Upload folder using huggingface_hub

README.md

@@ -0,0 +1,57 @@
+---
+license: cc-by-4.0
+pretty_name: "Hebrew Word Senses — polysemy inventory, Hebrew-context derived"
+language:
+  - hbo
+tags:
+  - biblical-hebrew
+  - lexicography
+  - word-sense-disambiguation
+configs:
+  - config_name: default
+    data_files: "hbo_lex.tsv"
+---
+
+# Hebrew Word Senses
+
+Which Biblical Hebrew lexemes are polysemous (have more than one distinct meaning), and how their
+occurrences cluster into each sense — 12,536 sense-rows across ~8,015 (lexeme, verbal-stem) groups.
+
+## Method
+
+**Sense-cluster structure** (which occurrences belong to which sense) is decided entirely from Hebrew:
+occurrences of a (lexeme, stem) are grouped by their English gloss, then merged whenever their Hebrew-
+clause embedding centroids are close (single-linkage clustering) — the question "are these the same
+sense?" is answered by how the word is actually used in Hebrew context, not by an English label. No
+UBS/SDBH data is involved in this step at any point.
+
+**Gloss labels** (the human-readable name for each sense) come from two sources:
+- The **dominant** sense of each group (its most common meaning) uses
+  [`bcv-commons/strongs`](https://huggingface.co/datasets/bcv-commons/strongs)'s own vetted gloss.
+- Every other (**sub-**)sense uses Cherith Glosses (Andi Wu, Copyright 2022, CC BY 4.0) via MACULA's
+  own `@gloss` attribute — confirmed via [MACULA-Hebrew's own
+  LICENSE](https://github.com/Clear-Bible/macula-hebrew/blob/main/LICENSE.md) to be a distinct,
+  separately-licensed source, **not** SDBH/UBS MARBLE.
+
+## Columns
+
+| column | meaning |
+|---|---|
+| `lex` | ETCBC/BHSA lexeme id |
+| `stem` | verbal stem (binyan) — blank for non-verbs |
+| `sense` | sense number within this (lex, stem) group; `1` is always the dominant sense |
+| `gloss` | the English label for this sense (see Method above for provenance) |
+| `count` | how many occurrences fall in this sense-cluster |
+| `share` | `count` / the group's total occurrences — `sense=1`'s share tells you how polysemous the word actually is in practice (a word with `sense=1` at 99% share is barely polysemous despite having other senses tagged) |
+
+## License
+
+CC BY 4.0. Sense-cluster structure: independently derived, no restricted source. Dominant-sense
+glosses: from `bcv-commons/strongs`. Sub-sense glosses: Cherith Glosses (Andi Wu, CC BY 4.0), via
+MACULA (Clear-Bible/macula-hebrew, CC BY 4.0) — not derived from or reproducing UBS MARBLE / SDBH.
+
+## Provenance
+
+Built by `bcv-RAG/scripts/cluster_senses_hebrew.py` (clustering) and
+`shoresh/macula/build_hbo_lex_export.py` (publish-ready gloss substitution) in
+[bcv-commons/bcv-query](https://github.com/bcv-commons/bcv-query).