@inproceedings{lopes-etal-2019-contributions,
    title = "Contributions to Clinical Named Entity Recognition in {P}ortuguese",
    author = "Lopes, F{\'a}bio  and
      Teixeira, C{\'e}sar  and
      Gon{\c{c}}alo Oliveira, Hugo",
    editor = "Demner-Fushman, Dina  and
      Cohen, Kevin Bretonnel  and
      Ananiadou, Sophia  and
      Tsujii, Junichi",
    booktitle = "Proceedings of the 18th BioNLP Workshop and Shared Task",
    month = aug,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/jlcl-multiple-ingestion/W19-5024/",
    doi = "10.18653/v1/W19-5024",
    pages = "223--233",
    abstract = "Having in mind that different languages might present different challenges, this paper presents the following contributions to the area of Information Extraction from clinical text, targeting the Portuguese language: a collection of 281 clinical texts in this language, with manually-annotated named entities; word embeddings trained in a larger collection of similar texts; results of using BiLSTM-CRF neural networks for named entity recognition on the annotated collection, including a comparison of using in-domain or out-of-domain word embeddings in this task. Although learned with much less data, performance is higher when using in-domain embeddings. When tested in 20 independent clinical texts, this model achieved better results than a model using larger out-of-domain embeddings."
}