{
  "id": "wikimedia-foundation",
  "title": "Wikimedia Foundation",
  "lang": "en",
  "source_lang": "en",
  "summary": "The Wikimedia Foundation is the nonprofit organization that hosts Wikipedia and related free-knowledge projects, providing infrastructure, legal support, fundraising, and coordination.",
  "category": "computer-science",
  "subcategory": "software",
  "type": "organization",
  "aliases": [
    "WMF",
    "Wikimedia Foundation, Inc."
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/wikimedia-foundation/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/wikimedia-foundation/",
      "json": "https://aiwiki.page/zh/wikimedia-foundation.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The Wikimedia Foundation is a United States–based nonprofit organization that operates Wikipedia and a group of related collaborative knowledge projects. Established in 2003, it provides the technical infrastructure, institutional support, and legal framework through which volunteers create and maintain freely reusable educational resources. It is incorporated in Florida, has its headquarters in San Francisco, and qualifies as a tax-exempt charitable organization under section 501(c)(3) of the United States Internal Revenue Code.\n\nThe foundation is distinct from both Wikipedia’s editing communities and the wider network of independent organizations that support Wikimedia activities. Its responsibilities include operating websites, developing software, raising funds, administering grants, protecting trademarks, and responding to legal threats. Most encyclopedia articles and other project content are written and reviewed by volunteers rather than foundation employees. This division between institutional operation and community editorial authority is central to Wikimedia’s organizational structure."
    },
    {
      "heading": "Origins and institutional development",
      "text": "Wikipedia began on 15 January 2001 as an openly editable encyclopedia associated with Nupedia, an earlier encyclopedia project supported by Jimmy Wales’s company, Bomis. Wales and Larry Sanger played central roles in Wikipedia’s establishment. Unlike Nupedia’s comparatively formal review process, Wikipedia used a wiki interface that allowed contributors to revise pages directly. Its rapid growth created requirements for hosting, administration, and stewardship that increasingly exceeded the arrangements of a small privately supported website.\n\nThe Wikimedia Foundation was incorporated in Florida on 20 June 2003. Its creation gave Wikipedia and related projects a separate charitable organization capable of accepting donations, holding assets, and entering contracts. It also established an institutional distinction between the projects and Bomis. The name “Wikimedia” identified a broader family of educational resources rather than an encyclopedia alone: Wiktionary had already begun in 2002, and additional projects were developing around source texts, textbooks, and quotations.\n\nDuring its early years, the foundation operated with a small staff and substantial volunteer assistance. Growing traffic required increasingly dependable servers, fundraising systems, and administrative procedures. The organization’s original base was in St. Petersburg, Florida; it moved its headquarters to San Francisco in 2008. As it professionalized, it added functions such as financial management, software engineering, grant administration, communications, and legal counsel. The transition did not replace volunteer authorship, but it enlarged the paid organization supporting that authorship.\n\nLeadership changes accompanied this expansion. Sue Gardner became executive director in 2007 and served until 2014. Lila Tretikov followed, and Katherine Maher subsequently led the organization, becoming chief executive officer in 2019. Maryana Iskander took office as chief executive officer in January 2022. These appointments concerned management of the foundation, not an editor-in-chief role for Wikipedia: the chief executive does not direct the ordinary editorial decisions of the encyclopedia’s language communities.\n\nThe distinction between foundation and movement became more pronounced as independent national and thematic organizations developed. Wikimedia Deutschland, founded in 2004, was an early national chapter and later became a major contributor to software development, including Wikidata. Other chapters, thematic organizations, and user groups developed their own memberships and programs. Their recognition as Wikimedia affiliates does not make them branch offices or subsidiaries of the foundation. The result is a distributed institutional network in which hosting, local outreach, software development, and editorial work may be performed by different organizations or communities."
