Datasets:

Nacryos
/

ancient-scripts-datasets

5 detailed changelog entries covering:
- 001: Initial dataset creation (sources, pipeline, 1,178 languages)
- 002: Database rectification (IPA fix, 18 ancient languages, adversarial audits)
- 003: Cognate pairs v2 rebuild (6 critical bug fixes)
- 004: Phylogenetic enrichment (Glottolog tree, phylo_pairs.tsv)
- 005: Parquet conversion + YAML dataset card

Each entry documents: objectives, scripts, sources, source reputability,
methodology, tests performed, cross-referencing, and academic references.

Files changed (6) hide show

docs/changelog/001_initial_dataset.md +99 -0
docs/changelog/002_database_rectification.md +87 -0
docs/changelog/003_cognate_pairs_v2.md +106 -0
docs/changelog/004_phylo_enrichment.md +208 -0
docs/changelog/005_parquet_conversion.md +111 -0
docs/changelog/INDEX.md +27 -0

docs/changelog/001_initial_dataset.md ADDED Viewed

	@@ -0,0 +1,99 @@

+# 001 — Initial Dataset Creation
+**Date**: 2026-03-08
+**Scope**: Full dataset construction from scratch
+---
+## 1. Objective
+Create a comprehensive multilingual lexicon and cognate pair dataset for training phonetic/linguistic models, covering 1,000+ languages with IPA transcriptions, sound class encodings, and cognate relationships.
+## 2. Scripts Used
+| Script | Purpose |
+|--------|---------|
+| `cognate_pipeline/src/cognate_pipeline/ingest/cldf_ingester.py` | Ingest CLDF-formatted source repos (languages.csv, forms.csv, cognates.csv, borrowings.csv) |
+| `cognate_pipeline/src/cognate_pipeline/ingest/csv_ingester.py` | Ingest CSV/TSV/.cog format files |
+| `cognate_pipeline/src/cognate_pipeline/ingest/wiktionary_ingester.py` | Ingest Wiktionary JSONL pronunciation data |
+| `cognate_pipeline/src/cognate_pipeline/normalise/ipa_normaliser.py` | Normalize IPA transcriptions (NFC, strip brackets, remove stress) |
+| `cognate_pipeline/src/cognate_pipeline/normalise/sound_class.py` | Encode IPA → SCA (Sound Correspondence Alphabet, List 2012) |
+| `cognate_pipeline/src/cognate_pipeline/normalise/epitran_backend.py` | Grapheme-to-phoneme conversion via Epitran |
+| `cognate_pipeline/src/cognate_pipeline/cognate/candidate_gen.py` | Generate pairwise candidates within concept sets |
+| `cognate_pipeline/src/cognate_pipeline/cognate/baseline_levenshtein.py` | SCA-weighted Levenshtein distance scoring |
+| `cognate_pipeline/src/cognate_pipeline/cognate/clustering.py` | Union-Find connected components + UPGMA clustering |
+| `cognate_pipeline/src/cognate_pipeline/export/cldf_exporter.py` | Export to CLDF Wordlist format |
+## 3. Data Sources
+| Source | Repository | License | Description |
+|--------|-----------|---------|-------------|
+| NorthEurAlex | `https://github.com/lexibank/northeuralex` | CC BY 4.0 | Northern Eurasian lexical database, 1,016 concepts × 107 languages |
+| IDS | `https://github.com/lexibank/ids` | CC BY 4.0 | Intercontinental Dictionary Series, 1,310 concepts × 329 languages |
+| ABVD | `https://github.com/lexibank/abvd` | CC BY 4.0 | Austronesian Basic Vocabulary Database, 210 concepts × 1,380+ languages |
+| WOLD | `https://github.com/lexibank/wold` | CC BY 4.0 | World Loanword Database, 41 donor languages, 21K+ borrowing events |
+| Sino-Tibetan | `https://github.com/lexibank/sinotibetan` | CC BY 4.0 | Sino-Tibetan cognate database, 6,159 entries with COGIDs |
+| WikiPron | `https://github.com/CUNY-CL/wikipron` | Apache 2.0 | IPA pronunciations mined from Wiktionary |
+## 4. Source Reputability
+- **CLDF repositories** (lexibank): Published by the Max Planck Institute for Evolutionary Anthropology. Cross-Linguistic Data Formats (CLDF) is the de facto standard for computational historical linguistics (Forkel et al. 2018, *Scientific Data*).
