In this article, we describe an element of a suite of computational tools for assigning word-class tags (as a preparation for part of speech (POS) tagging) to word forms in unrestricted Northern Sotho texts. POS-tagging is a step towards a linguistic analysis of the texts, which in turn allows for advanced data extraction.
The tool component that is described, identifies (and classifies) noun forms. Several types of linguistic knowledge are used to recognize nouns that are not contained in the noun lexicon of the system. These include the relationship between singular and plural noun prefixes, knowledge about noun derivation, and data about the co-occurrence of the candidate with concords, pronouns and adjectives in a local context.
Our implementation is a symbolic, voting-based process : together, all tests determine whether a candidate is a noun; accuracy on unseen test data is around 92%.