Developing an OCR - Wikibase Pipeline for Place Names in the RGTC Series

Description

Abstract (English)

I present the development of a semi-automated workflow that extracts and digitally stores certain ‘geographically relevant information’ related to place names inventoried in the publication series Répertoire géographique des textes cunéiformes. Geographically relevant information goes beyond the place names themselves and includes schematizing the spatial relations and other geographical information discussed in the prose section of each entry.

This schematized information is then imported in a Wikibase hosting a linguistically annotated corpus of Akkadian, adding an additional layer of cross-linked information. This workflow involves:

1. Training and fine-tuning of an OCR model tailored to the format of the RGTC print publications.

2. Applying text-processing scripts that correct some of the OCR results themselves, generate additional training data, and improve the results of the extraction script.

3. Structuring the extracted results so they can be stored as geoJSON objects.

4. Importing those objects into the Wikibase.
   
   

In particular, I focus on volume seven of that series, which deals with certain first millennium place names. The benefits of this work include a reasonably accurate digital copy of the place name entries within an RGTC volume, an OCR model which can be applied to similarly-structured volumes, and an example of how the spatial information within that volume can be represented as linked open data within Wikibase.