LexA-Index (AI Word Overuse Explorer)

Paper: AI-Associated Lexical Shifts Across 34 Languages: Cross-Lingual Convergence and Diachronic Uptake in News Writing — arXiv:2605.25358 · Live explorer: aiwordexplorer.com

The Explorer lets you explore words that are systematically overused by AI models compared to human baselines, across multiple languages, registers, and AI models.

This repository contains:

• the website (index.html, about.html)
• the CSV outputs (as a 7z, unzip in the top level folder)
• a small script to build website-ready json's from the cvs's (build_data.py).

What this is for

With this repo, you can:

• reproduce the website data build locally,
• inspect the underlying csv's,
• and interactively visualise results for all available language/register/model combinations.

Motivation and background are summarised on the About page.

Citation

If you use this code or data, a citation is appreciated (though not required; see the licence).

@article{juzek-2026-ai-34-languages,
  title   = {AI-Associated Lexical Shifts Across 34 Languages: Cross-Lingual Convergence and Diachronic Uptake in News Writing},
  author  = {Juzek, Thomas Stephan},
  journal = {arXiv preprint arXiv:2605.25358},
  year    = {2026},
  doi     = {10.48550/arXiv.2605.25358},
  url     = {https://arxiv.org/abs/2605.25358}
}

Quick start

• Clone
• Unpack the .7z
• Generate them from the CSVs with:

python3 build_data.py

• Serve locally

python3 -m http.server

Then open:

• http://localhost:8000/

Key metrics include:

• LAS Score: Laid out in our paper
• OPM: occurrences per million tokens (AI and human)
• Ratio: AI OPM / human OPM

Licence

• Code: MIT No Attribution (MIT-0). See LICENSE. Use it freely, no attribution required.
• Data and word lists: CC0 1.0 Universal (public domain dedication). See LICENSE-DATA.

A citation is not required but is appreciated; see the Citation section.

AI Assistance

Repository polished with Claude Code.

Contact

Thomas Stephan Juzek — FSU profile