Version 1.0
Created by Danslav Slavenskoj
Date: May 2025

Languages: 中文简体 中文繁體 Čeština Deutsch Español Français Hrvatski 日本語 한국어 Polski Português Русский Српски

• JSON Schema - Full validation schema for LVTag format
• Classifier Definitions - Machine-readable classifier specifications
• Specification - Jump to format details
• Examples - See LVTag in action

The Language Variant Tag (LVTag) format is a systematic approach to language classification that extends the BCP 47 standard using private-use subtags. It enables precise identification of language varieties across multiple dimensions including formality, politeness, domain, and orthography.

Classification Rigor: LVTag brings systematic organization to language tagging by providing clear, separate dimensions for different types of variation. Unlike existing subtags and systems that mix different categories at the same level, LVTag maintains strict separation between formality, politeness, domain, and other dimensions.

Standards Compatibility: LVTag is fully compliant with BCP 47 (RFC 5646) and works seamlessly with:

• IANA Language Subtag Registry
• ISO 639 language codes
• Unicode CLDR
• W3C language tags
• HTTP Accept-Language headers
• XML lang attributes
• HTML lang attributes

Technology Integration: LVTag tags can be used directly in:

• Natural Language Processing (NLP) pipelines
• Machine Translation systems
• Content Management Systems (CMS)
• Language detection libraries
• Search engines and information retrieval
• Web applications and APIs
• Localization workflows

Use Cases:

• Audience Targeting: Match content to appropriate audiences based on register and domain
• Translation Quality: Maintain appropriate formality and politeness levels in machine translation
• Language Learning: Teach learners appropriate register for different contexts
• Corpus Linguistics: Build precisely tagged corpora for research
• Social Media Analysis: Classify user-generated content by register and domain
• Customer Service: Route messages based on formality and domain to appropriate agents

While BCP 47 provides excellent support for identifying languages, scripts, and regions, it lacks standardized mechanisms for capturing sociolinguistic variation within a language. Current standards don't address:

• Register Variation: No way to distinguish between formal and informal varieties of the same language
• Politeness Levels: Critical for languages like Japanese, Korean, and Thai where politeness is grammatically encoded
• Domain-Specific Language: No standard for marking technical, medical, or legal language varieties
• Sociolects: No mechanism for identifying social group varieties (youth language, professional jargon)
• Historical Stages: Limited support for distinguishing classical from modern forms
• Formality Gradients: No numeric scale for computational processing of register
• Proto-Languages: Inconsistent encoding - some proto-languages have ISO codes (e.g., ine for PIE) while others don't, and ISO 639-5 family codes aren't valid in BCP 47 tags, creating a confusing landscape for historical linguistics
• Orthographic Variation: While BCP 47 handles scripts, it doesn't effectively capture variations within scripts (spelling reforms, romanization systems, competing standards) that fundamentally affect text processing, search, and spell-checking

LVTag fills these gaps using BCP 47's private-use extension mechanism (-x-), providing a systematic, machine-readable way to encode these critical dimensions of language variation while maintaining full backward compatibility.

The advent of large language models and sophisticated NLP tools has made precise language variety classification not just useful but essential. Modern systems need to:

• Generate text appropriate to specific contexts (formal vs. informal, polite vs. casual)
• Train on properly classified corpora to avoid mixing registers inappropriately
• Provide culturally and contextually appropriate responses
• Handle code-switching and mixed-language content accurately
• Preserve stylistic consistency when translating or transforming text
• Filter training data based on formality, domain, or other characteristics
• Adapt output to match user preferences or requirements

LVTag provides the granular metadata need to understand not just what language is being used, but how it's being used, enabling more nuanced and appropriate language processing pipelines.

language-x-[classifier]-[value]-[classifier2]-[value2]...

Where:

• language is a valid BCP 47 primary language subtag (e.g., en, ko, ja)
• x indicates the beginning of private-use subtags
• classifier is a category identifier (see Magic Tags below)
• value is the specific classification within that category

LVTag supports both long-form and short-form "magic" classifiers for flexibility:

Identifies specific orthographic conventions or writing system variants beyond standard script tags.

