Napovedovanje naglasa slovenskih besed z metodami strojnega učenja

Krsnik, Luka (author), Robnik Šikonja, Marko (mentor), Šef, Tomaž (comentor)

Za naglaševanje slovenskih besed ne obstaja preprost algoritem, naglasa slovenskih besed se namreč govorci naučimo med njihovim spoznavanjem. Metode strojnega učenja so se pri naglaševanju izkazale za uspešne. V magistrski nalogi smo na problemu preizkusili globoke nevronske mreže. Testirali smo različne arhitekture nevronskih mrež, več različnih predstavitev podatkov in ansamble mrež. Najboljše rezultate je vrnil ansambelski pristop, ki je pravilno napovedal 87,62 % besed iz testne množice. S predlaganim pristopom smo za več odstotkov izboljšal rezultate drugih metod strojnega učenja.

• Language: Slovenian
• Keywords: umetna inteligenca, strojno učenje, globoke nevronske mreže, procesiranje naravnega jezika, naglaševanje
• Work type: Master's thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2017
• PID: 20.500.12556/RUL-98276
• Publication date in RUL: 23.11.2017

Secondary language

• Language: English
• Title: Prediction of stress of Slovenian words with machine learning methods

Abstract

There is no simple algorithm for stress assignment of Slovene words. Speakers of Slovene are usually taught accents together with words. Machine learning algorithms give positive results on this problem, therefore we tried deep neural networks. We tested different architectures, data presentations and an ensemble of networks. We achieved the best results using the ensemble method, which correctly predicted 87,62 % of tested words. Our neural network approach improved results of other machine learning methods and proved to be successful in stress assignment.

• Keywords: artificial inteligence, data mining, machine learning, deep neural networks, natural language processing, accentuation