The development of a statistical spell checker for Maltese

Abstract

The detection and correction of spelling errors is an integral part of most modern word-processors. This is a consequence of the large amount of misspelled words that one commonly finds in typed text. The reasons for this frequent occurrence of misspellings are varied but, among these, the one which carries most weight is the nature of Natural Languages which is extremely complex and full of exceptions. This project is concerned with the development of a statistical spell checker for the Maltese Language. The very rich morphology of this language poses the problem that traditional methods of Spelling Correction, such as dictionary lookup, are not ideal approaches to the problem at hand. In this work, the approach taken is based on statistical N-Gram spelling models. Various other statistical and probabilistic models, concerned with the problem of context-sensitive spelling correction, are also analysed. These additional techniques aim to provide a solution for the detection of real-word errors in a document, that is, the detection of those misspellings that result in an existing word in the language. The approach followed in these techniques is that of developing a statistical model from an existing training corpus and then using that model to interpret new, unseen data. Implementation involved the development of two separate tools. The first tool offers all the functions necessary for generating statistics while the second tool is the actual spell checker which makes use of these generated statistics in order to analyse the given text. The results obtained were promising and, when experiments were carried out using both the Maltese and English Language, it was concluded that there are no obvious disparities between the results obtained in both cases.