Can computational models learn syntax? The learnability of the wh- and coordinate structure island constraints by artificial neural networks in Dutch
Keywords
Loading...
Authors
Issue Date
2022-07-25
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Recent research has shown that artificial neural networks (ANNs), i.e. general learning systems without any knowledge of language built in, can acquire human-like grammatical knowledge solely based on the input they receive. The current research project investigated whether an ANN can learn two syntactic island constraints, namely the wh- and coordinate structure island constraint, in Dutch in a way comparable to human native speakers. It was established whether the island constraints exist in Dutch by testing native speakers with an acceptability judgement task, and their performance was compared to that of a Long Short-Term Memory network; while native speakers demonstrate a clear sensitivity to wh- and coordinate structure island violations in Dutch, the LSTM network is not able to recognize these gap-resistant islands in Dutch. Therefore, input alone might not be enough to learn about these constraints, and internal language knowledge/abilities might be necessary to learn about them.
Description
Citation
Supervisor
Faculty
Faculteit der Letteren