Syntax Role for Neural Semantic Role Labeling

Semantic role labeling (SRL) is dedicated to recognizing the semantic predicate-argument structure of a sentence. Previous studies in terms traditional models have shown syntactic information can make remarkable contributions SRL performance; however, necessity was challenged by few recent neural that demonstrate impressive performance without backbones and suggest syntax becomes much less important for labeling, especially when paired with deep network large-scale pre-trained language models. Despite this notion, field still lacks systematic full investigation on relevance SRL, both dependency monolingual multilingual settings. This paper intends quantify importance learning framework. We introduce three typical frameworks (baselines), sequence-based, tree-based, graph-based, which are accompanied two categories exploiting information: pruning-based feature-based. Experiments conducted CoNLL-2005, -2009, -2012 benchmarks all languages available, results show benefit from under certain conditions. Furthermore, we quantitative significance together thorough empirical survey using existing