Advancing Dialogue Systems in Low-Resource Conditions: A Synthetic Data Generation Approach

This thesis examines how to improve low-resource dialogue systems using a synthetic data approach. It focuses on three types: flowchart-based systems that guide users through structured steps, multi-modal systems that combine images and text and socially-aware systems that follow social norms. New methods create synthetic data to fill gaps, cover rare situations, and link visual and textual information. New datasets help systems learn and apply social norms.