Data colonialism and indigenous languages in AI: a critical review of existing initiatives and their struggles with data sovereignty

May combine two claims.

The lack of co-governance and the unremunerated reuse of Guaraní texts echo Couldry and Mejias's account of data colonialism, in which extractive practices associated with historical colonialism are combined with the quantification methods of modern computing.

“The lack of co-governance and the unremunerated reuse of these texts echo Couldry and Mejias's account of data colonialism, in which extractive practices associated with”

• cites Couldry & Mejias 2019 sourceSupported
• cites Gray 2023 sourceSupported

May combine two claims.

By respecting the First Nations principles of OCAP, which assert that Indigenous peoples retain these rights over data about them, and partnering with Inuit organizations, the Canadian MT efforts sought to ensure that the tools met community needs and that the data remained accessible to those communities.

“By respecting the First Nations principles of OCAP, which assert that Indigenous peoples retain these rights over data about them, and partnering with Inuit organizations, the Canadian MT efforts sought to ensure that the tools (like an Inuktitut translator or a verb conjugator for Mohawk) met community needs and that the data remained accessible to those communities.”

• cites Kuhn et al. 2020 sourcePartly supported

  Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

UCCIX's creators explicitly frame it as a blueprint for other low-resource and Indigenous languages, with the dual goals of preserving Irish in the digital era and promoting Irish knowledge and culture globally.

“Its creators explicitly frame it as a blueprint for other low-resource and Indigenous languages, with the dual goals of preserving Irish in the digital era and promoting Irish knowledge and culture globally.”

• cites Tran et al. 2024 sourcePartly supported

  The abstract says the work aims to preserve and promote the Irish language, knowledge, and culture in the digital era while providing a framework for other Indigenous languages. The Introduction and Conclusion also describe it as a blueprint for other low-resource languages. Our copy does not specify a goal of promoting Irish knowledge and culture globally. (LLM, conf=0.85)

  Experimental — accuracy not yet measured

May combine two claims.

The NRC project was guided by an Indigenous advisory council and explicitly aimed to collaborate with communities, putting equal emphasis on the agenda of the linguist and of the community.

“It was guided by an Indigenous advisory council and explicitly aimed to collaborate with communities, putting”

• cites Kuhn et al. 2020 sourceSupported

May combine two claims.

UCCIX is an open-source Irish LLM introduced in 2024.

“A landmark effort is UCCIX, an open-source Irish LLM introduced in 2024 (Tran et al. 2024).”

• cites Tran et al. 2024 sourceSupported

Classified as probably not a substantial claim. May combine two claims.

FAIR principles focus on making data maximally open and shareable, while CARE principles focus on people and purpose, emphasizing that data use should align with the rights and needs of Indigenous Peoples.

“The key distinction is that while FAIR principles focus on making data maximally open and shareable-goals that align with scientific efficiency, CARE principles focus on people and purpose, emphasizing that data use should align with the rights and needs of Indigenous Peoples.”

• cites Carroll et al. 2020 sourcePartly supported

  The Abstract and Conclusion describe FAIR as data-centric and focused on access and reuse, while CARE adds attention to people, purpose, and Indigenous rights. They do not characterize FAIR as seeking to make data maximally open and shareable; the Conclusion instead advocates using FAIR and CARE together. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

May combine two claims.

The National Research Council Canada (NRC) spearheaded an Indigenous Languages Technology project (2017-2020) that included an Inuktitut MT system, among other tools.

“The National Research Council Canada (NRC) spearheaded an Indigenous Languages Technology project (2017-2020) that included an Inuktitut MT system, among other tools.”

• cites Kuhn et al. 2020 sourcePartly supported

  The Introduction identifies an NRC Indigenous Languages Technology project funded for a three-year first phase beginning in 2017. The Abstract reports experiments with MT systems trained on an Inuktut–English corpus, alongside other software subprojects. It does not establish that the project included a completed Inuktitut MT tool, rather than experimental systems for Inuktut. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

The CARE Principles for Indigenous Data Governance were formulated in 2019 by the Global Indigenous Data Alliance and related networks.

