en-ru-filtered-dict-20m-corpus

This dataset is a filtered version of the en-ru-statistical-dict-20m-corpus. While the original dataset was derived from a massive 20-million-sentence parallel corpus, this version focuses on the most reliable and frequent word pairs.

Dataset Description

The goal of this version is to provide a "noise-free" core dictionary for English-Russian translation tasks, specifically targeting the most common vocabulary found in modern corpora.

Filtering Criteria

To ensure high quality, the following strict filters were applied to the raw statistical data:

1. High Frequency: Only word pairs with a total count >= 1000 in the source corpus were kept.
2. Statistical Confidence: Only pairs with a probability >= 0.5 were included.

Structure

The dataset contains the following columns:

• english_word: The source word in English.
• russian_word: The corresponding translation in Russian.
• count: The number of times this pair appeared in the original 20m sentence corpus.
• probability: The statistical translation probability (0.0 to 1.0).

Usage

from datasets import load_dataset

dataset = load_dataset("KvaytG/en-ru-filtered-dict-20m-corpus", split="train")

License

This dataset is released under the Apache License 2.0.

Citation

@misc{kvaytg_en_ru_filtered_dict_20m_corpus,
  author       = {KvaytG},
  title        = {Filtered English-Russian Statistical Dictionary},
  year         = {2026},
  publisher    = {Hugging Face},
  journal      = {Hugging Face Datasets},
  url          = {https://huggingface.co/datasets/KvaytG/en-ru-filtered-dict-20m-corpus},
  note         = {Filtered version of en-ru-statistical-dict-20m-corpus (count ≥ 1000, prob ≥ 0.5)}
}