DEplain: A German Parallel Corpus with Intralingual Translations into Plain Language for Sentence and Document Simplification

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 16441–16463 July 9-14, 2023 ©2023 Association for Computational Linguistics DEPLAIN: A German Parallel Corpus with Intralingual Translations into Plain Language for Sentence and Document Simplification Regina Stodden, Omar Momen and Laura Kallmeyer Heinrich Heine University Düsseldorf, Germany {firstname.secondname}@hhu.de Abstract Text simplification is an intralingual transla- tion task in which documents, or sentences of a complex source text are simplified for a target audience. The success of automatic text simplification systems is highly depen- dent on the quality of parallel data used for training and evaluation. To advance sentence simplification and document simplification in German, this paper presents DEPLAIN, a new dataset of parallel, professionally written and manually aligned simplifications in plain Ger- man (“plain DE” or in German: ‘Einfache Sprache”). DEPLAIN consists of a news- domain (approx. 500 document pairs, approx. 13k sentence pairs) and a web-domain corpus (approx. 150 aligned documents, approx. 2k aligned sentence pairs). In addition, we are building a web harvester and experimenting with automatic alignment methods to facili- tate the integration of non-aligned and to be- published parallel documents. Using this ap- proach, we are dynamically increasing the web- domain corpus, so it is currently extended to approx. 750 document pairs and approx. 3.5k aligned sentence pairs. We show that using DEPLAIN to train a transformer-based seq2seq text simplification model can achieve promis- ing results. We make available the corpus, the adapted alignment methods for German, the web harvester and the trained models here: https://github.com/rstodden/DEPlain. 1 Introduction Automatic text simplification (TS) is the process of automatically generating a simpler version of complex texts while preserving the main informa- tion (Alva-Manchego et al., 2020b). Current TS re- search mostly focuses on English and on sentence- level simplification. This paper contributes to TS research on German. Compared to other European languages, German is more difficult to read due to complex sentence structures and many compound words (Marzari, 2010). According to Buddeberg and Grotlüschen (2020), roughly 6.2 mio. adults in Germany (ap- prox. 12.1%) have reading and writing problems on the character-level (0.6%), word-level (approx. 3.4%) or sentence-level (approx. 8.1%). To coun- teract and make texts accessible to more people, currently two dominant German variants for sim- plified language exist (Maaß, 2020): 1. easy-to-read German (de: “Leichte Sprache”): following strict rules the com- plexity of the language is maximally reduced (almost corresponds to CEFR level A1). The main target group is people with cognitive or learning disabilities or communication impairments. 2. plain German (de: “Einfache Sprache”): re- duced complexity with a mild to a strong ex- tent (almost corresponds to CEFR levels A2 and B1), which can be compared to texts for non-experts. The main target group is people with reading problems and non-native German speakers. This is also reflected in a rise in research and ap- plication of manual and automatic German text simplification: i) Many German web pages are pro- vided in standard German as well as in plain …

Content truncated to 500 words