An analysis of Siswati morphological analyser outputs

Abstract

Siswati is characterised by a limited availability of annotated corpora and natural language processing (NLP) tools. The few NLP tools for Siswati include morphological analysers, part-of-speech taggers, morphological decomposers, spelling checkers and machine translation systems. Given the existence of these tools, it is imperative to assess their effectiveness in producing reliable results that significantly contribute to the development of Siswati. This article aims to evaluate the accuracy of the outputs produced by the Siswati morphological analyser (SMA) developed by the Centre for Text Technology. Adopting a mixed-methods approach that is undergirded by the task technology fit theory, this article examines a corpus of 133 Siswati words processed by the SMA. Researchers manually verified the results, revealing that the analyser performed most effectively with adjectives (100%), conjunctions (90.9%), locatives (85.7%), and adverbs (70.9%). Its accuracy decreased with nouns (56.3%) and relatives (50%). The analyser struggled significantly with verbs (44%) and possessives (33.3%), indicating room for improvement in these word categories. Despite these discrepancies, the existence of an SMA is a positive step toward the language’s development in the realm of NLP. The findings underscore the need to refine the morphological analyser to contribute effectively to the linguistic advancement of Siswati.