Initial deep network architecture (T1.3)

Deep neural networks will be adapted to morphologically rich languages by using character-level inputs and additional information on morphology (suffixes, prefixes, separately trained POS tags) (report and source code) (T1.3).