README.md

metadata

license: apache-2.0
task_categories:
  - text-generation
language:
  - mg
size_categories:
  - 10K<n<100K

Collection Tononkalo Malagasy

Dataset Description

This dataset is a collection of Malagasy poems (Tononkalo) scraped from Vetso Serasera. It focuses on creative writing, rhymes, and artistic expression in the Malagasy language.

Important Note: This dataset has been rigorously filtered.

1. Language Filtering: Poems written primarily in French or English have been removed to ensure a high-quality Malagasy corpus.
2. Cleaning: Metadata, dates, author signatures inside the text, and website headers/footers have been stripped to leave only the raw poetic content.

• Curated by: Lo-Renz-O
• Language: Malagasy (mg)
• Source: Vetso Serasera
• License: User-generated content (Rights belong to original authors). Used for research/educational purposes.

Dataset Structure

The dataset is formatted in JSONL.

Data Fields

• url (string): The URL where the poem was found.
• title (string): The title of the poem.
• author (string): The name of the poet.
• content (string): The body of the poem, preserving line breaks (\n) for structure.

Use Cases

This dataset captures the creative, emotional, and metaphorical aspects of the Malagasy language. It contains vocabulary and sentence structures rarely found in formal news or wikipedia data.

• Creative Text Generation: Training LLMs to write poetry, song lyrics, or creative prose in Malagasy, adhering to rhyme schemes and rhythmic patterns.
• Vocabulary Expansion: Poetry often utilizes deep Malagasy vocabulary (teny gasy lalina), metaphors, and proverbs (ohabolana), helping models understand context beyond literal meanings.
• Cultural Preservation: Preserving the digital heritage of Malagasy contemporary and traditional poetry for future linguistic analysis.

Example

{
  "url": "https://vetso.serasera.org/tononkalo/author/title",
  "title": "Embona",
  "author": "Rado",
  "content": "Raha mbola misy embona\nny lasa tsy fanary\nny fo no aoka hiteny\n..."
}

Filtering Logic

To ensure quality for LLM training: Login Pages Removed: Redirects to login pages were filtered out. Signatures Removed: Lines at the end of poems containing the author's name or dates (e.g., "12/12/2025") were programmatically removed. Non-Malagasy Removed: A heuristic based on stop-words was used to exclude non-Malagasy texts.

How to Use

from datasets import load_dataset

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

# Print a poem
print(dataset['train'][0]['content'])