alexbrandsen/archaeo_ner_dutch

---
language:
- nl
license: other
task_categories:
- token-classification
pretty_name: Dutch Archaeology NER Dataset
license_name: hippocratic-license-3.0
license_link: https://firstdonoharm.dev/version/3/0/full.md
dataset_info:
features:
- name: tokens
sequence: string
- name: ner_tags
sequence:
class_label:
names:
'0': O
'1': B-ART
'2': I-ART
'3': B-CON
'4': I-CON
'5': B-LOC
'6': I-LOC
'7': B-MAT
'8': I-MAT
'9': B-PER
'10': I-PER
'11': B-SPE
'12': I-SPE
splits:
- name: fold1_train
num_bytes: 4490700
num_examples: 22150
- name: fold1_validation
num_bytes: 1579488
num_examples: 5852
- name: fold1_test
num_bytes: 1574291
num_examples: 5750
- name: fold2_train
num_bytes: 4685070
num_examples: 22465
- name: fold2_validation
num_bytes: 1379777
num_examples: 5431
- name: fold2_test
num_bytes: 1579700
num_examples: 5865
- name: fold3_train
num_bytes: 4762905
num_examples: 19560
- name: fold3_validation
num_bytes: 1501653
num_examples: 8757
- name: fold3_test
num_bytes: 1379769
num_examples: 5427
- name: fold4_train
num_bytes: 4533412
num_examples: 17029
- name: fold4_validation
num_bytes: 1609278
num_examples: 7963
- name: fold4_test
num_bytes: 1501649
num_examples: 8755
- name: fold5_train
num_bytes: 4460910
num_examples: 20039
- name: fold5_validation
num_bytes: 1574155
num_examples: 5747
- name: fold5_test
num_bytes: 1609342
num_examples: 7965
download_size: 7478347
dataset_size: 38222099
configs:
- config_name: default
data_files:
- split: fold1_train
path: data/fold1_train-*
- split: fold1_validation
path: data/fold1_validation-*
- split: fold1_test
path: data/fold1_test-*
- split: fold2_train
path: data/fold2_train-*
- split: fold2_validation
path: data/fold2_validation-*
- split: fold2_test
path: data/fold2_test-*
- split: fold3_train
path: data/fold3_train-*
- split: fold3_validation
path: data/fold3_validation-*
- split: fold3_test
path: data/fold3_test-*
- split: fold4_train
path: data/fold4_train-*
- split: fold4_validation
path: data/fold4_validation-*
- split: fold4_test
path: data/fold4_test-*
- split: fold5_train
path: data/fold5_train-*
- split: fold5_validation
path: data/fold5_validation-*
- split: fold5_test
path: data/fold5_test-*
tags:
- archaeology
---

# Dutch Archaeology NER Dataset

A selection of Dutch archaeology field reports, annotated by archaeology students from Leiden University.

## Labels

The following labels are included:

- ART, artefacts ('bijl', 'pijlpunt')
- MAT, materials ('vuursteen', 'ijzer')
- PER, time periods ('Middeleeuwen', '400 v. Chr.')
- CON, archaeological contexts ('greppel','beerput')
- LOC, locations ('Amsterdam', 'Oss')
- SPE, species ('Betula nana', 'koe')

## Folds

The reason I supply 5 folds is because I get wildly different F1 scores between folds, and because it's important to keep whole documents in folds: these are long documents, any document that's split between train and test instantly leads to a higher F1, as the model starts recognising specific tokens as entities, leading to overfitting. A micro average F1 over 5 folds with no split documents seems like the fairest evaluation, closest to real-world inference.

### Citation Information

```
@inproceedings{brandsen-etal-2020-creating,
title = "Creating a Dataset for Named Entity Recognition in the Archaeology Domain",
author = "Brandsen, Alex and
Verberne, Suzan and
Wansleeben, Milco and
Lambers, Karsten",
editor = "Calzolari, Nicoletta and
B{\'e}chet, Fr{\'e}d{\'e}ric and
Blache, Philippe and
Choukri, Khalid and
Cieri, Christopher and
Declerck, Thierry and
Goggi, Sara and
Isahara, Hitoshi and
Maegaard, Bente and
Mariani, Joseph and
Mazo, H{\'e}l{\`e}ne and
Moreno, Asuncion and
Odijk, Jan and
Piperidis, Stelios",
booktitle = "Proceedings of the Twelfth Language Resources and Evaluation Conference",
month = may,
year = "2020",
address = "Marseille, France",
publisher = "European Language Resources Association",
url = "https://aclanthology.org/2020.lrec-1.562",
pages = "4573--4577",
abstract = "In this paper, we present the development of a training dataset for Dutch Named Entity Recognition (NER) in the archaeology domain. This dataset was created as there is a dire need for semantic search within archaeology, in order to allow archaeologists to find structured information in collections of Dutch excavation reports, currently totalling around 60,000 (658 million words) and growing rapidly. To guide this search task, NER is needed. We created rigorous annotation guidelines in an iterative process, then instructed five archaeology students to annotate a number of documents. The resulting dataset contains {\textasciitilde}31k annotations between six entity types (artefact, time period, place, context, species {\&} material). The inter-annotator agreement is 0.95, and when we used this data for machine learning, we observed an increase in F1 score from 0.51 to 0.70 in comparison to a machine learning model trained on a dataset created in prior work. This indicates that the data is of high quality, and can confidently be used to train NER classifiers.",
language = "English",
ISBN = "979-10-95546-34-4",
}

