The article discusses the problem of complex texts hindering the ability of some people to read and understand them, even though they speak the language. To address this issue, the authors propose an approach for automatic text simplification of German narrative texts, with the aim of making these texts more accessible to a wider audience.
Motivation
The rise of the internet has made it easier and more convenient for people to access a vast amount of texts, but not everyone can fully read and understand them due to their complexity. Text simplification can help overcome this barrier by providing simpler versions of texts that are easier to comprehend. Narrative texts are an important means of how humans create meaning, so simplifying these texts can contribute to involving more people in the process.
Related Work
Previous research has focused on developing text simplification systems for various languages, including German. However, most of these approaches rely solely on automated methods, which may not always produce accurate results. The authors propose a new approach that combines both automated and human-based methods to improve the quality of simplified texts.
Methodology
The proposed approach consists of two stages: (1) automatic text simplification using a pre-trained language model, and (2) post-processing to further simplify the text while maintaining its meaning. The authors use a combination of automated metrics, such as BERTscore, BLEU, and ROUGE, to evaluate the quality of the simplified texts. They also employ human evaluations to ensure that the simplified texts are both accurate and easy to understand.
Results
The authors evaluate their approach on a dataset of German narrative texts and show that it produces high-quality simplified texts that are both accurate and easier to read than the original versions. They also compare their approach with existing text simplification systems and show that it outperforms them in terms of both automated metrics and human evaluations.
Conclusion
The article presents a new approach for automatic text simplification of German narrative texts, with the aim of making these texts more accessible to a wider audience. The proposed approach combines both automated and human-based methods to improve the quality of simplified texts, and shows that it outperforms existing systems in terms of both automated metrics and human evaluations. By providing simpler versions of narrative texts, this approach can contribute to involving more people in the process of creating meaning and shaping our environment.
