The goal of semantic role labeling (SRL) is to discover the predicate-argument structure of a sentence, which plays a critical role in deep processing of natural language. This paper introduces simple yet effective auxiliary tags for dependency-based SRL to enhance a syntax-agnostic model with multi-hop self-attention. Our syntax-agnostic model achieves competitive performance with state-of-the-art models on the CoNLL-2009 benchmarks both for English and Chinese.

Attentive Semantic Role Labeling with Boundary Indicator