Verbal extension sequencing: an examination from a computational perspective

Lexical transducers utilise a two-level finite-state network to simultaneously code morphological analysis and morphological generation rewrite rules. Multiple extensions following the verb root can be morphologically analysed as a closed morpheme class using different computational techniques. Analysis of a multiple extension sequence is achieved by trivial analysis, based on any combination of the closed class members, but this produces unnecessary over-generation of lexical items, many of which may not occur in a lexicon. Limiting the extension combinations, in an attempt to represent examples that may actually exist – in terms of both the possible number of extensions in a sequence and the relative ordering of the extensions – leads to a radical reduction in the generation of lexical items while the ability to analyse adequately is maintained. The article highlights details of an investigation based on both trivial analysis and an approach that prevents dramatic overgeneration. The article is based on test data reflecting possible extension sequences and the morphophonemic alternations of these extensions for Northern Sotho, garnered from literature research, lexicographic investigation and the computational morphological analysis of texts.