A*STAR Institute for Infocomm Research (A*STAR I2R) has developed the second version of MERaLiON (Multimodal Empathetic Reasoning and Learning in One Network), Southeast Asia’s empathetic Multimodal Large Language Model (MLLM), designed to reflect the region’s rich linguistic diversity and unique communication styles by understanding how people in Southeast Asia speak, emote, and interact.
First launched in December 2024, MERaLiON is one of two national large language models (LLMs) under Singapore’s S$70 million National Multimodal Large Language Model Programme (NMLP), supported by the National Research Foundation (NRF) and the Infocomm Media Development Authority (IMDA).
MERaLiON Version 2
MERaLiON Version 2 introduces significant improvements in contextual understanding and user experience. These include a broader language coverage that supports English, Mandarin, Malay, Tamil, Singlish, Bahasa Indonesia, Thai, and Vietnamese.
It also claims to have code-switching capabilities (e.g., English-Chinese, English-Malay, Singlish-Chinese), reflecting natural communication patterns across the region.
Furthermore, the MLLM claims to be emotionally intelligent. It achieves this by understanding emotions and paralinguistic cues. For instance, it can detect emotional tone, gender, and paralinguistic features in speech, enabling more nuanced and empathetic AI interactions.
With its culturally aware feature, MERaLiON understands and interprets different styles of spoken language in various languages.
MERaLiON Consortium
Spearheaded by A*STAR I2R and supported by IMDA, the MERaLiON Consortium is a collaborative platform that brings together industry and research partners, making them integral to the adoption and application of MERaLiON-powered solutions across Southeast Asia.
Its twelve-member alliance contributes domain and technical expertise, co-develops use cases and scales the deployment of culturally aware AI across sectors.
The consortium aims to accelerate the development, adoption, and impact of culturally contextualised AI to aggregate demand and reduce costs, enhance model capabilities, and co-develop use cases.