dbamman / NAACL2019-literary-entities

Code to support Bamman et al. (2019), "An Annotated Dataset of Literary Entities" (NAACL)

NAACL2019-literary-entities

Code to support Bamman et al. (2019), "An Annotated Dataset of Literary Entities" (NAACL 2019).

• litbank is a snapshot of the master LitBank repo used for these experiments.
• layered-bilstm-crf contains a clone of the Github repo for the nested NER model of Ju et al. 2018. train.py and test.py are both updated to take a command-line argument specifying a configuration file; and test.py now also writes predictions to file (specified in the configuration as predictions_path).
• ACEReader wraps code from Stanford CoreNLP for processing the XML files of the ACE 2005 data (LDC2006T06), including tokenization and sentence splitting.
• 80/10/10 train, dev and test splits (by document) for both ACE 2005 and LitBank can be found in data/ace/{train,dev,test}.ids and data/litbank/{train,dev,test}.ids, respectively.

pipeline

This pipeline requires access to ACE 2005. Download from LDC and specify the path in ACE2005_PATH below.

ACE2005_PATH=/path/to/LDC2006T06
cd scripts
./create_literary_data.sh
./create_ace_data.sh $ACE2005_PATH
./train_ner.sh
./test_ner.sh
./evaluate_gender.sh
./create_figures.sh