SyntaxNet Models for the CoNLL 2017 Shared Task

TL;DR

This paper introduces ParseySaurus, a dependency parsing system using DRAGNN that outperforms previous models on the CoNLL 2017 Shared Task by leveraging transition-based recurrent parsing and character-based representations.

Contribution

The paper presents a new baseline dependency parser, ParseySaurus, that combines transition-based recurrent parsing with character-based representations, achieving superior accuracy.

Findings

Outperforms state-of-the-art models by 3.47% LAS

Effective on 52 Universal Dependencies Treebanks

Demonstrates the strength of combining transition-based parsing with character embeddings

Abstract

We describe a baseline dependency parsing system for the CoNLL2017 Shared Task. This system, which we call "ParseySaurus," uses the DRAGNN framework [Kong et al, 2017] to combine transition-based recurrent parsing and tagging with character-based word representations. On the v1.3 Universal Dependencies Treebanks, the new system outpeforms the publicly available, state-of-the-art "Parsey's Cousins" models by 3.47% absolute Labeled Accuracy Score (LAS) across 52 treebanks.