This repository implements the semantic role labeler described in the paper A Full End-to-End Semantic Role Labeler, Syntax-agnostic Over Syntax-aware?
The codes are developed based on the Dynet implementation of biaffine dependency parser.
python preprocess-conll09.py --train /path/to/train.dataset --test /path/to/test.dataset --dev /path/to/dev.dataset
or
python preprocess-conll08.py --train /path/to/train.dataset --test /path/to/test.dataset --dev /path/to/dev.dataset
We use embedding pre-trained by GloVe (Wikipedia 2014 + Gigaword 5, 6B tokens, 100d)
cd run
python train.py --config_file ../config.cfg [--dynet-gpu]
cd run
python test.py --config_file ../config.cfg [--dynet-gpu]
All configuration options (see in run/config.py) can be specified by the configuration file config.cfg.