Paper: Structured Sparsity in Natural Language Processing: Models, Algorithms and Applications

ACL ID N12-4002
Title Structured Sparsity in Natural Language Processing: Models, Algorithms and Applications
Venue Annual Conference of the North American Chapter of the Association for Computational Linguistics
Session Tutorial Abstracts
Year 2012
Authors

This tutorial will cover recent advances in sparse modeling with diverse applications in natural language processing (NLP). A sparse model is one that uses a relatively small number of features to map an input to an output, such as a label sequence or parse tree. The advantages of sparsity are, among others, compactness and interpretability; in fact, sparsity is currently a major theme in statistics, machine learning, and signal processing. The goal of sparsity can be seen in terms of earlier goals of feature selection and therefore model selection (Della Pietra et al., 1997; Guyon and Elisseeff, 2003; McCallum, 2003). This tutorial will focus on methods which embed sparse model selection into the parameter estimation problem. In such methods, learning is carried out by minimizing...