A few clicks and anyone can convince themselves that machine translation (MT) software now performs well enough to offer a bona fide service, from which professional translators and lay users alike may benefit.
The users and uses of MT tools are many: primarily used to view material written in a language that the user does not know, it can also be used to provide information in other languages, communicate with speakers of another language, increase the production speed of human translations, help with writing in a foreign language, learn new languages, and so on. Some MT uses do not even involve any human reading or writing. For example, MT can be used to consistently index or categorize documents written in several languages for the purposes of information searching or detecting fake or hateful content. The material that is put through MT systems is no less varied, ranging from single words or terms to technical and literary texts, as well as all forms of content available online: websites, forum posts, tweets, and so on. It is therefore unsurprising that online translation services are being used on an ever-increasing scale, with billions of translation requests covering several thousand language pairs being served every day.
This improvement in MT is the result of relatively recent technological innovations in the field of language processing and stems from the convergence of several factors. First and foremost is the development of computational models, which were once based on probabilistic models and today use “neural” computing architectures…