README.md

metadata

license: apache-2.0
task_categories:
  - token-classification
language:
  - bg
pretty_name: Bulgarian Dictionary 2024
size_categories:
  - 1M<n<10M

Bulgarian Dictionary 2024

Dataset Summary

This is a dictionary of single-word Bulgarian tokens, tagged by the approproate part-of-speech tag.

Supported Tasks

• token-classification: The dataset can be used to train a model for token classification, which consists in applying a class to each token in a sequence.

Languages

• bg: Only Bulgarian is supported by this dataset.

Dataset Structure

Data Instances

Each instance contains a word, which is a single-worded token. Each of those also comes with a tag that is the part-of-speech` tag for that token which can be of 11 pre-defined classes.

...
{
  'word': 'кабел'
  'tag': 'N'
},
{
  'word': 'зелена'
  'tag': 'A'
},
{
  'word': 'над'
  'tag': 'R'
},
...

Data Fields

• token: a string sequence that's the single-worded token
• tag: a capitalised character that represent the part-of-speech tag, it can be one of 11 tags:
  1. N, referring to noun ('съществително име')
  2. A, referring to adjective ('прилагателно име')
  3. V, referring to verb ('глагол')
  4. D, referring to adverb ('наречие')
  5. P, referring to pronoun ('местоимение')
  6. T, referring to particle ('частица')
  7. M, referring to numeral ('числително име')
  8. C, referring to conjunction ('съюз')
  9. I, referring to interjection ('междуметие')
  10. R, referring to preposition ('предлог')
  11. H, referring to hybrid, or named entity ('лични имена, имена на държави, институции, и други имена')

The tag system is from the one used by the BulTreeBankTeam described in their[1] paper.

in the event of a word having multiple tags, it's recorded as two separate entries. For example the word 'син' can refer to a (1) noun, as the family relation of a son, or (2) adjective, the property of being blue. Such a word would have two rows:

• 'син', N
• 'син', A

Data Splits

No pre-defined split has been applied to the dataset so the developer has the freedom to choose one that suits the task.

Dataset Creation

Curation Rationale

Main motivations for the creating of this dataset:

• the official Bulgarian dictionary supplied by the Bulgarian Academy of Sciences is a great resource, however, it does not supply the user with a collection of valid tokens
• there is a lack of baseline models for rule-based approaches to POS-tagging, likely due to the lack of resources like this, this aims to encourage people to compare diverse models

Source Data

The source data for this dataset has been collected from:

• Rechko - The Bulgarian Dictionary
• Slovored
• Official Bulgarian Dicrionary from The Bulgarian Academy of Sciences

Initial Data Collection and Normalization

The data collection process was such:

1. Collect word data from dictionaries mentioned, along with their part-of-speech tag data.
2. Map tag data to pre-determined 11 tags.
3. Break up words into single tokens and apply tags to broken up tokens, for example:

• 'щеше да скачам' gets broken into 'щеше', 'да', 'скачам'
• 'щеше' and 'да' are already in the dicationary and are tagged as particles, and since the whole sequence is tagged as a verb by the source dictionary, the last token 'скачам' is also tagged as verb

4. Remove duplicates.

Personal and Sensitive Information

The source of this dataset is open-source data collections and carry the same amount of risk of personal and/or sensitive information as they do. However, the names and places which are tagged withe name tag refer to general names and are not associated with personal information of any particular individual or organisation.

Considerations for Using the Data

Social Impact of Dataset

A dataset like this can be beneficial for language learners and developers in the error correction POS-tagging community.

Discussion of Biases

Only the names that were already indexed in the source dictionaries have been used, so the biases in the collection of names and words in the source dictionaries carry over to this one as well.

Other Known Limitations

As language constantly evolves and changes many words, especially novel ones, may be missing from this collection.

Additional Information

Licensing Information

The license of the dataset is apache2.0.

[1]: Simov, K., Osenova, P. and Slavcheva, M., (2004). BTB-TR03: BulTree-Bank Morphosyntactic Tagset. BTB-TS version 2.0. Technical report, Bulgarian Academy of Sciences, Sofia, Bulgaria.