End-to-End Learning of Semantic Role Labeling using Structured Prediction with Neural Networks

This paper introduces a neural network model for semantic role labeling (SRL), a process of identifying the roles of words in a sentence with respect to a predicate. The model uses distributed word representations and structured prediction to learn SRL tasks directly from data, without handcrafted features or separate syntactic parsing. It demonstrates strong performance on the PropBank benchmark, showing the effectiveness of end-to-end learning for SRL.