“A concrete expression of Indigenous Data Sovereignty is found in the CARE Principles for Indigenous Data Governance, formulated in 2019 by the Global Indigenous Data Alliance and related networks (Carroll et al. 2020).”

• cites Carroll et al. 2020 sourcePartly supported

  The Abstract and Development section say that the Research Data Alliance’s International Indigenous Data Sovereignty Interest Group developed the CARE Principles with input from Indigenous networks and others. The Development section dates the workshop where the principles were first drafted to 2018; the References list a September 2019 version, while the Research Data Alliance section says the Global Indigenous Data Alliance hosts the principles. The paper therefore does not establish that GIDA formulated them in 2019. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

May combine two claims.

Since 1999, laws and legislative proceedings must be published in Inuktitut, leading to a substantial bilingual corpus of Inuktitut-English parliamentary records (the Nunavut Hansard).

“Since 1999, laws and legislative proceedings must be published in Inuktitut, leading to a substantial bilingual corpus of Inuktitut-English parliamentary records (the Nunavut Hansard).”

• cites Kuhn et al. 2020 sourcePartly supported

  Section 3.2 reports a substantial Inuktut-English Nunavut Hansard corpus drawn from Legislative Assembly proceedings dating from April 1999. It does not say that laws or proceedings must be published in Inuktitut, or that a legal requirement led to the corpus. The paper uses “Inuktut” for the language varieties represented, rather than specifying Inuktitut alone. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S2 (checkable).

Data colonialism refers to the massive appropriation of social life through data extraction, a new colonial move in which human experiences become a resource to be mined for economic value.

“Coined by Couldry and Mejias (2019), data colonialism refers to “the massive appropriation of social life through data extraction”, a process that represents “a new colonial move” in which human experiences become a resource to be mined for economic value (Couldry and Mejias 2019).”

• cites Couldry and Mejias 2019 sourceSupported

Classified as probably not a substantial claim.

Birhane (2021) argues that Western AI often imposes rigid, decontextualized systems that clash with the communal and contextual ways knowledge is held in Indigenous cultures.

“Nevertheless, Birhane (2021) argues that Western AI often imposes rigid, decontextualized systems that clash with the communal and contextual ways knowledge is held in Indigenous cultures.”

• cites Birhane 2021 sourcePartly supported

  In “Rationality: the dominant orthodoxy,” Birhane describes Western-influenced data science as treating knowledge as static and context-free. “Knowing that centers human relations” says technologies shaped by a limited Western perspective are forced onto marginalized groups. But the paper discusses Afro-feminist epistemology and ubuntu, not knowledge in Indigenous cultures specifically, so that part of the attribution is unsupported. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim.

Indigenous Data Sovereignty asserts that Indigenous Peoples have the right to control the collection, ownership, and application of data about their members, cultures, and lands.

“The IDS was proposed with the idea that Indigenous Peoples have the right to control the collection, ownership, and application of data about their members, cultures, and lands.”

• cites Carroll et al. 2020 sourceSupported

Classified as probably not a substantial claim. May combine two claims.

As Cooper, Heldreth, and Hutchinson argue, NLP for Indigenous languages should primarily serve Indigenous communities, prioritizing communities' values and experiences with respect to NLP projects, and the expressed opinions of those communities around data governance.

“As Cooper, Heldreth, and Hutchinson argue,”

• cites Cooper et al. 2024 sourceSupported

Probably not a substantial claim: may not pass S2 (checkable).

Data relating to Indigenous communities should be governed by those communities, according to their values and for their collective benefit.

“In essence, IDS extends the political notion of sovereignty into the realm of information: it asserts that data relating to Indigenous communities should be governed by those communities, according to their values and for their collective benefit (Carroll et al. 2020).”

• cites Carroll et al. 2020 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

CARE principles were developed to complement the FAIR principles (Findable, Accessible, Interoperable, Reusable).

“These principles were developed in consultation with Indigenous communities worldwide to complement the mainstream data management principles known as FAIR: Findable, Accessible, Interoperable, Reusable (Carroll et al. 2020).”

