generated content, a combination of technology, product, and relevant data is required.
Most neural network translators work in the cloud. The reason is obvious – a lot of computing power is required for neural translation. In addition, cloud developers offer the user a translation in exchange for their data.
This means that information that the user translates using cloud services can be made publicly available. Users, in turn, agree with this: some deliberately, others – because they inattentively read the terms of the user agreement.
For some business tasks, confidentiality is not critical, but for many corporate customers, it is absolutely impossible because company data is intellectual property, which is protected no less carefully than material property.
Neural Translation – Human Assistant
The quality of neural network translation has reached such a level that the use of technologies will enter the life of every modern specialist even more densely and even affect the translation profession.
The speed, quality, and confidentiality guarantees that machine translation will be used in all industries, including business and science. Professional translators will increasingly post-edit and prepare data for training machine translation systems.
Of course, there will not be a machine uprising shortly, and the computer will not be able to replace humans in translation completely. However, machine translation will be included in professionals’ toolboxes and will become an essential assistant that will speed up and simplify their work. Routine – to the machine, creativity – to the person.
Real-time website translation tools are getting better and better. But, as machine learning experts have noted, “the winner is not the one with the better algorithms, but the one with the most data,” so the quality and quantity of content are still one of the most important components of any successful translation system.