```

原始信息汇总

Dutch Archaeology NER Dataset

概述

• 语言: 荷兰语
• 许可: 其他（Hippocratic License 3.0）
• 任务类别: 词性标注
• 数据集名称: Dutch Archaeology NER Dataset

数据集信息

• 特征:

  • tokens: 字符串序列
  • ner_tags: 序列标签，包含以下类别:
    • 0: O
    • 1: B-ART
    • 2: I-ART
    • 3: B-CON
    • 4: I-CON
    • 5: B-LOC
    • 6: I-LOC
    • 7: B-MAT
    • 8: I-MAT
    • 9: B-PER
    • 10: I-PER
    • 11: B-SPE
    • 12: I-SPE

• 数据分割:

  • fold1_train: 22150个样本，4490700字节
  • fold1_validation: 5852个样本，1579488字节
  • fold1_test: 5750个样本，1574291字节
  • fold2_train: 22465个样本，4685070字节
  • fold2_validation: 5431个样本，1379777字节
  • fold2_test: 5865个样本，1579700字节
  • fold3_train: 19560个样本，4762905字节
  • fold3_validation: 8757个样本，1501653字节
  • fold3_test: 5427个样本，1379769字节
  • fold4_train: 17029个样本，4533412字节
  • fold4_validation: 7963个样本，1609278字节
  • fold4_test: 8755个样本，1501649字节
  • fold5_train: 20039个样本，4460910字节
  • fold5_validation: 5747个样本，1574155字节
  • fold5_test: 7965个样本，1609342字节

• 数据集大小:

  • 下载大小: 7478347字节
  • 总大小: 38222099字节

配置

• 默认配置:
  • 数据文件路径:
    • fold1_train: data/fold1_train-*
    • fold1_validation: data/fold1_validation-*
    • fold1_test: data/fold1_test-*
    • fold2_train: data/fold2_train-*
    • fold2_validation: data/fold2_validation-*
    • fold2_test: data/fold2_test-*
    • fold3_train: data/fold3_train-*
    • fold3_validation: data/fold3_validation-*
    • fold3_test: data/fold3_test-*
    • fold4_train: data/fold4_train-*
    • fold4_validation: data/fold4_validation-*
    • fold4_test: data/fold4_test-*
    • fold5_train: data/fold5_train-*
    • fold5_validation: data/fold5_validation-*
    • fold5_test: data/fold5_test-*

标签

• ART: 文物（如 bijl, pijlpunt）
• MAT: 材料（如 vuursteen, ijzer）
• PER: 时间时期（如 Middeleeuwen, 400 v. Chr.）
• CON: 考古背景（如 greppel, beerput）
• LOC: 地点（如 Amsterdam, Oss）
• SPE: 物种（如 Betula nana, koe）

引用信息

@inproceedings{brandsen-etal-2020-creating, title = "Creating a Dataset for Named Entity Recognition in the Archaeology Domain", author = "Brandsen, Alex and Verberne, Suzan and Wansleeben, Milco and Lambers, Karsten", editor = "Calzolari, Nicoletta and B{e}chet, Fr{e}d{e}ric and Blache, Philippe and Choukri, Khalid and Cieri, Christopher and Declerck, Thierry and Goggi, Sara and Isahara, Hitoshi and Maegaard, Bente and Mariani, Joseph and Mazo, H{e}l{`e}ne and Moreno, Asuncion and Odijk, Jan and Piperidis, Stelios", booktitle = "Proceedings of the Twelfth Language Resources and Evaluation Conference", month = may, year = "2020", address = "Marseille, France", publisher = "European Language Resources Association", url = "https://aclanthology.org/2020.lrec-1.562", pages = "4573--4577", abstract = "In this paper, we present the development of a training dataset for Dutch Named Entity Recognition (NER) in the archaeology domain. This dataset was created as there is a dire need for semantic search within archaeology, in order to allow archaeologists to find structured information in collections of Dutch excavation reports, currently totalling around 60,000 (658 million words) and growing rapidly. To guide this search task, NER is needed. We created rigorous annotation guidelines in an iterative process, then instructed five archaeology students to annotate a number of documents. The resulting dataset contains { extasciitilde}31k annotations between six entity types (artefact, time period, place, context, species {&} material). The inter-annotator agreement is 0.95, and when we used this data for machine learning, we observed an increase in F1 score from 0.51 to 0.70 in comparison to a machine learning model trained on a dataset created in prior work. This indicates that the data is of high quality, and can confidently be used to train NER classifiers.", language = "English", ISBN = "979-10-95546-34-4", }