A Corpus for Automatic Readability Assessment and Text Simplification of German

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3302–3311 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3302 A Corpus for Automatic Readability Assessment and Text Simplification of German Alessia Battisti, Dominik Pf¨utze, Andreas S¨auberli, Marek Kostrzewa, Sarah Ebling Institute of Computational Linguistics, University of Zurich Andreasstrasse 15, 8050 Zurich, Switzerland {alessia.battisti, dominik.pfuetze, andreas.saeuberli}@uzh.ch, marek kostrzewa@yahoo.co.uk, ebling@cl.uzh.ch Abstract In this paper, we present a corpus for use in automatic readability assessment and automatic text simplification for German. The corpus is compiled from web sources and consists of parallel as well as monolingual-only (simplified German) data amounting to approximately 6,200 documents (nearly 211,000 sentences). As a unique feature, the corpus contains information on text structure (e.g., paragraphs, lines), typography (e.g., font type, font style), and images (content, position, and dimensions). While the importance of considering such information in machine learning tasks involving simplified language, such as readability assessment, has repeatedly been stressed in the literature, we provide empirical evidence for its benefit. We also demonstrate the added value of leveraging monolingual-only data for automatic text simplification via machine translation through applying back-translation, a data augmentation technique. Keywords: Simplified German, automatic readability assessment, automatic text simplification, multimodal 1. Introduction Simplified language is a variety of standard language char- acterized by reduced lexical and syntactic complexity, the addition of explanations for difficult concepts, and clearly structured layout.1 Among the target groups of simplified language commonly mentioned are persons with cognitive impairment and learning disabilities, prelingually deaf per- sons, functionally illiterate persons, and foreign language learners (Bredel and Maaß, 2016). Two natural language processing tasks deal with the con- cept of simplified language: automatic readability assess- ment and automatic text simplification. Readability assess- ment refers to the process of determining the level of dif- ficulty of a text, e.g., along readability measures, school grades, or levels of the Common European Framework of Reference for Languages (CEFR) (Council of Europe, 2009). Readability measures, in their traditional form, take into account only surface features. For example, the Flesch Reading Ease Score (Flesch, 1948) measures the length of words (in syllables) and sentences (in words). While readability has been shown to correlate with such features to some extent (Just and Carpenter, 1980), a consensus has emerged according to which they are not sufficient to account for all of the complexity inherent in a text. As Kauchak et al. (2014, p. 2618) state, “the usability of read- ability formulas is limited and there is little evidence that the output of these tools directly results in improved un- derstanding by readers”. As a result, more sophisticated models employing (deeper) linguistic knowledge such as lexical, semantic, morphological, morphosyntactic, syntac- tic, pragmatic, discourse, psycholinguistic, and language model features have been proposed (Collins-Thompson, 2014; Heimann M¨uhlenbock, 2013; Pitler and Nenkova, 2008; Schwarm and Ostendorf, 2005; Tanaka et al., 2013). Automatic text simplification was initiated in the late 1990s 1The term plain language is avoided, as it refers to a specific level of …

Content truncated to 500 words