{
  "id": "localization",
  "title": "Localization",
  "lang": "en",
  "source_lang": "en",
  "summary": "Localization adapts products, services, and content to the linguistic, cultural, technical, and regulatory requirements of particular audiences.",
  "category": "language",
  "subcategory": "",
  "type": "concept",
  "aliases": [
    "Localisation",
    "l10n"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/localization/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/localization/",
      "json": "https://aiwiki.page/zh/localization.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Localization is the adaptation of a product, service, or body of content for a particular linguistic, cultural, or regional audience. It includes translation, but also encompasses changes to interfaces, typography, images, audio, data formats, payment methods, documentation, and other features that affect how something functions in a target setting. A localized application, for example, may require not only translated menus but also different address fields, plural forms, sorting rules, and methods of entering text. The abbreviation l10n represents the ten letters between the initial l and final n in localization.\n\nLocalization became a distinct industrial activity through the international distribution of software and digital products, although adaptation for different audiences has a much longer history in publishing, commerce, and entertainment. It now connects language work with software engineering, design, product management, testing, and regulatory compliance. Its central problem is not simply how to replace words in one language with words in another, but how to preserve appropriate meaning and usable functionality when a product crosses linguistic and institutional boundaries."
    },
    {
      "heading": "Scope and conceptual boundaries",
      "text": "The distinguishing feature of localization is its attention to an entire usable product or communicative experience. Translation normally concerns the relationship between source and target texts, although translation practice can itself involve extensive adaptation. Localization makes the surrounding system an explicit part of the task: whether text fits a button, whether a search function recognizes local spelling, whether a spoken instruction matches an on-screen action, and whether a purchase form accepts the customer’s address. An accurate sentence can therefore form part of an unsuccessful localization if its technical or situational context is wrong.\n\nA central distinction is between localization and internationalization. Internationalization prepares a product to support multiple languages and regional conventions without requiring its underlying design to be rebuilt for each one. It includes separating interface text from executable code, supporting diverse scripts, and using locale-aware formatting services. Localization then supplies and validates the material for particular audiences. The two activities are interdependent: a localization project may reveal an internationalization defect, such as a form that cannot accept a surname containing a space.\n\nA locale is a configuration of linguistic and regional preferences used by a computing system. It is not equivalent to a country, language, or ethnic identity. French-language users in Canada and France may share much vocabulary while encountering different administrative terminology, currency defaults, and address conventions. Conversely, people within one jurisdiction may use several languages and scripts. A locale may therefore combine a language with a region and, where relevant, a script or additional preferences.\n\nCommercial terminology also distinguishes localization from globalization, although usage varies. In product organizations, globalization may describe the overall arrangements for operating internationally, including internationalization, localization, market planning, and support. In the social sciences, the same word has a much broader meaning involving international economic and cultural integration. Transcreation is another industry term, usually referring to substantial creative adaptation of advertising or promotional material to reproduce a desired effect rather than closely follow source wording. Its boundaries with copywriting and translation are not standardized.\n\nThese distinctions do not make localization culturally unrestricted. Some material must preserve precise relationships to an original: contractual clauses, product identifiers, technical specifications, or quoted statements. Other material permits substantial redesign. Decisions depend on purpose, audience, medium, and institutional constraints. Translation studies, particularly skopos theory, provides one account of this purpose-oriented reasoning. Nevertheless, a commercial brief cannot by itself determine legal validity, technical correctness, or what members of an audience will find intelligible."
