HabibaAbderrahim/Tunisian-Proverbs-with-Image-Associations-A-Cultural-and-Linguistic-Dataset

Tunisian Proverbs with Image Associations: A Cultural and Linguistic Dataset

Description

This dataset explores the rich oral tradition of Tunisian proverbs mapped into text format, pairing each with contextual explanations, English translations both word-to-word and it's equivalent Target Language dynamic, Automated prompt and AI-generated visual interpretations. It bridges linguistic, cultural, and visual modalities making it valuable for tasks in cross-cultural NLP, generative art, and multi-modal learning for low-resourced Language such as the Arabic Tunisian Dialect.

Some Selections

ظل راجل ولا ظل حيط

الملح و الصحبة

كل قرده في عين امه غزال

قلبه أبيض كالحليب

اضرب القطوسة تتأدب العروسة

اللي وجهها يفجعها مال بوها ينفعها

Objectives

• Preserve and promote intangible cultural heritage from underrepresented languages.
• Create an open-access, FAIR-compliant resource to support Generative AI, NLP, and multimodal ML in low-resource languages like Tunisian Arabic.
• Provide a dataset suitable for translation, text and image generation, proverb understanding, visual grounding, and educational tools.

Dataset Structure

• A Tunisian proverb in dialectal Arabic.
• An explanation of its meaning.
• Contextual tags.
• English translations in different styles:
  • Informal translation
  • Formal interpretation
• A text-to-image automated prompt used to generate 4 unique images.
• Four associated image paths.
• A CLIP score indicating relevance of the images to the proverb.

Language & Cultural Focus:

• Dialect: Tunisian Arabic (Derja or Tounsi)
• Languag variety: Tunisian Arabic and English

How to Use

To load the dataset in Google Colab, you can use the datasets library from Hugging Face:

from datasets import load_dataset
import cv2
from google.colab.patches import cv2_imshow

# Load the dataset
dataset = load_dataset("Heubub/Tunisian-Proverbs-with-Image-Associations-A-Cultural-and-Linguistic-Dataset")

# Get the first sample from the 'train' split
sample = dataset["train"][0]

# Extract proverb and prompt and images e.g the first image
proverb = sample["tunisan_proverb"]
prompt = sample["prompt"]
image_path_1 = sample["image_path_1"]

print(f"Proverb: {proverb}")
print(f"Prompt: {prompt}")

img_bgr = np.array(image_path_1)[:, :, ::-1]
cv2_imshow(img_bgr)

##Citation

If you use this dataset, please cite it as follows:

@misc{abderrahim_habiba_2025,
    author       = { Abderrahim Habiba and Hedi Ayadi and Fadoua Ouamani },
    title        = { Tunisian-Proverbs-with-Image-Associations-A-Cultural-and-Linguistic-Dataset (Revision c524d31) },
    year         = 2025,
    url          = { https://huggingface.co/datasets/HabibaAbderrahim/Tunisian-Proverbs-with-Image-Associations-A-Cultural-and-Linguistic-Dataset },
    doi          = { 10.57967/hf/5189 },
    publisher    = { Hugging Face }
}