Datasets:

DFKI-SLT
/

copious

language:
  - en
license: cc-by-4.0
size_categories:
  - 10K<n<100K
tasks:
  - named-entity-recognition
  - token-classification
pretty_name: COPIOUS
tags:
  - named-entity-recognition
  - biodiversity
  - species-occurrence
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.jsonl
      - split: validation
        path: dev.jsonl
      - split: test
        path: test.jsonl

Dataset Card for COPIOUS

Dataset Description
Dataset Structure
Dataset Creation
Considerations for Using the Data
- Social Impact of Dataset
- Other Known Limitations
Additional Information

Dataset Description

Homepage: COPIOUS: A gold standard corpus of named entities... (Biodiversity Data Journal)
Paper: COPIOUS: A gold standard corpus of named entities towards extracting species occurrence from biodiversity literature
Point of Contact: Sophia Ananiadou

Dataset Summary

Species occurrence records are critical in the biodiversity domain. However, a key challenge has been the lack of a single, consolidated corpus that includes all the entity types needed to extract this information from scientific literature. To address this gap, the COPIOUS corpus (Corpus of named entities towards extracting Species Occurrence) was created.

This dataset serves as a gold standard, annotated with a wide range of entities relevant to biodiversity science. It was developed as part of the broader COPIOUS project, which aims to build a knowledge repository of Philippine biodiversity by applying text mining to scientific texts. The corpus is manually annotated with five key entity categories: Taxon, Geographical Location, Habitat, Temporal Expression, and Person names.

Supported Tasks and Leaderboards

Tasks: Named Entity Recognition, Token Classification
Leaderboards: N/A

Languages

The text in the dataset is in English.

Dataset Structure

Data Instances

An example from the train split:

{
  "id": "100785_English_120409_32367238_1954",
  "tokens": ["FAMILY", "SERRANIDAE", "—", "SCHULTZ", "..."],
  "ner_tags": [0, 1, 0, 0, ...]
}

Data Fields

id: A unique identifier for the example (string).
tokens: A list of tokens (list of strings).
ner_tags: A list of integer IDs corresponding to the NER tags. The mapping is:

{
  "0": "O",
  "1": "B-Taxon",
  "2": "I-Taxon",
  "3": "B-Geographical_Location",
  "4": "I-Geographical_Location",
  "5": "B-Habitat",
  "6": "I-Habitat",
  "7": "B-Temporal_Expression",
  "8": "I-Temporal_Expression",
  "9": "B-Person",
  "10": "I-Person"
}

Data Splits

The corpus consists of 668 documents split as follows:

Split	Documents
train	534
validation	67
test	67

Dataset Creation

Curation Rationale

The corpus was created to be a sufficiently reliable and sizeable resource for training and evaluating Named Entity Recognition (NER) tools in the biodiversity domain, specifically for the task of species occurrence extraction.

Source Data

The data consists of 668 documents randomly selected from over 169,000 English-language pages downloaded from the Biodiversity Heritage Library (BHL).

Annotations

The dataset was manually annotated by two experts in biology using the Argo workbench. 200 documents were double-annotated to ensure quality, achieving an overall Inter-Annotator Agreement (IAA) of 81.86% F-score.

Format Conversion

For this Hugging Face Hub version, the dataset was prepared as follows:

Local Sourcing: The data was sourced from a local copy because the original download link on the NaCTeM website was inactive.
Conversion to JSONL: The original data was annotated in the Brat Standoff Format (.txt + .ann files). It has been converted to the standard JSON Lines (.jsonl) format to align with Hugging Face Hub standards.

Considerations for Using the Data

Social Impact of Dataset

This corpus can be used to develop text mining tools that automatically locate species occurrences in literature, which is useful for monitoring species distribution and preserving biodiversity.

Other Known Limitations

Annotation Complexity: The IAA for the Habitat entity type was the lowest (47.07% F-score), indicating this category is complex and challenging for both human annotators and automated systems.
Source Text Quality: The source documents from BHL contain a large number of OCR errors, which may adversely affect the performance of NER models trained on this data.

Additional Information

Dataset Curators

Nhung T.H. Nguyen, Roselyn S. Gabud, and Sophia Ananiadou.

Licensing Information

The dataset is distributed under the Creative Commons Attribution License (CC BY 4.0).

Citation Information

@article{nguyen2019copious,
  author = {Nguyen, Nhung T.H. and Gabud, Roselyn S. and Ananiadou, Sophia},
  title = {{COPIOUS: A gold standard corpus of named entities towards extracting species occurrence from biodiversity literature}},
  journal = {Biodiversity Data Journal},
  volume = {7},
  pages = {e29626},
  year = {2019},
  doi = {10.3897/BDJ.7.e29626}
}