Basque-to-Spanish and Spanish-to-Basque machine translation for the health domain

Abstract

[EU]Master Amaierako Lan honek medikuntza domeinuko euskara eta gaztelera arteko itzulpen automatiko sistema bat garatzeko helburuarekin emandako lehenengo urratsak aurkezten ditu. Corpus elebidun nahikoaren faltan, hainbat esperimentu burutu dira Itzulpen Automatiko Neuronalean erabiltzen diren parametroak domeinuz kanpoko corpusean aztertzeko; medikuntza domeinuan izandako jokaera ebaluatzeko ordea, eskuz itzulitako corpusa erabili da medikuntza domeinuko corpusen presentzia handituz entrenatutako sistema desberdinak probatzeko. Lortutako emaitzek deskribatutako helbururako bidean lehenengo aurrerapausoa suposatzen dute.

[EN]This project presents the initial steps towards the objective of developing a Machine Translation system for the health domain between Basque and Spanish. In the absence of a big enough bilingual corpus, several experiments have been carried out to test different Neural Machine Translation parameters on an out-of-domain corpus; while performance on the health domain has been evaluated with a manually translated corpus in different systems trained with increasing presence of health domain corpora. The results obtained represent a first step forward to the described objective.