• cites Carroll et al. 2020 sourceSupported

Classified as probably not a substantial claim. May combine two claims.

It calls for what Couldry and Mejias (2019) term a decolonial reversal, reclaiming the datafied future and ensuring that Indigenous peoples are not just data subjects but data authors and owners.

“Instead, it calls for what Couldry and Mejias (2019) term a decolonial reversal, reclaiming the”

• cites Couldry & Mejias 2019 sourcePartly supported

  In the passage beginning “At this point, we can learn much by recalling the vision of the late Peruvian sociologist Aníbal” (page unavailable), the authors call for resistance to data colonialism and rejection of extractive data practices. Our copy does not use the term “decolonial reversal” or call specifically for Indigenous peoples to become data authors and owners. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim. May combine two claims.

The CARE Principles for Indigenous Data have been proposed to complement the open data movement, and the above examples justify the necessity of the framework.

“The CARE Principles for Indigenous Data have been proposed to complement the open data movement, and the above examples justify the necessity of the framework.”

• cites Carroll et al. 2020 sourceSupported

May combine two claims.

Despite an estimated 90% of Paraguay's population speaking and writing Guaraní, the language accounts for only about 20% of online content.

“Local advocates note that despite an estimated 90% of Paraguay's population speaking and writing Guaraní, the language accounts for only about 20% of online content (Rising Voices 2020).”

• cites Rising Voices 2020Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Te Hiku negotiated to lead the development and insisted on a Kaitiakitanga (guardianship) model for data, rather than allowing Big Tech to scrape Māori words for a generic model.

“Instead of allowing Big Tech to scrape Māori words for a generic model, Te Hiku negotiated to lead the development and insisted on a Kaitiakitanga (guardianship) model for data.”

• cites Lee 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Despite the existence of Paraguayan institutions engaged in standardizing Guaraní, Google's announcements do not reference formal partnerships with government bodies, academic collaborators, or national institutions.

“Despite the existence of Paraguayan institutions engaged in standardizing Guaraní,foot_3 these announcements do not reference formal partnerships with government bodies, academic collaborators, or national institutions either.”

May combine two claims.

Irish Gaelic is the national and first official language of Ireland, yet it is classified by UNESCO as a 'Definitely Endangered' language.

“Irish is the national and first official language of Ireland, yet it is classified by UNESCO as a “Definitely Endangered” language (UNESCO 2010), meaning it is no longer being learned naturally by most children.”

• cites UNESCO 2010Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Uí Dhonnchadha and van Genabith (2006) developed a finite-state morphology and constraint grammar disambiguation, together with a part-of-speech tagger for Irish.

“Early foundational work on Irish language NLP laid important groundwork; for example, Uí Dhonnchadha and van Genabith (2006) developed a finite-state morphology and constraint grammar disambiguation, together with a part-of-speech tagger for Irish, one of several “considerable efforts…made over the past decade to develop natural language processing resources” for Irish.”

• cites Uí Dhonnchadha and van Genabith 2006 sourceNot checked — no full text of the source

May combine two claims.

Ard-Intleacht na Gaeilge is a public-sector project announced in late 2024, led by Údarás na Gaeltachta, to create an Irish-centric AI system by 2026.

“Announced in late 2024, Ard-Intleacht na Gaeilge is a public-sector project led by Údarás na Gaeltachta (the Gaeltacht development authority) to create an Irish-centric AI system by 2026 (Caollaí 2024).”

• cites Caollaí 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

In 2022-2024, Google Translate added Inuktitut (Inuktut) to its supported languages, marking the first Canadian Indigenous language on the platform.

“In 2022-2024, Big Tech also entered the arena.Google Translate added Inuktitut (Inuktut) to its supported languages, marking the first Canadian Indigenous language on the platform (News 2024).”

• cites News 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Even with community input, Google's system may generate incorrect or culturally inappropriate translations, especially given dialectal variation in Inuktut; if users trust the AI too much, it could spread misinformation.

“Even with community input, Google's system may generate incorrect or culturally inappropriate translations (especially given dialectal variation in Inuktut).If users trust the AI too much, it could spread misinformation.”

