Developing a culturally intelligent chatbot suitable for Singapore

Abstract

This project presents the development of a culturally intelligent chatbot tailored for the Singaporean context. Recognizing the limitations of large language models (LLMs) in addressing cultural nuances, this work integrates natural language processing (NLP) techniques with a multi-stage architecture pipeline to ensure cultural sensitivity to Singapore’s multilingual and multi-ethnic society. The chatbot system incorporates preprocessing modules to clean and preserve local colloquialisms (e.g., Singlish), zero-shot classification for detecting cultural or sensitive content, intent analysis for response tone control, and contextual augmentation to guide the model response generation process.

SEA-LION, a south-east asian trained instruction-following model, is selected as the core LLM for this project due to its high degree of cultural and linguistic alignment with Southeast Asia which covers the context of Singapore (SEA-LION.AI, 2024). Safety layers such as sentiment, toxicity, and sensitivity checks are embedded into the chatbot pipeline to filter off inappropriate responses. Comparative experiments with other models, including GPT-2, GPT-Neo, and OPT are conducted so as to demonstrate the enhanced cultural relevance and response quality of this proposed system.

This work highlights the importance of cultural awareness in Artificial Intelligence (AI) systems and offers a practical framework for developing chatbots suited to multicultural environments like Singapore.