license: mit
The COHeN dataset is designed for training and fine-tuning Biblical Hebrew classification models. It consists of 11968 verses from the Hebrew Bible along with labels indicating to which of the four chronological phases of the language each verse belongs. The following verses were chosen to represent each stage:
Archaic Biblical Hebrew (ABH): Gen 49:2–27; Exod 15:1–18; Num 23:7–10, 18–24; 24:3–9; 15–24; Deut 32:1–43; Deut 33:2–39; Jud 5:2–31; 2 Sam 22:2–51; Psa 18:1–50; 68:2–35 (minus the prose introductions in Exod 15:1; Num 23:7, 18; Num 24:3, 15; Deut 33:2; 2 Sam 22:2; Ps 18:2)
Classical Biblical Hebrew (CBH): 1 Sam 1:1–31:13; 2 Sam 1:1–22:1; 23:1–24:25; 1 Kgs 1:1–22:53; 2 Kgs 1:1–25:30
Transitional Biblical Hebrew (TBH): Isa 40:1–54:17; Jer 1:1–10:10; 10:12–42:34; Ezel 1:1–48:35
Late Biblical Hebrew (LBH): Eccl 1:1–12:14; Esth 1:1–10:3; Dan 1:1–2:4a; Dan 7:29–12:13, Ezra 1:1–4:6; 6:19–7:11; 7:26–10:44; Neh 1:1–13:31; 1 Chr 1:1–9:44; 12:1–40; 15:1–24; 16:7–43; 21:1–29:19; 2 Chr 7:1–3; 14:9–15:7; 17:1–19; 21:12–17; 24:25–22; 26:6–21; 29:3–31:21; 33:10–20; 34:3–7; 36:22–23
Class balance was achieved by oversampling from the three minority classes (ABH, TBH, and LBH).
The script used to generate the COHeN dataset can be found here.