How this atlas is built, and how to cite it

Upstream datasets

Each source is used for a specific plane of the atlas. Versions and licences are exactly as released; where a versioned Zenodo DOI is best obtained from the dataset's own landing page we link the canonical page rather than assert an identifier we have not verified.

CC-BY vs CC-BY-NC.The Bouckaert et al. 2018 phylogeny is used here via the Phlorest CLDF release (CC-BY-4.0). The same tree is also distributed through D-PLACE, parts of which carry CC-BY-NC. The atlas deliberately sources the CC-BY-4.0 Phlorest release so the deep-time layer is free of a non-commercial clause. Share-alike sources — Wiktionary (CC-BY-SA-4.0) and PHOIBLE (CC-BY-SA-3.0) — keep their share-alike terms if you redistribute those fields.

Download the open data

A deterministic, reproducible snapshot of release 1.1.0. These are open data under the upstream licences — attribute accordingly. The atlas's own derived layer (the joins, tiers, provenance flags, coverage tables) is released CC-BY-4.0.

• languages.csv

  One row per genuine languoid: identity, family, coordinates + provenance, tier, endangerment, grammar/lexicon coverage, deep-time divergence and full classification chain.

  CSV · 166 KB · 980 rows

• languages.geojson

  GeoJSON FeatureCollection of the 809 located languages (Point geometry); the 171 unlocated languoids are honestly omitted, never given a fake point.

  GEOJSON · 467 KB · 809 rows

• grammar-values.csv

  Long-format grammatical feature codings (one row per language × feature). value/present/absent/unknown(?) states kept distinct — read the state column; unknown = not recorded.

  CSV · 4.3 MB · 30,351 rows

• theses.json

  The 8 "why did the languages move?" scholarly theses, contradiction-preserving — each with mechanism, proponents, evidence for AND against, hard facts, contestation level and full citations (with DOIs).

  JSON · 126 KB · 8 rows

• sources.bib

  BibTeX for the upstream datasets and the 48 scholarly references behind the movement theses (DOIs as published). Includes an entry for citing the atlas itself.

  BIB · 27 KB · 56 rows

• CITATION.cff

  Citation File Format metadata for the atlas itself (title, authors, version, licence, release date, upstream references) — machine-readable "cite this".

  CFF · 3 KB

• README.md

  Data dictionary: every file + column, the per-source licences, how to cite, and the honesty/uncertainty notes.

  MD · 8 KB

Prefer a single documented bundle? The README / data dictionary describes every file, column and licence. A CLDF package is not yet emitted; the CSV + GeoJSON + BibTeX + CITATION.cff set above is the interchange format for this release.

How to cite this atlas

Cite the atlas and the upstream datasets you rely on. The machine-readable CITATION.cff and the full sources.bib (upstream datasets + the scholarly references behind the movement theses) are in the downloads.

Mob Translate project & Claude (2026). The Atlas of Australian Languages, data release 1.1.0. https://mobtranslate.com/atlas (accessed <date>).

@misc{atlas_australian_languages,
  author       = {{Mob Translate project and Claude (Anthropic)}},
  title        = {The Atlas of Australian Languages},
  year         = {2026},
  version      = {1.1.0},
  howpublished = {\url{https://mobtranslate.com/atlas}},
  note         = {Data release 1.1.0, build 2026-07-12. Derived layer CC-BY-4.0. Cite the upstream datasets too (see sources.bib).}
}

Method & reproducibility

Coverage, shown as fractions

Out of 980 genuine languoids (1,029 registry nodes minus 49 appendixed non-language nodes). Fractions are never rounded up to imply completeness.

located (with a real or approximate coordinate)

809 / 980

with a full classification chain

647 / 980

grammatically profiled (Grambank/WALS/AUS)

203 / 980

with a dated deep-time position

224 / 980

with a coded endangerment level

442 / 980

with a Glottolog code

656 / 980

Uncertainty & honesty policy

• Coordinates carry provenance.

  Every point is flagged glottolog / austlang / derived_centroid / none. Approximate and derived points are labelled as such; unlocated languages are listed, never plotted at a fake point.

• Autonyms are unverified candidates.

  Names drawn from alt-name lists (which conflate endonyms with spelling variants) are shown as candidates, never asserted as a community-confirmed autonym.

• Grammar coverage is partial and honest.

  203 of 980 languoids are grammatically profiled; the rest are shown as not profiled, never as absence of a feature. The grammar lens reports grammar_recorded_agreement — agreement over jointly-recorded features only, not overall grammatical similarity and not genetic relatedness (always read n_joint).

• Deep-time dates a language lineage, not a population.

  The Bouckaert 2018 ages (224 of 980 leaves dated) reconstruct the movement of language lineages across a continent already populated for ~65,000 years — not a peopling event. The deepest Pama-Nyungan nodes have weak posterior support and are shown with their 95% HPD, not a bare number. See the deep-time spread.

• The movement theses preserve contradiction.

  The eight “why did the languages move?” theses are held side by side — each with mandatory evidence against and a contestation level. No single winner is declared. Read the thesis matrix.