README.md

---
license: cc-by-4.0
language:
  - hbo
  - en
task_categories:
  - sentence-similarity
  - text-retrieval
tags:
  - nuberea
  - nuberea-tier-0
  - biblical-studies
  - old-testament
  - hebrew-bible
  - tanakh
  - intertextuality
  - parallel-passages
  - text-reuse
  - etcbc
  - bhsa
  - morphology
size_categories:
  - 10K<n<100K
source_datasets: []
status: active
configs:
  - config_name: narrative_pairs_verse
    data_files:
      - split: train
        path: data/narrative_pairs_verse/train-*.parquet
  - config_name: narrative_pairs_word
    data_files:
      - split: train
        path: data/narrative_pairs_word/train-*.parquet
  - config_name: poetic_pairs_verse
    data_files:
      - split: train
        path: data/poetic_pairs_verse/train-*.parquet
  - config_name: poetic_pairs_word
    data_files:
      - split: train
        path: data/poetic_pairs_word/train-*.parquet
---

# T'OMIM: Tanakh Observable Matches of Intertextual Mimesis

T'OMIM (Tanakh Observable Matches of Intertextual Mimesis, from Hebrew תאומים,
"twins") is an open-access dataset of labeled parallel passages in the Hebrew
Bible, compiled for computational and literary research on inner-biblical
intertextuality. It pairs two distinct corpora of known parallels:

- **554 narrative verse pairs** drawn from the Chronicles synoptic tradition,
  cataloged by Bendavid (2013) and Endres et al. (1998).
- **256 poetic half-verse pairs** identified by Berlin (2008), Fokkelman (2001),
  Kugel (1981), Watson (1994), and Tsumura (2023).

Each corpus is provided at two levels of granularity. Verse-level tables contain
the paired Hebrew texts with their source citations. Word-level tables expand
each passage into its constituent tokens, preserving the full morphological
annotation of the ETCBC Biblia Hebraica Stuttgartensia Amstelodamensis (BHSA;
van Peursen, Sikkel, and Roorda 2015): part of speech, verbal stem and tense,
gender, number, person, lexeme, English gloss, and hierarchical syntactic
structure.

The dataset is suitable for training and evaluating models for semantic
similarity, text reuse detection, and intertextual retrieval in Biblical Hebrew.

## Dataset Overview

- **Total rows**: 28,819 (across all configs)
- **Configs**: 4 (verse-level and word-level views of two corpora)
- **Languages**: Biblical Hebrew (text), English (glosses)

| Config | Rows | Columns | Description |
|--------|------|---------|-------------|
| `narrative_pairs_verse` | 554 | 6 | Narrative parallel pairs with Hebrew verse text |
| `narrative_pairs_word` | 25,572 | 58 | Word-level ETCBC annotation for narrative pairs |
| `poetic_pairs_verse` | 256 | 6 | Poetic parallel pairs with Hebrew half-verse text |
| `poetic_pairs_word` | 2,437 | 58 | Word-level ETCBC annotation for poetic pairs |

## Source

| Attribute | Value |
|-----------|-------|
| **Title** | T'OMIM: Tanakh Observable Matches of Intertextual Mimesis |
| **Author** | Smiley, David |
| **DOI** | [10.5281/zenodo.19135731](https://doi.org/10.5281/zenodo.19135731) |
| **Publisher** | Zenodo |
| **Version** | v1 (March 22, 2026) |
| **License** | CC-BY-4.0 |

The Hebrew text and morphological annotations are derived from the ETCBC Biblia
Hebraica Stuttgartensia Amstelodamensis (BHSA) dataset (van Peursen, Sikkel, and
Roorda 2015; [DOI:10.17026/DANS-Z6Y-SKYH](https://doi.org/10.17026/DANS-Z6Y-SKYH)).
T'OMIM includes a curated subset of ETCBC word-level features relevant to
parallel-passage analysis. Researchers requiring additional layers (clause type,
phrase syntax, discourse domain, qere/ketiv variants) can join T'OMIM's
structural identifiers (`nd` and the `in_*` columns) to the full BHSA dataset via
the [Text-Fabric framework](https://github.com/ETCBC/bhsa).

## Verse-Level Schema (`narrative_pairs_verse`, `poetic_pairs_verse`)

| Column | Type | Description |
|--------|------|-------------|
| `pair_id` | int64 | Sequential identifier for each parallel pair (0-indexed) |
| `source_ref` | string | Reference for the source passage (e.g., `2 Sam 3:2`, `2 Sam 22:14a`) |
| `target_ref` | string | Reference for the target passage (e.g., `1 Chr 3:1`, `Ps 18:14a`) |
| `source_text` | string | Vocalized Hebrew text of the source passage |
| `target_text` | string | Vocalized Hebrew text of the target passage |
| `reference` | string | Scholarly source for the parallel identification |

Narrative pairs use full verse references; poetic pairs use half-verse references
(with an `a`/`b` suffix indicating the half-verse division).

