Neural Unsupervised Semantic Role Labeling

The task of semantic role labeling ( SRL ) is dedicated to finding the predicate-argument structure. Previous works on are mostly supervised and do not consider difficulty in each example which can be very expensive time-consuming. In this article, we present first neural unsupervised model for SRL. To decompose as two argument related subtasks, identification clustering, propose a pipeline that correspondingly consists modules. First, train syntax-aware statistically developed rules. gets relevance signal token sentence, feed into BiLSTM, then an adversarial layer noise-adding classifying simultaneously, thus enabling learn structure sentence. Then another done through clustering learned embeddings biased toward their dependency relations. Experiments CoNLL-2009 English dataset demonstrate our outperforms previous state-of-the-art baseline terms non-neural models classification.