Yorùbá Verb Sense Disambiguation using Semantic Similarity between Case Sentences of a Sense Inventory

Abstract

The development of a word sense disambiguation component of a machine translation (MT) system is faced with many challenges. One of these is the incidence of contextual tonal variation in the Yorùbá language which makes the use of statistical based approach highly expensive for resolving lexical ambiguity in the language. This study examined the procedures underlining the resolution of lexical ambiguity in the context of Yorùbá-to-English MT system and developed a knowledge based approach which makes use of path-based similarity measurement between two instances of an ambiguous word to determine its right sense. This model achieved an accuracy of 96.1% for transitive verbs, 90.2% for intransitive verbs, with an overall accuracy of 94.6%, which is comparable with the high-performing supervised WSD, and a coverage of 69.3%.
This study suggests a method that can be used to address ambiguity resolution in other low resource languages.