+- **NorthEurAlex**: Curated by Dellert et al. (2020), University of Tübingen. Peer-reviewed in *Language Resources and Evaluation*.
+- **ABVD**: Greenhill, Blust & Gray (2008). Maintained by ANU. 291K expert cognate judgements from field linguists.
+- **WOLD**: Haspelmath & Tadmor (2009), Oxford University Press. 21K+ borrowing events with donor language and certainty metadata.
+- **IDS**: Mary Ritchie Key & Bernard Comrie (2015). Max Planck Institute.
+- **Sino-Tibetan**: Sagart et al. (2019). Published in *PNAS*.
+## 5. Methodology
+### Lexicon Construction
+1. Clone all CLDF source repos to `sources/`
+2. Ingest via `cldf_ingester.py`: read `languages.csv` for ISO mapping, `forms.csv` for word forms, resolve IPA from Form/Value columns
+3. Normalize IPA: NFC Unicode normalization, strip `[/]` brackets, remove suprasegmentals
+4. Encode to SCA: `ipa_to_sound_class()` using List (2012) sound correspondence alphabet
+5. Deduplicate: per-language, per-concept, prefer IPA-bearing entries
+### Cognate Pair Generation (v1)
+1. Group entries by Concept_ID
+2. Generate all pairwise combinations within each concept set
+3. Score via SCA-weighted Levenshtein: substitution cost matrix from List (2012), gap penalty = 0.5
+4. Label relationship type from source metadata (expert_cognate from ABVD cognatesets, borrowing from WOLD)
+### Validation Set Construction
+- 9 language families (Germanic, Celtic, Balto-Slavic, Indo-Iranian, Italic, Hellenic, Semitic, Turkic, Uralic)
+- 40 shared concepts across 6 semantic domains (body, kinship, nature, animals, verbs, other)
+- Stratified by phylogenetic distance: L1 (same sub-branch), L2 (same branch), L3 (same family), L4 (cross-family)
+## 6. Tests Performed
+- **406 automated tests** (232 unit + 174 expanded validation) via `pytest`
+- `test_full_pipeline.py`: End-to-end on Ugaritic-Hebrew .cog format (ingest → score → cluster)
+- `test_validation_sets.py`: Schema validation (14 columns), phylogenetic stratification checks
+- `test_training_data.py`: Training data consistency (no empty IPA, valid ISO codes)
+- `test_expanded_validation.py`: Per-family branch validation (Austronesian, Balto-Slavic, Celtic, etc.)
+- Per-source CLDF validation tests: `sources/abvd/test.py`, `sources/ids/test.py`, `sources/wold/test.py`, `sources/iecor/test.py`
+## 7. Cross-Referencing
+- Each entry retains `Source_Record_ID` traceable to original CLDF record
+- IPA transcriptions spot-checked against WikiPron for 20 random languages
+- SCA encoding validated against known IPA→SCA mappings from List (2012) Table 2
+## 8. Output Summary
+| File | Rows | Size |
+|------|------|------|
+| `data/training/lexicons/` (per-language TSVs) | ~580K entries | ~195 MB total |
+| `data/training/cognate_pairs/cognate_pairs_inherited.tsv` | ~18.2M | ~1.18 GB |
+| `data/training/cognate_pairs/cognate_pairs_borrowing.tsv` | ~116K | ~8.0 MB |
+| `data/training/cognate_pairs/cognate_pairs_similarity.tsv` | ~170K | ~16.2 MB |
+| `data/training/metadata/languages.tsv` | 1,177 | 39 KB |
+| `data/training/validation/*.tsv` | 9 families + expanded | ~200 MB |
+## 9. Academic References
+- Forkel, R., et al. (2018). "Cross-Linguistic Data Formats." *Scientific Data*, 5:180205.
+- List, J.-M. (2012). "SCA: Phonetic alignment based on sound classes." *New Directions in Logic, Language, and Computation*, Springer.
+- Greenhill, S.J., Blust, R. & Gray, R.D. (2008). "The Austronesian Basic Vocabulary Database." *Evolutionary Bioinformatics*, 4:271-283.
+- Haspelmath, M. & Tadmor, U. (2009). *Loanwords in the World's Languages*. De Gruyter Mouton.
+- Sagart, L., et al. (2019). "Dated language phylogenies shed light on the ancestry of Sino-Tibetan." *PNAS*, 116(21):10317-10322.

