Doctoral Degrees (Afrikaans and Dutch)

Permanent URI for this collection

Now showing 1 - 1 of 1

A grammatical framework for the computational parsing of written Afrikaans sentences
(Stellenbosch : Stellenbosch University, 2019-12) Swarts, Johannes Jacobus; Gouws, R. H.; Van Rooyen, G-J.; Oosthuizen, J.; Stellenbosch University. Faculty of Arts and Social Sciences. Dept. of Afrikaans and Dutch.
ENGLISH ABSTRACT: This dissertation investigates which grammatical framework is best suited to computationally represent and parse written Afrikaans sentences. This knowledge is necessary to build a large scale Afrikaans treebank – a resource which does not yet exist, but is a critical prerequisite for advanced endeavours in Afrikaans natural language processing. To gain this knowledge, we formally describe the building blocks of written Afrikaans from the perspectives of two major grammatical frameworks: constituency grammar and dependency grammar. Using these formal descriptions, we construct the first linguistically motivated treebank for Afrikaans, annotated with both constituency and dependency graphs. We perform k-fold cross-validation on multiple variations of this treebank with four state of the art sentence parsers, and fine-comb the results. Combining insights from the formal descriptions of written Afrikaans with the data obtained during parser evaluation, we conclude that dependency grammar outperforms constituency grammar at computationally representing the syntactic structure of written Afrikaans sentences under the conditions tested.