Format:

• Long: language-x-ortho-[variant]
• Short: language-x-w-[variant]

Examples (combined with standard script tags):

• az-Latn-x-ortho-new or az-Latn-x-w-new - Azerbaijani Latin script, new orthography
• de-Latn-x-ortho-1901 or de-Latn-x-w-1901 - German Latin script, 1901 orthography
• zh-Hans-x-ortho-pinyin or zh-Hans-x-w-pinyin - Simplified Chinese with Pinyin
• yi-Hebr-x-ortho-yivo or yi-Hebr-x-w-yivo - Yiddish Hebrew script, YIVO orthography

Identifies the formality level of language use.

Format:

• Long: language-x-form-[1-5]
• Short: language-x-f-[1-5]

Formality scale:

• 1 = Most formal (written documents, official speeches)
• 2 = Formal (business meetings, academic writing)
• 3 = Neutral/standard (news, general conversation)
• 4 = Informal (casual conversation, emails to friends)
• 5 = Most casual (intimate conversation, slang)

Examples:

• ko-x-form-1 or ko-x-f-1 - Most formal Korean
• en-x-form-3 or en-x-f-3 - Neutral English
• ja-x-form-5 or ja-x-f-5 - Most casual Japanese

Identifies the politeness/respect level of language use.

Format:

• Long: language-x-polite-[1-5]
• Short: language-x-p-[1-5]

Politeness scale:

• 1 = Most respectful/deferential (royal address, religious contexts)
• 2 = Very polite (formal honorifics, respectful speech)
• 3 = Polite/neutral (standard politeness)
• 4 = Familiar (among equals, friends)
• 5 = Intimate/plain (family, very close friends)

Examples:

• ko-x-polite-1 or ko-x-p-1 - Highest respect Korean
• ja-x-polite-2 or ja-x-p-2 - Very polite Japanese
• th-x-polite-3 or th-x-p-3 - Standard polite Thai

Identifies specialized vocabulary or professional context.

Format:

• Long: language-x-domain-[domain_type]
• Short: language-x-d-[domain_type]

Examples:

• en-x-domain-legal or en-x-d-legal - Legal English
• ja-x-domain-med or ja-x-d-med - Medical Japanese
• ko-x-domain-business or ko-x-d-business - Business Korean
• ja-x-domain-tech or ja-x-d-tech - Technical Japanese
• en-x-domain-fin or en-x-d-fin - Financial English

Identifies regional or geographic language varieties.

Format:

• Long: language-x-geo-[region]
• Short: language-x-g-[region]

Examples:

• ko-x-geo-gyeong or ko-x-g-gyeong - Gyeongsang Korean (경상도)
• ko-x-geo-jeolla or ko-x-g-jeolla - Jeolla Korean (전라도)
• es-x-geo-riopla or es-x-g-riopla - Rioplatense Spanish
• pt-x-geo-nordeste or pt-x-g-nordeste - Northeastern Brazilian Portuguese

Identifies proto-languages or reconstructed historical languages.

Format:

• Long: x-proto-[iso639-5_code if available]
• Short: x-a-[iso639-5_code if available]

Rules:

• MUST use ISO 639-5 language family codes when available
• Use descriptive identifiers only when no ISO 639-5 code exists

Examples using ISO 639-5 codes:

• x-proto-ine or x-a-ine - Proto-Indo-European
• x-proto-gem or x-a-gem - Proto-Germanic
• x-proto-sla or x-a-sla - Proto-Slavic
• x-proto-sem or x-a-sem - Proto-Semitic
• x-proto-cel or x-a-cel - Proto-Celtic
• x-proto-ira or x-a-ira - Proto-Iranian
• x-proto-inc or x-a-inc - Proto-Indo-Aryan
• x-proto-bat or x-a-bat - Proto-Baltic
• x-proto-roa or x-a-roa - Proto-Romance
• x-proto-trk or x-a-trk - Proto-Turkic

Examples without ISO 639-5 codes (descriptive, longer than three characters):

• x-proto-baltslav or x-a-baltslav - Proto-Balto-Slavic (no ISO 639-5 code)

Note:

• Language family codes (ISO 639-5) are NOT valid as standard primary BCP 47 language tags which is why we have implemented them using x-proto
• They are valid and preferred within private-use extensions (after x-)
• Therefore all proto-language tags must start with x- to comply with BCP 47

Identifies historical periods or stages of a language.