May combine two claims.

Who builds and governs Gaelic AI is as important as what the AI does, and these remain unaddressed issues in the two Irish efforts.

“These scenarios underscore that who builds and governs Gaelic AI is as important as what the AI does, and remain unaddressed issues in these two efforts.”

May combine two claims.

In 2022 the Irish government launched a comprehensive Digital Plan for the Irish Language, aiming to integrate Irish into 'today's digital society' and enable the Irish-speaking community to 'build and revitalise the language' through technology.

“In 2022 it launched a comprehensive Digital Plan for the Irish Language, aiming to integrate Irish into “today's digital society” (Department of Tourism, Culture, Arts, Gaeltacht, Sport and Media, Government of Ireland 2023a) and enable the Irishspeaking community to “build and revitalise the language” through technology (Department of Tourism, Culture, Arts, Gaeltacht, Sport and Media, Government”

• cites Department of Tourism, Culture, Arts, Gaeltacht, Sport and Media, Government of Ireland 2023aNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Department of Tourism, Culture, Arts, Gaeltacht, Sport and Media, Government of Ireland 2023bNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The vision for Ard-Intleacht is an AI assistant that can comprehend and respond in Irish across dialects, perform real-time Irish speech recognition and synthesis, handle translation between Irish and other languages, and imbue AI outputs with authentic Irish cultural context.

“The vision for Ard-Intleacht is an AI assistant that can comprehend and respond in Irish across dialects, perform real-time Irish speech recognition and synthesis, handle translation between Irish and other languages, and crucially, imbue AI outputs with authentic Irish cultural context.”

• cites Caollaí 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Te Hiku partnered with NVIDIA to use its open-source NeMo toolkit and GPU hardware, but on Māori terms, with the core mission remaining 'the preservation, promotion and revitalization of te reo Māori' rather than profit.

“Indeed, Te Hiku partnered with NVIDIA to use its open-source NeMo toolkit and GPU hardware, but on Māori terms -the core mission remained “the preservation, promotion and revitalization of te reo Māori” rather than profit.”

• cites Lee 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Google's approach to Guaraní stands in marked contrast to Ireland's government-led ArdIntleacht initiative and to Te Hiku Media's community-governed Māori model.

“This stands in marked contrast to Ireland's government-led ArdIntleacht initiative and to Te Hiku Media's communitygoverned Māori model.”

May combine two claims.

The collaborative model used by NRC to co-develop technology with Indigenous input at every stage is a best-practice template; it resulted in tools that communities actually wanted, such as a Kanyen'kéha (Mohawk) verb conjugator specifically requested by an Indigenous educator.

“Furthermore, the collaborative model used by NRC to co-develop technology with Indigenous input at every stage is a best-practice template.It resulted in tools that communities actually wanted [such as a Kanyen'kéha (Mohawk) verb conjugator, which was specifically requested by an Indigenous educator (Canada NRC 2019)].”

• cites Canada NRC 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

These choices help prevent a scenario where Irish data is extracted to train proprietary AI that Irish people then have to pay to use.

“These choices help prevent a scenario where Irish data is extracted to train proprietary AI that Irish people then have to pay to use.”

May combine two claims.

The IBM Research project with the Nheengatu and Guarani communities in Brazil explicitly had Indigenous collaborators lead and set the terms of the project, pledging not to reuse the data elsewhere without community consent.

“Similarly, the IBM Research project with the Nheengatu and Guarani communities in Brazil explicitly had Indigenous collaborators lead and set the terms of the project, pledging not to reuse the data elsewhere without community consent (IBM Research 2021).”

• cites IBM Research 2021Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

As more Indigenous languages are added to big platforms, activists caution about a new wave of appropriation where respective languages might be used to enrich AI services without yielding commensurate investment back into those language communities.

“Indeed, as more Indigenous languages are added to big platforms, activists caution about a new wave of appropriation where respective languages might be used to enrich AI services without yielding commensurate investment back into those language communities.”

May combine two claims.

Such efforts underscore that Guaraní NLP is possible, but they often rely on a few experts and meagre funding.

“Such efforts underscore that Guaraní NLP is possible, but they often rely on a few experts and meagre funding.”

Classified as probably not a substantial claim. May combine two claims.

The Māori experience offers a template for other Indigenous communities: one that emphasizes sovereignty, cultural values, and longterm empowerment over quick commercial solutions, but with data ownership loopholes to fill.

“Overall, the Māori experience offers a template for other Indigenous communities: one that emphasizes sovereignty, cultural values, and longterm empowerment over quick commercial solutions, but with data ownership loopholes to fill.”

• AssumptionRests on an assumption: Māori advocates have pointed out gaps in New Zealand law for protecting Indigenous data, such as no sui generis IP for data of national taonga/treasures.

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

The inclusion of Inuktitut in Google Translate symbolically elevates Inuktitut's status with an acknowledgement on the world stage of its importance.

“It also symbolically elevates Inuktitut's status with an acknowledgement on the world stage of its importance.”

May combine two claims.

The practice of using Indigenous data to train commercial algorithms without consent has been likened to digital extraction and echoes historic patterns of dispossession and exploitation.

“This practice has been likened to “digital extraction” and indeed “echoes historic patterns of dispossession and exploitation”, as one policy analysis noted in the context of Indigenous data being used to train commercial algorithms without consent (Tu 2025).”

• cites Tu 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Other Indigenous languages with official status (e.g. Māori and Irish) similarly benefit from translation mandates that produce parallel texts; these corpora can seed LLM fine-tuning or MT systems.

“Other Indigenous languages with official status (e.g.Māori and Irish) similarly benefit from translation mandates that produce parallel texts.These corpora can seed LLM fine-tuning or MT systems.”

Probably not a substantial claim: may not pass S3 (load-bearing).

This consultation helped Google's model handle Inuktitut text in both scripts.

“This consultation helped Google's model handle Inuktitut text in both scripts.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

This raises trust issues, as some communities might prefer self-hosted tools to keep their data within their jurisdiction and under community governance.

“This raises trust issues, as some communities might prefer self-hosted tools to keep their data within their jurisdiction and under community governance.”

May combine two claims.

The case studies above, alongside the Irish Gaelic experience, underline several common themes in applying LLMs and AI to Indigenous languages; first, community control and participation are crucial.

“The case studies above, alongside the Irish Gaelic experience, underline several common themes in applying LLMs and AI to Indigenous languages.First, community control and participation are crucial.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Without local control and investment, technology for Guaraní will remain piecemeal and at risk of exploitation.

“The main takeaway is that without local control and investment, technology for Guaraní will remain piecemeal and at risk of exploitation.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Using Google Translate or other AI means interactions (text and voice) are logged on corporate servers, an evident data privacy concern.

“Additionally, there is a surveillance aspect: using Google Translate or other AI means interactions (text and voice) are logged on corporate servers, an evident data privacy concern.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The inclusion of Inuktitut in a global service like Google Translate has clear benefits: Inuit people can now readily translate documents or websites, and non-speakers can access Inuktitut content, aiding everything from health services to cultural exchange.

“The inclusion of Inuktitut in a global service like Google Translate has clear benefits.Inuit people can now readily translate documents or websites, and non-speakers can access Inuktitut content, aiding everything from health services to cultural exchange.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The Digital Plan backs research on speech synthesis, speech recognition, machine translation and other tools to ensure Irish 'keeps pace with major European languages' in the digital age.

“This plan backs research on speech synthesis, speech recognition, machine translation and other tools to ensure Irish “keeps pace with major European languages” in the digital age (Department of Tourism, Culture, Arts, Gaeltacht, Sport and Media, Government of Ireland 2023b).”

• cites Department of Tourism, Culture, Arts, Gaeltacht, Sport and Media, Government of Ireland 2023bNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Google's public materials refer to working with native speakers, teachers, and linguists, but present this as supporting a pre-existing corporate project, vaguely worded for keeping the development team inspired rather than involvement in co-design or shared governance.

