Lo-Renz-O/GBV-Malagasy

Dataset Description

This dataset contains news articles, stories, and cultural reports scraped from Global Voices Malagasy. It is intended to support Natural Language Processing (NLP) tasks for the Malagasy language, such as language modeling and text generation.

The data is automatically scraped and updated once a month to ensure freshness.

Dataset Structure

The dataset is formatted in JSONL (JSON Lines). Each entry represents a single article.

Data Fields

• url (string): The original URL of the article.
• title (string): The title of the article.
• date (string): The publication date (ISO format or similar).
• author (string): The name of the author or translator.
• content (string): The main body text of the article (cleaned).

Example

{
  "url": "https://mg.globalvoices.org/2025/12/15/example-story",
  "title": "Lohateny momba ny fiarahamonina",
  "date": "2025-12-15T10:00:00+03:00",
  "author": "Rakoto",
  "content": "Ity dia ohatra iray amin'ny lahatsoratra hita ao amin'ny Global Voices..."
}

Use Cases

This dataset is ideal for training models on formal and standard Malagasy. Because news articles follow specific grammatical rules and journalistic standards, they provide a clean baseline for NLP.

• Language Modeling (LLM Pre-training): To teach models the core grammar, syntax, and vocabulary of the Malagasy language in a formal context.
• Named Entity Recognition (NER): The dataset contains numerous mentions of people, locations, organizations, and dates relevant to Madagascar and the world, useful for training entity extractors.

How to use

from datasets import load_dataset

dataset = load_dataset("Lo-Renz-O/GBV-Malagasy")

# Print the first example
print(dataset['train'][0])