yaser-aj/qatar-1908

Qatar in 1908 (قطر عام 1908)

Every place in Qatar that J. G. Lorimer's Gazetteer of the Persian Gulf, 'Oman and Central Arabia (Vol. II, Geographical and Statistical, Calcutta 1908) describes, as one row per place: towns, villages, quarters of Doha, wells, forts, capes, bays, islands and hills, with what the Gazetteer records about each (houses, population, tribes, pearl boats, wells, date palms, livestock, ruler), coordinates, and a link to the printed source page. Each place also carries a present-day layer taken from Wikipedia, Wikidata, OpenStreetMap and Wikimedia Commons, so the 1908 description can be read beside the place today (version 2: see Present-day layer below).

Built for Qafal ("Bringing Qatar's past home") at the QAI x Hugging Face Hackathon 01, Doha, November 2026.

214 places (112 wells, 25 capes, 15 islands, 13 bays, 11 ruins, 10 villages, 9 quarters, 6 areas, 6 forts, 5 hills, 2 towns); 112 are located exact (a modern Wikidata/GeoNames feature judged to be the same place) and 102 approximate (a related modern feature nearby, or a point computed from Lorimer's own bearings and distances). 538 passages: 282 from the Vol. II entries and 256 from Vol. I's chapter "History of Qatar" (pp. 787–835), for search and citation. 26 tribes from Lorimer's table of Qatar's settled population. 27 places carry a Wikipedia lead, 18 an image.

Files

Config	File	One row per
places	places.jsonl	place described in 1908
passages	passages.jsonl	passage of Lorimer's text (mostly 60–110 words; table rows are shorter), for search and chat; each cites its page. volume is II (Geographical, 1908) or I (Historical, 1915, with heading = Lorimer's marginal note and years)
tribes	tribes.jsonl	tribe or community in Lorimer's table of Qatar's settled population

from datasets import load_dataset
places = load_dataset("yaser-aj/qatar-1908", "places", split="train")
passages = load_dataset("yaser-aj/qatar-1908", "passages", split="train")

Main fields of places

Field	Meaning
id	Stable slug, e.g. wakrah, khor-shaqiq, malaihah-1
name_1908, aliases	Lorimer's spelling and the other forms he gives
name_en, name_ar, name_ar_source	Present-day English name (from Wikidata when matched) and Arabic name, with where the Arabic came from (wikidata:Q…, geonames:…, or transliterated from Lorimer's spelling)
kind, section	town, village, quarter, well, fort, ruin, cape, bay, island, hill, area; and which part of the Gazetteer the row comes from (east coast, west coast, interior, article)
lat, lon, location_quality, coord_source	exact: a Wikidata/GeoNames feature judged to be the same place. approximate: computed from Lorimer's own position ("9 miles south-east of Ruwais"), see below. unknown: no position could be computed
houses, population_1908, tribes, tribe_details	As printed; population_1908 keeps Lorimer's wording ("about 8,000 souls")
pearl_boats, other_boats, date_palms, wells, well_depth_fathoms, water_quality, livestock, ruler_1908	As printed
resources_table_1908	The row for this place in Lorimer's table of the resources of Qatar (p. 1533), when there is one
nature_1908, position_1908, remarks_1908, notes_1908	The Gazetteer's own cells or sentences
summary_1908_en, summary_1908_ar	Short summary of the 1908 text; the Arabic is a machine translation of the English (summary_1908_ar_note)
today_name_en, today_name_ar, today_en, today_ar, today_source_en, today_source_ar	Present-day names and the opening sentences of the English and Arabic Wikipedia articles, copied unchanged, with their URLs
today_population	Latest population on Wikidata, with year and source
image	Wikidata's image (P18) from Wikimedia Commons, with licence, author and file page
wikidata_id, geonames_id	Identifiers of the matched modern feature
source_pages	Printed pages with links to the Qatar Digital Library page image (url_qdl, url_qdl_checked) and to the scanned volume on the Internet Archive (url_archive); url is the QDL link when its image number is confirmed, otherwise the archive link
ocr_uncertain	Every reading of the OCR that is in doubt, and every correction made to it
checks	Automatic checks, e.g. distance between the matched feature and Lorimer's position

Present-day layer (v2, 6 November 2026)

Version 2 adds two files that describe each 1908 place today, from four open sources. The 1908 files (places, passages, tribes) are unchanged from version 1 (tag v1.0).

• 212 of 214 places have a short present-day summary in English and Arabic (v1: 27).
• 1,811 present-day passages for search and citation: 1,086 from 312 English and Arabic Wikipedia articles, 278 from Wikidata, 291 from OpenStreetMap, and 156 from Qatar's Open Data Portal (v2.1). 209 places can be asked about today.
• 78 places carry the numbered zone they lie in today, with the zone's residents and area from the National Planning Council's table by zone (2015 figures). The figure is the whole zone's, and the passages say so.
• 58 places list landmarks near them today (171 links to 100 Wikidata items: Souq Waqif, the Museum of Islamic Art, Al Janoub Stadium, Lusail Stadium...).
• Every passage carries its own source, license, license_url, attribution and page_url.

Config	File	One row per
today_places	today_places.jsonl	1908 place: what is at its site today, summaries, landmarks nearby
today_passages	today_passages.jsonl	present-day passage (EN or AR), with its source and licence

today = load_dataset("yaser-aj/qatar-1908", "today_passages", split="train")

Main fields of today_passages: passage_id, place_id and place_ids (the 1908 places it is about), source (wikipedia, wikidata, openstreetmap or qatar_open_data), lang, heading, text, page_label, page_url, license, license_url, attribution, retrieved, about_qid (the Wikidata item described), relation (same place, nearby or location) and distance_km for landmarks.