## Word-Level Schema (`narrative_pairs_word`, `poetic_pairs_word`)

Each row is a single word token from one side of a parallel pair. The `pair_id`
and `side` columns link each word back to its pair in the corresponding
verse-level table.

| Column | Type | Description |
|--------|------|-------------|
| `pair_id` | int64 | Matches `pair_id` in the verse-level table |
| `side` | string | `source` or `target` |
| `abbrev_ref` | string | Abbreviated verse reference (e.g., `Gen 1:1`) |
| `verse_ref` | string | Full verse reference |
| `book` | string | Book name (English; e.g., `Genesis`, `1 Samuel`) |
| `chapter` | int64 | Chapter number |
| `verse` | int64 | Verse number |
| `nd` | int64 | ETCBC node id (join key to the full BHSA dataset) |
| `otype` | string | ETCBC object type (e.g., `word`) |
| `number` | int64 | Word position |
| `in_book` | int64 | Book index within the corpus |
| `in_chapter` | int64 | Chapter index |
| `in_verse` | int64 | Verse index |
| `in_lex` | string | Lexeme container node |
| `in_half_verse` | string | Half-verse container node |
| `in_sentence` | string | Sentence container node |
| `in_sentence_atom` | string | Sentence-atom container node |
| `in_clause` | string | Clause container node |
| `in_clause_atom` | string | Clause-atom container node |
| `in_phrase` | string | Phrase container node |
| `in_phrase_atom` | string | Phrase-atom container node |
| `in_subphrase` | string | Subphrase container node |
| `g_cons` | string | Consonantal word (ETCBC transliteration) |
| `g_cons_utf8` | string | Consonantal word (Hebrew UTF-8) |
| `g_lex` | string | Lexeme as it appears (ETCBC transliteration) |
| `g_lex_utf8` | string | Lexeme as it appears (Hebrew UTF-8) |
| `g_word` | string | Vocalized word (ETCBC transliteration) |
| `g_word_utf8` | string | Vocalized word (Hebrew UTF-8) |
| `lex` | string | Lexeme id (ETCBC transliteration) |
| `lex_utf8` | string | Lexeme id (Hebrew UTF-8) |
| `voc_lex` | string | Vocalized lexeme (ETCBC transliteration) |
| `voc_lex_utf8` | string | Vocalized lexeme (Hebrew UTF-8) |
| `phono` | string | Phonetic transcription |
| `phono_trailer` | string | Phonetic trailer (inter-word material) |
| `trailer` | string | Word trailer (ETCBC transliteration) |
| `trailer_utf8` | string | Word trailer (Hebrew UTF-8) |
| `gloss` | string | English gloss of the lexeme |
| `language` | string | Language code (`Hebrew`/`Aramaic`) |
| `sp` | string | Part of speech |
| `pdp` | string | Phrase-dependent part of speech |
| `ls` | string | Lexical set |
| `freq_lex` | int64 | Corpus frequency of the lexeme |
| `rank_lex` | int64 | Frequency rank of the lexeme |
| `gn` | string | Gender |
| `nu` | string | Number |
| `ps` | string | Person |
| `st` | string | State (construct/absolute/emphatic) |
| `vs` | string | Verbal stem |
| `vt` | string | Verbal tense |
| `nme` | string | Nominal ending |
| `pfm` | string | Preformative |
| `prs` | string | Pronominal suffix |
| `prs_gn` | string | Pronominal-suffix gender |
| `prs_nu` | string | Pronominal-suffix number |
| `prs_ps` | string | Pronominal-suffix person |
| `uvf` | string | Univalent final |
| `vbe` | string | Verbal ending |
| `vbs` | string | Root formation |

## Usage

```python
from datasets import load_dataset

# Verse-level narrative parallels (Chronicles synoptic tradition)
verses = load_dataset("NuBerea/tomim", "narrative_pairs_verse", split="train")
print(verses[0]["source_ref"], "||", verses[0]["target_ref"])
print(verses[0]["source_text"])
print(verses[0]["target_text"])

# Word-level annotation for poetic parallels
words = load_dataset("NuBerea/tomim", "poetic_pairs_word", split="train")

# All tokens on the source side of pair 0
pair0_source = words.filter(lambda x: x["pair_id"] == 0 and x["side"] == "source")
for w in pair0_source:
    print(w["g_word_utf8"], w["sp"], w["gloss"])
```

## Citation

```
Smiley, D. (2026). T'OMIM: Tanakh Observable Matches of Intertextual Mimesis
[Data set]. Zenodo. https://doi.org/10.5281/zenodo.19135731
```

### Underlying source data

```
van Peursen, W.T., Sikkel, C., and Roorda, D. (2015). Hebrew Text Database
ETCBC4b. DANS. https://doi.org/10.17026/DANS-Z6Y-SKYH
```

## License

Distributed under the [Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/)
(CC-BY-4.0) license, matching the upstream Zenodo release.