Deep bidirectional transformers for Italian question answering

Deep learning continues to achieve state-of-the-art results in several NLP tasks, such as Question Answering (QA). Unfortunately, the requirements of neural QA systems are very strict in the size of the involved training datasets. Recent works show that the application of Automatic Machine Translation is an enabling factor for the acquisition of large scale QA training sets in resource poor languages such as Italian. In this work, we show how these resources can be used to train a state-of-the-art deep architecture, based on effective techniques recently proposed within the Bidirectional Encoder Representations from Transformers (BERT) paradigm.