docs/changelog/002_database_rectification.md ADDED Viewed

	@@ -0,0 +1,87 @@

+# 002 — Database Rectification & Ancient Language Expansion
+**Date**: 2026-03-12
+**Scope**: IPA pipeline bug fixes, 18 ancient language additions, adversarial audits
+---
+## 1. Objective
+Fix a critical IPA pipeline bug that caused 45 languages to have `Word == IPA` (i.e., raw orthography stored as IPA), add 18 ancient/reconstructed languages with proper IPA transcriptions, and conduct adversarial audits on all ancient language entries to ensure no AI-generated or hallucinated data.
+## 2. Scripts Used
+| Script | Purpose |
+|--------|---------|
+| `scripts/fix_ipa_pipeline.py` | Detect and fix Form/Value column swap in CLDF ingestion |
+| `scripts/add_ancient_languages.py` | Ingest ancient language data from specialized CLDF sources |
+| `scripts/audit_ancient_languages.py` | Adversarial audit: random sampling, Word==IPA ratio analysis, source verification |
+## 3. Data Sources
+Same CLDF sources as 001, plus:
+| Source | Repository | License | Description |
+|--------|-----------|---------|-------------|
+| IE-CoR | `https://github.com/lexibank/iecor` | CC BY 4.0 | Indo-European Cognate Relationships database |
+| ACD | `https://github.com/lexibank/acd` | CC BY 4.0 | Austronesian Comparative Dictionary |
+| DiACL | `https://github.com/lexibank/diacl` | CC BY 4.0 | Diachronic Atlas of Comparative Linguistics |
+## 4. Source Reputability
+- **IE-CoR**: Maintained by the Department of Linguistic and Cultural Evolution, MPI-EVA. Expert cognate judgements for Indo-European. Peer-reviewed methodology.
+- **ACD**: Robert Blust's lifetime work on Austronesian reconstruction. Widely cited as the authoritative reference (2,000+ citations).
+- **DiACL**: University of Oslo. Curated diachronic cognate data with expert annotations.
+- All sources are CLDF-compliant, hosted on lexibank (MPI-EVA), CC BY 4.0.
+## 5. Methodology
+### IPA Pipeline Fix
+- **Root cause**: CLDF `forms.csv` has both `Form` (orthographic) and `Value` (IPA) columns. Original ingester used `Form` as IPA for 45 languages.
+- **Detection**: Computed `Word == IPA` ratio per language. Healthy languages: ~5-15% overlap (transliteration scripts). Broken languages: 80-100% overlap.
+- **Fix**: Corrected column mapping. Re-ingested affected languages. Overall dataset Word==IPA ratio improved from 48.9% → 36.7%.
+### Ancient Language Additions (18 languages)
+Proto-Indo-European (ine-pro), Hittite (hit), Ugaritic (uga), Elamite (elx), Avestan (ave), Old Persian (peo), Phoenician (phn), Lycian (xlc), Lydian (xld), Luwian (xle), Phrygian (xpg), Eteocretan (xcr), Hattic (xur), Proto-Semitic (sem-pro), Proto-Dravidian (dra-pro), Proto-Caucasian (ccs-pro), Classical Mongolian (cms), Rhaetic (xrr).
+### Adversarial Audit Protocol
+For each of the 18 ancient languages:
+1. **Source verification**: Confirm entries trace to published CLDF records
+2. **Format check**: Valid ISO 639-3 code, non-empty IPA, proper SCA encoding
+3. **Content sampling**: Random 10-entry sample verified against source repo
+4. **Word==IPA ratio**: Must be within expected range for script type (transliteration: 30-60%, alphabetic: 5-20%)
+5. **Hallucination detection**: No entries may be AI-generated; all must trace to CLDF `Source_Record_ID`
+## 6. Tests Performed
+- **18 individual adversarial audit reports** (one per ancient language), stored in `docs/ADVERSARIAL_AUDIT_*.md`
+- **Master audit report**: `docs/ADVERSARIAL_DATABASE_AUDIT_2026-03-12.md`
+- All 18 languages achieved **PASS** verdict
+- Example: Avestan (ave) — 157 entries, 42.7% Word==IPA (expected for transliteration), IPA conversions verified against Skjærvø (2003)
+- Random sampling: 10 entries per language × 18 languages = 180 manual checks, all matched source
+## 7. Cross-Referencing
+- Each audit report includes a random 10-entry sample with source verification
+- Word==IPA ratios cross-checked against expected values for each script type
+- Ancient language IPA verified against published transcription conventions:
+  - Hittite: Kloekhorst (2008) *Etymological Dictionary of the Hittite Inherited Lexicon*
+  - Avestan: Skjærvø (2003) *An Introduction to Young Avestan*
+  - Ugaritic: Tropper (2000) *Ugaritische Grammatik*
+  - Proto-Indo-European: Fortson (2010) *Indo-European Language and Culture*
+## 8. Output Summary
+| Metric | Before | After |
+|--------|--------|-------|
+| Total languages | 1,159 | 1,177 |
+| Word==IPA ratio | 48.9% | 36.7% |
+| Ancient languages | 0 | 18 |
+| Audit verdicts | N/A | 18/18 PASS |
+| Entries corrected (IPA fix) | — | 23,904 (ABVD G2P rule-based) |
+## 9. Commits
+- IPA pipeline fix + ancient language additions
+- 18 adversarial audit reports
+- Master database audit report

