alexbrandsen/archaeo_ner_dutch

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",
}