T'OMIM: Tanakh Observable Matches of Intertextual Mimesis

T'OMIM (Tanakh Observable Matches of Intertextual Mimesis, from Hebrew תאומים meaning "twins") is an open-access dataset of labeled parallel passages in the Hebrew Bible, compiled for computational and literary research on inner-biblical intertextuality. The archive 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), and 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 (van Peursen, Sikkel, and Roorda 2015): part of speech, verbal stem and tense, gender, number, person, lexeme, English gloss, and hierarchical syntactic structure. The four resulting tables are distributed as Apache Parquet files under a CC-BY-4.0 license, suitable for training and evaluating models for semantic similarity, text reuse detection, and intertextual retrieval in Biblical Hebrew.

Files

File Name
Rows
Description

narrative_pairs_verse.parquet
554
Narrative parallel pairs with Hebrew verse text

narrative_pairs_word.parquet
25,572
Word-level ETCBC annotation for narrative pairs

poetic_pairs_verse.parquet
256
Poetic parallel pairs with Hebrew half-verse text

poetic_pairs_word.parquet
2,437
Word-level ETCBC annotation for poetic pairs

Verse-Level Tables

Both narrative_pairs_verse.parquet and poetic_pairs_verse.parquet share the same schema:

Column
Type
Description

pair_id
int
Sequential identifier for each parallel pair (0-indexed)

source_ref
string
Verse or half-verse reference for the source passage (e.g., "2 Sam 3:2" or "2 Sam 22:14a")

target_ref
string
Verse or half-verse reference for the target passage (e.g., "1 Chr 3:1" or "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 a/b suffix indicating the half-verse division).

Word-Level Tables

Both narrative_pairs_word.parquet and poetic_pairs_word.parquet share the same 58-column schema. Each row represents 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.

Pair linkage columns

Column
Type
Description

pair_id
int
Matches pair_id in the verse-level table

side
string
"source" or "target"

Verse and book locators

Column
Type
Description

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
int
Chapter number

verse
int
Verse number

Data Source and Extended Annotation

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