# Dormant Digital Collections, AI Triage: Lessons from the Islam West Africa Collection and Beyond

- **Type:** Lecture
- **Speakers:** Frédérick Madore
- **Event:** Deutsche Gesellschaft für Asienforschung (DGA)
- **Series:** DGA-Roundtable-Reihe Sommersemester 2026
- **Location:** Leibniz-Zentrum Moderner Orient
- **Country:** Germany
- **Date:** 6 May 2026
- **Language:** English
- **Project:** [Islam's 'Peripheries': Digital Humanities, Algorithmic Analysis, and AI in West Africa and Central Asia](https://www.frederickmadore.com/research/islams-peripheries-dh-ai-west-africa-central-asia.md)
- **Tags:** AI, Digital Humanities, IWAC, Large Language Models, OCR, Named Entity Recognition, Sentiment Analysis, West Africa, Archives
- **Web page:** https://www.frederickmadore.com/communications/dormant-digital-collections-ai-triage-dga-2026

## Abstract

Recent advances in large language models (LLMs) have significantly improved AI capabilities for non-Western languages and scripts. Drawing on my experience building and curating the Islam West Africa Collection (IWAC), a digital repository of over 14,500 newspaper articles and archival documents in French, English, Arabic, Hausa, and Ewe, I present a practical walkthrough of an AI-assisted research pipeline: from raw scans to interactive visualisation. The challenges I encountered with West African sources have direct parallels for scholars working with Asian languages and scripts.

I outline three stages of a "datafication" workflow: (1) text extraction through OCR and HTR, where multimodal LLMs increasingly outperform traditional engines on degraded documents in non-Western scripts; (2) data enrichment, including named entity recognition; and (3) data visualisation, where scholars without programming expertise can use AI-assisted coding to build interactive dashboards and explore their data.

I demonstrate this workflow through a sentiment analysis dashboard analysing how 12,000+ IWAC newspaper articles portray Islam across five West African countries. Featuring temporal trends, cross-country comparisons, and multi-model agreement metrics, the dashboard was developed without writing manually a single line of code, through iterative dialogue with AI assistants. I also introduce Google's NotebookLM for source analysis and synthesis.

Where Arlette Farge (2013) located the "allure" of the archive in the slow, tactile encounter with the document, AI offers a pragmatic complement: rapid triage of the ever-growing personal collections of scans that scholars accumulate, serving as an intuition pump for thinking with sources rather than automating their consumption. Rather than a theoretical discussion, this talk is an invitation to experiment, turning dormant digital collections into active research instruments.

## Links

- [Presentation](https://asienforschung.de/events/roundtable-2-digital-humanities-und-kuenstliche-intelligenz-in-den-regionalwissenschaften/)
- [Slides](https://slides.frederickmadore.com/talks/2026-05-06-dga-dormant-collections/)