    },
    {
      "heading": "Hosted projects and their relationships",
      "text": "Wikipedia is the foundation’s best-known project, but the Wikimedia portfolio encompasses several kinds of educational material. Wikipedia organizes explanatory articles into separate language editions, each with its own community processes. These editions are not simply translations of a central encyclopedia: articles, inclusion decisions, sourcing practices, and editorial priorities develop independently within shared technical and legal arrangements. Language editions can therefore differ substantially in size and coverage.\n\nSeveral sister projects separate tasks that would not fit an encyclopedia’s purpose. Wiktionary contains dictionary and linguistic information; Wikisource provides source texts that are in the public domain or available under compatible free licenses; and Wikibooks develops textbooks and instructional materials. Wikiquote collects attributed quotations, while Wikivoyage, which joined Wikimedia in 2013, provides travel guides. Wikinews publishes collaboratively produced news, Wikiversity supports learning resources and activities, and Wikispecies concentrates on taxonomic information. Their different purposes require distinct content policies, even where they share software.\n\nWikimedia Commons, launched in September 2004, supplies a shared repository for images, audio, video, and other media. A file uploaded there can be embedded in multiple Wikimedia projects without maintaining a separate copy in each language edition. Commons records information such as authorship, source, licensing, and descriptions, making it both a distribution system and a repository of rights-related documentation. Its scope is not identical to every local Wikipedia’s file policy: some editions permit limited local use of non-free material that Commons does not accept.\n\nWikidata, launched in October 2012, provides a collaboratively maintained structured knowledge base. Rather than storing only prose, it represents entities through statements, properties, qualifiers, and references. A person, for example, may be associated with a date of birth, an occupation, and identifiers used by external databases. Its multilingual labels and links connect pages about the same subject across projects. Software can also retrieve its data for infoboxes, queries, and external applications, although the presence of a statement in Wikidata does not automatically establish its reliability.\n\nOther Wikimedia sites support collaboration rather than serving primarily as reference works. Meta-Wiki hosts movement-wide discussions, coordination, and documentation. MediaWiki.org documents the software platform and its extensions. Wikimedia Incubator provides a workspace for developing certain new language editions before they receive separate sites. These support environments make the portfolio an interconnected publishing system: shared media, structured data, software documentation, and organizational coordination underpin the more visible educational projects."
    },
    {
      "heading": "Governance and the wider Wikimedia movement",
      "text": "The foundation’s legal authority rests with its Board of Trustees. The board has fiduciary responsibilities for the organization’s assets, mission, and long-term direction. It approves major policies and budgets, appoints and oversees the chief executive, and works through committees on matters such as audit, governance, and talent and culture. These corporate responsibilities are distinct from reviewing individual encyclopedia articles or choosing which interpretation an article should present.\n\nUnder the board structure established through reforms beginning in 2020, its authorized composition can reach 16 trustees: up to eight community- and affiliate-selected seats, up to seven board-selected seats, and one founder seat. The actual number of serving trustees can vary as vacancies and selection cycles occur. Community selection processes provide an important channel of participation, but trustees serve the foundation’s charitable purposes rather than acting simply as delegates bound to particular voters or national organizations.\n\nThe chief executive and staff implement the foundation’s operational responsibilities. Staff functions include product development, engineering, fundraising, legal affairs, finance, communications, and support for volunteers and affiliates. Annual plans describe intended work and allocate resources, generally around the foundation’s fiscal year from July through June. Public consultation on plans and strategies allows volunteers to comment, but consultation does not eliminate the board’s legal authority or management’s responsibility for operational decisions.\n\nThe wider Wikimedia movement includes contributors, recognized affiliates, informal groups, technical volunteers, and partner institutions. Chapters usually organize activities associated with a country or territory. Thematic organizations work around a subject or other non-geographical focus. User groups provide a more flexible form of recognition, often suited to smaller or newer communities. Affiliates may arrange editing workshops, photography projects, institutional partnerships, or advocacy, while maintaining their own legal and financial responsibilities where separately incorporated.\n\nCoordination across these bodies is negotiated rather than governed through a single chain of command. Affiliates do not ordinarily instruct local Wikipedia administrators, and language communities are not departments of the foundation. Proposed changes can consequently raise questions about which body has authority, who is represented, and how decisions should be reviewed. The Movement Charter process illustrated these difficulties: a draft intended to define movement-wide roles and structures underwent ratification voting in 2024 but was not ratified. The process exposed substantive differences over the allocation of authority and resources among the foundation, affiliates, and volunteer communities."
