From linguistic diversity to cognitive sovereignty? A critical study of data practices in Latin American language technology development

The proliferation of language technologies, including chatbots, speech-to-text systems, and machine translation, has revealed significant biases adversely affecting speakers of minoritized languages, particularly in the Global South. These biases stem in part from technology development practices that prioritize languages with extensive digital textual representation, leading to digital exclusion for speakers of minoritized and Indigenous languages. Data collection, curation, and classification have all been identified as critical junctures where biases emerge. Such practices often inadvertently replicate colonial extraction processes, prevent the full expression of linguistic diversity and impose ontologies that may impede cognitive sovereignty. This dissertation investigates these issues through a focus on Mexican and Central American Indigenous languages, utilizing a combination of historiography, archival research and ethnographic methods. The focus is on the question of how Indigenous languages become data in the first place, and what this means for linguistic diversity and cognitive sovereignty in the digital environment today. Historiography situates the development of language technology tools, showing how a combination of ideological presumptions and structural constraints in the 1940-60s resulted in computational approaches to language that continue to impact how language as a concept is constituted today. Research in archival collections documenting the transition from analog to digital methods in linguistics and linguistic anthropology highlights the sociopolitical and ideological factors informing the transformation of Indigenous languages into digital data and early computational approaches to working with such texts. Ethnographic work examining the development of a digital repository of Mesoamerican language content builds upon this historical research to investigate the ongoing influence of historical data practices on contemporary digital projects seeking to enhance access for heritage speaker communities and researchers, as well as the complexities of operationalizing contemporary ethical research protocols around legacy datasets. Understanding the situated factors informing the language technology development process, including data histories and the complexities of working with such data for transformative uses today, suggests both theoretical and practical insights with the potential to more deeply incorporate culturally sensitive protocols and thus contribute to greater linguistic equity for speakers of Indigenous languages.