Annotation projection with pattern based neural networks for antonym and synonym detection in the Yorùbá Language

Abstract

In this thesis, I explore the task of automatically distinguishing between synonyms and antonyms in the Yorùbá language. Previous work by Nguyen et al. [16] relies on linguistic resources such as WordNet to provide supervision for this task in English, however at this time there is no WordNet for Yorùbá. I investigate whether we can bridge this resource gap by utilizing machine translation systems, parallel corpora and natural language parsers, following the annotation projection paradigm [10, 13, 18]. In Chapter 1, I provide an overview of the entire work and my technical contributions. Next, I state the philosophical motivations of this work and its larger goals. In Chapter 2, I provide a background on annotation projection and on previous work on the task of distinguishing between antonyms and synonyms. In Chapter 3, I present the preliminary results of this work and explain its limitations as well as how they can be overcome in future work. I conclude with a summary about this work and its contributions.