    },
    {
      "heading": "Editorial independence and community administration",
      "text": "The foundation operates publishing platforms, but routine content decisions are made through project communities. On Wikipedia, volunteers draft articles, examine sources, discuss disputed wording, and apply local policies. Principles such as verifiability, neutral presentation, and restrictions on original research are implemented through community processes rather than through a centrally employed editorial staff. Other projects use different standards appropriate to dictionaries, source texts, media repositories, or structured data.\n\nA typical editorial dispute begins with revisions and discussion on an article’s talk page. Depending on the project, unresolved questions may proceed to noticeboards, requests for comment, mediation-like processes, or specialized dispute-resolution bodies. On some large editions, arbitration committees address serious conduct disputes. Their function is generally to regulate behavior and participation, not to certify the truth of every statement in an article. Smaller projects may rely more heavily on global assistance because they lack enough active participants to maintain equivalent institutions.\n\nTechnical permissions are distributed among community members. Administrators can perform actions such as deleting pages, blocking accounts, and protecting pages from editing, subject to local rules. Bureaucrats and other permission holders perform additional tasks where their projects authorize them. Global stewards can handle certain cross-project permissions and emergencies. These roles use access control mechanisms, but administrative status does not make a volunteer an employee or give that person unlimited editorial authority.\n\nThe foundation retains responsibilities that cannot be delegated entirely to local consensus. It can act in response to legal obligations, serious safety concerns, threats, or violations of global terms. Some actions involve private information that cannot appropriately be disclosed in a public discussion. Such interventions differ from ordinary editing decisions, but their boundaries have sometimes been contested. The temporary foundation ban of an English Wikipedia administrator in 2019, commonly associated with the “Fram” dispute, prompted substantial disagreement about institutional authority and procedural safeguards.\n\nA Universal Code of Conduct, approved by the board in 2021, established baseline expectations for behavior across Wikimedia spaces. Enforcement guidelines were subsequently approved in 2023 after community voting and revision. The framework addresses matters such as harassment, intimidation, and abuse while interacting with existing local rules. Its implementation illustrates a recurring institutional problem: establishing minimum protections across projects with different languages and governance traditions without unnecessarily displacing effective local processes. Public rules, appeal mechanisms, and the division of cases among local and global bodies are therefore significant elements of Wikimedia administration."
    },
    {
      "heading": "Technical infrastructure and software development",
      "text": "The foundation’s central operational task is keeping a large, multilingual publishing platform available and editable. Its sites depend principally on MediaWiki, a software system originally developed for Wikipedia and released as free software. MediaWiki manages page revisions, discussion interfaces, user permissions, templates, and extensions. Because the software is publicly available, it is also used by organizations unrelated to Wikimedia; conversely, operating a MediaWiki site does not make that site a Wikimedia project.\n\nWikimedia’s production environment combines databases, application services, caching layers, search systems, media storage, and monitoring. Pages that are requested repeatedly can be served through caches rather than reconstructed from underlying data on every request. A geographically distributed content delivery network reduces delays for readers and limits pressure on central infrastructure. Load balancing distributes work among servers, while multiple data centers and tested failover procedures reduce dependence on a single location.\n\nThe storage model must preserve both current content and extensive revision histories. A Wikipedia article is not merely one text document: it can have thousands of historical versions, associated discussion pages, templates, categories, and links. Relational database systems, including MariaDB, support much of this state. Search indexes and other derived systems provide functions that would be inefficient to obtain directly from the primary databases. Media files introduce additional storage and delivery requirements, particularly for large images, audio, and video.\n\nThe software environment also supports automated and semi-automated work. Bots perform tasks such as identifying certain forms of vandalism, maintaining links, and updating structured information, normally under applicable community approval rules. Volunteer developers use application programming interfaces to build editing tools, research applications, and maintenance systems. Toolforge and Wikimedia Cloud Services provide infrastructure for many community-developed tools, with operational arrangements distinct from the core production websites.\n\nChanges to user-facing software must accommodate an unusually diverse environment. A feature may need to function across hundreds of languages, multiple writing systems, mobile and desktop devices, and communities with different workflows. Unicode handling, right-to-left interfaces, accessibility, and translation are practical engineering requirements. Projects such as VisualEditor and the Parsoid parsing system sought to make editing less dependent on direct manipulation of wikitext while preserving compatibility with complex existing pages.\n\nSecurity and reliability are connected to the projects’ openness. The foundation must manage software vulnerabilities, abusive traffic, account compromise, and large-scale scraping without preventing legitimate public access. Its open development processes make much code and technical documentation inspectable, but sensitive security information and personal data require restricted handling. Public incident reports and service documentation provide visibility into operations without making every operational detail public."
