AI Fights for Cantonese as Mandarin Dominance and Data Scarcity Loom

A linguistic shift is underway as Mandarin gradually replaces Cantonese, a trend mirrored in the digital world where artificial intelligence models are failing to keep pace. The world's leading AI systems are predominantly trained on English data, creating significant challenges for languages with less digital content. This data scarcity means even major languages like Cantonese risk falling further behind, as machines struggle with translation nuances and cultural context that are intuitive to human speakers. Jacky Chan, founder of Hong Kong-based Votee AI, highlights specific challenges such as translating terms like 'dim sum' into English. Machines often produce unnatural translations that lack the cultural authenticity a human translator would provide. 'When you see it, you know it's not something a human writes,' Chan notes, emphasizing the translation troubles that form a growing list of issues for today's AI models which are strongest in English. In response, Votee AI has developed a specialized Cantonese large language model to address these deficiencies. The company, which has a decade of experience helping local companies with research, recognized that most existing LLMs can't handle Cantonese well due to its fundamental differences from Mandarin and English. Votee now aims to replicate its model for other low-resource languages in Southeast Asia and Africa, positioning itself as an 'OpenAI for neglected languages' and opening generative AI to underserved parts of the world.