Jabberwocky Parsing: Dependency Parsing with Lexical Noise

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F19%3A10427152" target="_blank" >RIV/00216208:11320/19:10427152 - isvavai.cz</a>

Result on the web

<a href="https://www.aclweb.org/anthology/W19-0112" target="_blank" >https://www.aclweb.org/anthology/W19-0112</a>

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Jabberwocky Parsing: Dependency Parsing with Lexical Noise

Original language description

Parsing models have long benefited from the use of lexical information, and indeed currentstate-of-the art neural network models for dependency parsing achieve substantial improvementsby benefiting from distributed representations of lexical information. At the sametime, humans can easily parse sentences with unknown or even novel words, as in Lewis Carroll’s poem Jabberwocky. In this paper, we carry out jabberwocky parsing experiments, exploring how robust a state-of-the-art neural network parser is to the absence of lexicalinformation. We find that current parsing models, at least under usual training regimens, are in fact overly dependent on lexicalinformation, and perform badly in the jabberwocky context. We also demonstrate that the technique of word dropout drastically improves parsing robustness in this setting, and also leads to significant improvements in out-of-domain parsing.

Czech name

—

Czech description

—

Type

O - Miscellaneous

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

—

Continuities

—

Publication year

2019

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů