About

Idrīsī is an open digital library of historical geography for the premodern Islamic world. It gathers the placenames and geographic coordinates preserved across the Islamic geographical tradition and organizes them as structured, searchable data, enabling scholars to locate a place, identify every source that records it, and compare how it was named and situated over time.

The library takes its name from the twelfth-century geographer al-Idrīsī, who compiled one of the medieval world's most ambitious descriptions of the inhabited earth. It forms part of the Dabīrān Project on the Intellectual History of the Islamic World, based at the Roshan Institute for Persian Studies, University of Maryland.

The challenge

Identifying a place in a premodern source is seldom straightforward. A single city may appear under several spellings, or under altogether different names, from one author to the next; one name may denote more than one place; and the coordinates assigned to a location often shifted as geographic knowledge advanced. These difficulties are especially pronounced for the premodern Islamic world, given the extent of the territories and the range of languages its geographers described.

At the same time, libraries have digitized more than a million Islamic manuscripts. This vast corpus is now visible online, yet it remains largely unusable as data. Conventional text-recognition tools are designed for modern print and falter on handwritten Arabic-script pages, with their varied hands, dense tables, and numerals written as letters.

How it works

Idrīsī treats the manuscript page not merely as text to transcribe but as structure to recover. Using computer vision, we detect the tables on a page and segment them into their constituent cells. Where a reliable printed edition exists, we align it to the manuscript cell by cell, preserving the accuracy of a vetted text while remaining anchored to the original source. Natural-language processing then assists in identifying the placenames and coordinates within the extracted text.

Every automated step is reviewed by scholars, and their corrections are fed back into the models, so that accuracy improves over time. The result is a corpus in which each value is at once machine-readable and traceable to the precise page from which it was drawn.

Points of departure

The project began with a printed reference work: E. S. and M. H. Kennedy'sGeographical Coordinates of Localities from Islamic Sources (1987), which assembled the coordinates recorded for roughly 1,250 localities across some seventy-four sources. Its highly regular tables made it an ideal proving ground for our methods, and yielded a vetted dataset of approximately 45,000 entries, each tagged with its provenance.

From there the project turned to the manuscripts themselves. A central source is the Taqwīm al-buldān ("Reckoning of Cities") of Abū al-Fidāʾ, a fourteenth-century geographer who arranged the known world into tables of names, coordinates, and descriptions — a gazetteer that anticipates the logic of the modern database. Working directly from the manuscript preserves detail and nuance that printed editions frequently omit.

Openness and reuse

Idrīsī is designed for reuse. We release our datasets and documentation to open repositories and contribute placename and coordinate data to established open gazetteers. We also develop and publish standards for encoding geographic data in premodern, multilingual sources, so that other researchers may build upon the same foundation.

Idrīsī has been supported by a 2024 Digital Humanities Advancement Grant from the National Endowment for the Humanities. Read the University of Maryland's announcement of the award.