The chatbot lets patients use their own language to describe their symptoms and for doctors to establish a diagnosis and discuss treatment in that language.
It won machine translation experts Systran and artificial intelligence specialist Pertimm an award at the AI Paris artificial intelligence symposium as it keeps patients’ medical details private by avoiding human translators.
The software was dubbed the “e-health agent of the future” as the program continually learns and improves its translations.
It raises the question of whether conversational programs will one day replace the need for human translators.
Artificial intelligence has already made online translation better than before with Google Translate translating web pages or blocks of text with reasonable accuracy.
While still sometimes clunky and with occasional howlers, translations were so bad when Google launched Translate they were funny.
Early systems were based on statistical translation (ST) systems recognising and translating words and texts using a set of rules. However neural machine translation (NMT) – where computers learn without being programmed – saw a move from a focus on words and phrases to their use in context, and gives much better and appropriate translations.
Like Systran’s chatbot, Google Translate and other online translation software such as DeepL uses NMT to give reasonably accurate translations most of the time, at least in the main European languages.
However, AI translation cannot yet improve upon the work of human translators.
A test in South Korea this year pitted machine translation against humans, translating texts from Korean to English and vice versa.
The results said “90% of the neural machine translated text was ‘grammatically awkward,’ or “never the kind of translation produced by an educated native speaker.”
Even so, Systran is upbeat on the potential of its chatbot. Sales and marketing director Gaëlle Bou said: “What’s exciting about this project is its humanity: the best of technology in semantic research, AI and neural translation, serving patients and giving quality medical care.”
Despite this, human translators do not see their jobs threatened, especially in legal and other complex fields – including medical areas.
“Artificial intelligence is being used to get cars to drive by themselves, but that is simple compared to having to translate legal documents,” translator Myriam Barbier from La Rochelle told Connexion. “There is so much knowledge and history which has to be understood that I am sure humans will be used for many years to come for legal and technical translation.”
She said only one client had come to her with a machine-translated text for correction, and says most of the time assermenté (sworn) translators work using original documents.
Advances in translation software and systems were highlighted when the French travel guide Petit Futé sold four million books in Mandarin with the translation work carried out by Systran’s automatic system.
The guide publishers said 70% of the machine-translated text was accurate, allowing them to speed up the work of the human translators, who remain ultimately in charge. In addition, the system is designed to ‘learn’ from any human corrections so it can only improve with time.