metadata

language:
  - dv
  - en
tags:
  - translation
  - summarization
  - dhivehi
license: mit
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: source
      dtype: string
    - name: en_translation
      dtype: string
    - name: en_summary_short
      dtype: string
    - name: en_summary_medium
      dtype: string
    - name: en_summary_detail
      dtype: string
    - name: dv_summary_short
      dtype: string
    - name: dv_summary_medium
      dtype: string
    - name: dv_summary_detail
      dtype: string
  splits:
    - name: train
      num_bytes: 376655894
      num_examples: 58738
  download_size: 158396140
  dataset_size: 376655894

Dataset Card for alakxender/dv-en-parallel-corpus

Dataset Description

This dataset contains Dhivehi news articles with their English translations and summaries at different lengths.

Languages

Dhivehi (dv)
English (en)

Dataset Structure

{
    "source": "string",            # Original article source
    "en_translation": "string",    # English translation
    "en_summary_short": "string",  # Short English summary
    "en_summary_medium": "string", # Medium English summary
    "en_summary_detail": "string", # Detailed English summary
    "dv_summary_short": "string",  # Short Dhivehi summary
    "dv_summary_medium": "string", # Medium Dhivehi summary
    "dv_summary_detail": "string"  # Detailed Dhivehi summary
}

Summary Types

Short Summary (20-50 words)

Best for: Executive overviews, abstracts, or capturing the essence of a text briefly.

Medium Summary (50-100 words)

Best for: Detailed summaries, reports, or providing a comprehensive overview of a text.

Detailed Summary (100-200 words)

Best for: In-depth analysis, detailed reports, or providing a comprehensive overview of a text.

Dataset Statistics

Number of examples: 58738

Dataset Creation

Source Data

The dataset consists of news articles collected from various Dhivehi news sources.

Preprocessing

Articles are translated to English
Multiple levels of summarization are generated for both languages, following the specified length guidelines
Each article includes both English and Dhivehi summaries at three different detail levels

Note

All summarization and translation tasks were performed using both fine-tuned and base models. Because the process was fully automated and unsupervised, AI-generated content can sometimes produce undesirable results.