DEplain/DEplain-web-doc

DEplain-web-doc: A corpus for German Document Simplification

DEplain-web-doc is a subcorpus of DEplain Stodden et al., 2023) for document simplification. The corpus consists of 396 (199/50/147) parallel documents crawled from the web in standard German and plain German (or easy-to-read German). All documents are either published under an open license or the copyright holders gave us the permission to share the data. If you are interested in a larger corpus, please check our paper and the provided web crawler to download more parallel documents with a closed license.

Human annotators also sentence-wise aligned the 147 documents of the test set to build a corpus for sentence simplification. For the sentence-level version of this corpus, please see https://huggingface.co/datasets/DEplain/DEplain-web-sent.

The documents of the training and development set were automatically aligned using MASSalign. You can find this data here: https://github.com/rstodden/DEPlain/. If you use the automatically aligned data, please use it cautiously, as the alignment quality might be error-prone.

Dataset Card for DEplain-web-doc

Table of Contents

• —Dataset Description
• —Dataset Summary
• —Supported Tasks and Leaderboards
• —Languages
• —Dataset Structure
• —Data Instances
• —Data Fields
• —Data Splits
• —Dataset Creation
• —Curation Rationale
• —Source Data
• —Annotations
• —Personal and Sensitive Information
• —Considerations for Using the Data
• —Social Impact of Dataset
• —Discussion of Biases
• —Other Known Limitations
• —Additional Information
• —Dataset Curators
• —Licensing Information
• —Citation Information
• —Contributions

Dataset Description

• —Repository: DEplain-web GitHub repository
• —Paper: "DEplain: A German Parallel Corpus with Intralingual Translations into Plain Language for Sentence and Document Simplification."
• —Point of Contact: Regina Stodden

Dataset Summary

DEplain-web (Stodden et al., 2023) is a dataset for the evaluation of sentence and document simplification in German. All texts of this dataset are scraped from the web. All documents were licenced with an open license. The simple-complex sentence pairs are manually aligned. This dataset only contains a test set. For additional training and development data, please scrape more data from the web using a web scraper for text simplification data and align the sentences of the documents automatically using, for example, MASSalign by Paetzold et al. (2017).

Supported Tasks and Leaderboards

The dataset supports the evaluation of text-simplification systems. Success in this task is typically measured using the SARI and FKBLEU metrics described in the paper Optimizing Statistical Machine Translation for Text Simplification.

Languages

The texts in this dataset are written in German (de-de). The texts are in German plain language variants, e.g., plain language (Einfache Sprache) or easy-to-read language (Leichte Sprache).

Domains

The texts are from 6 different domains: fictional texts (literature and fairy tales), bible texts, health-related texts, texts for language learners, texts for accessibility, and public administration texts.

Dataset Structure

Data Access

• —The dataset is licensed with different open licenses dependent on the subcorpora.

Data Instances

• —document-simplification configuration: an instance consists of an original document and one reference simplification.
• —sentence-simplification configuration: an instance consists of an original sentence and one manually aligned reference simplification. Please see https://huggingface.co/datasets/DEplain/DEplain-web-sent.
• —sentence-wise alignment configuration: an instance consists of original and simplified documents and manually aligned sentence pairs. In contrast to the sentence-simplification configurations, this configuration contains also sentence pairs in which the original and the simplified sentences are exactly the same. Please see https://github.com/rstodden/DEPlain

Data Fields

Data Splits

DEplain-web contains a training set, a development set and a test set. The dataset was split based on the license of the data. All manually-aligned sentence pairs with an open license are part of the test set. The document-level test set, also only contains the documents which are manually aligned. For document-level dev and test set the documents which are not aligned or not public available are used. For the sentence-level, the alingment pairs can be produced by automatic alignments (see Stodden et al., 2023).

Document-level:

Sentence-level:

: Documents per Domain in DEplain-web.

: Inter-Annotator-Agreement per Domain in DEplain-web-manual.

: Information regarding Simplification Operations in DEplain-web-manual.

Dataset Creation

Curation Rationale

Current German text simplification datasets are limited in their size or are only automatically evaluated. We provide a manually aligned corpus to boost text simplification research in German.

Source Data

Initial Data Collection and Normalization

The parallel documents were scraped from the web using a web scraper for text simplification data. The texts of the documents were manually simplified by professional translators. The data was split into sentences using a German model of SpaCy. Two German native speakers have manually aligned the sentence pairs by using the text simplification annotation tool TS-ANNO by Stodden & Kallmeyer (2022).

Who are the source language producers?

The texts of the documents were manually simplified by professional translators. See for an extensive list of the scraped URLs see Table 10 in Stodden et al. (2023).

Annotations

Annotation process

The instructions given to the annotators are available here.

Who are the annotators?

The annotators are two German native speakers, who are trained in linguistics. Both were at least compensated with the minimum wage of their country of residence. They are not part of any target group of text simplification.

Personal and Sensitive Information

No sensitive data.

Considerations for Using the Data

Social Impact of Dataset

Many people do not understand texts due to their complexity. With automatic text simplification methods, the texts can be simplified for them. Our new training data can benefit in training a TS model.

Discussion of Biases

no bias is known.

Other Known Limitations

The dataset is provided under different open licenses depending on the license of each website were the data is scraped from. Please check the dataset license for additional information.

Additional Information

Dataset Curators

DEplain-APA was developed by researchers at the Heinrich-Heine-University Düsseldorf, Germany. This research is part of the PhD-program `Online Participation'', supported by the North Rhine-Westphalian (German) funding scheme `Forschungskolleg''.

Licensing Information

The corpus includes the following licenses: CC-BY-SA-3, CC-BY-4, and CC-BY-NC-ND-4. The corpus also include a "saveuseshare" license, for these documents the data provider permitted us to share the data for research purposes.

Citation Information

@inproceedings{stodden-etal-2023-deplain,
    title = "{DE}-plain: A German Parallel Corpus with Intralingual Translations into Plain Language for Sentence and Document Simplification",
    author = "Stodden, Regina  and
      Momen, Omar  and
      Kallmeyer, Laura",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    notes = "preprint: https://arxiv.org/abs/2305.18939",
}

This dataset card uses material written by Juan Diego Rodriguez and Yacine Jernite.