Using Semantic Networks for Question Answering - Case of Low-Resource Languages Such As Swahili

Question Answering is an important aspect of language processing. While much research has been done for other languages, low-resourced languages such as Swahili have not been studied much, despite it being an important language spoken in East Africa. Due to lack of language resources, we rely on syntactic analysis of raw text to create rules that extract Subject-Predict-Object to then create a semantic network of Swahili, since Swahili is Subject-Verb-Object (SVO) language. We then use SPARQL to query the network. We got our test corpus from public websites with Swahili Frequently Asked Questions (FAQ) and some Swahili reading comprehension story books. Results for story books show that the semantic network can correctly answer on average three of the five set questions for each sample text. Shortcomings of our method is mainly related to inadequacies of preprocessing tools, where further research is needed. Our method should be applicable in creating networks for any low-resourced language that follows the SVO language pattern, which can then be queried.