Decolonizing LLMs for Indigenous languages: a functional or critical interculturality perspective?

Large language models (LLMs) are increasingly promoted as decolonial tools for preserving Indigenous languages, yet whether such methods can be decolonial remains contested. This paper conceptualizes that debate through the distinction between functional and critical interculturality, separating recognition of cultural difference from the redistribution of power over knowledge production. Rather than resolving this distinction from literature alone, we investigate how it is translated into the perspectives of an Indigenous community itself, drawing on twelve interviews with Zapotec speakers in Oaxaca, Mexico, about AI and LLM-based tools for language preservation. We find that participants’ own accounts divide along these same lines: some value the technical recognition LLMs offer their language, while a majority articulate a critical position, arguing that these technologies are designed within and for the ideological frameworks of culturally dominant countries and that the resources they require make independent community development virtually impossible. Building on this division, we propose critical interculturality of data as an analytical framework that shifts attention from whether Indigenous languages are represented in digital and computational methods to who controls the terms of that representation, pointing to community-led alternatives practiced beyond AI.