AfroSentiments

AfroSentiments is a comprehensive multilingual sentiment and emotion corpus covering hundreds of African languages, built using English as a proxy via Google Gemini (gemini-2.5-flash-lite) and aligned on verse keys from the Africa Corpus Builder.

Dataset Structure

Each language subset contains sentence-aligned text with the following features:

• verse_key: Book, chapter, and verse identifier (e.g., GEN.1.1).
• text: Text in the target African language.
• sentiment: Binary sentiment label (positive or negative).
• emotion: Primary emotion (e.g., awe, hope, joy, reverence, despair).
• english_text: The aligned English proxy text.

Usage

Load any language subset easily using the Hugging Face datasets library:

from datasets import load_dataset

# Load Swahili sentiment dataset
dataset = load_dataset("AfriSpeech/afrosentiments", "swh")
print(dataset["train"][0])

# Load Twi sentiment dataset
twi_dataset = load_dataset("AfriSpeech/afrosentiments", "twi")