“Google's public materials refer to”

• cites MercoPress 2022Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Inuit communities cannot fine-tune Google's translation model for local needs or prevent Google from using the data for other purposes.

“They cannot fine-tune Google's translation model for local needs or prevent Google from using the data for other purposes.”

May combine two claims.

This observation is supported by Hogan-Brun & O'Rourke's (2019) research about minority languages showing resilience, as the survival and growth of these languages stem from local initiative and valuing of diversity.

“This observation is supported by Hogan-Brun & O'Rourke's (2019) research about minority languages showing resilience, as the survival and growth of these languages stem from local initiative and valuing of diversity.”

• cites Hogan-Brun & O'Rourke 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

According to Google, they worked with Inuit translators and the Inuit Tapiriit Kanatami (ITK) to verify translation quality and address issues like the two different writing systems.

“According to Google, they worked with Inuit translators and the Inuit Tapiriit Kanatami (ITK, the national Inuit organization) to verify translation quality and address issues like the two different writing systems (Latin alphabet qaliujaaqpait and syllabics qaniujaaqpait).”

• cites News 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Te Hiku's ASR model transcribes te reo Māori with 92% accuracy, rivaling commercial systems.

“The result was an ASR model that transcribes te reo Māori with 92% accuracy, rivaling commercial systems.”

• cites Lee 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The existence of a large bilingual corpus was a game-changer; it shows the value of governments adopting policies (like bilingual records, or open data) that inadvertently create the raw material for language tech.

“The existence of a large bilingual corpus was a game-changer.It shows the value of governments adopting policies (like bilingual records, or open data) that inadvertently create the raw material for language tech.”

May combine two claims.

Te Hiku's project embodied Indigenous data governance, as the Māori community retained ownership of the speech corpus and the models, and they set the terms for any external collaboration.

“In practice, Te Hiku's project embodied Indigenous data governance, as the Māori community retained ownership of the speech corpus and the models, and they set the terms for any external collaboration.”

May combine two claims.

Although Google's addition of Guaraní to Google Translate was the first appearance of Guaraní in a major translation application and a notable technical achievement, it was also largely an external, top-down inclusion.

“Although this was the first appearance of Guaraní in a major translation application and a notable technical achievement, it was also largely an external, top-down inclusion.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The Māori project rejected Big-Tech, top-down hegemony.

“Similar to the Irish AI efforts, the Māori project also rejected Big-Tech, top-down hegemony.”

Probably not a substantial claim: may not pass S2 (checkable).

The risk in this context is that Guaraní might become an object of study or a checkbox for AI companies without empowering Guaraní communities to develop their own AI solutions.

“The risk in this context is thus that Guaraní might become an object of study or a checkbox for AI companies without empowering Guaraní communities to develop their own AI solutions.”

May combine two claims.

The State's role in Ard-Intleacht brings funding and access to the State's existing pool of Irish language content, including archive audio and video and text corpora.

“The State's role in Ard-Intleacht also brings funding and access to “the State's existing pool of Irish language content, which includes archive audio and video” and text corpora (Caollaí 2024).”

• cites Caollaí 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The Irish Gaelic AI efforts are conscious of safeguarding data, but the application of an Anglophone LLM suggests these maneuvers may be susceptible to influences of colonial or Anglophone ideologies.

“The Irish Gaelic AI efforts are conscious of the issue of safeguarding data, but the application of an Anglophone LLM suggests that these maneuvers may be susceptible to influences of colonial or Anglophone ideologies.”

May combine two claims.

Te Hiku Media is a Māori-owned media organization that expanded from radio broadcasting into AI development.

“A standout example is the work of Te Hiku Media, a Māori-owned media organization that expanded from radio broadcasting into AI development.”

• cites Lee 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Under the OCAP framework, access and possession of data are as important as ownership, meaning communities must have the practical ability to get and use their data whenever they want.

“Under the OCAP framework, access and possession of data are as important as ownership, meaning communities must have the practical ability to get and use their data whenever they want.”

May combine two claims.

