README.md

---
license: mit
task_categories:
- text-classification
language:
- ne
tags:
- metaphor
- figurative
- nepali
- rupak
- simile
pretty_name: Rupak by Nabin
size_categories:
- 1K<n<10K
---
# Nepali Metaphor Detection Dataset

Here metaphor includes all kind of figurative speech that can be interpreted diferrently while reading literal and mean differently in meaning. The AarthaAlankaars like Simile,oxymoron, paradox, juxtaposition, personification, proverbs and idioms/phrases are included
as a metaphor and thus annotated as metaphor. The classification of these subtypes is not done in dataset.
---

## Dataset Card

### Dataset Summary
This dataset consists of Nepali-language sentences annotated with **literal** and **metaphorical** labels. It aims to foster research in figurative language understanding, metaphor detection, and linguistic creativity modeling for low-resource languages like Nepali.

- **Language**: Nepali
- **Task**: Sentence-level metaphor detection (binary classification)
- **Total Sentences**: [4502]
- **Total Metaphorical**: [1335]
- **Total Literal**: [3057]
- **Validation Set**: [110]
- **Labels**:
  - `0`: Literal
  - `1`: Metaphorical


### Motivation
Existing datasets for metaphor detection predominantly focus on English and other resource-rich languages. Nepali, being a morphologically rich, under-resourced language, lacks such curated resources. This dataset addresses this gap to:

- Enable fine-tuning and benchmarking of language models for figurative tasks.
- Encourage exploration of metaphorical nuances in Nepali culture and literature.


### Dataset Structure
Each sample contains:
- `Sentence`: Sentence in Nepali.
- `label`: 0 (Literal) or 1 (Metaphorical)

Example:
```json
{"text": "सूर्यले पर्वतपछाडि आफ्नो मुख लुकाए पनि तिनको लाली कालो पहाडका टुप्पाहरूमा फैलिँदै थियो। ", "label": 1}
```


### Data Splits
Recommended Split:
- **Training set**: 60%
- **Validation set**: 20%
- **Test set**: 20%

**Stratified sampling** ensures class balance across splits.


### Statistics Overview
- Class Distribution:
  - Literal: [3057,69.60%]
  - Metaphorical: [1335,30.39%]
- Avg. Sentence Length: [7.86] words
- Vocabulary Size: [12246] words
- Longest sentence: [37 words]
```json
{"text": "जसरी खोलामा बाढी आउनु भन्दा अघि खोलाछेउको आफ्नो घरको पर्खाललाई मजबुत बनाइयो भने बाढी आउने डरबाट सुरक्षित भइ व्यक्ति अविरल झरीको रातमा पनि आनन्दको निद्रा लिन्छ, त्यस्तै गरी भयानक दुखको परिस्थितिमा पनि यस्तो साधकले धैर्य नहारेको हुन्छ।"}
```
- Shortest sentence: [1 words]
  ```json
  {"text": "चन्द्रमुखी"}, {"text": "अर्धाङ्गिनी"}, {"text": "मृगनयनी"}
  ```


### Annotation Process
- Annotators: Nabin Bhusal,Manoj Bhandari and Sudesh Satyal.
              [Manually Annotated using webapp https://meta4app.web.app/]
- Guidelines: Based on literary, cultural, and linguistic cues in Nepali Grammar.
- Quality Control: Dual annotation with consensus resolution.


### Licensing
- MIT
- Attribution Required: Yes / No


### Citation
If you use this dataset, please cite:

```
@dataset{rupakbynabin,
  title={Nepali Sentence-level Metaphor Detection Dataset},
  author={Nabin Bhusal},
  year={2025},
  url={https://huggingface.co/datasets/bnabin/nepalimetaphorcorpus/}
}
```


### Ethical Considerations
- Ensure that sentences do not propagate bias or harmful stereotypes.
- Acknowledge limitations.
- Dataset is intended for research and educational purposes.


### Contact
- Maintainer: [Nabin Bhusal]
- Email: [[email protected]]

---