    },
    {
      "heading": "Historical development and industrial organization",
      "text": "Practices resembling localization predate computers. Publishers adapted editions for different readerships; manufacturers supplied labels and instructions for export markets; and distributors altered advertising, packaging, and product names. Film distribution developed specialized forms of subtitling and dubbing, with their own timing, performance, and censorship constraints. These activities established recurring problems—terminology control, consistency across media, and the relationship between cultural adaptation and fidelity—that later became central to software localization.\n\nPersonal computing in the 1980s gave these problems a new technical configuration. Software sold internationally contained menus, messages, documentation, keyboard assumptions, and character-handling routines that often reflected its original market. Adapting a program could require modification of source code as well as translation. Limited display space, incompatible character encodings, and restricted fonts made support for additional scripts particularly demanding. Japanese software distribution, for example, required attention to multibyte character processing and text input rather than merely the substitution of translated strings.\n\nDuring the 1990s, commercial software publishers increasingly treated international releases as coordinated production programs. Specialist vendors combined linguistic services with engineering, desktop publishing, and testing. The Localization Industry Standards Association, founded in 1990, was one organization associated with the professionalization of this work; it ceased operating in 2011. Its activities included support for exchange formats and industry discussion. The first version of the Unicode Standard appeared in 1991, providing an increasingly important foundation for multilingual text processing, although adoption across operating systems and applications took many years.\n\nThe expansion of the internet changed both the scale and timing of localization. Websites could be updated without manufacturing a new physical edition, while international audiences could encounter a service before its operator had deliberately entered their markets. Web interfaces, online documentation, electronic commerce, and search visibility created continuing rather than purely release-based language requirements. Later, mobile application stores and cloud-hosted services further reduced the importance of the single, finished localized release. A service might introduce new text every day while supporting dozens of language versions.\n\nIndustrial organization consequently ranges from large centralized programs to decentralized community projects. A product owner may employ language specialists directly, contract with a multilingual vendor, or coordinate separate providers for individual languages. Open-source projects often combine volunteer translation with professional engineering and review. These arrangements distribute authority differently: developers control technical behavior, linguists judge contextual meaning, and local product teams may influence terminology or market-specific features. The practical unit of work is often not a document but a changing set of interdependent interface strings, help articles, legal notices, media assets, and support materials."
    },
    {
      "heading": "Locale identifiers, standards, and regional data",
      "text": "Reliable localization depends on representing language and regional preferences in a form that software can exchange. The Internet Engineering Task Force’s BCP 47 framework, incorporating RFC 5646 and RFC 4647, defines language-tag syntax and procedures for matching tags. Tags can identify a language and add script, region, and other subtags. For example, pt-BR identifies Portuguese associated with Brazil, while zh-Hant identifies Chinese written with traditional characters. A tag need not contain a region, and a region subtag does not establish the citizenship or location of an individual user.\n\nThe distinction between language, script, and region is operationally important. Chinese content cannot always be handled by treating simplified and traditional Chinese characters as mechanically interchangeable: vocabulary, usage, and institutional terminology may also differ. Likewise, languages written in more than one script require an explicit script policy where automatic inference would be unreliable. A product’s available content variants may be less detailed than the user’s preferences, making a documented matching strategy necessary.\n\nThe Unicode Consortium’s Common Locale Data Repository supplies structured data used by many software libraries and platforms. Its contents include number and date patterns, language and territory names, plural rules, and other locale-dependent information. The International Components for Unicode libraries, commonly called ICU, implement many related services. These resources allow products to reuse maintained linguistic data rather than embed separate assumptions in each application. They do not eliminate the need to test a product’s particular content or determine which regional conventions its audience expects.\n\nRegional formatting involves more than changing separators. The value one thousand two hundred thirty-four and one half can appear as 1,234.5 or 1.234,5, among other conventions. A currency symbol such as $ is ambiguous without context, and changing its display does not convert the underlying amount. Dates require attention to field order, month names, calendars, and time zones. The date 03/04/2025 is ambiguous across common numerical conventions. A time zone is also not simply a fixed offset: daylight-saving rules and government decisions can change the relationship between local time and coordinated time.\n\nA further requirement is fallback: deciding what to show when the preferred locale or individual resource is unavailable. A system might fall back from a regional language version to a broader language version and finally to a default. Such chains must account for script differences and available content rather than blindly delete subtags. Stored data should also remain distinct from its presentation. A transaction timestamp, for instance, can retain a machine-readable value while being displayed according to a user’s time zone and preferences. This separation permits reformatting without repeatedly interpreting ambiguous human-readable strings."