    },
    {
      "heading": "Free licensing, ownership, and reuse",
      "text": "Wikimedia’s mission depends on making educational content reusable, not merely readable without payment. Most Wikipedia text is available under a Creative Commons Attribution-ShareAlike license, subject to the terms applicable to the particular content. This permits copying, redistribution, and adaptation, including commercial reuse, provided the license conditions are met. Attribution and share-alike requirements distinguish this arrangement from an unrestricted dedication to the public domain.\n\nThe legal structure developed over time. Wikipedia originally relied on the GNU Free Documentation License, a license designed primarily for documentation. In 2009, Wikimedia carried out a major licensing transition that made eligible text available under Creative Commons Attribution-ShareAlike 3.0, while much text also remained available under the earlier license. Following the 2023 update to Wikimedia’s Terms of Use, the principal Creative Commons license for contributed text became Attribution-ShareAlike 4.0. The licensing history of particular material still matters when determining its permitted reuse.\n\nContributors generally retain copyright in their original contributions while licensing them under the applicable project terms. The foundation does not acquire universal ownership of everything volunteers submit. Nor can a contributor validly grant rights over material copied without permission from an incompatible source. Community review, deletion procedures, and legal notice processes consequently play important roles in identifying copyright violations. A page’s availability on a Wikimedia site is not by itself proof that every element was uploaded lawfully.\n\nDifferent projects and media types have different rights arrangements. Wikidata’s structured data is released under the Creative Commons CC0 public-domain dedication. Commons accepts material that is freely licensed or in the public domain under its rules, but files can carry different licenses and attribution requirements. Some Wikipedia editions host limited non-free content under local exemption policies and applicable law. Such a file may be viewable in an article without being freely reusable elsewhere.\n\nCopyright also differs from trademark and other legal restrictions. The foundation controls Wikimedia names and logos through trademark policies designed to identify the organization and its projects. A freely licensed article does not authorize a third party to suggest that its product is officially endorsed by Wikipedia. Photographs may additionally implicate privacy, personality, or other rights that a copyright license does not resolve.\n\nThe foundation’s licensing framework supports educational publishing, translation, offline distribution, and reuse by other digital services. It also means the foundation cannot normally promise exclusive access to ordinary project content: compatible reuse is part of the system’s design. The practical challenge is preserving attribution, provenance, and license information as material moves through increasingly complex digital distribution chains."
    },
    {
      "heading": "Fundraising, expenditure, and financial stewardship",
      "text": "The foundation’s operations are funded primarily through donations rather than reader subscriptions or conventional advertising on encyclopedia pages. Fundraising banners, email appeals, recurring contributions, major gifts, and grants all contribute to revenue. Small online donations are a prominent part of this model. Payment processing, donor communication, fraud prevention, and compliance across jurisdictions make fundraising itself a substantial operational function.\n\nThe foundation publishes annual reports, audited financial statements, and United States tax filings. These documents serve different purposes. Audited statements report financial position and activity under accounting standards; the Internal Revenue Service Form 990 provides standardized information about finances, compensation, governance, and activities; annual plans describe prospective priorities and budgets. Amounts in these documents are not always directly interchangeable because budgets, cash receipts, investment changes, and recognized accounting revenue measure different things.\n\nExpenditure extends beyond purchasing servers. It includes engineering and product staff, legal work, fundraising, finance, administrative systems, security, community support, and grants. Much of the expense of operating a global collaborative service is personnel rather than hardware. At the same time, much content production takes place outside the foundation’s payroll: volunteer writing, reviewing, photography, and administration are not represented as employee expenditure.\n\nThe Wikimedia Endowment, launched in 2016, was established to provide long-term support for Wikimedia projects. It reached its initial fundraising goal of US$100 million in 2021. An endowment differs from an annual operating reserve because its purpose is sustained support over an extended period, governed by investment and distribution arrangements. Endowment assets and the foundation’s immediately spendable operating resources therefore require separate interpretation. The existence of a large long-term fund does not mean that its entire balance is treated as the current year’s operating budget.\n\nA separate revenue initiative, Wikimedia Enterprise, launched commercially in 2021, offers services designed for organizations that reuse Wikimedia content at scale. These include delivery and support arrangements suited to commercial search, information services, and other large consumers. The service does not turn Wikipedia articles into exclusive proprietary material: the underlying free licenses remain in effect, and ordinary access routes continue to exist. The distinction is between paying for a service level and paying for exclusive ownership of the knowledge.\n\nFundraising practices have attracted discussion within and outside Wikimedia, particularly over the tone of donation appeals, reserve levels, and the distribution of resources between the foundation and communities. Financial evaluation requires separating these questions. The cost of operating the websites, the desired size of reserves, investment in future capabilities, and the wording used to solicit donations are related but analytically distinct matters."
