Signwrite: A corpus of written Swedish texts and keystroke data from deaf and hard-of-hearing children and adolescents

The Signwrite corpus This repository contains the Signwrite corpus of written Swedish texts and keystroke data from deaf and hard-of-hearing children and adolescents. The data is partly longitudinal, with a total of 74 texts from 48 different participants. Furthermore, all texts have been manually corrected and the corrections annotated with error types, following the SweLL project guidelines. The intended use of the corpus is in research on the writing development of children, as well as for the creation and evaluation of natural language processing tools for writing support and assessment. The Signwrite corpus is further described in the following publication: Moa Gärdenfors, Robert Östling, Lisa Rudebeck and Mats Wirén. A corpus of written Swedish texts and keystroke data from deaf and hard-of-hearing children and adolescents. To appear in Proceedings of the 15th Workshop on Natural Language Processing for Computer Assisted Language Learning (NLP4CALL 2026). November 2026. Gothenburg, Sweden. The dataset has the following DOI: 10.5281/zenodo.23803517 Acknowledgments This work has been supported by Språkbanken, a Swedish national research infrastructure consortium jointly funded by its 10 partner institutions and the Swedish Research Council (current phase 2025--28, project id 2023-00161). License The data is licensed under the terms of Creative Commons Attribution 4.0(CC-BY). Normalized files The file texts.json contains text and participant metadata, texts corrected to standard Swedish writing conventions, as well as correction annotations (error types). The data is using the same JSON format as the SweLL corpus. Keylogs The keylogs subdirectory contains ScriptLog files in JSON format. The file names match the text keys (and essay_id field) in texts.json. Each name contains two numbers, where the first one is a participant identifier and the second one (1 or 2) indicates the session number. Not all participants produced a text in both sessions.