    },
    {
      "heading": "Text architecture, writing systems, and interface design",
      "text": "A foundation of multilingual software is Unicode, which assigns code points to characters and defines numerous properties and text-processing rules. Unicode is not a font, and using a Unicode encoding does not guarantee correct display. A working system also needs suitable fonts, shaping engines, input methods, and text layout. Arabic letters change visible form according to their joining context; many South Asian scripts require reordering and positioning of glyphs. Correct storage and correct rendering are therefore related but separate technical requirements.\n\nThe relationship between a user-perceived character and stored text is another source of errors. A visible character may consist of a base letter and a combining diacritic, and some emoji sequences contain several code points. Counting code units or code points is not always an appropriate way to enforce a visible-character limit. Unicode normalization addresses certain equivalent character representations, but it does not make all visually similar text identical. Search, truncation, cursor movement, and validation each need a defined policy suited to their function.\n\nRight-to-left text introduces additional layout problems. The Unicode Bidirectional Algorithm handles the ordering of mixed-direction text, including combinations of Arabic or Hebrew with Latin-script product names and numbers. Interface layout may also need mirroring, such as placing a navigation panel on the opposite side. Yet not everything should be mirrored: logos, photographs, mathematical notation, and some media controls have independent directional meanings. Directional isolation is important when user-generated text is inserted into surrounding interface text, because punctuation and adjacent numbers can otherwise appear in unexpected positions.\n\nInternationalized interfaces separate translatable resources from program logic and provide enough context for translators to understand them. The English word “Open” may be a command, an adjective describing a status, or part of a longer phrase. Reusing one resource for all these functions can force incorrect translations. Sentence construction through concatenation is similarly fragile: joining a name, a verb, and an object in a fixed sequence assumes a particular word order. Languages with grammatical agreement may require the entire message to be selected or generated as a coherent unit.\n\nLayout must accommodate differences in text length, line breaking, and type metrics. There is no universal percentage by which translation expands text: a short label may grow dramatically, while a longer passage may change little. Chinese and Japanese line breaking does not rely on spaces in the same way as English, and different writing traditions impose restrictions on punctuation at line boundaries. Flexible components, meaningful wrapping, and realistic font testing are more reliable than fixed-size boxes designed around one language. Accessibility adds further requirements, including correct language metadata, usable reading order, and localized labels for controls encountered through assistive technology."
    },
    {
      "heading": "Linguistic adaptation, terminology, and cultural interpretation",
      "text": "Localization applies translation to material whose meaning is distributed across text, interaction, and context. A warning may be understandable only if the translator knows what action triggered it and what the user can do next. Screenshots, developer notes, character information, and access to a working build can therefore be essential linguistic inputs. Isolated strings conceal distinctions such as singular and plural, grammatical gender, whether a name refers to a person or a file, and whether a button begins an action or confirms its completion.\n\nTerminology management seeks consistency without assuming that every source word has one permanent equivalent. A terminology database may record a concept, approved terms by language, definitions, domain information, and restrictions on use. This differs from a simple bilingual word list. In a graphics application, “layer” denotes a specific software object; in a logistics system, the same source word may refer to a physical arrangement. Consistency must operate at the level of concepts and contexts, not indiscriminate string replacement. Product names and trademarks may remain unchanged while descriptive terms require translation.\n\nPragmatics is especially relevant to instructions, politeness, and brand voice. Forms of address can encode distinctions absent from the source language. A German interface may adopt formal or informal second-person address, with consequences for pronouns, verb forms, and the relationship projected toward users. Some languages allow subject omission where repeated English-style pronouns would sound unnatural. A style guide can specify register, capitalization, punctuation, and preferred constructions, but it must leave room for grammatical differences rather than require superficial imitation.\n\nCultural adaptation may affect examples, illustrations, humor, gestures, and assumptions about everyday activities. A baseball metaphor may communicate poorly to an audience unfamiliar with the sport, while an image of a particular electrical plug may be technically inappropriate outside its original market. Nevertheless, cultural categories are not uniform national traits. Age, profession, education, and familiarity with a product can matter as much as geography. Research with intended users is more informative than treating a symbol, color, or phrase as having one universal meaning throughout an entire country.\n\nNames, addresses, and personal identity reveal the limits of source-market assumptions. Forms that demand exactly one given name and one family name, require a postal code in every country, or reject non-Latin characters can exclude valid users. Transliteration may be required by a particular administrative system, but it is not equivalent to translation and does not justify discarding a person’s preferred spelling. Similar care applies to sex and gender fields, honorifics, and kinship terminology. Localization can expose these design problems, although solving them may require changes to data models and business processes rather than linguistic revision alone."