docs/changelog/003_cognate_pairs_v2.md ADDED Viewed

	@@ -0,0 +1,106 @@

+# 003 — Cognate Pairs v2 Rebuild
+**Date**: 2026-03-13
+**Scope**: Complete rebuild of all cognate pair files, fixing 6 critical pipeline bugs
+---
+## 1. Objective
+The original cognate pair extraction pipeline (v1) had 6 critical bugs that produced incorrect or incomplete data. This entry documents the complete rebuild from scratch using corrected v2 extraction scripts.
+## 2. Bugs Fixed
+| # | Bug | Impact | Root Cause |
+|---|-----|--------|------------|
+| 1 | ABVD never read `cognates.csv` | 291K expert cognate judgements missing | Script read `forms.csv:Cognacy` column instead of the dedicated `cognates.csv` CognateTable |
+| 2 | Multi-set cognacy truncation | 37K lost set memberships | `Cognacy` field like `"1,2"` was truncated to `"1"` instead of splitting |
+| 3 | WOLD fabricated borrowing pairs | False borrowing relationships | Script generated pairs from concept co-occurrence instead of reading `borrowings.csv` donor-recipient table |
+| 4 | Concept-aligned pairs mislabeled as inherited | Inflated inherited count | Pairs from concept alignment (no expert cognacy) were labeled `expert_cognate` |
+| 5 | Sino-Tibetan Word field = Concept string | Wrong word forms | `Word_A`/`Word_B` contained concept labels, not actual word forms |
+| 6 | Alphabetical 50-entry hard truncation | Large families truncated | Families with >50 entries in a concept set had entries past the 50th alphabetically discarded |
+## 3. Scripts Used
+| Script | Purpose |
+|--------|---------|
+| `scripts/extract_abvd_cognates_v2.py` | Read authoritative `cognates.csv` (291K entries), handle multi-set membership, include Doubt column |
+| `scripts/extract_wold_borrowings_v2.py` | Read authoritative `borrowings.csv` (21K donor-recipient events), extract Target_Form_ID + Source_Form_ID |
+| `scripts/extract_sinotibetan_cognates_v2.py` | Read `sinotibetan_dump.tsv` (6,159 entries with COGID), separate inherited vs borrowing by BORROWING column |
+| `scripts/extract_iecor_cognates.py` | Process IE-CoR CLDF CognateTable |
+| `scripts/extract_acd_cognates.py` | Process Austronesian Comparative Dictionary |
+| `scripts/merge_cognate_pairs.py` | Deduplicate across sources with priority ordering, produce 3 output files |
+## 4. Data Sources
+Same 6 CLDF sources as 001, plus IE-CoR and ACD from 002.
+## 5. Source Reputability
+- **ABVD `cognates.csv`**: 291,000 expert cognate judgements made by field linguists over 20+ years. Each entry has a `Doubt` column (certain/doubtful). This is the gold standard for Austronesian cognacy.
+- **WOLD `borrowings.csv`**: 21,000+ explicit donor→recipient borrowing events curated by 41 specialist authors. Each has a certainty score (1-5 scale).
+- **IE-CoR**: Expert cognate sets for Indo-European, curated by computational historical linguists at MPI-EVA.
+- All entries have `Source_Record_ID` for traceability back to original CLDF records.
+## 6. Methodology
+### 14-Column Output Schema
+```
+Lang_A | Word_A | IPA_A | Lang_B | Word_B | IPA_B | Concept_ID |
+Relationship | Score | Source | Relation_Detail | Donor_Language |
+Confidence | Source_Record_ID
+```
+### Scoring
+- **SCA-weighted Levenshtein**: Normalized edit distance on Sound Class Alphabet (List 2012) encodings
+- Substitution cost: 0 for same class, 0.5 for similar classes (e.g., voiced↔voiceless stop), 1.0 for different classes
+- Gap penalty: 0.5
+- Score = 1.0 − (normalized_distance), range [0.0, 1.0], rounded to 4 decimal places
+### Deduplication Priority
+When the same (Lang_A, Lang_B, Concept_ID) tuple appears in multiple sources:
+1. `expert_cognate` (priority 0) — keeps expert-labelled pair
+2. `borrowing` (priority 1)
+3. `concept_aligned` (priority 2)
+4. `similarity_only` (priority 3) — lowest priority, only kept if no better evidence
+Pair key is order-independent: `min(side_a, side_b) || max(side_a, side_b) || concept`
+### ABVD v2 Extraction (Bug #1 + #2 fix)
+- Reads `cognates.csv` (CLDF CognateTable), not `forms.csv:Cognacy`
+- Splits multi-set membership: `"1,2"` → entries in both cognateset 1 and 2
+- Preserves `Doubt` column as `Confidence` field
+### WOLD v2 Extraction (Bug #3 fix)
+- Reads `borrowings.csv` directly: each row = one donor→recipient event
+- Extracts `Target_Form_ID` and `Source_Form_ID`, resolves to language/word/IPA
+- No fabricated pairs — only explicit borrowing relationships
+## 7. Tests Performed
+- All 406 existing tests pass after rebuild
+- `test_full_pipeline.py`: End-to-end Ugaritic-Hebrew cognate detection verified
+- `test_training_data.py`: Schema validation on all 3 output files
+- Row count verification against source repos:
+  - ABVD: `wc -l cognates.csv` matches expected extraction count
+  - WOLD: `wc -l borrowings.csv` matches expected extraction count
+- Deduplication audit: Verified no duplicate (Lang_A, Lang_B, Concept_ID) tuples in output
+## 8. Cross-Referencing
+- 20 random inherited pairs traced back to ABVD `cognates.csv` by `Source_Record_ID`
+- 10 random borrowing pairs traced back to WOLD `borrowings.csv` by `Source_Record_ID`
+- Verified WOLD borrowing pairs have correct `Donor_Language` by cross-referencing with source
+- Verified Sino-Tibetan Word fields now contain actual word forms (not concept labels)
+## 9. Output Summary
+| File | Rows | Size | Change from v1 |
+|------|------|------|-----------------|
+| `cognate_pairs_inherited.tsv` | 22.9M | 2.2 GB | +291K expert cognates recovered |
+| `cognate_pairs_borrowing.tsv` | 17K | 1.9 MB | Completely rebuilt (v1 was fabricated) |
+| `cognate_pairs_similarity.tsv` | 465K | 49.9 MB | Relabeled (was falsely `inherited` in v1) |
+## 10. PRD Reference
+Full specification: `docs/prd/PRD_COGNATE_PAIRS_V2.md`

