Towards a More Intuitive Sinhala Chatbot: Leveraging NLU for Enhanced Intent Identification and Entity Extraction

Abstract

In the fast-paced and ever-changing world of conversational AI, chatbots have become essential interfaces for user interactions in various domains, especially when supporting different languages. This study delves into the development of chatbots and their ability to understand language, explicitly focusing on the Sinhala language. The effectiveness of two platforms, Rasa NLU and Microsoft LUIS, were compared in identifying and extracting intents. Both platforms showed proficiency, but Rasa stood out for its flexibility, cost-effectiveness and accurate intent recognition. A case study in the restaurant domain was conducted to demonstrate the system’s capabilities. An architecture was created that can interpret Sinhala expressions and analyze intents using the NLU engine. The study defined four intents: Food Ordering, Get In Touch, About Restaurant and None. The findings highlight how this architecture has the potential to accurately interpret intents during chatbot development regardless of the conversational language used. This research aims to contribute insights to developers, linguists and AI enthusiasts involved in language-specific chatbot development by emphasizing its promises and challenges.