---

license: cc-by-4.0 task_categories:

• question-answering
• visual-question-answering language:
• nl
• fr
• es size_categories:
• 1K-10K annotations_creators:
• machine-generated source_datasets:
• original

Colonial Newspaper QA & VQA Dataset

Overview

This dataset contains Question–Answer (QA) and Visual Question Answering (VQA) data derived from 19th-century colonial newspapers.
The corpus focuses on administrative, legal, and social content, with particular emphasis on slavery-related discourse, such as manumission requests, slave sales, official announcements, and public notices.

The dataset is designed for research in:

• Historical NLP
• Multilingual Question Answering
• Visual Question Answering (VQA)
• Multimodal document understanding
• Semantic and discourse analysis of historical texts

Dataset Structure

The dataset is released as a single Hugging Face dataset with multiple configurations (configs).
Each configuration corresponds to a specific language or task.

Configurations

Config name	Description
qa_nl	Dutch textual QA
qa_fr	French textual QA
qa_es	Spanish textual QA
vqa_nl	Dutch visual question answering

Each configuration can be loaded independently using the Hugging Face datasets library.

File Organization

dataset/
├── README.md
├── images/                # Page-level newspaper images (for VQA)
├── QA/
│   ├── DutchQA.json
│   ├── FrenchQA.json
│   └── SpanishQA.json
└── VQA/
    └── DutchVQA.json