The LESSEN project (Low-Resource Chat-based Conversational Intelligence) is the inaugural project within the LATTICE Lab, focused on making conversational AI accessible for lower-resource contexts such as the Dutch language.

The project’s research is carried out through several Work Packages (WPs), each addressing a specific challenge:

WP1.1: Neural Architectures – Focuses on designing neural architectures for chat agents that are efficient in terms of training, memory, and inference.

WP1.2: Efficient Data Use – Develops ways to use existing data from different languages and forms, such as entity graphs, to empower AI even when data is scarce.

WP2.1: Domain-Specific Challenges – Aims to solve the hurdles of developing conversational agents in specific fields by transferring knowledge from other, data-rich domains.

WP2.2: Synthetic Data Generation – Focuses on creating methods to generate artificial data for training the data-hungry models needed for conversational agents.

WP3.1: Safety, Security, and Privacy – Dedicated to detecting and preventing unsafe interactions in dialogues while investigating critical privacy and security issues.

WP3.2: Agent Transparency – Works on algorithms that can generate model explanations for stakeholders to ensure the AI’s operations are transparent.

WP4: Robust Evaluation – Focused on developing rigorous evaluation methods and tools specifically for Large Language Models (LLMs) and conversational agents.

WP5: Knowledge Integration and Dissemination – Aimed at integrating and sharing the knowledge and research outputs generated across the lab with both consortium members and external parties.