Learning Constraints for Structured Prediction Using Rectifier Networks
This addresses the need for automated constraint learning in NLP to reduce reliance on domain expertise, though it is incremental as it builds on existing constraint-based methods.
The paper tackles the problem of learning constraints for structured prediction in NLP tasks, where domain knowledge is typically required, by training a two-layer rectifier network to identify valid structures and converting it into linear constraints, resulting in improved prediction accuracy, particularly with limited training data.
Various natural language processing tasks are structured prediction problems where outputs are constructed with multiple interdependent decisions. Past work has shown that domain knowledge, framed as constraints over the output space, can help improve predictive accuracy. However, designing good constraints often relies on domain expertise. In this paper, we study the problem of learning such constraints. We frame the problem as that of training a two-layer rectifier network to identify valid structures or substructures, and show a construction for converting a trained network into a system of linear constraints over the inference variables. Our experiments on several NLP tasks show that the learned constraints can improve the prediction accuracy, especially when the number of training examples is small.