Format:

• Long: language-x-hist-[period]
• Short: language-x-h-[period]

Examples:

• en-x-hist-old or en-x-h-old - Old English period
• en-x-hist-middle or en-x-h-middle - Middle English period
• ja-x-hist-kobun or ja-x-h-kobun - Classical Japanese (古文)
• ko-x-hist-hunmin or ko-x-h-hunmin - Middle Korean (훈민정음 period)
• el-x-hist-koine or el-x-h-koine - Koine Greek (Κοινή)
• sa-x-hist-vedic or sa-x-h-vedic - Vedic Sanskrit (वैदिक)

Identifies text genre or literary style.

Format:

• Long: language-x-genre-[genre_type]
• Short: language-x-e-[genre_type]

Examples:

• en-x-genre-news or en-x-e-news - News English
• ja-x-genre-manga or ja-x-e-manga - Manga Japanese (漫画)
• ko-x-genre-webtoon or ko-x-e-webtoon - Korean webtoon (웹툰)
• zh-x-genre-shi or zh-x-e-shi - Chinese poetry (詩)
• fr-x-genre-bd or fr-x-e-bd - French comics (bande dessinée)
• de-x-genre-marchen or de-x-e-marchen - German fairy tales (Märchen)

Identifies the communication medium.

Format:

• Long: language-x-medium-[medium_type]
• Short: language-x-m-[medium_type]

Examples:

• en-x-medium-spoken or en-x-m-spoken - Spoken English
• ko-x-medium-digital or ko-x-m-digital - Digital/online Korean
• ja-x-medium-written or ja-x-m-written - Written Japanese
• hi-x-medium-bcast or hi-x-m-bcast - Broadcast Hindi
• zh-x-medium-sms or zh-x-m-sms - SMS/text message Chinese

Identifies sociolect or social group varieties.

Format:

• Long: language-x-socio-[social_group]
• Short: language-x-s-[social_group]

Examples:

• en-x-socio-academic or en-x-s-academic - Academic sociolect
• en-x-socio-urban or en-x-s-urban - Urban sociolect
• es-x-socio-juvenil or es-x-s-juvenil - Spanish youth sociolect (jerga juvenil)
• fr-x-socio-jeune or fr-x-s-jeune - French youth sociolect
• de-x-socio-jugend or de-x-s-jugend - German youth sociolect (Jugendsprache)
• ko-x-socio-online or ko-x-s-online - Korean online sociolect

Identifies the fundamental mode of language production.

Format:

• Long: language-x-modality-[mode]
• Short: language-x-o-[mode]

Examples:

• en-x-modality-spoken or en-x-o-spoken - Spoken English
• en-x-modality-written or en-x-o-written - Written English
• asl-x-modality-signed or asl-x-o-signed - American Sign Language
• en-x-modality-multi or en-x-o-multi - Multimodal English (speech + gestures)
• fr-x-modality-tactile or fr-x-o-tactile - Tactile French (for deafblind)

Identifies the linguistic register or functional variety of language use.

Format:

• Long: language-x-register-[register_type]
• Short: language-x-r-[register_type]

Examples:

• en-x-register-frozen or en-x-r-frozen - Frozen register (prayers, pledges)
• en-x-register-formal or en-x-r-formal - Formal register (academic papers)
• en-x-register-consult or en-x-r-consult - Consultative register (professional)
• en-x-register-casual or en-x-r-casual - Casual register (friends)
• en-x-register-intimate or en-x-r-intimate - Intimate register (family)

Identifies the communicative function or speech act.

Format:

• Long: language-x-pragma-[function]
• Short: language-x-u-[function]

Examples:

• en-x-pragma-request or en-x-u-request - Request function
• ja-x-pragma-apology or ja-x-u-apology - Apology function
• es-x-pragma-complmnt or es-x-u-complmnt - Compliment function
• ar-x-pragma-greeting or ar-x-u-greeting - Greeting function
• zh-x-pragma-refusal or zh-x-u-refusal - Refusal function

Identifies temporal aspects or tense usage patterns.

