Neurona-sareetan oinarritutako euskararako korreferentzia-ebazpena

Abstract

Lan honek euskararako korreferentzia-ebazpenean egindako lanari jarraipena ematea du helburu, korreferentzia-ebazpenerako neurona-sareetan oinarritutako sistema bat eraikiz. Horretarako polonierarako eraikitako sistemabat hartu da abiapuntutzat, eta euskarara egokitu.EPEC-KORREF corpusetik abiatuta, aipamen-bikoteaketa hauen ezaugarriak erauzi dira eta neurona-sarea entrenatu da aipamen-bikoteak korreferenteak ote direnerabakitzeko. Jarraian, neurona-sarearen iragarpenetatik korreferentzia-klusterrak sortu eta ebaluatu egin dira.CoNLL metrikan % 41,20 puntuko F1 balioa lortu da eta lortutako emaitza baxuak corpusaren tamaina txikiagatikdirela ondorioztatu da.

This work aims to continue with the previous work done in coreference resolution for Basque language, building asystem for coreference-resolution based on neural networks. For this purpose, a system for Polish has been usedand adjusted for Basque. Mention-pairs and their features were extracted from the EPEC-KORREF corpus, totrain the neural network model at the task of deciding if the pair is coreferent or not. After that, the coreferenceclusters were built from the output of the neural network, and these were evaluated. 41.20% F1 points of theCoNLL metric were obtained and it was concluded that the low results obtained were due to the small size of thecorpus.