SỌ̀RỌ̀: A Yorùbá Language Task Oriented Dialogue System

This paper presents Sọ̀rọ̀: a task-oriented dialogue system, developed utilising a rule-based technique. Sọ̀rọ̀ is fit for having a trade of four key sorts: greetings, small talk, basic number-crunching, and time/date. This dialogue framework has been fabricated following a linguistic investigation of the Yorùbá language and the rules defined from the analysis of publicly supported information. Sọ̀rọ̀'s conversational capacities are restricted to text-based trade and revolved only around a small domain of topics due to its limited vocabulary data sets. The framework involves three primary and auxiliary scripts each. The primary scripts are the bag-of-words, natural language understanding and natural language processing scripts while the auxiliary scripts are the task manager, the properties script, and the main script where dialogue sessions occur. This study developed rules that identify sentence types in utterances, split sentences into intent and entity, perform a list of tasks as identified in utterances and provide a response to this effect. This paper characterises rules that relate 66.7% accuracy the type of sentence contained within a sample utterance, with a precision of 94.7%. This study demonstrates the practicality of a Yorùbá language dialogue framework and simultaneously, design a dynamic dialogue system architecture likely to be improved upon with additional data.