Please use this identifier to cite or link to this item: https://www.um.edu.mt/library/oar/handle/123456789/70203
Title: Morphology in the Maltese language : a computational perspective
Authors: Borg, Claudia (2015)
Keywords: Maltese language -- Morphology
Machine learning
Computational linguistics
Issue Date: 2015
Citation: Borg, C. (2015). Morphology in the Maltese language: a computational perspective (Doctoral dissertation).
Abstract: This thesis presents the first comprehensive and systematic treatment of Maltese morphology using machine learning techniques. Maltese is considered as a ‘mixed’ language reflected in the hybrid nature of the morphological system, which has elements of both templatic systems typical of Semitic languages, and stem-based systems typical of Indo-European ones. The research looked at three different aspects of computational morphology, namely segmentation, relations and labelling. The segmentation task first explored unsupervised techniques to learn potential stems and affixes. The results were then used as the basis of the relations task, through the clustering of words on the basis of their orthographic and semantic similarity. The clustering technique was also unsupervised and used a metric to measure the disparity or similarity of a group of words so as to improve the clusters. An evaluation of the clusters was carried out using both experts and non-experts. The results of the non-expert group focused on the quality of the clusters, whilst the analysis of the expert responses focused on the differences between the concatenative and non-concatenative word clusters. Morphological labelling of words was viewed as a classification problem and approached using supervised techniques. Initially, the research focused on the classification of verbal inflections, resulting in a sequence of classifiers that represented different morphological properties. Cascade classifiers were then built for the noun and adjective categories, and integrated into a single classification system. The classification of grammatical category was also explored, questioning whether the morphological labels outputted by the different cascades could be used to reinforce the classification of the grammatical category. A final evaluation tested the full classification system on gold standard data from the MLRS corpus. The research resulted in a morphological classification system for verbs, nouns and adjectives. Although it has not yet achieved a sufficiently high accuracy, it provides the foundations for a more complete morphological analyser with broader coverage. The scope of the research was not merely a technological one, to create a morphological analyser, but rather to investigate the hybridity of the morphological system in Maltese and how this impacts the results of different techniques.
Description: PH.D.LINGUISTICS
URI: https://www.um.edu.mt/library/oar/handle/123456789/70203
Appears in Collections:Dissertations - InsLin - 2015

Files in This Item:
File Description SizeFormat 
15PHDLIN002.pdf4.21 MBAdobe PDFView/Open


Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.