docs/changelog/004_phylo_enrichment.md ADDED Viewed

	@@ -0,0 +1,208 @@

+# 004 — Phylogenetic Metadata Enrichment
+**Date**: 2026-03-14
+**Scope**: Added `phylo_pairs.tsv` — phylogenetic relationship metadata for all language pairs, derived from Glottolog CLDF
+---
+## 1. Objective
+Add phylogenetic metadata to the dataset so that downstream models can filter cognate pairs by evolutionary relationship (e.g., "train only on close sister pairs" or "exclude cross-family noise"). The existing cognate pairs have no phylogenetic information — they are flat (Lang_A, Lang_B) pairs with no tree context.
+## 2. Scripts Used
+| Script | Lines | Purpose |
+|--------|-------|---------|
+| `scripts/ingest_glottolog.py` | ~50 | Download Glottolog CLDF repo from GitHub |
+| `scripts/build_glottolog_tree.py` | 333 | Parse Glottolog NEXUS classification → build ancestry paths → output `glottolog_tree.json` |
+| `scripts/build_phylo_pairs.py` | 310 | Cross-reference cognate pair language pairs against tree → classify phylogenetic relationship → output `phylo_pairs.tsv` |
+| `scripts/validate_phylo_pairs.py` | 288 | Automated known-answer tests (14 cases), near-ancestral integrity, coverage audit, random pair audit |
+## 3. Data Sources
+| Source | Repository | License | Description |
+|--------|-----------|---------|-------------|
+| Glottolog CLDF v5.x | `https://github.com/glottolog/glottolog-cldf` | CC BY 4.0 | Comprehensive language classification: 27,177 languoids, 8,184 with ISO 639-3 codes |
+### Files Used from Glottolog
+- `cldf/languages.csv` — 27,177 languoids with ISO 639-3 mapping, family assignment, Glottocodes
+- `cldf/classification.nex` — NEXUS format containing Newick trees for each language family
+## 4. Source Reputability
+**Glottolog** is the authoritative reference for language classification in computational linguistics:
+- **Maintainers**: Harald Hammarström, Robert Forkel, Martin Haspelmath, Sebastian Bank (Max Planck Institute for Evolutionary Anthropology)
+- **Citation**: Hammarström, H., Forkel, R., Haspelmath, M. & Bank, S. (2026). *Glottolog 5.x*. Leipzig: Max Planck Institute. DOI: 10.5281/zenodo.15640174
+- **Usage**: Standard reference in all major computational historical linguistics publications
+- **Scope**: Every known human language and dialect (8,000+ languages)
+- **Peer review**: Continuously curated by professional linguists; corrections submitted via GitHub issues
+## 5. Methodology
+### Step 1: Tree Construction (`build_glottolog_tree.py`)
+1. **Parse NEXUS**: Read `classification.nex`, extract Newick strings per family tree
+2. **Parse Newick**: Character-by-character parser handles nested parentheses, node labels, branch lengths (stripped — topological only)
+3. **Build ancestry paths**: BFS from root to each leaf, recording full path (e.g., `["Indo-European", "Germanic", "West Germanic", "Anglic", "English"]`)
+4. **Map ISO codes**: Cross-reference `languages.csv` to map ISO 639-3 → (Glottocode, family_id, ancestry_path)
+5. **Output**: `glottolog_tree.json` — JSON index mapping every ISO code to its tree position
+### Step 2: Phylogenetic Classification (`build_phylo_pairs.py`)
+For each unique (Lang_A, Lang_B) pair in the cognate dataset:
+1. **Look up ancestry paths** for both languages in the tree index
+2. **Compute MRCA** (Most Recent Common Ancestor): Longest common prefix of the two ancestry paths
+3. **Compute tree distance**: `edges_A_to_MRCA + edges_B_to_MRCA`
+4. **Classify relationship** using this decision tree:
+```
+IF one or both languages not in tree:
+    → "unclassified"
+ELIF different top-level families:
+    → "cross_family"
+ELIF one language is in NEAR_ANCESTOR_MAP
+  AND the other's ancestry passes through the listed descendant clade
+  AND the other is NOT itself an ancient/medieval language:
+    → "near_ancestral" (Ancestor_Lang = the ancient language)
+ELIF MRCA_Depth >= 3:
+    → "close_sister" (share a specific sub-branch)
+ELSE:
+    → "distant_sister" (share family or major branch only, depth 1-2)
+```
+**Threshold**: `CLOSE_SISTER_DEPTH_THRESHOLD = 3` — chosen because depth 3 guarantees shared sub-branch classification (e.g., "West Germanic" under "Germanic" under "Indo-European"), while depth 1-2 captures only broad family membership.
+### Step 3: Near-Ancestral Curated Map
+**Why a curated map is necessary**: Glottolog treats ALL attested languages as leaf nodes. Latin is classified as a *sibling* of Romance languages, not a *parent*. Without correction, Latin↔French would be classified as `close_sister` instead of `near_ancestral`.
+**Full curated map** (19 entries):
+| Ancestor | ISO | Descendant Clade (Glottocode) | Example Descendant |
+|----------|-----|-------------------------------|-------------------|
+| Latin | lat | roma1334 (Romance) | French, Spanish, Italian |
+| Ancient Greek | grc | koin1234 (Koineic Greek) | Modern Greek |
+| Sanskrit | san | indo1321 (Indo-Aryan) | Hindi, Bengali |
+| Old English | ang | angl1265 (Anglic) | Modern English |
+| Middle English | enm | angl1265 (Anglic) | Modern English |
+| Old French | fro | oila1234 (Oïl French) | Modern French |
+| Old Spanish | osp | cast1243 (Castilian) | Modern Spanish |
+| Old Norse | non | nort3160 (North Germanic) | Swedish, Danish, Norwegian |
+| Old High German | goh | high1289 (High German) | Modern German |
+| Middle Dutch | dum | mode1257 (Modern Dutch) | Dutch, Afrikaans |
+| Old Irish | sga | goid1240 (Goidelic) | Irish, Scottish Gaelic |
+| Middle Irish | mga | goid1240 (Goidelic) | Irish, Scottish Gaelic |
+| Old Church Slavonic | chu | sout3147 (South Slavic) | Bulgarian, Serbian |
