Nacryos/ancient-scripts-datasets

Ancient Scripts Decipherment Datasets

Collated datasets for the paper:

Deciphering Undersegmented Ancient Scripts Using Phonetic Prior Jiaming Luo, Frederik Hartmann, Enrico Santus, Regina Barzilay, Yuan Cao Transactions of the Association for Computational Linguistics, 2021 arXiv:2010.11054

This repository gathers the training datasets used in the paper — both those hosted in the authors' GitHub repos and the external cited sources.

---

Repository Structure

data/
├── gothic/                          # Gothic language data
│   ├── got.pretrained.pth           # Pretrained phonological embeddings (PyTorch)
│   ├── segments.pkl                 # Phonetic segment data (Python pickle)
│   ├── gotica.txt                   # Gothic Bible plain text (Wulfila project)
│   └── gotica.xml.zip               # Gothic Bible TEI XML (Wulfila project)
│
├── ugaritic/                        # Ugaritic-Hebrew cognate data
│   ├── uga-heb.no_spe.cog          # Full cognate pairs (TSV, ~7,353 tokens)
│   └── uga-heb.small.no_spe.cog    # Small training subset (~10% of full)
│
├── iberian/                         # Iberian inscription data
│   └── iberian.csv                  # Cleaned Hesperia epigraphy (3,466 chunks)
│
├── religious_terms/                 # ** CURATED SUBSET: Religious vocabulary **
│   ├── README.md                    # Methodology and category definitions
│   ├── ugaritic_hebrew_religious.tsv  # ~170 Ug-Heb cognate pairs (deity, ritual, sacred)
│   ├── gothic_religious.tsv         # ~65 Gothic Bible religious terms
│   └── iberian_religious.tsv        # ~40 Iberian votive/religious elements
│
├── linear_b/                        # Linear B (Mycenaean Greek) dataset
│   ├── README.md                    # Sources, methodology, limitations
│   ├── linear_b_signs.tsv           # 211 signs (88 syllabograms + 123 ideograms)
│   ├── sign_to_ipa.json             # 74 syllabogram → IPA mappings
│   └── linear_b_words.tsv           # 2,484 words with IPA, glosses, sources
│
├── validation/                      # Phylogenetic validation dataset (9 branches)
│   ├── README.md                    # Format, sources, concept list
│   ├── concepts.tsv                 # 40 shared concept IDs
│   ├── germanic.tsv                 # got, ang, non, goh (~160 entries)
│   ├── celtic.tsv                   # sga, cym, bre (~120 entries)
│   ├── balto_slavic.tsv             # lit, chu, rus (~120 entries)
│   ├── indo_iranian.tsv             # san, ave, fas (~120 entries)
│   ├── italic.tsv                   # lat, osc, xum (~120 entries)
│   ├── hellenic.tsv                 # grc, gmy (~80 entries)
│   ├── semitic.tsv                  # heb, arb, amh (~120 entries)
│   ├── turkic.tsv                   # otk, tur, aze (~120 entries)
│   └── uralic.tsv                   # fin, hun, est (~120 entries)
│
└── cited_sources/                   # External datasets cited in the paper
    ├── genesis/
    │   ├── Hebrew.xml               # Hebrew Bible (Christodouloupoulos & Steedman 2015)
    │   └── Latin.xml                # Latin Bible (same corpus)
    ├── basque/
    │   ├── Basque-NT.xml            # Basque New Testament (same corpus)
    │   └── Trask_Etymological_Dictionary_Basque.pdf  # Trask's Basque etymological dictionary
    └── iberian_names/
        └── RodriguezRamos2014.pdf   # Iberian onomastic index (personal names)

---

Dataset Details

Gothic (data/gothic/)

File	Source	Description
got.pretrained.pth	DecipherUnsegmented	Pretrained phonological embeddings trained on Gothic IPA data
segments.pkl	DecipherUnsegmented	Serialized phonetic segment inventory
gotica.txt	Wulfila Project	Plain text of the Gothic Bible (4th century CE translation by Bishop Wulfila)
gotica.xml.zip	Wulfila Project	TEI P5 XML encoding with linguistic annotations

The Gothic Bible is the primary source of Gothic text. The paper uses unsegmented Gothic inscriptions from the 3rd-10th century AD period.

Ugaritic (data/ugaritic/)

File	Source	Description
uga-heb.no_spe.cog	NeuroDecipher	Full Ugaritic-Hebrew cognate pairs
uga-heb.small.no_spe.cog	NeuroDecipher	~10% training subset

Format: Tab-separated values. Each row is a cognate pair. Column 1 = Ugaritic transliteration, Column 2 = Hebrew transliteration. | separates multiple cognates; _ marks missing entries. Originally from Snyder et al. (2010), covering 7,353 segmented tokens from the 14th-12th century BC.

Iberian (data/iberian/)

File	Source	Description
iberian.csv	DecipherUnsegmented	Cleaned epigraphic inscriptions

Format: CSV with columns REF. HESPERIA (inscription reference code) and cleaned (transcribed text). Contains 3,466 undersegmented character chunks from the 6th-1st century BC. Sourced from the Hesperia database and cleaned via the authors' Jupyter notebook.

