This repository hosts code for experiments on multilingual event detection mainly benchmarking the tasks of Trigger and Argument Identification/Classification in ACE 2005 dataset.

The list of required packages are under requirements.txt. To install them all, run this command:

 pip install -r requirements.txt

In addition to that, preprocessing requires Stanford Core NLP Parser, for that you will need to download CoreNLP, add it to your classpath and start StanfordCoreNLPServer using the command:

 java -mx4g -cp "*" edu.stanford.nlp.pipeline.StanfordCoreNLPServer -port 9001 -timeout 150000000

Under DataModule, run

python data_preprocessor.py 

which is adapted to work primarily on ACE05 and split it into train, dev and test using the doc_split and generates the data in BIO annotation scheme. The full data cannot be included as it is under license. Be sure, to change get_args.py script to comply with the desired options for processing:

• languages: choose the language that you would like to focus on separated by comma
• root-dir: change the root directory of the dataset
• data-ace-path: change the release of the data
• method: jmee for data like the following or tagging to produce BIO annotation
• pre-dir: output directory for saving the preprocessed files

Now that you have the dataset in BIO Annotation scheme for which examples are provided in DatasModule/sample/, you can train and evaluate either monolingually or multilingually by running:

 python train.py