The set of tags constructed according to RFC 4646 for the identification of languages.
RFC 4646 obsoletes RFC 3066.
Language tags have changed rules three times in the IETF’s history, and DCMI has a term for each version. dcterms:RFC4646 covers the middle one: tags “constructed according to RFC 4646 for the identification of languages”. The official comment adds one fact, that RFC 4646 obsoletes RFC 3066. Anyone tracing records back through older profiles will meet RFC 3066 values too, which is why the two numbers are so often searched together.
A short lineage of language tag rules
| RFC | Published | Key change |
|---|---|---|
| RFC 1766 | 1995 | first language tag format, primary tag plus optional subtags |
| RFC 3066 | 2001 | allowed three-letter ISO 639-2 codes as primary tags |
| RFC 4646 | 2006 | structured subtags, IANA Language Subtag Registry |
| RFC 5646 | 2009 | extended language subtags, ISO 639-3 and 639-5 coverage |
DCMI defines a syntax encoding scheme for each of the four (RFC1766, RFC3066, RFC4646, RFC5646). Each is an rdfs:Datatype, applied to the literal of dcterms:language.
What RFC 4646 introduced
Before 2006, a tag was mostly a language code optionally followed by a country code, with a separate route for registering whole special tags. RFC 4646 replaced that with a grammar of positioned subtags:
- language: two or three letters, from ISO 639.
- script: four letters from ISO 15924, such as
Latn,Cyrl,Hant. - region: two letters from ISO 3166-1 or three digits from UN M49, such as
419for Latin America. - variant: registered subtags for dialects or orthographies.
- extension and private use: introduced by a single letter.
The second big change was administrative. RFC 4646 created the IANA Language Subtag Registry, one file listing every valid subtag with its type and dates. Validating a tag became a matter of checking each part against that registry, which software could do automatically.
Script subtags were the practical win for many collections. Serbian in Cyrillic and Serbian in Latin script, Chinese in simplified or traditional characters, Azerbaijani in Latin or Cyrillic: under RFC 3066 these were awkward to separate, and RFC 4646 handled them with sr-Cyrl, zh-Hans, az-Latn.
RFC 4646 also introduced a rule that is easy to miss: a script subtag should be used only when it adds information. en-Latn is valid but redundant, since English is almost always written in Latin script, and the registry records a “suppress-script” value for such languages. Validators flag redundant script subtags as warnings, not errors.
Reading older records
Collections cataloged around 2006–2009 may declare RFC4646 in their XML schemas or in an application profile. Those tags almost always remain valid under RFC 5646, since the newer rules were written to keep existing well-formed tags working. Migration usually means changing the declared scheme, not the tags themselves.
Records from the RFC 3066 era need a closer look. Some tags registered whole under the old regime, such as i-klingon or zh-guoyu, were later marked as grandfathered or deprecated in the registry, with a preferred replacement (tlh, cmn). A batch check against the current registry finds them quickly.
Common slips
Underscores are the classic error: en_US is a POSIX locale identifier, not a language tag, and BCP 47 uses hyphens. Case is the second. Tags are case-insensitive, but the registry recommends lowercase language, title-case script and uppercase region (zh-Hant-TW), and consistent case makes duplicates easier to spot.
A last one is specific to Dublin Core: typing a three-letter code like deu as RFC4646 is legal, but RFC 4646 prefers the shortest available code, so de is the canonical form. For bare three-letter codes, dcterms:ISO639-3 is the more natural scheme to declare, and it avoids implying that the tag was checked against the subtag registry.
Example in RDF
@prefix dcterms: <http://purl.org/dc/terms/> .
<https://example.org/editions/poems-1931-cyrillic.epub>
dcterms:language "sr-Cyrl"^^dcterms:RFC4646 .
<https://example.org/editions/poems-1931-latin.epub>
dcterms:language "sr-Latn"^^dcterms:RFC4646 .
Questions about dcterms:RFC4646
What is RFC 4646?
RFC 4646, published by the IETF in September 2006, defined how to build language tags such as en-GB or zh-Hant. It replaced RFC 3066 and was itself replaced by RFC 5646 in 2009. All three have served as BCP 47, the stable name for the current rules.
What changed between RFC 3066 and RFC 4646?
RFC 4646 introduced a formal subtag structure with separate slots for language, script, region and variant, and created the IANA Language Subtag Registry as the single list of valid subtags. Under RFC 3066 tags were less structured and many were registered whole.
Is a tag valid under RFC 4646 still valid today?
In nearly all cases yes. RFC 5646 was designed to be backward compatible, so well-formed RFC 4646 tags remain well formed. RFC 5646 added extended language subtags and drew on newer ISO 639 parts.
Should new records use dcterms:RFC4646?
No. For new records dcterms:RFC5646 is the current scheme. RFC4646 is useful when describing older records accurately, for instance when migrating a collection cataloged between 2006 and 2009 under a profile that named it.
What does 'sr-Cyrl' mean?
It is Serbian (sr) written in Cyrillic script (Cyrl). Script subtags use the four-letter codes of ISO 15924. They matter for languages written in more than one script, where text in each script needs a different font and search handling.