    },
    {
      "heading": "Production workflows and language technologies",
      "text": "A localization workflow usually begins with defining the target audiences, content inventory, quality requirements, and release dependencies. Material must then be extracted into a form that preserves identifiers and distinguishes translatable text from code, variables, and markup. A translation management system can coordinate assignments, version changes, review, terminology, and delivery. Its usefulness depends on the quality of the information supplied: an efficiently distributed but context-free string remains difficult to translate accurately.\n\nTranslation memory stores previously translated segments together with their source counterparts and relevant metadata. When new content arrives, the system proposes exact or approximate matches. This can reduce repetitive work and improve consistency in documentation and recurring interface text. A match is not automatically correct for a new occurrence, however. The same sentence may refer to a different object, use a different tone, or contain changed underlying terminology. Approval status, domain, product version, and contextual identifiers can be as important as textual similarity.\n\nExchange standards allow content and linguistic assets to move between systems. XLIFF, developed through OASIS, represents localizable content and associated information for interchange; its first version was published in 2002. Translation Memory eXchange, or TMX, supports the exchange of translation-memory data. Such formats reduce dependence on proprietary storage, but interoperability is not guaranteed merely by a file extension. Tools may implement different versions, handle inline codes differently, or preserve different subsets of metadata. Round-trip testing checks whether a file can be exported, processed, and reimported without damaging its structure.\n\nMachine translation can provide draft output, particularly for large volumes of repetitive or lower-risk material. Neural systems became prominent in production during the 2010s, and large language models have added capabilities for contextual rewriting, terminology application, and linguistic assistance. Their output can nevertheless omit content, alter numbers, mishandle placeholders, or generate plausible but unsupported wording. Human post-editing is therefore a distinct activity with defined quality expectations, not merely a final glance at automatically generated text. Confidential material also introduces questions about where data is processed and retained.\n\nContinuous delivery requires coordination between localization and frequent software changes. Automated pipelines may detect modified resources, submit them for translation through an application programming interface, run checks, and return approved files to a repository. Stable identifiers help preserve translations when source wording changes, while version tracking prevents obsolete text from reappearing. Release policies must decide what happens when a language version is incomplete: retain earlier approved wording, display a fallback, delay the feature, or restrict its availability. Each choice has consequences for consistency and for the user’s ability to complete a task."
    },
    {
      "heading": "Quality assurance and evaluation",
      "text": "Localization quality has several dimensions that cannot be reduced to whether a translation sounds fluent. Linguistic quality concerns accuracy, terminology, grammar, and suitability for context. Functional quality concerns whether the localized product behaves correctly. Visual quality includes typography, layout, clipping, and alignment. Locale correctness concerns dates, numbers, sorting, input, and other regional behavior. A polished sentence in a broken checkout flow is not a successful result, while a functional interface with misleading instructions may be equally unusable.\n\nLinguistic review compares the translated content with its source and intended function. Error typologies, including the Multidimensional Quality Metrics framework, distinguish categories such as mistranslation, omission, terminology, and style, often with severity levels. A severity judgment depends on consequence: a minor punctuation inconsistency and an incorrect instruction that causes data loss should not carry the same weight. Evaluation also needs a distinction between an actual error and an acceptable alternative. Without agreed terminology and style expectations, review can become a succession of individual preferences.\n\nAutomated checks are effective for some classes of defect. They can detect missing placeholders, altered markup, inconsistent numbers, forbidden terms, and empty translations. If a source message contains {accountName}, losing or changing that token may cause a runtime failure. Yet an automated check cannot generally establish that the surrounding sentence describes the right account or action. Technical validation and linguistic review address different failure modes and are most useful when their responsibilities are explicit.\n\nPseudolocalization tests internationalization readiness before real translation is available. Software transforms source strings by adding accented characters, expanding their length, or surrounding them with markers. Separate pseudo-right-to-left tests can expose directionality problems. These techniques reveal hard-coded text, clipping, encoding failures, and assumptions about layout, but they do not simulate the grammar or cultural conventions of a real language. In-context testing with actual localized builds remains necessary, particularly for dynamic messages, keyboard navigation, and interactions with external services.\n\nEvaluation at scale also uses sampling, user testing, and operational measures. Task-completion rates and support incidents can reveal problems that sentence-level review misses, although they have many causes besides language. Automatic translation metrics such as BLEU and COMET can assist certain comparisons, but a single aggregate score does not certify a complete localized product. Test coverage must also reflect risk. The number of combinations of language, device, operating system, user role, and account state can grow rapidly, so teams commonly prioritize critical flows and representative configurations rather than test every possible combination equally."
