Regular expressions: enabling the development of computational aids for Zulu natural language processing
Laurette Pretorius, Sonja E. Bosch · South African Institute of Computer Scientists and Information Technologists · 2002
The computational treatment of morphologically complex languages, such as those belonging to the Bantu language family, requires as basic computational aids a machine-readable lexicon containing a list of all the word roots in the language as well as a computational morphological analyser/generator.In Zulu, where words are formed by productive affixations of derivational and inflectional suffixes to roots or stems, the constant core element in a word is the root. For example a single verb root in Zulu may have hundreds of thousands of different inflected/derived forms. At present no (comprehensive) machine-readable lexicon exists for Zulu. Since a machine-readable lexicon constitutes a key language resource in human language technology and natural language processing, and should therefore be readily accessible, reusable and portable, we discuss our development of such a resource as an XML document. In order to perform this task of maintaining and updating the lexicon in a systematic and exhaustive way, we elaborate on the automation of the identification and extraction of new roots from real natural language corpora. We explain that in order to do this for a language such as Zulu a computational morphological analyser/generator is required.In Pretorius and Bosch [2002] we reported on a first prototype of a computational morphological analyser/generator for Zulu in which the morphology of Zulu nouns was covered, using the Xerox finite-state calculus and tools to model and implement the appropriate morphotactics and alternation rules for a selected Zulu lexicon.In this paper we report on subsequent work, focussing on the design, building, and continued updating of a Zulu XML lexicon, using a variant of the morphological analyser/generator, called a guesser. We highlight the fundamental role that regular expressions play in XML, in the morphological analyser/generator, and in the guesser variant of the analyser as a tool for mining, exploring and analysing electronically available Zulu corpora.