PlainMedScale - A Corpus of Multi-Level Simplified Medical Texts in German and English

Oct 22, 2026

Uploaded by OKfine

PlainMedScale introduces a topic-aligned medical corpus covering four comprehensibility levels for German and English. The tiers map to distinct communicative functions: reference, explanation, decision support, and access, moving beyond the simple expert–lay split. Two pilot studies show readability metrics from two-register settings do not generalize across the full continuum, and a SOTA open-weight LLM prompted for plain language only partially preserves input difficulty. Code and data are released for research.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

Related Work

Dataset and Parallel Passages

Table 1: Etiology of Appendicitis across Four Sources

PlainMedScale: A Corpus of Multi-Level Simplified Medical Texts in German and English Bruno Brocai Heidelberg University German Linguistics Ilaria Papagno Heidelberg University Translation Studies Mayumi Ohta Heidelberg University SFB 1671 {bruno.brocai@gs, ilaria.papagno@iued, ohta@cl}.uni-heidelberg.de arXiv :2608 .0 1 158v 1 [ cs .CL] 2 Aug 2026 Abstract We introduce PlainMedScale, a topic-aligned medical corpus spanning four levels of comprehensibility in German and English, drawn from MSD Manuals (professional and consumer), Gesund.Bund, Apotheken Umschau (Einfache Sprache), and the NHS. The four tiers correspond to distinct communicative functions—reference, explanation, decision support, and access — and move beyond the binary expert–lay contrast of prior corpora. In two pilot studies enabled by the alignments, we show that many readability metrics established on two registers fail to generalize across the full gradient, and that a SOTA open-weight LLM prompted for Plain Language still partially preserves the difficulty of its input. Code ([https://github.com/GS](https://github.com/GS)Uni-Heidelberg/PlainMedScale) and data (10 .5281/zenodo.21728290) are made available. 1 Introduction Providing high-quality, accessible medical information has been widely acknowledged as a key determinant of enhancing health literacy, defined as individuals’ capability to locate, comprehend, and utilize health-related information in their daily lives (Sørensen et al., 2012 ; Council of Europe, 2023) . Simplification of medical texts into Plain Language should provide access to relevant medical information for a wider audience, including citizens with low literacy skills, cognitive impairments, or a second language background (Inclusion Europe, n.d. ; Laban et al., 2021) . Recent work on Plain Language (Einfache Sprache) guidelines (Maaß, 2020 ; Bock, 2015) supports the translation of specialized content into lay-friendly texts. Atthe same time, automatic text simplification has advanced considerably (Lyu and Pergola, 2024), offering a potential solution for making specialized knowledge more accessible at scale—provided suitable corpora. However, existing German medical text corpora exhibit some limitations. Many present sentence alignment, which enables a detailed analysis of linguistic transformations, but fail to capture content transposition, tending to produce same-length document pairs, without considering the summarization nature of the simplification process, as pointed out by Aumiller and Gertz (2022) . Most of the existing sources align two registers, a professional version and a single simplified counterpart, thereby reducing medical text simplification to a binary expert-lay contrast. Yet, medical text simplification is not a uniform process but rather a continuum spanning multiple degrees of accessibility. Capturing this gradient requires aligning more than two levels of difficulty. To this end, we extend the English and German texts of the multilingual multiMSDCorpus (Horiguchi et al., 2025), with additional lay-oriented resources in both languages. The resulting document-aligned corpus integrates four complementary sources per language, covering different degrees of medical language difficulty. This corpus serves two key purposes: (1) to demonstrate, through a comparison of different steps of linguistic and content transformations across registers, that conventional patient versions no longer meet the required level of acceptability; and (2) to provide four aligned sources at different difficulty levels, offering a more comprehensive picture of the many facets of medical text simplification. 2 Related Work For English, Med-EASi (Basu et al., 2023) provides short expert–lay text pairs annotated with simplification strategies (elaboration, replacement, insertion, deletion), while JEBS (Xia et al., 2025) is a larger resource targeting fine-grained lexical simplification of complex biomedical term

FAQs

What is PlainMedScale and what languages does it cover?

PlainMedScale is a topic-aligned medical corpus designed for multi-level text simplification. It covers German and English, using content from multiple medical information sources.

How do the four comprehensibility levels relate to communicative functions?

The four tiers correspond to different communicative functions: reference, explanation, decision support, and access. They are intended to reflect increasing degrees of accessibility rather than a binary expert–lay contrast.

What do the pilot studies show about readability metrics and LLM simplification?

The studies indicate that readability metrics trained or established on two-register settings fail to generalize across the full gradient. They also show that an open-weight LLM prompted for plain language only partially preserves the difficulty of its input.

Oct 22, 2026

Uploaded by OKfine