Challenge

Zentralbibliothek Zürich provides a text corpus of printed travelogues from the 16th to 19th century. Can you identify and extract colonial traces in these French and German texts? For instance, you could try describe the gaze on the "other" by tracking down mentions of geographic regions, different languages, certain ethnicities or one of the following semantic fields with NLP methods:

• the concept of the "noble savage"
• slavery
• representation of dominance
• hierarchies and structures of control

For instance, the last field could be looked at from an economic, military, cultural or political perspective. Starting from a military perspective, word clusters like "Truppe – Fortifikation – Kriegszug - verfolgt – Beute – Treueeid – schwören – Vasall – kniend - Tribut" could be interesting to search for inside the German texts.

What perspectives on colonialism can be gained by the extracted entities, that are normally not considered in travel narratives?

- Shift the focus away from the positioning and the role of the author in the colonial project

- Undertake a schematic grouping: Identify intermediaries, rand-and-file employees or marginalized voices

- Map the colonial infrastructure in an arena

- Compare the different naming for places in time

We used the SBB-NER-Tagger, developed by the Staatsbibliothek zu Berlin (SBB), for Named Entity Recognition in the travelogues. We were able to extract person, organization and place entities from the OCR texts. The SBB-NER-Tagger contains a BERT-based model which has been trained on the SBB collections of early modern prints in German, French and English language, and (at first sight) seems to produce some decent results.

We created one JSON file per text page, starting from the original JSON file of the dataset, and enhancing each page object with the identified entity strings and types (person, place, organization).

Result

We have a prototypical frontend displaying the entities present per book page: https://luminous-speculoos-e2e07f.netlify.app/

For a reference of the entity categories used, please see the BERT documentation.

A possible next step would be to use the Named Entity Linking Tool SBB-NED from SBB to link the entities found in the texts to Wikidata objects.