Naija grammatical classes and constructions

Abstract

n this talk, we show how a treebank can be used to extract the distributional classes of lexemes and the main constructions of a language, viz Naija. We focus on grammatical lexemes whose categorization poses particular problems due to the isolating nature of Naija. In particular, we study the elements that revolve around the verb and the noun.The treebank we use, SUD_Naija-NSC, is a 150 000 word corpus manually annotated in dependency syntax: words are linked by syntactic links and each word receives a syntactic function from its governor and is annotated with POS and morphological features. A powerful query system, Grew-Match, allows us to quantify simple and complex patterns in the treebank and to extract relevant configurations.