Irish Gaelic is an Indigenous language of its own land now at risk, sustained by community efforts and policy support, but threatened by the dominance of English.

“In effect, Irish Gaelic is an Indigenous language of its own land now at risk, sustained by community efforts and policy support, but threatened by the dominance of English.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

By investing in a bespoke Irish AI, the government aims to safeguard the position of Irish in the digital realm and ensure the bilingual public service goal by 2030 is bolstered by technology.

“By investing in a bespoke Irish AI, the government aims to safeguard the position of Irish in the digital realm and ensure the bilingual public service goal by 2030 is bolstered by technology.”

• cites Caollaí 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

The community's voice needs to be central, or else Guaraní could be included in AI in form but not in spirit: the language will be reduced to a token presence, rather than a living, evolving language.

“The community's voice needs to be central, or else Guaraní could be included in AI in form but not in spirit: the language will be reduced to a token presence, rather than a living, evolving language.”

Probably not a substantial claim: may not pass S3 (load-bearing).

From a data-colonialism perspective, publicly available Guaraní texts were appropriated to fuel proprietary models now accessed on corporate terms.

“From a data-colonialism perspective, one could further argue that publicly available Guaraní texts were appropriated to fuel proprietary models now accessed on corporate terms.”

Probably not a substantial claim: may not pass S2 (checkable).

The analytically sound frame for Indigenous claims is stewardship and governance over datasets, annotations, alignments, and trained models that encode collective knowledge and risk.

“Consequently, the analytically sound frame for Indigenous claims is not property in the language itself but stewardship and governance over datasets, annotations, alignments, and trained models that encode collective knowledge and risk.”

May combine two claims. The quote only partly states this claim.

Google's announcement foregrounds that the techniques introduced supplement massively multilingual models with a self-supervised task to enable zero-resource translation, emphasizing methodological innovation and scale rather than community-defined aims.

“By contrast, Google's announcement foregrounds that the techniques introduced”

• cites Google Research 2022Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

When outsiders unilaterally control Indigenous data, they repeat the injustices of colonialism; when Indigenous groups direct how their data are used, they can empower sustainable development, cultural preservation, and innovation on their own terms.

“When outsiders unilaterally control Indigenous data, they repeat the injustices of colonialism; conversely, when Indigenous groups direct how their data are used, they can empower sustainable development, cultural preservation, and innovation on their own terms.”

Probably not a substantial claim: may not pass S2 (checkable).

This infringes on the ideal that Inuit should own and control their language data as per data sovereignty standards and practices.

“This infringes on the ideal that Inuit should own and control their language data as per data sovereignty standards and practices.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

By open-sourcing the Irish LLM and benchmarks, the researchers ensured that the Irish public and scholars collectively benefit, a step showing awareness of open data principles and arguably with Indigenous data governance.

“By open-sourcing the Irish LLM and benchmarks, the researchers ensured that the Irish public and scholars collectively benefit-a step showing awareness of open data principles and arguably with Indigenous data governance, since no single company can lock up the model or its underlying Irish corpus.”

May combine two claims.

In 2022, Google added Guaraní to Google Translate as one of twenty-four under-resourced languages, using a zero-shot approach trained only on monolingual text given the scarcity of parallel corpora.

“In 2022, Google added Guaraní to Google Translate as one of twenty-four under-resourced languages, using a zero-shot approach trained only on monolingual text given the scarcity of parallel corpora (Google Research 2022).”

• cites Google Research 2022Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Google's model was trained on publicly available Inuit data, but now the model itself is proprietary to Google, and Inuit communities do not have direct control over it.

“Google's model was undoubtedly trained on publicly available Inuit data, but now the model itself is proprietary to Google.Inuit communities do not have direct control over it.”

• AssumptionRests on an assumption: Google's Inuktitut translation model was trained on publicly available Inuit data.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

IBM researchers found that open-sourcing the tools for Indigenous language AI is easier than open-sourcing the data, since the latter raises questions of who owns a language and how to protect community rights.

“For example, IBM researchers found that opensourcing the tools for Indigenous language AI is easier than open-sourcing the data, since the latter raises questions of who owns a language and how to protect community rights.”