Format:

• Long: language-x-temporal-[aspect]
• Short: language-x-t-[aspect]

Examples:

• en-x-temporal-past or en-x-t-past - Past-oriented discourse
• ja-x-temporal-nonpast or ja-x-t-nonpast - Non-past focus
• id-x-temporal-atemprl or id-x-t-atemprl - Timeless/atemporal
• fr-x-temporal-future or fr-x-t-future - Future-oriented
• zh-x-temporal-aspect or zh-x-t-aspect - Aspectual focus

Identifies information source marking.

Format:

• Long: language-x-evidence-[source]
• Short: language-x-v-[source]

Examples:

• qu-x-evidence-direct or qu-x-v-direct - Direct witness
• tr-x-evidence-hearsay or tr-x-v-hearsay - Hearsay/reported
• ja-x-evidence-infer or ja-x-v-infer - Inferential
• en-x-evidence-assume or en-x-v-assume - Assumed
• de-x-evidence-quote or de-x-v-quote - Quotative

Identifies emotional tone or affect.

Format:

• Long: language-x-affect-[emotion]
• Short: language-x-k-[emotion]

Examples:

• en-x-affect-angry or en-x-k-angry - Angry tone
• ja-x-affect-humble or ja-x-k-humble - Humble affect
• es-x-affect-joyful or es-x-k-joyful - Joyful expression
• ko-x-affect-sad or ko-x-k-sad - Sad/melancholic
• fr-x-affect-neutral or fr-x-k-neutral - Neutral affect

Identifies age-related or generational language varieties.

Format:

• Long: language-x-age-[generation]
• Short: language-x-n-[generation]

Examples:

• en-x-age-child or en-x-n-child - Child speech
• ja-x-age-teen or ja-x-n-teen - Teenager language
• ko-x-age-elder or ko-x-n-elder - Elder speech
• es-x-age-genz or es-x-n-genz - Generation Z
• zh-x-age-millenl or zh-x-n-millenl - Millennial speech

Identifies gender related language varieties.

Format:

• Long: language-x-gender-[identity]
• Short: language-x-i-[identity]

Examples: (Examples removed)

Identifies level of domain expertise on a 0-10 scale.

Format:

• Long: language-x-expert-[0-10]
• Short: language-x-b-[0-10]

Expertise scale:

• 0 = No knowledge
• 1-2 = Beginner
• 3-4 = Intermediate
• 5-6 = Advanced
• 7-8 = Expert
• 9-10 = Master/Authority

Examples:

• en-x-expert-0 or en-x-b-0 - No expertise
• de-x-expert-3 or de-x-b-3 - Intermediate level
• ja-x-expert-7 or ja-x-b-7 - Expert level
• es-x-expert-9 or es-x-b-9 - Master level
• zh-x-expert-5 or zh-x-b-5 - Advanced level

Identifies conversational or interactional patterns.

Format:

• Long: language-x-interact-[structure]
• Short: language-x-2-[structure]

Examples:

• en-x-interact-turn or en-x-2-turn - Turn-taking
• ja-x-interact-overlap or ja-x-2-overlap - Overlapping speech
• es-x-interact-monolog or es-x-2-monolog - Monologic
• ar-x-interact-dialog or ar-x-2-dialog - Dialogic
• zh-x-interact-multi or zh-x-2-multi - Multi-party

Identifies prosodic or suprasegmental features.

Format:

• Long: language-x-prosody-[feature]
• Short: language-x-y-[feature]

Examples:

• en-x-prosody-stress or en-x-y-stress - Stress-timed
• ja-x-prosody-pitch or ja-x-y-pitch - Pitch-accent
• fr-x-prosody-syllable or fr-x-y-syllable - Syllable-timed
• zh-x-prosody-tone or zh-x-y-tone - Tonal patterns
• es-x-prosody-rhythm or es-x-y-rhythm - Rhythmic patterns

Identifies lexical density as a numeric value (0-100).

