Exploring German Multi-Level Text Simplification

Proceedings of Recent Advances in Natural Language Processing, pages 1339–1349 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_150 1339 Exploring German Multi-Level Text Simplification Nicolas Spring, Annette Rios, Sarah Ebling Department of Computational Linguistics, University of Zurich, Switzerland {spring,arios,ebling}@cl.uzh.ch Abstract We report on experiments in automatic text simplification (ATS) for German with multiple simplification levels along the Common Euro- pean Framework of Reference for Languages (CEFR), simplifying standard German into lev- els A1, A2 and B1. For that purpose, we inves- tigate the use of source labels and pretraining on standard German, allowing us to simplify standard language to a specific CEFR level. We show that these approaches are especially effective in low-resource scenarios, where we are able to outperform a standard transformer baseline. Moreover, we introduce copy labels, which we show can help the model make a dis- tinction between sentences that require further modifications and sentences that can be copied as-is. 1 Introduction Simplified language is a variety of standard lan- guage characterized by reduced lexical and syn- tactic complexity, the addition of explanations for difficult concepts, and clearly structured layout.1 Among the target groups of simplified language are people with cognitive impairment and autism spectrum disorder, prelingually deaf and function- ally illiterate people, and sometimes also foreign language learners and children (Bredel and Maaß, 2016). Automatic text simplification (ATS), the process of automatically producing a simplified version of a standard-language text, was initiated in the late 1990s (Carroll et al., 1998; Chandrasekar et al., 1996) and has since then been approached by means of rule-based and statistical approaches. As part of the rule-based paradigm, the operations carried out typically include replacing complex lexical and 1The term plain language is avoided here, as it refers to a specific level of simplification. Simplified language subsumes all efforts of reducing the complexity of a text. syntactic units by simpler ones. The statistical paradigm so far has mainly conceptualized the sim- plification task as a case of monolingual (sentence- based) machine translation (MT), i.e., as one of con- verting standard-language into simplified-language sentences using MT techniques (Specia, 2010). However, while in bilingual parallel texts used for MT, the relation between source and target sen- tences is mostly 1:1, ATS usually requires n:m alignments with unaligned parts in-between. ATS research has been documented for English (Zhu et al., 2010), Spanish (Saggion et al., 2015), Portuguese (Aluisio and Gasperin, 2010), French (Brouwers et al., 2014), Italian (Barlacchi and Tonelli, 2013), and other languages. Research on German is still sparse but has gained momentum in recent years due to a number of legal and polit- ical developments in German-speaking countries, such as the introduction of a set of regulations for accessible information technology (Barrierefreie- Informationstechnik-Verordnung, BITV 2.0) in Ger- many, the approval of rules for accessible informa- tion and communication (Barrierefreie Information und Kommunikation, BIK) in Austria, and the rat- ification of the United Nations Convention on the Rights of Persons with Disabilities (UN CRPD) in Switzerland. In this paper, we report on work in automatic simplification of …

Content truncated to 500 words