Year: 2023
Type: dissertation
Abstract: In this study, we train and evaluate state-of-the-art Transformer models for abstractive text summarization in Icelandic. We introduce the first publicly available abstractive summarization dataset for Icelandic, RÚV Radio News (RRN), which we use for training and evaluating our models. Our findings reveal that when fine-tuned for a low-resource summarization setting, the multilingual mT5 model outperforms a similar capacity PEGASUS model that is pre-trained exclusively on Icelandic text. Additionally, we discover that data augmentation, using machine-translated text, can result in ungrammatical output text.
Source: Skemman
Author: Þór Sverrisson
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7045297103