Poenostavljanje besedil v slovenščini z velikimi jezikovnimi modeli

Bone, Blaž (author), Robnik Šikonja, Marko (mentor)

V diplomski nalogi smo raziskali poenostavljanje besedil v slovenščini z uporabo velikih jezikovnih modelov. Cilj naloge je bil razviti modele, ki lahko učinkovito poenostavijo slovenska besedila. Uporabili smo obstoječe angleške učne množice, jih strojno prevedli v slovenščino, nato na teh podatkih naučili modele, kot so SloT5, mT5 in mBART. Izvedli smo kvantitativno in kvalitativno analizo rezultatov, pri čemer smo uporabili metrike, kot so BLEU, SARI, BERTScore in LaBSE Similarity. Rezultati so pokazali, da so modeli uspešno poenostavili besedila, ohranili ključne informacije in smiselno poenostavili strukturo in jezik. Kljub uspešnim poenostavitvam so modeli pogosto ponovili izvirne povedi brez večjih sprememb.

• Language: Slovenian
• Keywords: obdelava naravnega jezika, poenostavljanje besedila, strojno učenje, veliki jezikovni modeli
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-160702
• COBISS.SI-ID: 209156355
• Publication date in RUL: 03.09.2024

BONE, Blaž, 2024, Poenostavljanje besedil v slovenščini z velikimi jezikovnimi modeli [online]. Bachelor’s thesis. [Accessed 5 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=160702

Secondary language

• Language: English
• Title: Text simplification for Slovene using large language models

Abstract

In this thesis, we explored text simplification in Slovene using large language models. The goal of the thesis was to develop models that can effectively simplify Slovene texts. We used existing English training datasets, which we machine-translated into Slovene, and then trained models such as SloT5, mT5, and mBART on these data. We conducted quantitative and qualitative analysis of the results, using metrics such as BLEU, SARI, BERTScore, and LaBSE Similarity. The results showed that the models can successfully simplify texts, retain key information, and meaningfully simplify the structure and language. Despite the successful simplifications, the models often repeat the original sentences without significant changes.

• Keywords: natural language processing, text simplification, machine learning, large language models