When launched in the fifties, machine translation was initially used for military purposes, specifically to spy on the exchanges made between the Soviet Union and its allies. In 1966, the American Automatic Language Processing Advisory Committee (ALPAC) – in charge of evaluating the progress in computational linguistics – published a report showing its skepticism about the research done so far in machine translation. This report triggered the US Government decision to cut funds for this field and machine translation did not evolve for several decades.
The way the pioneers of machine translation designed their system was “rule based”, they literally programmed dictionaries and thousands of linguistic rules to set up their translation systems. A machine translation model developed with this approach is by definition, long and complicated to create, difficult to modify and not very flexible. In the 2000s, a radically new technological approach was found, with the emergence of statistical machine translation. The idea was then to use the increased processing power of modern servers and the massive amount of available data in order to “train” machine translation engines through a statistical process. Nowadays we go even further by applying machine learning to set up very fine-tuned Machine translation engines for a specific linguistic field or document type. This level of specialisation allows, therefore, the machine to produce much better quality translations.
Within the language industry, the question about the future of translators is now arising more and more. Will machine translation tools fully replace human translation?
Lingua Custodia has developed a customised machine translation solution, called Verto, which is specialised in the financial domain and our response to this question is very clear: Our tools help professionals performing the translation work faster: they gain in productivity, translate more in less time, but machine translation remains just a tool to help them !
Within the technical field, new technological approaches are able to do most of the work, thanks to a sharp machine learning solution. However, an expert will always be needed to validate the text and make some corrections before publication. In this case, the machine is complementary and appears to be a true ally for professional translators, who can handle larger volumes of texts in less time.
Regarding literary texts, they require a strong local adaptation and a pretty “loose” or pictorial translation: the use of automatic translation is not going to really help the translator, who is more a “trans-creator” in the target language. Therefore, the translators of “creative” texts have their place and we are still far from the day when poetry would be automatically translated. Machines are very efficient to “learn” but teaching them how to be creative is still very complex!
To continue the discussion on automated translation engines, please join us on our LinkedIn page.
For more information about VERTO, our customised translation tool specialised in the financial field, please contact us.