---
dataset_info:
  features:
  - name: Language
    dtype:
      class_label:
        names:
          '0': Arabic
          '1': Bulgarian
          '2': Chinese
          '3': Danish
          '4': Dutch
          '5': English
          '6': Estonian
          '7': French
          '8': German
          '9': Greek
          '10': Hindi
          '11': Indonesian
          '12': Italian
          '13': Japanese
          '14': Kannada
          '15': Korean
          '16': Latin
          '17': Malayalam
          '18': Modern Greek
          '19': Persian
          '20': Polish
          '21': Portugeese
          '22': Portugese
          '23': Portuguese
          '24': Pushto
          '25': Romanian
          '26': Russian
          '27': Spanish
          '28': Swahili
          '29': Swedish
          '30': Sweedish
          '31': Tamil
          '32': Thai
          '33': Turkish
          '34': Urdu
          '35': Vietnamese
  - name: Text
    dtype: large_string
  splits:
  - name: train
    num_bytes: 19720651.44222097
    num_examples: 83358
  - name: validation
    num_bytes: 4225752.489994374
    num_examples: 17862
  - name: test
    num_bytes: 4225989.067784654
    num_examples: 17863
  download_size: 20360216
  dataset_size: 28172393
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: mit
task_categories:
- text-classification
pretty_name: multilang-detect
---
# Dataset Card for Multilingual Language Detection

## Dataset Details


This dataset is a comprehensive resource for **multilingual text classification**, specifically designed for language identification. It contains over 100,000 text samples from 36 different languages, sourced from various public datasets and meticulously cleaned for machine learning applications.

The primary goal of this dataset is to train models that can accurately predict the language of a given text snippet. It is pre-split into training (70%), validation (15%), and test (15%) sets, with stratification on the `Language` column to ensure a balanced distribution of languages across all splits.

- **Language(s) (NLP):** Arabic, Bulgarian, Chinese, Danish, Dutch, English, Estonian, French, German, Greek, Hindi, Indonesian, Italian, Japanese, Kannada, Korean, Latin, Malayalam, Modern Greek, Persian, Polish, Portuguese, Pushto, Romanian, Russian, Spanish, Swahili, Swedish, Tamil, Thai, Turkish, Urdu, Vietnamese.
- **License:** `mit`

- **Repository:** [minhleduc/multilang-classify-dataset-02](https://huggingface.co/datasets/minhleduc/multilang-classify-dataset-02)

## Uses

### Direct Use

This dataset is ideal for:
- Training and fine-tuning language identification models.
- Benchmarking multilingual text classifiers.
- Research in multilingual representation learning.

### Out-of-Scope Use

This dataset is not suitable for:
- Evaluating the grammatical correctness or fluency of a text.
- Training models for tasks that require pristine, uncleaned text with original punctuation and casing.
- Making judgments about the author of the text. The data has been anonymized and processed, and may contain biases from its original sources.

## Dataset Structure

The dataset is provided in a `DatasetDict` containing `train`, `validation`, and `test` splits. Each sample has two fields:

- `Text` (`string`): The cleaned text sample.
- `Language` (`ClassLabel`): The language of the text, encoded as an integer. The mapping from integer to language name is stored in the feature metadata.

**Example:**
```json
{
  "Text": "this is an example of english text",
  "Language": English
}
```