README.md

---
dataset_info:
  features:
  - name: word
    dtype: string
  - name: definition
    dtype: string
  - name: df
    dtype: string
  splits:
  - name: train
    num_bytes: 37600665
    num_examples: 295504
  download_size: 15206081
  dataset_size: 37600665
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: mit
language:
- ru
---
This dataset contains the Russian-language data I collected for training reverse dictionaries. The data consists of Russian words and their definitions. Each word-definition pair is also labeled with its source, of which there are three:

- 'efremova' (circa 211K) refers to the Efremova's New Explanatory-Morphological Dictionary (2000), which is an authoritative Russian dictionary that was chosen for its lack of examples (allowing for easier data collection) and the large amount of words represented (circa 140K);
- 'codwoe' (circa 50K) refers to the dataset created by the organizers of the CODWOE (COmparing Definitions and WOrd Embeddings) track of SemEval-2022, available here: https://codwoe.atilf.fr/. This part of the dataset only contains definitions for nouns, verbs, adjectives and adverbs. Notably, the original dataset also contains (usually several) examples of use for every word; I have not retained them here, but if you need examples of use in your training (for instance to generate embeddings) they are freely available there;
- 'absite' (circa 35K) refers to absite.com, a Russian-language crossword website, from where I scraped words and clues for them. Unlike the other parts of the dataset, 'absite' contains only definitions for nouns; but since the definitions here are crossword clues and not dictionary definitions, they are written in a more everyday style of Russian, which corresponds to how a hypothetical user of a reverse dictionary would likely phrase their queries.

There are circa 296K datapoints in total.

Note: this dataset is _not_ filtered from the dictionary definitions of words in the test data that I collected (available here: https://huggingface.co/datasets/snagbreac/russian-reverse-dictionary-test-data). This allows you to work with the full volume of data I collected; however, use of the test data may be ill-advised, as some of it is contained in the training dataset. The filtered dataset is available here: https://huggingface.co/datasets/snagbreac/russian-reverse-dictionary-train-data.

I sincerely hope that someone finds this dataset useful for training reverse dictionaries, both Russian-language and multilingual.