Regresar

Ecuadorian Kichwa to Spanish automatic translation: An initial approximation based on LSTM Neural Networks for educational applications

Abstract:

According to the latest estimates of UNESCO, currently, the Kichwa is experiencing a rapid loss among young generations. Some studies pbkp_redict that approximately half of 7000 World languages will be extinct by the end of the XXI century. For these reasons, it is fundamental to develop efforts to rescue and revitalize the indigenous languages as a fundamental part of the cultural heritage of ancient nations. In this paper, we describe an initial approximation to develop an automatic translation tool from the Ecuadorian Kichwa variant to Spanish. However, this task is challenging, because currently there are not enough digital bilingual texts (Ecuadorian Kiwcha-Spanish) to design automatic translation tools. Our proposal compared Long Short-Term Memory Neural Network (LSTM-NN) and Transformer NN as is described in state of the art. Although the initial results do not have the precision required to have a suitable translator (due to the lack of enough bilingual texts), this first approximation shows that it is feasible to use LSTM NN to continue the research and development process in this area.