Joint bi-affine parsing and semantic role labeling

摘要

We propose a simple encoder-decoder model for joint learning of dependency parsing and semantic role labeling (SRL). Experiments on CoNLL-2009 datasets show that our model is competitive with the state-of-the-art ensemble model on SRL task and significantly outperforms state-of-the-art joint models on joint evaluation metrics. Results show that with the implicit encoding, the syntax information can further improve a state-of-the-art semantic role labeler.