Format:

• Long: language-x-lexical-[0-100]
• Short: language-x-l-[0-100]

Examples:

• en-x-lexical-20 or en-x-l-20 - Low density (20%)
• de-x-lexical-55 or de-x-l-55 - Medium density (55%)
• ja-x-lexical-75 or ja-x-l-75 - High density (75%)
• es-x-lexical-40 or es-x-l-40 - Moderate density (40%)
• zh-x-lexical-85 or zh-x-l-85 - Very high density (85%)

Identifies syntactic complexity as a numeric value (0-100).

Format:

• Long: language-x-syntax-[0-100]
• Short: language-x-z-[0-100]

Examples:

• en-x-syntax-15 or en-x-z-15 - Simple syntax (15%)
• de-x-syntax-70 or de-x-z-70 - Complex syntax (70%)
• ja-x-syntax-45 or ja-x-z-45 - Moderate complexity (45%)
• es-x-syntax-30 or es-x-z-30 - Low complexity (30%)
• zh-x-syntax-60 or zh-x-z-60 - High complexity (60%)

Identifies the start date of language use (ISO 8601 format without punctuation).

Format:

• Long: language-x-start-[YYYYMMDD]
• Short: language-x-0-[YYYYMMDD]

Date formats:

• Full date: YYYYMMDD
• Year-month: YYYYMM
• Year only: YYYY

Examples:

• en-x-start-20240315 or en-x-0-20240315 - English starting March 15, 2024
• ja-x-start-19890108 or ja-x-0-19890108 - Japanese starting January 8, 1989
• es-x-start-202403 or es-x-0-202403 - Spanish starting March 2024

Identifies the end date of language use (ISO 8601 format without punctuation).

Format:

• Long: language-x-end-[YYYYMMDD]
• Short: language-x-1-[YYYYMMDD]

Date formats:

• Full date: YYYYMMDD
• Year-month: YYYYMM
• Year only: YYYY

Examples:

• en-x-end-20240415 or en-x-1-20240415 - English ending April 15, 2024
• ja-x-end-20190430 or ja-x-1-20190430 - Japanese ending April 30, 2019
• es-x-end-202412 or es-x-1-202412 - Spanish ending December 2024

Identifies level of taboo, vulgar, or offensive content.

Format:

• Long: language-x-taboo-[0-5]
• Short: language-x-j-[0-5]

Examples:

• en-x-taboo-0 or en-x-j-0 - No taboo content
• en-x-taboo-3 or en-x-j-3 - Moderate taboo level
• ja-x-form-5-taboo-4 or ja-x-f-5-j-4 - Very casual Japanese with high taboo level

Indicates confidence score for the immediately preceding classifier.

Format:

• Long: language-x-[classifier]-[value]-conf-[0-100]
• Short: language-x-[classifier]-[value]-c-[0-100]

Special behavior:

• The confidence score applies to the classifier immediately before it
• Multiple confidence scores can be used for different classifiers
• If no classifier precedes it, the confidence applies to the base language tag

Examples:

• en-x-form-3-conf-95 or en-x-f-3-c-95 - Neutral formality with 95% confidence
• ko-x-polite-2-conf-80-domain-med-conf-60 or ko-x-p-2-c-80-d-med-c-60 - Very polite (80% confidence) medical Korean (60% confidence)
• ja-x-hist-kobun-conf-100 or ja-x-h-kobun-c-100 - Classical Japanese with 100% confidence
• x-proto-ine-conf-75 or x-a-ine-c-75 - Proto-Indo-European with 75% confidence

LVTag supports multiple classifiers in a single tag to provide precise language identification. Both long and short forms can be mixed:

ko-x-form-4-domain-business
ko-x-f-4-d-business
ko-x-form-4-polite-2-domain-business
ko-x-f-4-p-2-d-business

Examples above show Korean with informal formality (4) but polite speech (2) in business context.

Note: All values must be 8 characters or shorter to comply with BCP 47 subtag length restrictions. While specific values for many classifiers are to be established through expert usage and community consensus, the numeric scales, date formats, and basic values listed below are defined in this standard.

