---
language:
- en
- nl
- it
license: mit
datasets:
- PolySarc
---

# **PolySarc Dataset: Multilingual Sarcasm Detection Corpus**

## **Overview**
PolySarc is a **multilingual sarcasm detection dataset** that contains labeled sentences in **English, German, Italian, Dutch, and Hinglish**.  
It is designed for **training and evaluating sarcasm detection models** across multiple languages, making it a valuable resource for **cross-lingual NLP research**.

The dataset includes three key columns:
- **`sentence`**: The textual content of the sentence.
- **`label`**: Binary sarcasm label (`0` = Non-Sarcastic, `1` = Sarcastic).
- **`language`**: Language code corresponding to the sentence.

---

## **Dataset Statistics**
### **Language Distribution**
The dataset contains **79,428 sentences** distributed across four languages:

| **Language** | **Total Sentences** |
|-------------|------------------|
| 🇮🇹 **Italian (it)** | 23,871 |
| 🇬🇧 **English (en)** | 22,837 |
| 🇳🇱 **Dutch (nl)** | 20,771 |
| 🇮🇳 **Hinglish (hi-en)** | 11,949 |

---

### **Label Distribution**
The dataset is **imbalanced**, with more non-sarcastic sentences than sarcastic ones:

| **Label** | **Meaning** | **Count** |
|----------|------------|-----------|
| **0** | Non-Sarcastic | 45,921 |
| **1** | Sarcastic | 33,507 |

- **57.8% of sentences are non-sarcastic**.
- **42.2% of sentences are sarcastic**, making it a substantial dataset for sarcasm detection.

---

## **Why PolySarc?**
✅ **Multilingual**: Covers multiple languages, allowing cross-lingual sarcasm analysis.  
✅ **Balanced Representation**: While slightly imbalanced, it still contains a significant number of sarcastic sentences.  
✅ **Real-World Applications**: Useful for **social media analysis, sentiment analysis, and humor detection** in NLP.  

This dataset provides a **challenging yet valuable** benchmark for **sarcasm detection models** across different linguistic structures.