Paris and Stanford at EPE 2017: Downstream Evaluation of Graph-based Dependency Representations - Archive ouverte HAL Access content directly
Conference Papers Year : 2017

Paris and Stanford at EPE 2017: Downstream Evaluation of Graph-based Dependency Representations

(1) , (2) , (3) , (2) , (1, 4) , (2, 5)
1
2
3
4
5

Abstract

We describe the STANFORD-PARIS and PARIS-STANFORD submissions to the 2017 Extrinsic Parser Evaluation (EPE) Shared Task. The purpose of this shared task was to evaluate dependency graphs on three downstream tasks. Through our submissions, we evaluated the usability of several representations derived from English Universal Dependencies (UD), as well as the Stanford Dependencies (SD), Predicate Argument Structure (PAS), and DM representations. We further compared two parsing strategies: Directly parsing to graph-based dependency representations and a two-stage process of first parsing to surface syntax trees and then applying rule-based augmentations to obtain the final graphs. Overall, our systems performed very well and our submissions ranked first and third. In our analysis, we find that the two-stage parsing process leads to better downstream performance, and that enhanced UD, a graph-based representation, consistently outperforms basic UD, a strict surface syntax representation, suggesting an advantage of enriched representations for downstream tasks.
Fichier principal
Vignette du fichier
epe17.pdf (553.43 Ko) Télécharger le fichier
Origin : Publisher files allowed on an open archive
Loading...

Dates and versions

hal-01592051 , version 1 (31-12-2017)

Identifiers

  • HAL Id : hal-01592051 , version 1

Cite

Sebastian Schuster, Éric Villemonte de La Clergerie, Marie D Candito, Benoît Sagot, Christopher D Manning, et al.. Paris and Stanford at EPE 2017: Downstream Evaluation of Graph-based Dependency Representations. EPE 2017 - The First Shared Task on Extrinsic Parser Evaluation, Sep 2017, Pisa, Italy. pp.47-59. ⟨hal-01592051⟩
209 View
181 Download

Share

Gmail Facebook Twitter LinkedIn More