弥合语言鸿沟：移动网络运营商在AI生态系统中的角色

The Role of Mobile NetworkOperators in Al Ecosystems GSMA EmergingTech Programme The GSMA is a global organisation unifying the mobileecosystem to discover, develop and deliver innovationfoundational to positive business environments andsocietal change. Our vision is to unlock the full power ofconnectivity so that people, industry, and society thrive.Representing mobile operators and organisations acrossthe mobile ecosystem and adjacent industries, the GSMAdelivers for its members across three broad pillars:Connectivity for Good, Industry Services and Solutions,and Outreach. This activity includes advancingpolicy, tackling today’s biggest societal challenges,underpinning the technology and interoperability thatmake mobile work, and providing the world’s largestplatform to convene the mobile ecosystem at the MWCand M360 series of events. The GSMAEmergingTech programmeaccelerates impactand climate action by fostering the adoption of AI andemerging technologies in low- and middle-income countries(LMICs) by working with public, private and third sectorinnovators to develop scalable and sustainable solutionsthat have inclusive and responsible AI at the core. TheEmerging Tech programme works closely with the GSMAAI for Impactinitiative to drive real-world, impact-focusedimplementation with telcos in LMICs. To get in touch with the Emerging Tech team, please email:emergingtech@gsma.com We invite you to find out more atwww.gsma.com Authors: Eugénie Humeau, GSMA Mobile for DevelopmentZarah Udwadia, GSMA Mobile for Development This material has been funded by UK InternationalDevelopment from the UK government and is supported bythe GSMA and its members. The views expressed do notnecessarily reflect the UK Government’s official policies. Contributors: Kimberly Brown, GSMA Mobile for DevelopmentIbrahim Sajid, GSMA Mobile for DevelopmentMaureen Imiegha, GSMA Mobile for Development(Marketing) Acknowledgements: We would like to thank the many individuals andorganisations that contributed to this research. This includes Digital Umuganda, gheero, GIZ FAIR Forward– Artificial Intelligence for All, Indosat Ooredoo Hutchison,Karya, Mozilla Foundation, Pindo, Reverie LanguageTechnologies and RAIght.ai. Published:March 2026 Contents Definitions4 Acronyms and abbreviations5 List of figures, spotlights and tables6 1.1.The language divide111.2.The importance of cultural and linguistic diversity151.3.The opportunity: models in local languages161.4.Research objectives17 2.Insights from the ecosystem182.1.Existing initiatives and approaches192.2.Challenges faced by local language initiatives222.3. Implications for digital sovereignty24 3.Mobile network operators and local language AI253.1.AI adoption trends263.2.Case studies281. Orange:Supporting Senegal’s customers in local languages282. Dialog Axiata:Creating inclusive digital services in Sri Lanka303.Beeline (VEON Group):Bridging the AI language gap in Kazakhstan 334.Indosat:Building sovereign AI for Indonesia37 4.Lessons and implications414.1.Key lessons from the case studies424.2.Pathways for MNOs to contribute to local language AI464.3. Conclusion49 Definitions Acronyms and abbreviations List of figures List of spotlights List of tables Executive summary Language remains one of the biggest barriers to the equitabledevelopment of artificial intelligence (AI) in low- and middle-incomecountries (LMICs). The digital world is dominated by a small numberof “high-resource” languages, particularly English, with abundantdigital data resources available. The vast majority of the world’slanguages, by contrast, are “low resource” and lack the machine-usable data that can be used for training natural language processing(NLP) models, particularly large language models (LLMs), whichrequire massive amounts of data. incentives around shared objectives for languageinclusion, national priorities and scale. Third, someMNOs are emerging as sovereign AI enablers, investingin compute, cloud platforms and model-hostingenvironments that position local language AI as part ofbroader national digital infrastructure. In this pathway,MNOs do not just deploy AI in their own services butprovide the infrastructure layer that enables bothprivate-sector innovation and public-sector digitaltransformation. Models trained on data that does not represent theworld’s vast linguistic and cultural diversity are notaccessible, relevant, reliable or impactful for peoplewho live their lives in low-resource languages. This riskswidening existing digital divides while also threateningthe preservation of languages across the world. A growing number of efforts are addressing thislinguistic imbalance. Startups, innovators, researchersand communities in LMICs are building and applyinglocally relevant AI models, curating and crowdsourcinglinguistically diverse datasets and creating enablingenvironments for greater AI language inclusion. However,these efforts are operationally demanding, resourceinte