+| Old Russian | orv | east1426 (East Slavic) | Russian, Ukrainian |
+| Old Chinese | och | clas1255 (Classical Chinese) | Mandarin, Cantonese |
+| Ottoman Turkish | ota | oghu1243 (Oghuz) | Modern Turkish |
+| Classical Arabic | arb-cla | arab1395 (Arabic) | Modern Arabic varieties |
+| Middle Persian | pal | west2794 (Western Iranian) | Modern Persian |
+| Old Japanese | ojp | japo1237 (Japonic) | Modern Japanese |
+**Justification**: This map is small (19 entries), exhaustively verifiable by any linguist, and covers all well-attested ancestor-descendant relationships in the dataset. Each mapping follows the consensus in historical linguistics.
+### Output Schema (9 columns)
+```
+Lang_A | Lang_B | Phylo_Relation | Tree_Distance | MRCA_Clade |
+MRCA_Depth | Ancestor_Lang | Family_A | Family_B
+```
+| Column | Type | Description |
+|--------|------|-------------|
+| `Lang_A` | ISO 639-3 | First language (alphabetically ordered) |
+| `Lang_B` | ISO 639-3 | Second language |
+| `Phylo_Relation` | enum | `near_ancestral`, `close_sister`, `distant_sister`, `cross_family`, `unclassified` |
+| `Tree_Distance` | int | Number of edges between the two languages via MRCA |
+| `MRCA_Clade` | Glottocode | Glottocode of the most recent common ancestor node |
+| `MRCA_Depth` | int | Depth of MRCA from tree root (root = 0) |
+| `Ancestor_Lang` | ISO or `-` | For `near_ancestral`: the ancestor language's ISO code. `-` otherwise |
+| `Family_A` | string | Top-level family name for Lang_A |
+| `Family_B` | string | Top-level family name for Lang_B |
+### Design Decision: Separate Lookup Table
+The phylo metadata is stored as a separate `phylo_pairs.tsv` rather than as 7 additional columns on the 22.9M-row `cognate_pairs_inherited.tsv`. Rationale:
+- 22.9M pairs reference only ~385K unique (Lang_A, Lang_B) combinations
+- Adding 7 columns to 22.9M rows = 160M+ redundant cells
+- Downstream join at query time: O(1) dict lookup by `(Lang_A, Lang_B)` key
+- Keeps the cognate pairs file focused on linguistic data
+## 6. Tests Performed
+### Automated Known-Answer Tests (14 cases in `validate_phylo_pairs.py`)
+| Lang_A | Lang_B | Expected | Rationale |
+|--------|--------|----------|-----------|
+| deu | eng | close_sister | Both West Germanic |
+| eng | fra | distant_sister | Germanic vs Italic under IE |
+| fra | lat | near_ancestral | Latin → Romance → French |
+| lat | spa | near_ancestral | Latin → Romance → Spanish |
+| lat | osc | distant_sister | Latino-Faliscan vs Sabellic |
+| ang | eng | near_ancestral | Old English → Modern English |
+| got | swe | distant_sister | East vs North Germanic |
+| hin | san | near_ancestral | Sanskrit → Indo-Aryan → Hindi |
+| eng | jpn | cross_family | IE vs Japonic |
+| ceb | tgl | close_sister | Central Philippine |
+| dan | swe | close_sister | North Germanic |
+| ita | spa | close_sister | Italo-Western Romance |
+| rus | ukr | close_sister | East Slavic |
+| gle | sga | near_ancestral | Old Irish → Goidelic → Irish |
+**Result**: All 14 tests PASS.
+### Coverage Audit
+- ≥95% of ISO codes in the cognate dataset found in Glottolog tree index
+- Unclassified pairs logged for manual review
+### Near-Ancestral Integrity
+- Every `near_ancestral` pair verified: ancestor is in curated map AND descendant's ancestry path passes through the specified clade
+### Random Pair Audit
+- 20 random pairs selected, each verified:
+  - Family_A and Family_B match Glottolog's top-level classification
+  - MRCA_Depth and Tree_Distance consistent with ancestry paths
+  - Phylo_Relation follows the classification decision tree correctly
+## 7. Cross-Referencing
+- All tree positions cross-referenced against Glottolog online (https://glottolog.org/glottolog/language/[ISO])
+- Known-answer pairs verified against published family trees in:
+  - Campbell, L. (2013). *Historical Linguistics: An Introduction*. MIT Press.
+  - Ringe, D. (2006). *From Proto-Indo-European to Proto-Germanic*. Oxford University Press.
+- Near-ancestral relationships verified against standard reference grammars for each ancestor language
+## 8. Output Summary
+| File | Rows | Size |
+|------|------|------|
+| `data/training/metadata/phylo_pairs.tsv` | ~386K | 24.7 MB |
+| `data/training/raw/glottolog_cldf/glottolog_tree.json` | ~8K entries | ~2 MB |
+### Relationship Distribution
+| Relation | Count | % |
+|----------|-------|---|
+| close_sister | ~45% | Most pairs share a sub-branch |
+| distant_sister | ~25% | Same family, different major branches |
+| cross_family | ~20% | Different language families |
+| near_ancestral | ~5% | Ancestor-descendant |
+| unclassified | ~5% | One/both languages not in Glottolog |
+## 9. Limitations (Phase 1)
+- **Topological only**: No branch lengths (years of divergence). The Glottolog tree encodes relationships but not temporal distances.
+- **Phase 2 planned**: Import Bayesian phylogenies from Phlorest (Chang et al. 2015, Bouckaert et al. 2012) to add `Divergence_Years` column.
+- **Near-ancestral map is manually curated**: Only covers 19 well-attested ancestor-descendant pairs. Other potential ancestor-descendant relationships (e.g., within Austronesian) are classified as `close_sister`.
+## 10. Academic References
+- Hammarström, H., Forkel, R., Haspelmath, M. & Bank, S. (2026). *Glottolog 5.x*. DOI: 10.5281/zenodo.15640174.
+- Campbell, L. (2013). *Historical Linguistics: An Introduction*. 3rd ed. MIT Press.
+- Ringe, D. (2006). *From Proto-Indo-European to Proto-Germanic*. Oxford University Press.
+## 11. PRD Reference
+Full specification: `docs/prd/PRD_PHYLO_ENRICHMENT.md`

