Supplementary Materials: Generative Language Models and the Transformation of Scientific D

Edlira Gugu's supplementary materials for a COMPTEXT 2026 paper include six appendices supporting a computational text analysis of 15,154 empirical research articles. The corpus spans Linguistics, Social Sciences, and Computer Science, partitioned into pre-LLM (2020-2022) and post-LLM (2023-2024) periods. Appendices contain corpus metadata, preprocessing code, extended statistical results, topic model documentation, a rhetorical template catalogue, and analysis code.

Use Cases

• Replicate a computational text analysis pipeline based on the provided Python preprocessing and analysis code.
• Analyze changes in scientific discourse features like epistemic markers and passive voice across disciplines and time periods.
• Examine the 23 identified rhetorical templates and their coverage rates in academic writing.
• Repurpose the metadata and sampling protocol for building comparable corpora in related fields.

Strengths

• Includes a clearly defined corpus of 15,154 articles with a temporal partition for pre- and post-LLM analysis.
• Provides complete Python code for both preprocessing and analysis, covering techniques like MATTR, LDA, and Sentence-BERT.
• Documents specific analytical outputs, including 30 LDA topics and 23 rhetorical templates.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for individual data tables is unknown, which may limit suitability assessment.
• The dataset is small (2.2 MB), suggesting it contains documentation and code rather than the full primary corpus data.

Provenance

Source

figshare, author Edlira Gugu.

Collection Method

Created as supplementary materials for a conference paper; the corpus was likely assembled from journal articles.

Time Range

Article corpus spans January 2020 to September 2024.

Freshness

Last updated 2026-05-26 21:42:55; freshness should be verified.