Master Tesia

Rol semantikoen etiketatzea (SRL) garrantzia handia hartzen ari den hizkuntzalaritza konputazionalaren barneko alorra da. Izan ere, Association for Computational Linguistics (ACL) erakundeak oinarrizko NLP (Natural Language Processing) ikerketa lerrotako dauka. Lengoaia Naturalaren Prozesamenduaren alorrean kokatzen diren aplikazioen garapena aurrera eraman ahal izateko, askotan, beharrezkoa izango da informazio semantikoa, eta hain zuzen ere, rol semantikoak etiketatuta dauzkaten corpusak eskura izatea. Makina bidezko itzulpenean eta testu-laburpen automatikoan, esaterako, lagungarria izan daiteke rolek eskaintzen duten informazioa, besteak beste emaitza hobeak lortu ahal izateko. Lan honek euskarazko aditzen rol semantikoak era automatikoan etiketatzeko gaitasuna izango duen sistema baten garapen prozesua deskribatzen du. Honetan, VerbNet eredua jarraitzen duen EPEC-ROLSEM corpusa erabili da. Hasiera batean, bi hurbilpen ezberdin aztertzen dira: Lehen hurbilpena erregela linguistikoetan oinarrituta rol semantikoak etiketatzen dituen sisteman datza. Bigarrena, ordea, ikasketa automatikoko teknikak (Machine Learning) erabilita garatutako sistema bat inplementatzean datza. Ondoren, azken etiketatzailea garatu da hurbilpen bakoitza inplementatzen duten sistemen gainean egindako esperimentuetatik lortu diren emaitzetan oinarrituta. Abstract: Semantic Role Labeling (SRL) is a research area on the rise in Natural Language Processing and is listed as a core NLP task by the Association for Computational Linguistics (ACL). As a matter of fact, having a large corpus with annotation of semantic roles is crucial for the development of applications and advanced systems for machine translation, language learning, text summarization and many others. This work describes the process followed to develop a system for the automatic labeling of semantic roles for Basque verbs. EPEC-ROLSEM, a corpus labeled at predicate level following the VerbNet model has been used for this purpose. At first, two different approaches are considered: The first one consists of a system that will label semantic roles based on a set of linguistic rules; on the other hand, a second approach consisting of a system developed using Machine Learning (ML) techniques is considered. Afterwards, the final labeler is implemented based on the results that were obtained from experiments, which had been performed on systems that used both approaches.