Hugging Face's logo

Datasets:
nectec
/
best2009

Tasks:
Token Classification
Modalities:
Text
Formats:
parquet
Languages:
Thai
Size:
100K - 1M
Tags:
word-tokenization
Libraries:
Datasets
pandas
License:
Dataset card Viewer Files Community
4
best2009 / README.md
julien-c's picture
julien-c HF staff
Align more metadata with other repo types (models,spaces) (#4607)
38c1d5f
preview code
|
raw
history blame
8.03 kB
metadata
annotations_creators:
  - expert-generated
language_creators:
  - found
language:
  - th
license:
  - cc-by-nc-sa-3.0
multilinguality:
  - monolingual
size_categories:
  - 100K<n<1M
source_datasets:
  - original
task_categories:
  - token-classification
task_ids:
  - token-classification-other-word-tokenization
paperswithcode_id: null
pretty_name: best2009

Dataset Card for best2009

Table of Contents

Dataset Description

Dataset Summary

best2009 is a Thai word-tokenization dataset from encyclopedia, novels, news and articles by NECTEC (148,995/2,252 lines of train/test). It was created for BEST 2010: Word Tokenization Competition. The test set answers are not provided publicly.

Supported Tasks and Leaderboards

word tokenization

Languages

Thai

Dataset Structure

Data Instances 

{'char': ['?', 'ภ', 'ู', 'ม', 'ิ', 'ป', 'ั', 'ญ', 'ญ', 'า', 'ช', 'า', 'ว', 'บ', '้', 'า', 'น', '\n'], 'char_type': [4, 1, 10, 1, 10, 1, 4, 1, 1, 10, 1, 10, 1, 1, 9, 10, 1, 4], 'fname': 'encyclopedia_00031.txt', 'is_beginning': [1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1]}
{'char': ['ภ', 'ู', 'ม', 'ิ', 'ป', 'ั', 'ญ', 'ญ', 'า', 'ช', 'า', 'ว', 'บ', '้', 'า', 'น', ' ', 'ห', 'ม', 'า', 'ย', 'ถ', 'ึ', 'ง', ' ', 'ค', 'ว', 'า', 'ม', 'ร', 'ู', '้', 'ข', 'อ', 'ง', 'ช', 'า', 'ว', 'บ', '้', 'า', 'น', ' ', 'ซ', 'ึ', '่', 'ง', 'เ', 'ร', 'ี', 'ย', 'น', 'ร', 'ู', '้', 'ม', 'า', 'จ', 'า', 'ก', 'พ', '่', 'อ', 'แ', 'ม', '่', ' ', 'ป', 'ู', '่', 'ย', '่', 'า', 'ต', 'า', 'ย', 'า', 'ย', ' ', 'ญ', 'า', 'ต', 'ิ', 'พ', 'ี', '่', 'น', '้', 'อ', 'ง', ' ', 'ห', 'ร', 'ื', 'อ', 'ผ', 'ู', '้', 'ม', 'ี', 'ค', 'ว', 'า', 'ม', 'ร', 'ู', '้', 'ใ', 'น', 'ห', 'ม', 'ู', '่', 'บ', '้', 'า', 'น', 'ใ', 'น', 'ท', '้', 'อ', 'ง', 'ถ', 'ิ', '่', 'น', 'ต', '่', 'า', 'ง', 'ๆ', '\n'], 'char_type': [1, 10, 1, 10, 1, 4, 1, 1, 10, 1, 10, 1, 1, 9, 10, 1, 5, 3, 1, 10, 1, 1, 10, 1, 5, 1, 1, 10, 1, 1, 10, 9, 1, 1, 1, 1, 10, 1, 1, 9, 10, 1, 5, 1, 10, 9, 1, 11, 1, 10, 1, 1, 1, 10, 9, 1, 10, 1, 10, 1, 1, 9, 1, 11, 1, 9, 5, 1, 10, 9, 1, 9, 10, 1, 10, 1, 10, 1, 5, 1, 10, 1, 10, 1, 10, 9, 1, 9, 1, 1, 5, 3, 1, 10, 1, 3, 10, 9, 1, 10, 1, 1, 10, 1, 1, 10, 9, 11, 1, 3, 1, 10, 9, 1, 9, 10, 1, 11, 1, 1, 9, 1, 1, 1, 10, 9, 1, 1, 9, 10, 1, 7, 4], 'fname': 'encyclopedia_00031.txt', 'is_beginning': [1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1]}

Data Fields

  • fname: file name; also marks if article is articles, news, encyclopedia or novels
  • char: characters
  • char_type: character types as adopted from by deepcut
  • is_beginning: is beginning of word

Data Splits

train test
# lines 148,995 2,252
avg words per line 39.05 NA
total words 5,818,521 NA
avg characters per line 140.39 202.79
total characters 20,918,132 456,684
# lines articles 16,990 NA
# lines encyclopedia 50,631 NA
# lines novels 50,140 NA
# lines news 31,234 NA

Dataset Creation

Curation Rationale

The dataset was created for BEST 2010: Word Tokenization Competition by NECTEC.

Source Data

Initial Data Collection and Normalization

[More Information Needed]

Who are the source language producers?

Respective authors of the articles, news, encyclopedia and novels

Annotations

Annotation process

Detailed annotation guidelines can be found in BEST_Guideline_Release1.pdf as part of the uncompressed files. Word tokenization standard used was InterBEST2009

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

All data are curated from public sources. No personal and sensitive information is expected to be included.

Considerations for Using the Data

Social Impact of Dataset

  • word tokenization dataset from articles, news, encyclopedia and novels

Discussion of Biases

  • texts are relatively formal ones from articles, news, encyclopedia and novels.
  • word tokenization standard used was InterBEST2009.

Other Known Limitations

  • some tags unrelated to word tokenization (<NE> and <AB>) are cleaned out.
  • no word boundary provdied for the test set

Additional Information

Dataset Curators

NECTEC

Licensing Information

CC-BY-NC-SA 3.0

Citation Information

Dataset:

@inproceedings{kosawat2009best,
  title={BEST 2009: Thai word segmentation software contest},
  author={Kosawat, Krit and Boriboon, Monthika and Chootrakool, Patcharika and Chotimongkol, Ananlada and Klaithin, Supon and Kongyoung, Sarawoot and Kriengket, Kanyanut and Phaholphinyo, Sitthaa and Purodakananda, Sumonmas and Thanakulwarapas, Tipraporn and others},
  booktitle={2009 Eighth International Symposium on Natural Language Processing},
  pages={83--88},
  year={2009},
  organization={IEEE}
}
@inproceedings{boriboon2009best,
  title={Best corpus development and analysis},
  author={Boriboon, Monthika and Kriengket, Kanyanut and Chootrakool, Patcharika and Phaholphinyo, Sitthaa and Purodakananda, Sumonmas and Thanakulwarapas, Tipraporn and Kosawat, Krit},
  booktitle={2009 International Conference on Asian Language Processing},
  pages={322--327},
  year={2009},
  organization={IEEE}
}

Character type features:

@inproceedings{haruechaiyasak2009tlex,
  title={TLex: Thai lexeme analyser based on the conditional random fields},
  author={Haruechaiyasak, Choochart and Kongyoung, Sarawoot},
  booktitle={Proceedings of 8th International Symposium on Natural Language Processing},
  year={2009}
}

Contributions

Thanks to @cstorm125 for adding this dataset.