Unsilencing Colonial Archives via Automated Entity Recognition

TL;DR

This paper introduces an automated entity recognition approach to enhance access to colonial archives by identifying marginalized individuals often omitted in traditional indexes, supported by a new annotated corpus and baseline models.

Contribution

It develops a specialized annotation typology, creates a large annotated corpus from the Dutch East India Company archives, and provides baseline neural network models for automated entity recognition.

Findings

Nearly 70,000 annotations released as shared task

Baseline neural models demonstrate effective entity recognition

Automated recognition can help broaden access to colonial archives

Abstract

Colonial archives are at the center of increased interest from a variety of perspectives, as they contain traces of historically marginalized people. Unfortunately, like most archives, they remain difficult to access due to significant persisting barriers. We focus here on one of them: the biases to be found in historical findings aids, such as indexes of person names, which remain in use to this day. In colonial archives, indexes can perpetuate silences by omitting to include mentions of historically marginalized persons. In order to overcome such limitations and pluralize the scope of existing finding aids, we propose using automated entity recognition. To this end, we contribute a fit-for-purpose annotation typology and apply it on the colonial archive of the Dutch East India Company (VOC). We release a corpus of nearly 70,000 annotations as a shared task, for which we provide baselines using state-of-the-art neural network models. Our work intends to stimulate further contributions in the direction of broadening access to (colonial) archives, integrating automation as a possible means to this end.