Main fields of today_places: place_id, today_qids, today_name_en/ar, today_kind_en/ar, area_en/ar and municipality_en/ar (from OpenStreetMap), country_en/ar, summary_en/ar with summary_source_en/ar, population (value, year, source), heritage, landmarks (name, kind, distance, coordinates, Wikipedia and Wikidata links, Commons image), zones (number, name, residents, year, km²) and sources.

How it was made: Wikidata items with an English or Arabic article inside Qatar's bounding box were matched to the 1908 places by their Wikidata link or by name (scripts/names.py), and landmarks by distance and how well known they are. Article text comes from the October 2026 Wikipedia dump on the Hub (omarkamali/wikipedia-monthly); sections that only tell pre-1950 history are left out, since the 1908 side comes from Lorimer. OpenStreetMap's Nominatim gave the district and municipality of every site. Sentences whose numbers were lost with Wikipedia's unit templates are repaired or left out, and one sentence (Al Wakrah's lead) is reordered so its 1900s governor is not read as today's. No language model wrote any present-day fact. Rebuild steps: docs/present-day.md.

Not included: the UNESCO site could not be reached from the build machine, and the open data portal's heritage table only counts sites by category; two places (Ghubaibah, Jazirat Rubadh) have no present-day text.

How it was made

1. Text. The OCR of Vol. II from the Digital Library of India scan on the Internet Archive (in.ernet.dli.2015.206964) was split into printed pages by its running heads (scripts/segment_pages.py). Printed pages were mapped to Qatar Digital Library image numbers (scripts/qdl_pages.py). The Qatar entries are pp. 1505–1535 (the QATAR article and its tables of the east coast, west coast and interior) plus 18 separate articles (Doha, Wakrah, Khor Shaqiq, Zubarah, Fuwairat and others).
2. Extraction. Eight Claude agents, working in parallel on parts of the text, wrote one row per place and the passages, following scripts/extraction_prompt.md: only what the text says, OCR errors fixed but not rewritten, every doubtful reading recorded in ocr_uncertain.
3. Checks. A script checks every number against the cited pages and every passage's words against the OCR (scripts/check_rows.py). A second set of agents, which had not written the rows, re-read each row against the text (scripts/verify_prompt.md). Remaining corrections were made by hand, each with its reason (data/raw/corrections.jsonl); for example the pearl-boat figures of Doha (350) and Fuwairat (35), which the OCR gives as 850/360 and 85 and which are confirmed by the printed column total of 817.
4. Places today. Each place was compared with Wikidata items and GeoNames entries in Qatar by a transliteration-aware name match and by distance from Lorimer's position (scripts/match_candidates.py); agents chose a match or none, and the build re-checks the distance. Places without a match are located from Lorimer's own bearings and distances ("9 miles SE of Ruwais", "3 miles below Dohah" along the coast) using the Natural Earth coastline (scripts/geocode.py); these are approximate and can be several kilometres off.
5. History. Four more agents turned Vol. I, Part I, Chapter IV "History of Qatar" (pp. 787–835, Internet Archive scan in.ernet.dli.2015.206962) into passages under the same rules (scripts/history_prompt.md); each keeps Lorimer's marginal note as heading and links the places it is about (place_ids).
6. Present-day text and images come from Wikipedia's REST API and Commons' metadata unchanged (scripts/fetch_today.py). No language model wrote any present-day fact.

Rebuild: python scripts/build_dataset.py data/work data/ref data/processed in the Qafal repository; build_report.json lists every fix applied and every warning (e.g. matches far from Lorimer's position).

Limitations

• The OCR is noisy (3/8, 5/6 and 1/l confusions; tables read column by column). Doubtful readings are kept and flagged rather than silently resolved; check ocr_uncertain and the page image before relying on a figure.
• Lorimer's spellings, tribal names and descriptions are those of a British colonial official in 1908 and reflect that viewpoint and its terminology.
• Approximate coordinates are computed from statute miles and compass bearings and can be several kilometres off, especially for wells far from any matched place.
• Internet Archive links (url_archive) open the scan at the printed page number; the Qatar Digital Library link (url_qdl) is exact when url_qdl_checked is true and may be a page or two off otherwise.
• summary_1908_ar is a machine translation; Arabic names marked transliterated from Lorimer's spelling are a transliteration, not an attested modern spelling.

Sources and licences

• J. G. Lorimer, Gazetteer of the Persian Gulf, 'Oman and Central Arabia, Vol. II (1908), public domain. Page images: Qatar Digital Library (British Library, IOR/L/PS/20/C91/4; Open Government Licence). OCR: Digital Library of India scan on the Internet Archive.
• Wikidata (CC0), GeoNames via the Hub dataset yuiseki/geonames (CC BY 4.0), Natural Earth coastline (public domain).
• Wikipedia text (today_en, today_ar): CC BY-SA 4.0, from the article linked in today_source_en/ar.
• Present-day layer (today_places, today_passages): each row names its own licence. Wikipedia passages are CC BY-SA 4.0 (Wikipedia contributors, article in page_url); Wikidata passages CC0 1.0; OpenStreetMap passages and the area/municipality fields contain information from OpenStreetMap, © OpenStreetMap contributors, available under the Open Database License (ODbL 1.0). Zone rows (source qatar_open_data, zones): Qatar Open Data Portal, National Planning Council, CC BY 4.0.
• Images: each under the licence and author given in its image field (Wikimedia Commons).

Because the present-day layer includes Wikipedia text, the dataset as a whole is released under CC BY-SA 4.0. Rows derived from OpenStreetMap remain under the ODbL 1.0, as each row's license says.