Mythoscope
Toward a computational framework for comparative mythology.
Mythoscope is a computational framework for comparative mythology. It builds a corpus of
myth and folklore texts, embeds them, and turns the result into an explorable semantic
space — with character / place / time graphs extracted per text by LLMs, a
geographic view, and full-text search by meaning. Alongside this unsupervised layer it
assembles a motif database that integrates the three traditional folklore indexes —
Thompson (TMI), Aarne–Thompson–Uther (ATU) and Berezkin's areal catalogue — into one
cross-linked, browsable whole, with an automatic cross-walk between them and
lexical/semantic parallel-finding on top. Everything is served through one web UI.
Live demo · Documentation · Roadmap
Features
- Semantic space — whole-corpus embeddings projected (UMAP) into an interactive, coloured
scatter you can explore by tradition and region.
- Search by meaning — full-text semantic search and nearest-neighbour "similar fragments"
across traditions.
- Knowledge graphs — per-text character/relation, place, and time graphs (Ages / Realms /
Beings), extracted by LLMs.
- Atlas — a geographic view of the traditions in the corpus.
- Motif crosswalk — a browsable catalogue integrating TMI, ATU, and Berezkin with a
cross-index crosswalk and lexical/semantic parallels.
- One UI + API — a single web app plus a REST/OpenAPI service; export bundles let you
browse prebuilt data with no heavy dependencies.
Quickstart
git clone https://github.com/OWNER/mythoscope
cd mythoscope
python -m venv .venv && source .venv/bin/activate
pip install -e ".[viewer]" # lightweight: browse prebuilt data. Use ".[all]" for the full pipeline.
mytho --help
mytho server # serve the web UI + API at http://localhost:8000
Install profiles (viewer / search / all), environment setup, and the end-to-end pipeline are
documented in docs/how-to.md.
How it works
The pipeline is linear, idempotent, and resumable:
corpus → embeddings → { projections, graphs } → server
motifs (independent of the corpus) → server
- corpus — download & clean source texts (Project Gutenberg + local files), write cleaned
.txt + a catalog.
- embeddings — chunk, optionally LLM-preprocess, encode (e.g.
bge-m3), store in ChromaDB.
- projections — reduce vectors with UMAP into the semantic-space views.
- graphs — LLM-extract characters/relations, places, and a narrative timeline per text.
- motifs — scrape and cross-link the TMI, ATU, and Berezkin indexes into one catalogue.
- server — a FastAPI app reads
outputs/ and serves the SPA + REST/OpenAPI API.
Documentation
- How to — setup, CLI, and the end-to-end pipeline. Start here.
- Deployment — how the live site runs, the fabric tasks, and server performance.
- Research context — surveys of the field (computational folkloristics, motif induction).
- Motif indexes — how TMI, ATU and Berezkin are sourced, parsed and cross-linked.
- Papers — the working paper drafts and bibliography.
- Proposals — design notes, the public-docs plan, and go-to-market.
- Mockups — standalone feature prototypes over the motif data.
- Awesome Computational Mythology — a curated field resource list, maintained as its own standalone repository.
Project status & roadmap
Early research software (v0.1). Experiments, candidate data sources, potential collaborations,
and submission targets are tracked in docs/roadmap.md.
Contributing
Contributions of code, corpus, annotations, and docs are welcome — see
CONTRIBUTING.md and the Code of Conduct.
Start with docs/how-to.md for setup and the pipeline. Security issues:
see SECURITY.md.
Citation
If you use Mythoscope in academic work, please cite it — see CITATION.cff
(a DOI will be added on the first archived release). GitHub renders a "Cite this repository"
button from that file.
License
Source code is under the MIT license. Documentation and prose are under
CC BY-SA 4.0, and derived datasets (e.g. the motif crosswalk) under CC BY 4.0; see
LICENSE for details. Third-party source texts and indexes retain their own terms.
