Let's Talk, Not Type: An Oral-First Multi-Agent Architecture for Guaran\'i

A team of researchers has published a position paper challenging the fundamental text-centric design of modern AI and HCI systems. Using Guarani, an official and widely spoken language of Paraguay, as a case study, Samantha Adorno, Akshata Kishore Moharir, and Ratna Kandala argue that current language support remains insufficient because it forces oral languages into a text-based pipeline. Their work highlights how this approach fails to align with lived oral practices and the linguistic reality of diglossia, where a community uses different languages for different social contexts.

In response, the researchers propose a radical alternative: an 'oral-first' multi-agent architecture specifically for Guarani. Instead of the standard 'speech-to-text-to-AI-to-text-to-speech' chain, their framework treats spoken conversation as the primary locus of interaction. The architecture decouples core natural language understanding from specialized agents that manage conversation state, repair (handling misunderstandings), and crucially, community-led governance. This design prioritizes features like turn-taking and shared context, which are central to oral communication.

The proposed technical framework is a direct challenge to the AI industry's universalist assumptions. By embedding principles of indigenous data sovereignty and designing for oral practices from the ground up, the paper contends that AI can only become truly culturally grounded by making this fundamental shift. The conclusion is that for digital ecosystems to empower rather than overlook diverse communities, spoken interaction must be a first-class design requirement, not an adaptation of a text-centric model.