ArmTDP-NER / README.md
Yeva's picture
Update README.md
6fedc1d verified
|
Raw
History Blame Contribute Delete
3.54 kB
metadata
license: apache-2.0
task_categories:
  - token-classification
language:
  - hy
tags:
  - named-entity-recognition
pretty_name: ArmTDP-NER
dataset_info:
  homepage: https://github.com/myavrum/ArmTDP-NER
size_categories:
  - n<1K

Dataset Summary

ArmTDP-NER is a manually annotated named entity corpus for Modern Eastern Armenian. It contains 900 newspaper articles across various domains with 17,632 entity annotations. The annotation follows a modified OntoNotes 5.0 markup scheme adapted for Eastern Armenian-specific considerations, covering 18 named entity types.

Dataset Structure

Data Fields

The corpus is distributed as newline-delimited JSON (JSONL). Each line is one document with character-offset span annotations.

Field Type Description
id int Document identifier
text string Raw document text
labels list of [start, end, label] Character-offset spans with string label names
annotation_approver string Annotator identifier

Annotation Scheme

The corpus uses 18 named entity types derived from the OntoNotes 5.0 scheme:

Entity Type Count Description
PERSON 3,502 People, including fictional characters
ORGANIZATION 3,388 Companies, agencies, institutions
GPE 3,290 Countries, cities, states
DATE 1,995 Absolute or relative dates/periods
CARDINAL 1,145 Numerals not fitting other categories
NORP 823 Nationalities, religious or political groups
ORDINAL 608 Ordinal numbers
FACILITY 601 Buildings, airports, highways, bridges
LAW 466 Named documents made into laws
TIME 405 Times smaller than a day
QUANTITY 343 Measurements (weight, distance, etc.)
MONEY 267 Monetary values, including units
LOCATION 166 Non-GPE locations, water bodies, mountain ranges
WORK OF ART 156 Titles of books, songs, etc.
PERCENT 112 Percentages, including the % symbol
PRODUCT 107 Vehicles, weapons, foods, etc.
EVENT 211 Named hurricanes, battles, wars, sports events
LANGUAGE 47 Named languages

Total entity annotations: 17,632

Dataset Statistics

Metric Value
Documents 900
Entity types 18
Total annotations 17,632

Dataset Creation

Source Data

Newspaper articles in Modern Eastern Armenian sourced from various domains.

Annotations

Annotations were produced manually by the ArmTDP team. The annotation scheme adapts the OntoNotes 5.0 standard to account for Eastern Armenian-specific linguistic considerations.

Additional Information

Dataset Curators

Marat M. Yavrumyan and the ArmTDP team at Yerevan State University, in collaboration with Armenia National SDG Innovation Lab and UC Berkeley's Armenian Linguists network.

Contact: marat.yavrumyan@ysu.am

Licensing Information

Apache License 2.0

Citation Information

@misc{ArmTDP-NER,
  title        = {{ArmTDP-NER}: Eastern Armenian Named Entity Corpus, 1.0},
  author       = {Marat Yavrumyan},
  year         = {2020},
  doi          = {https://doi.org/10.5281/zenodo.17999570}
  url          = {https://github.com/myavrum/ArmTDP-NER},
  note         = {Corpus available from https://github.com/myavrum/ArmTDP-NER}
}

Related Resources

A pre-trained NER model for Eastern Armenian trained on this corpus is available through the Stanza framework.