    },
    {
      "heading": "Media, commerce, and specialized domains",
      "text": "Different media place different constraints on adaptation. In games, text may interact with branching narrative, character identity, audiovisual timing, and gameplay rules. A pronoun chosen for a character can affect many later lines; an item name may appear in dialogue, inventory lists, and voiced instructions. Localization sometimes includes redesigning graphics, recording new performances, or changing the implementation of puzzles based on language. Testing must follow the game’s states and interactions, since reading a script alone cannot reveal every mismatch between dialogue and action.\n\nAudiovisual localization includes subtitling, dubbing, voice-over, and accessibility-related versions. Subtitles must coordinate readable text with shot changes, speech timing, and available screen space. Dubbing adds constraints involving performance duration, visible articulation, and the relationship between voice and character. Audio description and captions serve distinct access functions and should not be treated as interchangeable with translated dialogue. Rights agreements, credits, and the availability of separate dialogue and music tracks can determine which adaptations are technically and contractually possible.\n\nElectronic commerce links language to transaction infrastructure. A localized storefront may require locally usable payment methods, delivery options, product measurements, tax presentation, and return information. Displaying a price in a local currency is different from charging in that currency, and both differ from calculating taxes or import duties. Size labels and units may require conversion, but conversions must preserve the relevant precision and product specification. The International System of Units provides standardized units; it does not eliminate market-specific conventions for describing screen sizes, clothing, or building materials.\n\nSpecialized domains place stricter limits on linguistic freedom. Pharmaceutical information, medical-device interfaces, and financial disclosures operate under jurisdiction-specific requirements. Relevant obligations may concern approved wording, document structure, language availability, traceability, or the relationship between translated and authorized versions. For example, European Union rules for medical devices include language requirements that depend in part on the Member State where a device is made available. Localization in such settings is tied to documented review and change control rather than an unrestricted effort to make content more persuasive or familiar.\n\nPublic services and education introduce different priorities. A government portal may need terminology aligned with the institutions that actually administer a benefit, not an approximate foreign equivalent. Educational materials may require changes to curriculum references, notation, and examples while preserving the concept being taught. Emergency messages must remain consistent across channels and understandable under time pressure. In these domains, the intended audience may include people with limited literacy or limited access to digital services, making language choice, presentation, and delivery method inseparable parts of the adaptation."
    },
    {
      "heading": "Cultural agency, inclusion, and governance",
      "text": "Localization influences which languages become usable in digital environments and which users must operate through a second language. Supporting a language involves more than adding translated pages: it can require fonts, keyboards, terminology, search behavior, documentation, and long-term maintenance. Communities with fewer commercial resources may have substantial translation expertise but limited access to the engineering infrastructure needed for full product support. This connects localization to language policy, multilingualism, and the practical distribution of technological access.\n\nCommercial prioritization usually reflects expected demand, operating costs, and strategic commitments, but language coverage is not a simple function of speaker numbers. A language may have many speakers yet limited purchasing access to a particular service, while a smaller language may be supported because of institutional obligations or established demand. Regional labels can also obscure variation within a language. Decisions about standard language, spelling, and terminology may privilege one established form over alternatives used by other communities. Documented consultation and editorial responsibility make these choices more visible, even when one interface cannot accommodate every variant.\n\nAdaptation also raises questions about cultural representation. Replacing an unfamiliar reference can improve comprehension, but repeated substitution may erase the setting or identity of a work. This tension is particularly visible in literature, games, and audiovisual media, where foreignness may be part of the intended experience. There is no universal boundary between necessary explanation, creative adaptation, and distortion. The relevant considerations include genre, audience expectations, the creator’s aims, and the contractual authority granted to translators and editors.\n\nPolitical and legal restrictions must be distinguished from ordinary linguistic adjustment. A distribution market may impose requirements affecting maps, names, symbols, or available content. Product teams then face separate questions about factual representation, compliance, availability, and disclosure. These are governance decisions, even when implemented through the same files and workflows used for translation. Calling every such alteration “localization” can conceal who authorized it and why; conversely, ordinary differences in regional terminology do not by themselves establish political interference.\n\nData handling and labor arrangements add further responsibilities. Translation platforms may process unreleased products, personal information, or confidential correspondence, making data privacy and access control relevant throughout the workflow. Machine-generated drafts can change the work assigned to language professionals without eliminating the need for contextual judgment. Volunteer communities require decisions about contributor permissions, review authority, and continuity when maintainers leave. Across these arrangements, accountability depends on identifiable ownership of language resources, preserved revision histories, and procedures for correcting errors after publication."
    }
  ],
  "links": [
    "translation",
    "language",
    "locale-computing",
    "translation-studies",
    "skopos-theory",
    "film",
    "internet",
    "chinese-language",
    "traditional-chinese-characters",
    "unicode",
    "word-order",
    "accessibility",
    "pragmatics",
    "machine-translation",
    "large-language-model",
    "application-programming-interface",
    "bleu",
    "international-system-of-units",
    "language-policy",
    "multilingualism",
    "standard-language",
    "data-privacy"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}