Joint Semantic Analysis with Document-Level Cross-Task Coherence Rewards

Coreference resolution and semantic role labeling are NLP tasks that capture different aspects of semantics, indicating respectively, which expressions refer to the same entity, what roles serve in sentence. However, they often closely interdependent, both generally necessitate natural language understanding. Do form a coherent abstract representation documents? We present neural network architecture for joint coreference English, train graph networks model 'coherence' combined shallow graph. Using resulting coherence score as reward our analyzer, we use reinforcement learning encourage global over document between annotations. This leads improvements on multiple datasets from domains, across range encoders expressivity, calling, believe, more holistic approach semantics NLP.