metadata

license: apache-2.0

Konkani Figurative Language Corpus (Idioms + Metaphors) 📚

A dataset for idiom and metaphor classification in low-resource Konkani.

🚀 Overview

This dataset extends the Konidioms Corpus (Shaikh et al., 2024) by adding metaphor annotations. It supports binary classification for both idioms and metaphors in Konkani, a multi-script language spoken by approximately 2.5 million people.

📊 Dataset Description

Language: Konkani (Devanagari script)
Tasks:
- Idiom Detection (Yes/No)
- Metaphor Detection (Yes/No)
Format: CSV with the following columns:
- id: Sentence identifier
- sentence: The Konkani sentence
- idiom: Yes or No
- metaphor: Yes or No
- split: train or test
Size:
- 6,520 idiom-annotated sentences
- 500 metaphor-annotated sentences
Splits:
- ~80% training
- ~20% testing

🎯 Motivation

Konkani is a low-resource language with significant dialect and script variation. Figurative language, particularly metaphors, remains understudied. This dataset allows exploration of idioms and metaphors within a single corpus and supports efficient modeling efforts in underrepresented languages.

🧰 Baseline Model

From the paper: Pruning for Performance: Efficient Idiom and Metaphor Classification in Low-Resource Konkani Using mBERT

Model: mBERT + BiLSTM
Optimization: Gradient-based attention head pruning
Result: Comparable performance to full mBERT with fewer parameters

🧪 Example Use Cases

Figurative language analysis in Indic languages
Efficient multilingual transformer training
Evaluation of pruning strategies for model compression
Cross-lingual transfer learning in low-resource contexts

📚 Citation

If you use this dataset, please cite:

@misc{do2025pruningperformanceefficientidiom,
      title={Pruning for Performance: Efficient Idiom and Metaphor Classification in Low-Resource Konkani Using mBERT}, 
      author={Timothy Do and Pranav Saran and Harshita Poojary and Pranav Prabhu and Sean O'Brien and Vasu Sharma and Kevin Zhu},
      year={2025},
      eprint={2506.02005},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.02005}, 
}

📝 License

Released under the CC-BY 4.0 License.

⚠️ Limitations

Limited metaphor annotations (500 examples)
Only Devanagari script supported
Domain-specific language bias possible

🙌 Contributions

Contributions welcome! Please open an issue or pull request for improvements or additional data.