# Most formal Korean
ko-x-form-1

# Very polite Japanese
ja-x-polite-2

# Legal English
en-x-domain-legal

# Gyeongsang Korean
ko-x-geo-gyeong

# Proto-Indo-European
x-proto-ine

# Most formal Korean
ko-x-f-1

# Very polite Japanese
ja-x-p-2

# Legal English
en-x-d-legal

# Gyeongsang Korean
ko-x-g-gyeong

# Proto-Indo-European
x-a-ine

# Informal but polite Korean business language
ko-x-form-4-polite-2-domain-business
ko-x-f-4-p-2-d-business

# Formal and respectful Japanese medical language
ja-x-form-1-polite-1-domain-med
ja-x-f-1-p-1-d-med

# Southern Vietnamese with neutral formality, polite speech, technical domain
vi-x-geo-southern-form-3-polite-2-domain-tech
vi-x-g-southern-f-3-p-2-d-tech

# Complex classification with multiple dimensions
en-x-h-middle-e-poetry-m-written-f-1
ja-x-f-2-p-1-d-med-h-kobun-m-written

# Language varieties showing formality/politeness distinction
ko-x-f-5-p-2  # Very casual but polite (to older friend)
ko-x-f-1-p-4  # Very formal but familiar (written to peer)
ja-x-f-4-p-1  # Casual formality but highest respect
en-x-f-5-j-4  # Very casual English with high taboo level

1. Language Learning Applications

   • Teach appropriate register for different social contexts
   • Provide domain-specific vocabulary training

2. Machine Translation

   • Maintain register consistency in translations
   • Apply domain-specific terminology

3. Content Classification

   • Automatically categorize text by formality and domain
   • Route content to appropriate reviewers or systems

4. Corpus Linguistics

   • Build tagged corpora for linguistic research
   • Study register and domain variation

1. Subtag Length: Each subtag after x- must be 8 characters or fewer
2. Order: Classifiers can appear in any order after x-
3. Uniqueness: Each classifier type should appear only once per tag (except conf which can appear multiple times)
4. Case: Tags should be lowercase (case-insensitive per BCP 47)
5. Magic Tags: Short form tags are single characters; q, 3-9 are reserved for future use
6. Mixing: Long and short forms can be mixed within the same tag
7. Proto Tags: Must start with x- and SHOULD use ISO 639-5 codes when available (e.g., x-proto-sla not x-proto-slavic)
8. Confidence: The conf/c classifier applies to the immediately preceding classifier
9. Numeric Values: Must be within defined ranges (0-5 for taboo, 0-10 for expertise, 0-100 for percentage values)
10. Date Format: Dates use ISO 8601 without punctuation (YYYY, YYYYMM, or YYYYMMDD)

LVTag format is fully compatible with:

• BCP 47 (RFC 5646)
• ISO 639 language codes
• IANA Language Subtag Registry
• Unicode CLDR

1. Precision: Enables fine-grained language variety identification
2. Extensibility: New registers and domains can be added
3. Standards-based: Built on established BCP 47 private-use mechanism
4. Machine-readable: Systematic format enables automated processing
5. Human-readable: Clear, descriptive subtags
6. Flexibility: Support for both verbose long-form and concise short-form tags
7. Brevity: Short magic tags enable compact representation while maintaining clarity

LVTag is designed to evolve with the needs of the language technology community. We welcome suggestions for new classifiers, improvements to existing ones, and real-world implementation feedback.

To propose extensions or contribute to the specification:

• Open an issue at github.com/lvtag/spec
• Join the discussion on existing proposals
• Share your implementation experiences
• Submit pull requests for documentation improvements

Reserved single-character codes (q, 3-9) are available for future standardized extensions.

• BCP 47: Tags for Identifying Languages
• IANA Language Subtag Registry

This specification is released under the CC0 1.0 Universal (Public Domain Dedication).

Why CC0: To ensure maximum adoption and implementation freedom, LVTag is placed in the public domain. This means:

• No permission needed to use, implement, or modify
• No attribution required (though appreciated)
• No legal barriers for commercial or governmental use
• Compatible with all software licenses
• Used by major standards like Unicode CLDR

Patent Grant: Any patents covering the LVTag specification are hereby licensed royalty-free for any implementation that complies with this specification.

No Endorsement: Use of LVTag does not imply endorsement by the specification authors.

To the extent possible under law, Danslav Slavenskoj has waived all copyright and related or neighboring rights to the Language Variant Tag (LVTag) Format Specification. This work is published from: United States of America.