This repository provides the code for the work in NAACL2019: Event Detection without Triggers.

Because of the copyright issue of ACE2005 Corpus, we can not release the corpus. For test, we give 10 samples in data/test_corpus_10.txt.

Each line represents a testing sample, whose format is as follows:

w1 e1 \t w2 e2 \t ... ... wn en \t evt1 evt2 ... evtm

where, [w1, w2, ..., wn] are tokens of a testing sentence, [e1, e2, ..., en] are the corresponding entity type of each token, [evt1, evt2, ..., evtm] are the types of events mentioned in this sentence (if m is 0, this block will be replaced with a single 'NEGATIVE' label).

We provide a trained model, which can be downloaded here: model files.

You can run this code to evaluate the trained model using the following command:

python run_model.py evaluation 

or train the model using your own traininng corpus:

python run_model.py train 

Required running environment:
1. python 2.7
2. tensorflow 1.4 or higher