    },
    {
      "heading": "Grants, partnerships, and participation",
      "text": "The foundation supports contributors not only by operating websites but also through grants and programs. Funding may support affiliate operations, community events, technical work, content improvement, or efforts to increase participation. Grant mechanisms have changed over time, but a recurring objective is to direct resources toward work that cannot be sustained entirely through unpaid contributions. This can include coordination, training, travel, equipment, and locally appropriate organizational support.\n\nPartnerships with galleries, libraries, archives, and museums are often described through the acronym GLAM. Such institutions can contribute digitized collections, specialist knowledge, catalog information, and access to source materials. A museum might release photographs of objects under a compatible license, while volunteers improve related articles and add identifiers to Wikidata. These activities are frequently organized by affiliates or community groups rather than by foundation staff directly. Institutional collaboration does not automatically override a project’s sourcing or neutrality requirements.\n\nEducational programs connect Wikimedia participation with schools and higher education. Students may improve articles, translate material, or evaluate sources as part of supervised coursework. The educational value lies partly in learning how public knowledge is documented, challenged, and revised. However, assignments must operate within project rules: participation in a university course does not guarantee that an article will be retained or that its claims satisfy an encyclopedia’s sourcing standards.\n\nContent coverage is shaped by inequalities in access to sources, time, skills, and connectivity. The digital divide affects who can contribute consistently, particularly where connections are expensive or unreliable. Published source availability also matters. Communities working in languages with fewer digitized references may face difficulties that software translation alone cannot solve. Reliable coverage depends on the existence of usable evidence as well as the availability of an editing interface.\n\nWikimedia’s movement strategy process, launched in 2017, emphasized “knowledge equity” and “knowledge as a service.” The first phrase concerned participation and gaps in representation; the second concerned making knowledge available through useful infrastructures and interfaces. These goals influenced discussions of regional funding, capacity building, and technical development. They did not establish a single centrally administered content plan for every project.\n\nEfforts to broaden participation also involve retention and safety. Newcomers must learn editing conventions, citation requirements, copyright rules, and local discussion practices. Mentorship, newcomer tools, community training, and anti-harassment processes address different parts of this problem. Measuring success is difficult: account creation, first edits, sustained contribution, and the production of well-supported content are different outcomes. A program that recruits many participants does not necessarily produce a durable editing community."
    },
    {
      "heading": "Legal responsibilities, privacy, and access restrictions",
      "text": "The foundation occupies a legal position different from that of an individual author or a traditional publisher employing its writers. It hosts material contributed by users while maintaining terms of use and processes for legal complaints. In the United States, its activities intersect with intermediary protections such as Section 230 of the Communications Decency Act and copyright procedures under the Digital Millennium Copyright Act. These frameworks have different scopes, exceptions, and requirements; neither is a general exemption from all legal obligations.\n\nLegal requests may concern alleged defamation, copyright infringement, disclosure of user information, or removal of content. The foundation evaluates such requests under applicable law and its policies, and publishes transparency reporting about categories of demands. A complaint to the foundation is not identical to an editorial request on an article’s discussion page. Community mechanisms can correct unsupported content independently of whether a legal demand is valid, while a binding legal order can create obligations beyond ordinary editorial procedures.\n\nPrivacy is especially important because participation can expose contributors to personal or political risks. Registered users may edit under pseudonyms, but technical records can still contain information relevant to security and abuse investigations. Access to nonpublic personal information is restricted through policies and permissions. Public revision histories, meanwhile, preserve extensive records of contributions. This creates a continuing tension between maintaining the integrity of a collaborative historical record and limiting unnecessary disclosure of personal information.\n\nGovernments and network operators can restrict access even when the foundation does not remove content. Turkey blocked Wikipedia in April 2017. In December 2019, Turkey’s Constitutional Court ruled that the block violated freedom of expression, and access was restored in January 2020. The case demonstrated that a project’s availability depends not only on its hosting infrastructure but also on national legal decisions and network-level controls.\n\nTechnical choices can affect these conflicts. Wikimedia adopted mandatory HTTPS for its sites in 2015, encrypting connections between readers and Wikimedia servers. Encryption protects traffic from some forms of observation and alteration, but it does not prevent a state or network operator from blocking a domain or service. Availability, confidentiality, and resistance to selective interference are related technical concerns rather than interchangeable guarantees.\n\nThe foundation also engages in public-policy advocacy concerning copyright, intermediary liability, surveillance, and access to knowledge. Its legal work can include defending the organization, supporting eligible contributors, and challenging measures that affect the projects’ operation. The practical reach of these efforts remains jurisdiction-dependent: a United States nonprofit can provide institutional support, but contributors and readers remain subject to differing legal environments around the world."