docs/changelog/005_parquet_conversion.md ADDED Viewed

	@@ -0,0 +1,111 @@

+# 005 — Parquet Conversion & HuggingFace Dataset Card
+**Date**: 2026-03-15
+**Scope**: Convert 5 key TSV files to Parquet (ZSTD), add YAML dataset card for `datasets` library integration
+---
+## 1. Objective
+Enable programmatic access via `datasets.load_dataset("Nacryos/ancient-scripts-datasets", "config_name")` by:
+1. Converting the 5 largest TSV files to Parquet format (columnar, compressed, typed)
+2. Adding a YAML frontmatter dataset card to `README.md` with named configs pointing to the Parquet files
+3. Preserving all original TSV files as legacy backups (no data removed)
+### Why Parquet?
+- The HuggingFace `datasets` library v4.0 removed support for custom loading scripts (they cause `RuntimeError`). YAML-based configs pointing to Parquet files are the recommended replacement.
+- Parquet enables: columnar reads (load only needed columns), predicate pushdown (filter without full scan), proper null handling, typed schemas.
+- TSV sentinel values (`-` for missing data) are converted to proper nulls.
+## 2. Scripts Used
+| Script | Lines | Purpose |
+|--------|-------|---------|
+| `scripts/convert_to_parquet.py` | ~100 | Read TSV via pyarrow, force all columns to string on initial read, convert `-` sentinels to null, cast typed columns (int/float), write Parquet with ZSTD compression |
+### Key Implementation Details
+- **Type safety**: All columns initially read as `pa.string()` to avoid pyarrow type inference issues (e.g., Score column auto-detected as `double` conflicting with `-` sentinel)
+- **Null handling**: All `-` values converted to proper Parquet nulls via `pc.if_else(pc.equal(col, "-"), None, col)`
+- **Compression**: ZSTD (Zstandard) — best compression ratio for this data profile
+- **Column casting**: Int columns (Tree_Distance, MRCA_Depth) cast to `pa.int32()`, float columns (Score) cast to `pa.float64()`
+## 3. Data Sources
+No new external data. This change only converts existing TSV files already in the dataset.
+### Input Files (preserved as legacy)
+| File | Rows | TSV Size |
+|------|------|----------|
+| `cognate_pairs_inherited.tsv` | 22.9M | 2.2 GB |
+| `cognate_pairs_similarity.tsv` | 465K | 49.9 MB |
+| `cognate_pairs_borrowing.tsv` | 17K | 1.9 MB |
+| `phylo_pairs.tsv` | 386K | 24.7 MB |
+| `languages.tsv` | 1,177 | 39.3 KB |
+## 4. Source Reputability
+N/A — format conversion only, no new external data. All data integrity preserved from source TSVs.
+## 5. Methodology
+### Conversion Process
+1. Read TSV with `pyarrow.csv.read_csv()`, forcing all columns to `pa.string()` schema
+2. For each column, replace `-` sentinel with `null`
+3. Cast numeric columns to proper types (`int32`, `float64`)
+4. Write Parquet with `pq.write_table(table, path, compression='zstd')`
+### YAML Dataset Card
+Added to `README.md` frontmatter:
+```yaml
+configs:
+  - config_name: cognate_pairs_inherited
+    data_files:
+      - split: train
+        path: data/training/cognate_pairs/cognate_pairs_inherited.parquet
+  - config_name: cognate_pairs_borrowing
+    data_files:
+      - split: train
+        path: data/training/cognate_pairs/cognate_pairs_borrowing.parquet
+  # ... (5 total configs)
+```
+### Why NOT a Loading Script
+HuggingFace `datasets` v4.0 (released late 2025) removed custom loading script support entirely. Any repo with a `*.py` loading script triggers `RuntimeError`. YAML-based configuration is the official replacement.
+## 6. Tests Performed
+- **Row count verification**: Each Parquet file verified to have identical row count to source TSV
+- **Schema verification**: Column names and types match expected schema
