Beyond documentation: Indigenous-informed design knowledge for language revitalization

The paper opens with an encounter in Guatemala. Near Tikal, a local guide described his situation in a sentence that condenses the whole problem: Spanish is not his language, English is not his language, and he does not speak his own. He was referring to Itza, a Maya language with fewer than fifty fluent speakers left (University of Toronto, 2025). A language can remain part of who you are even after its transmission from one generation to the next has already broken.

The encounter is treated as what it is: a personal, sensitizing moment, not evidence about Maya communities and not a license to speak for them. What it motivates is a design question. Digital systems have become part of language work everywhere. Can they do more than store what is left?

The literature separates three things. Documentation records a language: dictionaries, grammars, recordings, stories (Austin & Sallabank, 2018; Grenoble & Whaley, 2005). Preservation safeguards those resources for later. Revitalization is broader. It restores and expands the use of a language across social domains and generations, and its central benchmark is whether children acquire and actually use it (Fishman, 1991; Grenoble & Whaley, 2005).

Digital archives, corpora, and online dictionaries are documentation infrastructure (Galla, 2016; Garrett, 2018). They matter, and with the last fluent speakers aging, recording is time-critical. But none of it revives a language on its own. Revitalization-oriented systems have to support language as lived practice inside a larger ecology of speakers, learners, families, schools, media, institutions, and governance (Fishman, 1991; Hinton et al., 2018).

That makes it an information systems problem. A system embeds assumptions about access, openness, classification, authorship, ownership, and authority. If those assumptions come from dominant Western defaults, the system may reproduce exactly the marginalization that endangered the language, reducing it to decontextualized words and phrases (Meighan, 2021). The design-relevant question is therefore not what should be digitized, but who defines the problem, who controls representation and access, and whose criteria determine whether the system is valuable (Chung et al., 2024; Smith, 2021).

No community was studied directly. Instead, the paper reads eleven documented initiatives as sources of design knowledge. They were identified from the literature on digital language revitalization (Galla, 2016; Hinton et al., 2018; Llanes-Ortiz, 2023) and from project documentation, and selected to vary in the role the system plays and in how far Indigenous knowledge shapes it. Welsh is included deliberately as a boundary case: a minoritized but not Indigenous language where state institutions rather than Indigenous knowledge dominate (Williams, 2014).

The analysis ran in four steps between April and June 2026. From 38 documents, 55 situated features were extracted: concrete functions, governance arrangements, access structures, participation mechanisms, and reported uses. Each initiative was then profiled. Features were compared across cases and abstracted into 21 cross-case design features, which were finally grouped into six clusters. The move from situated feature to cross-case design feature follows the logic of design science research (Hevner et al., 2004; Strohmann et al., 2026), and the paper is candid that this very move carries the risk of decontextualization it warns against.

The design vocabulary is the paper’s main result. A few features illustrate what it captures. In Mukurtu, access is governed by community-defined cultural protocols that are more granular than the usual public, registered, and administrator roles (Christen, 2015): authority is a structural property of the system, not a moderation policy layered on top. Te Hiku Media keeps Māori language data and the speech technologies derived from it under community control, attaches a Kaitiakitanga license that travels with the data, and refuses extractive partnerships as a design decision, a stance that matches the call for relational rather than extractive language technology (Bird, 2024). RACHEL libraries in Guatemala work offline because connectivity cannot be assumed. And several cases show that revitalization needs new words for new domains, because a language that cannot name a browser tab cannot be lived in.

A system related to an Indigenous language is not automatically Indigenous-informed. Across the cases, Indigenous knowledge becomes consequential for design in five distinct modes, and initiatives combine them differently. The modes build on the argument that Indigenous knowledge can itself serve as justificatory knowledge for design (Chung et al., 2024). Many foreground content, access, and visibility. Mukurtu and Te Hiku Media show the deeper end, where cultural protocols shape the access architecture and a concept such as kaitiakitanga, a Māori principle of guardianship, informs licensing and infrastructure choices.

Three tensions run through the findings and will confront any future project. Openness against protection: wide access supports learning and visibility, but unrestricted openness can violate cultural protocols and data sovereignty, which is the tension the CARE Principles for Indigenous Data Governance were formulated to address (Carroll et al., 2020). Standardization against plurality: digital systems want stable orthographies and metadata, while languages carry regional variation, oral traditions, and contested authority. Scalability against situatedness: reusable platforms can serve many communities only if they stay adaptable to local histories and governance.

The study relies on public documentation, which shows what systems intend and offer but little about actual use or community-defined outcomes. And I write as a Western European researcher who does not represent the communities involved, which also means that the categories I use to reconstruct design knowledge are not culturally neutral (Chughtai & Young, 2025; Smith, 2021). The 21 design features are offered as contestable openings, not conclusions. What comes next has to be participatory and community-led, with communities defining the problem, the categories, and what counts as success (Hermes et al., 2012; Meighan, 2021; Smith, 2021).