• cites IBM Research 2021Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Replicability faces practical hurdles; many communities lack the funding, high-quality data, or expert teams to replicate the experience of Ireland or New Zealand.

“However, replicability faces practical hurdles.Many communities lack the funding, high-quality data, or expert teams to replicate the experience of Ireland or New Zealand.”

• AssumptionRests on an assumption: Many Indigenous communities lack the funding, high-quality data, or expert teams to replicate the experience of Ireland or New Zealand.

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

Solidarity and knowledge-sharing will therefore become important; networks such as the Indigenous AI network and workshops (e.g., the ACL Workshop on Indigenous Languages) help spread these ideas.

“Solidarity and knowledge-sharing will therefore become important.Networks such as the Indigenous AI network and workshops (e.g., the ACL Workshop on Indigenous Languages) help spread these ideas.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

An extractive approach sees data as gold to be mined; this can lead to exploitation and a loss of trust.

“An extractive approach, by contrast, sees data as gold to be mined.This can lead to exploitation and a loss of trust.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Implementing these rights in the AI space might mean requiring Free, Prior and Informed Consent (FPIC) before using a community's language data in a system, or ensuring that any AI product involving an Indigenous language has an oversight committee from that community.

“Implementing these rights in the AI space might mean requiring Free, Prior and Informed Consent (FPIC) before using a community's language data in a system, or ensuring that any AI product involving an Indigenous language has an oversight committee from that community.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

There is a clear difference between relational and extractive ethics in data; a relational approach, inspired by Indigenous worldviews, treats language data as more than a commodity: it is a taonga or treasure, imbued with the soul of the people who speak it.

“Second, there is a clear difference between relational and extractive ethics in data.A relational approach, inspired by Indigenous worldviews, treats language data as more than a commodity: it is a taonga or treasure, imbued with the soul of the people who speak it (Hernández 2025).”

• cites Hernández 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable).

Without ethical guardrails, even well-intentioned AI can end up recolonizing knowledge.

“Without ethical guardrails, even well-intentioned AI can end up recolonizing knowledge.”

May combine two claims.

Model replicability and accessibility emerged as both a hope and a challenge; the success of one language's LLM (like UCCIX or Te Hiku's ASR) can serve as a template, allowing adaptation by other languages with the right resources and expertise provided.

“Third, model replicability and accessibility emerged as both a hope and a challenge.The success of one language's LLM (like UCCIX or Te Hiku's ASR) can serve as a template, allowing adaptation by other languages with the right resources and expertise provided.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The cases highlight an evolving role for policy and governance; governments can be enablers, but they need to work in true partnership with Indigenous communities to avoid top-down control.

“Finally, the cases highlight an evolving role for policy and governance.Governments can be enablers, but they need to work in true partnership with Indigenous communities to avoid top-down control.”

Probably not a substantial claim: may not pass S2 (checkable).

The future of Gaelic, Māori, Guaraní, Inuktitut, and other Indigenous languages in the digital sphere should not be left to chance or solely to corporate algorithms.

“Whether it is Gaelic, Māori, Guaraní, or Inuktitut, the future of these and other Indigenous languages in the digital sphere should not be left to chance or solely to corporate algorithms.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The comparative perspective reinforces the central thesis emerging from the Irish case: LLMs for Indigenous and endangered languages hold great promise for revitalization, but realizing that promise without peril requires aligning AI development with the self-determination and values of the language communities.

“In conclusion, the comparative perspective reinforces the central thesis emerging from the Irish case: LLMs for Indigenous and endangered languages hold great promise for revitalization, but realizing that promise without peril requires aligning AI development with the self-determination and values of the language communities.”

Probably not a substantial claim: may not pass S2 (checkable).

By learning from each other's successes and missteps, and by centering principles like Indigenous data sovereignty, AI can realistically be harnessed for encouraging linguistic diversity and not against it.

“By learning from each other's successes and missteps, and by centering principles like Indigenous data sovereignty, AI can realistically be harnessed for encouraging linguistic diversity and not against it.”