Diversity in Spectral Learning for Natural Language Parsing

TL;DR

This paper introduces a method to generate diverse spectral models for natural language parsing by adding noise to features, leading to improved parsing accuracy for English and German.

Contribution

It presents a novel approach to create multiple spectral models with noise, enhancing diversity and performance in latent-variable PCFG parsing.

Findings

Achieved 90.18 F1 score for English parsing

Achieved 83.38 F1 score for German parsing

Significant improvement over baseline models

Abstract

We describe an approach to create a diverse set of predictions with spectral learning of latent-variable PCFGs (L-PCFGs). Our approach works by creating multiple spectral models where noise is added to the underlying features in the training set before the estimation of each model. We describe three ways to decode with multiple models. In addition, we describe a simple variant of the spectral algorithm for L-PCFGs that is fast and leads to compact models. Our experiments for natural language parsing, for English and German, show that we get a significant improvement over baselines comparable to state of the art. For English, we achieve the $F_1$ score of 90.18, and for German we achieve the $F_1$ score of 83.38.