evilfreelancer's picture
Update README.md
6d13e61 verified
metadata
dataset_info:
  features:
    - name: English
      dtype: string
    - name: Russian
      dtype: string
  splits:
    - name: train
      num_bytes: 372104
      num_examples: 1607
    - name: eval
      num_bytes: 19356
      num_examples: 100
  download_size: 206613
  dataset_size: 391460
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: eval
        path: data/eval-*
license: mit
task_categories:
  - translation
language:
  - ru
  - en
tags:
  - programming
  - tech
  - php
  - parallel-corpora
  - bert
size_categories:
  - 1K<n<10K

OPUS PHP (ru-en) Parallel Corpora

The dataset contains parallel corpora, featuring pairs in Russian and English.

Initially, the original OPUS PHP v1 (en&ru) dataset was intended to be used for training the enbeddrus project. However, due to its poor quality and high level of noise, it was decided to use only the English corpus from this dataset. The English text was then automatically translated using LibreTranslate, followed by manual translation and quality improvement using Google Translate.

Thus, the entire dataset was reviewed by the author of this project.

Subsequently, all lines shorter than 15 characters were removed, and lines such as "Example 1. function_name()" were manually deleted. Additionally, all duplicate lines, such as untranslated code samples, were removed. As a result, the original OPUS PHP v1 dataset was reduced from ~2k lines to ~1.6k lines.