alakxender/dv-synthetic-errors

DV Text Errors

Dhivehi text error correction dataset containing correct sentences and synthetically generated errors. The dataset aims to test Dhivehi language error correction models and tools.

About Dataset

• Task: Text error correction
• Language: Dhivehi (dv)

Dataset Structure

Input-output pairs of Dhivehi text:

• correct: Original correct sentences
• incorrect: Sentences with synthetic errors

Statistics

• Train set: {train_examples} examples ({0.7999997975429817}%)
• Test set: {test_examples} examples ({0.10000010122850919}%)
• Validation set: {val_examples} examples ({0.10000010122850919}%)

Details:

• Unique words: {448628}

{
  "total_examples": {
    "train": 3161164,
    "test": 395146,
    "validation": 395146
  },
  "avg_sentence_length": {
    "train": 11.968980097204701,
    "test": 11.961302910822836,
    "validation": 11.973824864733542
  },
  "error_distribution": {
    "min": 0,
    "max": 2411,
    "avg": 64.85144965588626
  }
}

Usage

from datasets import load_dataset

dataset = load_dataset("alakxender/dv-synthetic-errors")

Dataset Creation

Created using:

• Source: Collection of Dhivehi articles
• Error generation: Character and diacritic substitutions
• Error rate: 30% per word probability