Manivald listens to what you say in contemporary Estonian and tries to express it in his own XVIII century Estonian. Due to his respectable 300-year old age he does not know most of the words that the contemporary youth use, and his hearing could be better. So please be patient with him and don't get upset if he makes a mistake!
Manivald's "brain" includes two independent systems: hearing (speech recognition) and interpretation (machine translation). Both systems are built using machine learning. This means that his behavious was not described by programmers, rather he learned it himself based on a large number (tens and hundreds of thousands) of examples. More specifically we used the algorithms called neural networks. After learning for days or even weeks Manivald can generalize based on the examples he has seen and process sentences that he has not heard before.
The speech recognition module was developed by the lab of phonetics and speech technology of the cybernetics institute, Tallinn University of Technology. Machine translation models were trained using the texts from the contemporary and XVIII century bible, collected by the Institute of the Estonian Language. Manivald was brought to life by us!, the research group of natural language processing, institute of computer science, University of Tartu.
Come meet Manivald at the Robotex exhibition in Tallinn, December 2nd-4th!