---
pretty_name: Kurdish Parallel Corpus — Raw Source
license: odc-by
language:
- ku
- kmr
- ckb
- tr
- en
- ar
tags:
- kurdish
- multilingual
- news
- raw-html
- web-crawl
configs:
- config_name: "2026_11_06"
  data_files:
  - split: train
    path: "2026_11_06/*.jsonl.gz"
---

# Kurdish Parallel Corpus — Raw Source

Whole-domain crawls of Kurdish / multilingual news sites, stored as **raw HTML**
(no extraction, no filtering). This is a **raw source corpus** — a candidate for
building aligned/parallel data downstream, not parallel data itself. One JSON
object per line:

| field | description |
|-------|-------------|
| `url` | final page URL (after redirects) |
| `html` | raw HTML of the page, unchanged |
| `content_type` | the response `Content-Type` header |
| `fetched_at` | ISO 8601 UTC fetch timestamp |

Each dated subset (e.g. `2026_11_06`) is one crawl snapshot. Extraction, metadata
parsing and alignment are all done downstream against this raw HTML, so the
source is never lost and can be re-processed without re-crawling.

## Pipeline

This dataset was produced by the open-source crawling pipeline at
[kurdish_parallel_crawl](https://github.com/cikay/kurdish_parallel_crawl) — a
Scrapy crawler that walks whole news domains and stores each page's raw
HTML.