ESURF: Simple and Effective EDU Segmentation

TL;DR

This paper introduces a simple, effective method for EDU segmentation using lexical and character n-gram features with random forest classification, outperforming existing methods and enhancing discourse parsing efficiency.

Contribution

The paper presents a novel, straightforward approach for EDU segmentation that leverages lexical and character n-grams, demonstrating superior performance over previous techniques.

Findings

Outperforms existing segmentation methods

Enhances discourse parser accuracy

Highlights importance of lexical features

Abstract

Segmenting text into Elemental Discourse Units (EDUs) is a fundamental task in discourse parsing. We present a new simple method for identifying EDU boundaries, and hence segmenting them, based on lexical and character n-gram features, using random forest classification. We show that the method, despite its simplicity, outperforms other methods both for segmentation and within a state of the art discourse parser. This indicates the importance of such features for identifying basic discourse elements, pointing towards potentially more training-efficient methods for discourse analysis.