---
license: cc-by-4.0
language:
- hi
- en
tags:
- code-mixing
- Hinglish
- expert-annotated
size_categories:
- 1M<n<10M
configs:
  - config_name: LID
    data_files:
      - split: train
        path: "LID_train.csv"
      - split: test
        path: "LID_test.csv"  
  - config_name: POS
    data_files:
      - split: train
        path: "POS_train.csv"
      - split: test
        path: "POS_test.csv" 
  - config_name: NER
    data_files:
      - split: train
        path: "NER_train.csv"
      - split: test
        path: "NER_test.csv"
  - config_name: Translation
    data_files:
      - split: train
        path: "Translation_train.csv"
      - split: test
        path: "Translation_test.csv"
---

## Dataset Details

**COMI-LINGUA** (**CO**de-**MI**xing and **LING**uistic Insights on Natural Hinglish **U**sage and **A**nnotation) is a high-quality Hindi-English code-mixed dataset, manually annotated by three annotators. It serves as a benchmark for multilingual NLP models by covering multiple foundational tasks.

**COMI-LINGUA provides annotations for several key NLP tasks:**

* **Language Identification (LID):** Token-wise classification of Hindi, English, and other linguistic units.<br>
  Initial predictions were generated using the [Microsoft LID tool](https://github.com/microsoft/LID-tool), which annotators then reviewed and corrected.
* **Matrix Language Identification (MLI):** Sentence-level annotation of the dominant language.
* **Part-of-Speech (POS) Tagging:** Syntactic categorization for linguistic analysis.<br>
  Tags were pre-assigned using the [CodeSwitch NLP library](https://github.com/sagorbrur/codeswitch), which annotators then reviewed and corrected.
* **Named Entity Recognition (NER):** Identification of entities in Hinglish text.<br>
  Each token is pre-assigned a language tag using the [CodeSwitch NLP library](https://github.com/sagorbrur/codeswitch), which annotators then reviewed and corrected.
* **Translation:** Parallel translation of sentences in Romanized Hindi and Devanagari Hindi and English languages.<br>
  Initial translation predictions were generated using the [Llama 3.3 LLM](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) and refined by human annotators for accuracy.

### Dataset Description

- **Curated by:** [Lingo Research Group at IIT Gandhinagar, India](https://lingo.iitgn.ac.in/)
- **Funded by:** SERB
- **Language(s) (NLP):** Bilingual (Hindi [hi], English [en])
- **License:** cc-by-4.0