Welcome

to the Machines Reading Maps Tutorials and Wiki! We have developed digital tools for the detection and semantic enrichment of text on maps, and you are very welcome to use them.

If you find any mistake in these tutorials, or if you had a question and you couldn't find an answer in the documentation, please get in touch with us by emailing [email protected] or by opening an issue in this repository.

This wiki has been prepared as a complement to Machines Reading Maps annotation tasks and events.

Its main aim is to introduce our semantic annotation platform that combines features of the manual image annotation tool Recogito, originally developed by Rainer Simon for Pelagios, with mapKurator, an automatic map text spotting tool developed by Yao-Yi Chiang, Zekun Li, and Jina Kim at the University of Minnesota.

The wiki highlights basic information in the main section, but also offers extra material for those who are interested in exploring the tools further. For some of the features of Recogito that were already part of its main, public instance at pelagios.recogito.org, we have pointed to existing resources.

Another core objective of this wiki is to document our map annotation practices. These reflect the choices we have made in annotating gold standard data for evaluating mapKurator performance.

Table of Contents

• Integrated Recogito + mapKurator Tutorials
• Tutorials for Specific Events
• Map Text Annotation Guidelines

Integrated Recogito + mapKurator Tutorials

This version of Recogito enables two types of annotation: manual and automatic. The two approaches are complementary and can be used as stand alone or in combination. If you are interested in manual annotation only, you can read the tutorial Manual annotation with Recogito. If you want to perform a preliminary automatic annotation and then verify and enrich it manually, you can read the tutorial Automatic annotation with mapKurator and Recogito.

Manual annotation

• Tutorial on how to annotate scanned maps using bespoke Recogito interface created for Machines Reading Maps.

Automatic annotation

• Tutorial on how to use the advanced features of the integrated Recogito + mapKurator to enable "automatic annotation" of scanned maps.

Using mapKurator within Recogito can speed up the process of manual annotation for certain kinds of maps. Please note that performance of mapKurator will vary widely depending on the features of the digitized map image (coloration, background texture, language of text on map, font/script of text, image resolution, etc).

Tutorials for Specific Events

• Edinburgh Transcription Project (National Library of Scotland)

Map Text Annotation Guidelines

We document the process for the creation of Gold Standard (GS) data, extended to both our case-study collections: the Ordinance Survey Maps of Great Britain, and the Sanborn Maps. These guidelines work internally, as a way to reflect on the rational of the manual annotation, but can also be seen by future collaborators as a reference in the application of the methodology to other case studies.

Some of the criteria are specific for the creation of gold standard data for the evaluation of mapKurator computer vision and entity linking models. Combining the concerns of humanistic and machine learning research, these guidelines ensure that map annotations are captured in a way that is useful for computational research in multiple domains. Researchers simply interested in deep annotation of few maps for other research purposes (for example, where evaluation of inferred data is not an issue) don't necessarily need to adhere to these guidelines.

• Annotating for the Gold Standard