Abstract: Foundation models have fundamentally transformed natural language processing, evolving from task-specific systems to powerful universal language learners that demonstrate remarkable multilingual capabilities. This talk examines how these large-scale models are breaking down language barriers and potentially democratizing AI access across diverse linguistic communities worldwide. We explore the technical foundations that enable effective multilingual scaling, including cross-lingual transfer learning mechanisms, advanced tokenization strategies for diverse writing systems, and the curation of massive multilingual training datasets. The discussion covers how these models develop shared linguistic representations that facilitate knowledge transfer between languages, enabling robust performance even for low-resource languages with limited training data.

Key topics include the theoretical principles underlying multilingual representation learning, practical challenges in supporting the world's linguistic diversity, and the emergent cross-lingual reasoning capabilities that arise from scale. We examine current evaluation methodologies and their limitations in capturing true multilingual competence across different cultural contexts. Beyond technical considerations, the talk addresses critical questions of linguistic equity and cultural representation in AI systems. We discuss how current models may perpetuate existing digital divides and explore strategies for building truly inclusive AI that respects linguistic autonomy and cultural diversity. This includes examining bias in multilingual models and approaches for ensuring equitable performance across language communities.

The presentation concludes with a forward-looking perspective on creating sustainable multilingual AI ecosystems that serve all language communities fairly, highlighting both the immense promise and the ongoing challenges in realizing the vision of universal language AI.

About the Speaker: Sudeshna Sarkar is a Professor in the Department of Computer Science & Engineering, IIT Kharagpur. She is a Fellow of INAE and former Head, Department of Computer Science & Engineering, and Former (Founding) Head, Centre of Excellence in Artificial Intelligence at IIT Kharagpur. She did her BTech and PhD in Computer Science & Engineering form IIT Kharagpur and MS from University of California, Berkeley. Her research interests are in Natural Language Processing and in applications of Artificial Intelligence and Machine Learning.

We look forward to your active paerticipation.