    },
    {
      "heading": "Large-scale reuse, artificial intelligence, and institutional boundaries",
      "text": "Wikimedia content circulates far beyond its own websites. Search engines use encyclopedia material to support information panels and direct answers; applications reuse Commons files; researchers analyze editing histories; and organizations integrate Wikidata into information systems. Public data dumps and APIs support these uses. Consequently, Wikimedia’s influence cannot be measured solely through visits to its web pages: information may reach users through systems that do not visibly resemble a Wikimedia project.\n\nThis distribution creates attribution and sustainability questions. Free licenses permit extensive reuse, but downstream services must still comply with applicable conditions. Summaries, transformed text, and machine-generated answers can make provenance difficult for users to identify. At the same time, readers who obtain answers elsewhere may not encounter donation appeals or opportunities to edit. The foundation’s interest in reliable reuse services therefore intersects with the economic and social arrangements that sustain the original projects.\n\nArtificial intelligence adds both production tools and new forms of reuse. Wikipedia has been used in training datasets and evaluation resources for natural language processing. Its structured articles, links, multilingual coverage, and revision records make it useful for language and information-processing research. However, public availability does not remove licensing obligations, and inclusion in a dataset does not establish that every statement is accurate or suitable for every downstream task.\n\nWithin Wikimedia, machine-learning tools have supported activities such as identifying potentially damaging edits and assisting content assessment. These systems provide signals rather than authoritative judgments. A model may classify a constructive edit as suspicious or fail to recognize a subtle factual distortion. Community acceptance therefore depends on how tools expose uncertainty, support review, and fit established workflows. Automated translation presents similar issues: fluent output can still contain mistranslations, unsupported claims, or source-language biases.\n\nGenerative artificial intelligence introduces additional risks when plausible-looking text or citations are submitted to an encyclopedia. The relevant editorial question remains whether a contribution is verifiable and properly sourced, not merely whether it reads coherently. Policies governing automated contributions are developed within projects and can differ. The foundation’s provision of technology does not replace local responsibility for deciding what content meets a project’s standards.\n\nLarge-scale automated access also has operational consequences. Repeated downloading, scraping, and model-related collection can consume bandwidth and processing resources, especially when clients bypass efficient access methods or request expensive page variants. Public access does not make infrastructure capacity unlimited. Managing such demand requires distinguishing ordinary readers, legitimate research, responsible commercial reuse, and abusive traffic while preserving the availability of the underlying knowledge."
    }
  ],
  "links": [
    "wikipedia",
    "wiki",
    "wikispecies",
    "wikimedia-commons",
    "wikidata",
    "knowledge-base",
    "access-control",
    "mediawiki",
    "free-software",
    "load-balancing",
    "data-center",
    "application-programming-interface",
    "unicode",
    "creative-commons",
    "gnu-free-documentation-license",
    "copyright",
    "museum",
    "higher-education",
    "digital-divide",
    "data-privacy",
    "https",
    "search-engine",
    "artificial-intelligence",
    "training-data",
    "natural-language-processing",
    "machine-translation",
    "generative-artificial-intelligence"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}