dictionaries: move to separate repo Asakiri/langdex-dictionaries (clean license separation)

README.md

@@ -73,15 +73,9 @@ The following tables are obtained from the Etymology Atlas (Luke Steuber; CC BY-
 
 Translation coverage extends to approximately 5,700 languages through PanLex; definitions and IPA are available for approximately 5,000 languages through Wiktionary.
 
-## Dictionaries layer (`data/dictionaries/`)
+## Dictionaries
 
-A modular, per-source layer that digitizes public-domain scanned dictionaries (from the Internet Archive) into LangDex, extending coverage for under-resourced languages no commercial resource reaches.
-
-- `data/dictionaries/catalog.parquet` — an index of scanned dictionaries on archive.org, resolved to Glottolog codes and joined against LangDex coverage (`source_id`, `title`, `year`, `glottocode`, `language`, `family`, `downloads`, `public_domain`, `langdex_lexemes`). Ranks languages that have a public-domain dictionary but thin LangDex coverage.
-- `data/dictionaries/entries/<source_id>.parquet` — the **digitized dictionary** (source of truth): `glottocode`, `gloss_lang`, `glosses[]`, `forms[]`, `pos`, `ipa`, `confidence`, `extract_model`, `source_id`. Independently useful as a dictionary; the LangDex `lexemes`/`senses`/`lexeme_concept` rows are a regenerable projection of it via deterministic gloss→concept linking.
-- `data/dictionaries/ocr/<source_id>.parquet` — raw layout-preserving OCR per page (optional), so entries can be re-extracted without re-OCR.
-
-Each entry's target-language forms become `lexemes`; the metalanguage gloss maps to the canonical `concept` hub, weaving the language into the pivot-free translation graph.
+Digitized public-domain and openly-licensed scanned dictionaries for under-resourced languages live in a **separate companion dataset**, [Asakiri/langdex-dictionaries](https://huggingface.co/datasets/Asakiri/langdex-dictionaries) — kept apart so their mixed per-dictionary licensing never complicates this dataset's clean CC BY-SA core. The `open`-scope dictionaries there link into LangDex's shared concept hub.
 
 ## Access
 

data/dictionaries/catalog.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:7aaa39f1a436029678e5da705fb85371dde8d538c43a8e9f1bd056c6b3424531
-size 85624