Euskarazko elkarrizketa sistema automatikoa sare neuronalen bidez

Abstract

This work presents a neural dialogue system capable of learning Basque. To this end, we build upon generative adversarial networks which implement the idea of the Turing test. We demonstrate that training such a dialogue system with corpora two orders of magnitude smaller than usual English corpora is feasible. Finally, we also found that preprocessing the Basque language according to its morphology helps training these neural models. To the best of our knowledge, this is the rst attempt to develop a neural dialogue system in Basque.