+- **Random sample verification**: 10 random rows from each Parquet file compared byte-for-byte against corresponding TSV rows
+- **Null handling**: Verified `-` sentinels are proper nulls (not empty strings) in Parquet
+- **LFS integrity**: Verified no LFS pointer files were accidentally converted (detected and re-pulled `cognate_pairs_borrowing.tsv` which was a 3-line LFS pointer)
+- **HuggingFace API verification**: Confirmed `https://huggingface.co/api/datasets/Nacryos/ancient-scripts-datasets` returns all 5 configs
+## 7. Cross-Referencing
+- Parquet file sizes verified to be strictly smaller than TSV (compression working correctly)
+- Column count matches: 14 columns for cognate pairs, 9 for phylo_pairs, 5 for languages
+- HuggingFace Dataset Viewer activated and showing data correctly
+## 8. Output Summary
+| File | TSV Size | Parquet Size | Compression Ratio |
+|------|----------|-------------|-------------------|
+| `cognate_pairs_inherited.parquet` | 2.2 GB | 23.6 MB | 93.8x |
+| `cognate_pairs_similarity.parquet` | 49.9 MB | 5.0 MB | 10.0x |
+| `cognate_pairs_borrowing.parquet` | 1.9 MB | 487 KB | 4.0x |
+| `phylo_pairs.parquet` | 24.7 MB | 416 KB | 59.3x |
+| `languages.parquet` | 39.3 KB | 13.7 KB | 2.9x |
+| **Total** | **2.28 GB** | **29.5 MB** | **77.2x** |
+## 9. Legacy Preservation
+All original TSV files are preserved in their original locations. The Parquet files are stored alongside them:
+- `data/training/cognate_pairs/cognate_pairs_inherited.tsv` (original, kept)
+- `data/training/cognate_pairs/cognate_pairs_inherited.parquet` (new, added)
+No data was removed or modified. The Parquet files are a read-optimized mirror.
+## 10. SDK Package
+A companion Python SDK (`ancient-scripts-data`) was also created in a separate repo (`Project-Phaistos/ancient-scripts-datasets-NEW`) to provide typed APIs for accessing this data. See that repo's README for API documentation.

docs/changelog/INDEX.md ADDED Viewed

	@@ -0,0 +1,27 @@

+# Dataset Changelog Index
+All changes to the `Nacryos/ancient-scripts-datasets` HuggingFace dataset are logged here in reverse chronological order. Each entry is a self-contained document explaining **what** changed, **why**, **how** (scripts, sources, methodology), and **how it was validated**.
+## Changelog Entries
+| Date | Entry | Summary |
+|------|-------|---------|
+| 2026-03-15 | [005_parquet_conversion.md](005_parquet_conversion.md) | Added Parquet files + YAML dataset card for HF `datasets` library integration |
+| 2026-03-14 | [004_phylo_enrichment.md](004_phylo_enrichment.md) | Added phylogenetic metadata (`phylo_pairs.tsv`) derived from Glottolog CLDF |
+| 2026-03-13 | [003_cognate_pairs_v2.md](003_cognate_pairs_v2.md) | Rebuilt all cognate pairs from scratch (v2), fixing 6 critical pipeline bugs |
+| 2026-03-12 | [002_database_rectification.md](002_database_rectification.md) | IPA pipeline fixes, ancient language expansion, adversarial audits |
+| 2026-03-08 | [001_initial_dataset.md](001_initial_dataset.md) | Initial dataset creation: 1,178 languages, lexicons, cognate pairs v1, validation sets |
+## Log Standard
+Every changelog entry MUST include:
+1. **Objective** — Why was the data changed?
+2. **Scripts** — What scripts were written/run? What does each do?
+3. **Sources** — What external data sources were used?
+4. **Source Reputability** — Academic credentials, peer-review status, citation counts, cross-referencing
+5. **Methodology** — Algorithms, formulas, thresholds, and academic references
+6. **Tests Performed** — Automated tests, known-answer checks, random sampling
+7. **Cross-Referencing** — How were random samples verified against original sources?
+8. **Output Summary** — Row counts, file sizes, schema changes
+9. **Commits** — Git commit hashes for traceability