A Corpus of Keystroke Logging From Children and Adult Writing

Study Design

The initial dataset investigated the effect of the writing modality when processing demands of higher-level writing processes vary. For all experimental conditions, each writer produced two argumentative texts, one in each modality (handwriting versus typewriting). Data collection followed a within-participant design for the writing modality, with each session separated by approximately two weeks. Notice that the online datasets accompanying this publication only include the typed data. We do not present the handwriting data since there is no solution available for automatic processing and extraction of variables.

Three sub-corpora are distinguished by the systematic manipulation of processing demands through a between-participants design:

• – Experiment 1 focused on the planning process by manipulating topic familiarity.

• – Experiment 2 targeted the translating process by saturating the phonological loop through the presence of background speech, compared to a silent control condition.

• – Experiment 3 addressed the monitoring and control processes by manipulating visual feedback (whether the text is visible or not on the computer screen).

Sample

The sample included students from the University of Poitiers (see Table 1), all enrolled in the second year of a Psychology degree. All participants were native French speakers and reported no attentional, linguistic or learning disorders. Data were collected from 2017 to 2018.

Data Anonymization and Ethical Issues

The study was conducted in accordance with the ethical principles of the Declaration of Helsinki. Participants were recruited through a course credit system and provided free and informed consent before starting the experiment.

Existing Use of Data

These data were originally collected and analyzed in the frame of Sirine Bouriga (2020)’s doctoral thesis, entitled “Papier-crayon vs. écran-clavier : Effets sur le coût cognitif et sur la dynamique de la production de textes” (Pen-and-paper vs. keyboard-and-screen: Effects on cognitive cost and text production dynamics). Data corresponding to bursts characteristics such as duration, size, fluency within bursts and type of burst (Production, Delayed-Revision or Immediate-Revision) have been used in Miletic et al. (2022) and Cislaru et al. (2025a; 2025b).

Study Design

Data were collected in the children’s schools. The children completed writing tasks using a computer equipped with an AZERTY keyboard. Participants produced two types of texts: a narrative and an expository text (20 minutes each, with counterbalanced order).

Sample

Data were collected across three schools in the Académie de Poitiers (86000, France; see Table 2 for details). All participants were native French speakers who did not present any language or learning disorders as reported by parents.

Data Anonymization and Ethical Issues

This study was conducted in accordance with General Data Protection Regulation (GDPR). Participation required informed consent from both parents and children. At the very beginning of the experiment, children were explicitly informed, orally and in writing, that participation was entirely voluntary and that they could withdraw at any time without any consequences. Ethical approval was granted by the Comité d’Ethique pour la Recherche sur la Personne des Universités de Tours et de Poitiers (CER-TP 2025-04-07). Participants were anonymized using a unique code combining grade (C6 for Grade 6, CM2 for Grade 5, and CE2 for Grade 3), participant number (e.g., P01), and text condition (E for Expository, N for Narrative). For example, C6P01E identifies the first Grade 6 participant in the expository condition.

Existing Use of Data

Data corresponding to burst characteristics such as duration, size, fluency within bursts and type of burst (Production, Delayed-Revision or Immediate-Revision) have been used in a poster presented in the Writing Words Workshop at Oxford Brookes University in June 2025 (Houssein, Olive & Bordes, 2025) and in a study comparing the data produced by typical children and children with dyslexia (Cislaru et al., 2024).

Data Processing: Extraction of Bursts of Writing and Burst Segmentation

All bursts of writing were extracted using Bursted!, a Python-based application developed within the Pro-TEXT project (Bordes, Olive & Cislaru, 2025). This tool automates the extraction of bursts and their associated variables from raw keylogging files (Inputlog .idfx format) through a multi-stage pipeline. First, the application processes each key log file to filter out non-writing activities. It retains only keystrokes that change the current state of the text by adding or deleting characters (including spaces), while correcting complex events such as French accents or erroneous inserts. This stage produces a cleaned event log used as the basis for further processing. In a second step, a segmentation procedure identifies the textual segments with either a fixed corpus-wide threshold or an individualized threshold. Third, each burst is categorized based on its textual function and is associated with its temporal data. Finally, an output table (in .csv format) is constructed with all bursts’ metrics (see Table 3).

Because the pause threshold used to segment writing into bursts directly affects their characteristics, we employed an individualized threshold (see Bouriga, 2020). Specifically, we first calculated the percentile that corresponded to the standard 2-second literature threshold in the overall pause distribution of all participants in an experiment. Next, this percentile was applied to each individual pause distribution. The pause length that corresponded to this percentile was used as the individual threshold. Then, extracted bursts were categorized by textual function into Production bursts (P-bursts, creating new text at the leading edge) and Revision bursts (R-bursts, modifying existing text). The latter were subdivided into Immediate revision-bursts (IR-bursts, altering the text directly to the left of the cursor) and Delayed revision-bursts (DR-bursts, modifying text outside the most recent production zone).