Linear B / Mycenaean Greek (data/linear_b/)

File	Source	Description
linear_b_signs.tsv	Unicode UCD	211 signs: 88 syllabograms + 123 ideograms with Bennett numbers and IPA
sign_to_ipa.json	Ventris & Chadwick (1973)	74 syllabogram transliteration → IPA mappings
linear_b_words.tsv	Multiple (see below)	2,478 words with IPA, glosses, and source attribution

Format: Tab-separated values. The word list contains columns: Word (transliteration), IPA, SCA (sound class), Source, Concept_ID, Cognate_Set_ID, Gloss, Word_Type, IPA_Source. Words come from three CC-BY-SA compatible sources: jhnwnstd/shannon Linear B Lexicon (MIT, 2,272 entries), Wiktionary Mycenaean Greek lemmas (CC-BY-SA, 170 entries with 46 expert IPA), and IE-CoR cognate pairs (42 entries). The sign inventory is from the Unicode Character Database.

Cited Sources (data/cited_sources/)

These are external datasets referenced in the paper for known-language vocabularies and comparison:

File	Citation	Usage in Paper
genesis/Hebrew.xml	Christodouloupoulos & Steedman (2015)	Hebrew vocabulary for Ugaritic comparison
genesis/Latin.xml	Christodouloupoulos & Steedman (2015)	Latin vocabulary for cross-linguistic comparison
basque/Basque-NT.xml	Christodouloupoulos & Steedman (2015)	Basque vocabulary for Iberian comparison
basque/Trask_Etymological_Dictionary_Basque.pdf	Trask (2008)	Basque etymological data
iberian_names/RodriguezRamos2014.pdf	Rodriguez Ramos (2014)	Iberian personal name lists with Latin correspondences

The Bible texts are from the Massively Parallel Bible Corpus (CC0 licensed).

---

Additional Data Sources (Not Included)

The following sources were cited in the paper but are not machine-readable or freely downloadable:

• Wiktionary descendant trees for Proto-Germanic, Old Norse, and Old English vocabularies — extracted by the authors from Wiktionary's structured data
• Original Hesperia epigraphy (hesperia_epigraphy.csv) — referenced in the DecipherUnsegmented README but not present in the repository

---

Cognate Detection Pipeline

The cognate_pipeline/ directory contains a full Python package for cross-linguistic cognate detection, built on the datasets in this repository. It provides:

• Ingestion of CSV/TSV/COG, CLDF, Wiktionary JSONL, and generic JSON sources
• Phonetic normalisation with transcription type tracking (IPA, transliteration, orthographic)
• SCA sound class encoding (List 2012) for phonological comparison
• Family-aware cognate candidate generation (tags cognate_inherited vs similarity_only)
• Weighted Levenshtein scoring with SCA-class-aware substitution costs
• Clustering via connected components or UPGMA
• PostgreSQL/PostGIS database with 8 normalised tables and Alembic migrations
• Export to CLDF Wordlist and JSON-LD formats
• Full provenance tracking through every pipeline stage

Supports 36 languages across 9 phylogenetic branches (Germanic, Celtic, Balto-Slavic, Indo-Iranian, Italic, Hellenic, Semitic, Turkic, Uralic) plus isolates, with Glottocode resolution and IPA transcriptions.

See data/validation/README.md for the phylogenetic validation dataset.

See cognate_pipeline/README.md for installation and usage.

---

Original Repositories

• j-luo93/DecipherUnsegmented — main code for the paper
• j-luo93/NeuroDecipher — predecessor (Ugaritic/Linear B decipherment)
• j-luo93/xib — earlier Iberian codebase

Programmatic Access

Via HuggingFace datasets

from datasets import load_dataset

# Load cognate pairs (Parquet, fast)
ds = load_dataset("Nacryos/ancient-scripts-datasets", "cognate_pairs_inherited")

# Available configs: cognate_pairs_inherited, cognate_pairs_borrowing,
# cognate_pairs_similarity, phylo_pairs, languages

Via Python SDK

For typed access with phylogenetic filtering, IPA parsing, and validation sets:

pip install git+https://github.com/Project-Phaistos/ancient-scripts-datasets-NEW.git

from ancient_scripts_data import AncientScriptsDataset

ds = AncientScriptsDataset()  # auto-downloads from HF
pairs = ds.cognate_pairs("inherited", phylo_filter="close_sister", limit=1000)
lex = ds.lexicon("lat")
rel = ds.phylo_relation("lat", "spa")  # near_ancestral

See the SDK documentation for full API reference.

---

Paper Citation

@article{luo2021deciphering,
  title={Deciphering Undersegmented Ancient Scripts Using Phonetic Prior},
  author={Luo, Jiaming and Hartmann, Frederik and Santus, Enrico and Barzilay, Regina and Cao, Yuan},
  journal={Transactions of the Association for Computational Linguistics},
  volume={9},
  pages={69--81},
  year={2021},
  doi={10.1162/tacl_a_00354}
}