Supervised and Unsupervised Neural Approaches to Text Readability

Abstract

We present a set of novel neural supervised and unsupervised approaches for determining readability of documents. In the unsupervised setting, we leverage neural language models, while in the supervised setting three different neural architectures are tested in the classification setting. We show that the proposed neural unsupervised approach on average produces better results than traditional readability formulas and is transferable across languages. Employing neural classifiers, we outperform current state-of-the-art classification approaches to readability which rely on standard machine learning classifiers and extensive feature engineering. We tested several properties of the proposed approaches and showed their strengths and possibilities for improvements.

Predicting Text Readability from Scrolling Interactions

“…We compare our result to a stateof-the-art readability system by Vajjala and Lučić (2018) referred to as VAJJALA-2018. This system uses a multilayer perceptron classifier and has been shown to outperform BERT-based approaches on the OneStopEnglish dataset (Martinc et al, 2019). The system relies on 155 hand-crafted features which are grouped into six categories: traditional metrics, word features, psycholinguistic, lexical richness, syntactic and discourse features.…”