README.md

---
license: cc-by-nc-sa-4.0
pretty_name: "islamlab — Persian Islamicate Prose"
language:
  - fa
task_categories:
  - text-generation
  - text-retrieval
size_categories:
  - 10K<n<100K
tags:
  - persian
  - islamicate
  - historiography
  - classical-persian
  - islam
configs:
  - config_name: works
    default: true
    data_files:
    - split: train
      path: works/train-*.parquet
  - config_name: units
    data_files:
    - split: train
      path: units/train-*.parquet
  - config_name: passages
    data_files:
    - split: train
      path: passages/train-*.parquet
---

# islamlab — Persian Islamicate Prose

**237 Persian works** by authors who died between the 4th and the
13th Hijri century, segmented into **73,734 units** and **127,089 retrieval
passages** — 125 million characters.

Persian is the second language of Islamicate learning, and this is the part of
it that survives the same gates the Arabic corpus is held to. What survives is
chiefly **historiography** — Bayhaqī's *Tārīkh*, Mīrkhwānd's *Rawḍat al-ṣafā*,
Sharaf al-Dīn Yazdī's *Ẓafar-nāma*, Abū al-Faḍl's *Akbar-nāma*, Firishta's
*Tārīkh* — together with the *tadhkira* literature of biographical notices.

| work | author | d. (AH) | size |
|---|---|---:|---:|
| Habib Siyar | Khwandamir | 942 | 5.8M |
| ʿArafāt al-ʿāšiqīn | Awhadi Balyani | 1050 | 5.0M |
| Tafsir | Abu Bakr Surabadi | 484 | 3.9M |
| Tārīḫ | Muhammad Qasim Firishta | 1053 | 3.4M |
| Tārīḫ Bayhaqī | Abu Fadl Bayhaqi | 470 | 2.7M |
| Shāh Ǧahān Nāmah | Kanbuh Lahawri | 1082 | 2.4M |
| Rawdat Safa | Mir Khwand | 903 | 2.1M |
| Zafar Nama | Sharaf Din Yazdi | 858 | 1.9M |
| Akbar Nama | Abu Fadl Callami | 1011 | 1.9M |
| Khulasat Tawarikh | Ahmad Ibrahimi Husayni | 1026 | 1.8M |
| Matlac Sacdayn | Cabd Razzaq Samarqandi | 887 | 1.8M |
| Haft Iqlim | Amin Ahmad Razi | 1022 | 1.6M |

## What is not here, and why

This is a **separate dataset rather than a ninth shelf** of
[`islamlab/islamic-sciences`](https://huggingface.co/datasets/islamlab/islamic-sciences),
for two honest reasons.

The first is metadata. OpenITI's Persian subcorpus is thinly catalogued: two
thirds of it carries no genre tag at all, and a third has no death year in the
identifier. The Arabic corpus files every work under a science because the
sources say which one; here they do not, so no science is claimed. There are
no shelves in this dataset — only works, units and passages.

The second is the gate. The same rule that keeps Ibn ʿArabī and Ibn al-Fāriḍ
out of the Arabic corpus keeps out the Persian Sufi poets — ʿAṭṭār, Rūmī,
Sanāʾī, Jāmī — along with the Ismaili dāʿī Nāṣir-i Khusraw, the Twelver
scholars, and the secular dīwāns, which are not an Islamic science in any
language. That refusal is by **author name**, which is a blunter instrument
than the Arabic side's tag-and-collection gate. Read this dataset knowing
that: it is Persian Islamicate prose that passed a name-based filter, not a
corpus each of whose authors has been vetted one by one.

## Columns

`works` carries `work_id`, the transliterated title and author,
`death_year_ah` / `death_year_ce`, `century_ah`, size, and `text_quality`.
`units` is one heading-delimited block per row with `page_start` / `page_end`
anchors; `passages` re-cuts the same text to ~1,200 characters with a ready
citation. Keys join across all three on `work_id` and `unit_id`.

## Sources and licence

The texts come from the **Open Islamicate Texts Initiative**, release
2025.1.9 (CC BY-NC-SA 4.0), and this dataset carries the same terms. The works
themselves are public domain; the gate admits nothing whose author died after
1350 AH.

Corrections and removal requests are welcome on the **Community** tab.