Minor changes make a difference: a case study on the consistency of\n UD-based dependency parsers

2021/11/30 by Dmytro Kalpakchi, Kalpakchi, Dmytro, Johan Boye +1
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Text Readability and Simplification #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2111.15413

openalex publication_date 2021/11/30 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28

Abstract

Many downstream applications are using dependency trees, and are thus relying\non dependency parsers producing correct, or at least consistent, output.\nHowever, dependency parsers are trained using machine learning, and are\ntherefore susceptible to unwanted inconsistencies due to biases in the training\ndata. This paper explores the effects of such biases in four languages -\nEnglish, Swedish, Russian, and Ukrainian - though an experiment where we study\nthe effect of replacing numerals in sentences. We show that such seemingly\ninsignificant changes in the input can cause large differences in the output,\nand suggest that data augmentation can remedy the problems.\n

Citations

• Algorithm 457: finding all cliques of an undirected graph
• All Roads Lead to UD: Converting Stanford and Penn Parses to English Universal Dependencies with Multilayer Annotations

Related

• Dependency Language Models for Transition-based Dependency Parsing
• Revisiting the Effects of Leakage on Dependency Parsing
• Applying Occam's Razor to Transformer-Based Dependency Parsing: What Works, What Doesn't, and What is Really Necessary
• Fill it up: Exploiting partial dependency annotations in a minimum spanning tree parser
• Inherent Dependency Displacement Bias of Transition-Based Algorithms