Skip to content
Syntax Encoding Scheme · DCMI classes and encoding schemes

dcterms:ISO639-3: what the ISO 639-3 language code scheme covers

Published Formatory editorial team
http://purl.org/dc/terms/ISO639-3

The set of three-letter codes listed in ISO 639-3 for the representation of names of languages.

A sound archive holding recordings in Scottish Gaelic, Manx and Irish needs three distinct codes, and the older code lists do not always provide them. ISO 639-3 was built for that situation. The DCMI term dcterms:ISO639-3 is a syntax encoding scheme that declares a value to be one of the three-letter codes listed in ISO 639-3.

Where ISO 639-3 sits in the ISO 639 family

The ISO 639 standard grew in parts, each with a different scope:

Part Code length Scope
639-1 2 letters major languages, about 180 codes
639-2 3 letters about 500 codes, including language groups
639-3 3 letters individual languages, several thousand codes
639-5 3 letters language families and groups

DCMI defines schemes for both ISO 639-2 and ISO 639-3. The two share codes for languages present in both, so eng, fra and deu are valid either way. Where 639-2 offered a bibliographic and a terminology form (fre/fra, ger/deu), ISO 639-3 adopts the terminology form only. That detail matters when a library catalog using MARC bibliographic codes is merged with linguistic data, because fre and fra will otherwise appear as two languages.

Individual languages versus macrolanguages

ISO 639-3 introduced the idea of a macrolanguage: a code that stands for a cluster of closely related individual languages that are often treated as one. Arabic (ara) is a macrolanguage covering codes such as arz for Egyptian Arabic. Chinese (zho) covers cmn for Mandarin and yue for Cantonese, among others.

For a cataloger this creates a choice. The macrolanguage code is easier to search and matches older records. The individual code is more accurate for a spoken recording, where the variety is the whole point. Picking one rule per collection, and documenting it, is more useful than mixing both. Searchers rarely know the individual codes, so an interface that maps each individual language back to its macrolanguage for display gets the precision of one and the familiarity of the other.

Using the scheme with dcterms:language

The scheme types the literal of dcterms:language. That property expects a dcterms:LinguisticSystem, and a typed code literal is the conventional way to supply one. For web delivery and many software stacks, a BCP 47 tag under RFC 5646 is the format downstream systems expect; RFC 5646 allows ISO 639-3 codes as primary language subtags when no two-letter code exists, so gla and gd both appear in the wild for Scottish Gaelic. The shortest available code is preferred in a BCP 47 tag, which gives gd.

Format registries record languages the same way; the language code type is an example.

Errors that turn up in language fields

Two-letter codes typed as ISO639-3 are the most common defect, usually because a template default was copied. Language names in English (“German”) typed as codes are the next. Then come the special codes, used interchangeably when they mean different things:

  • zxx: no linguistic content.
  • und: undetermined, someone looked and could not tell.
  • mul: several languages present.
  • mis: a language without a code yet.

The ISO 639-3 tables are updated through an annual change request cycle, and codes are sometimes retired or split. A code that validated in 2010 may now be marked as retired, with a note naming its replacement.

The code tables are free to download from SIL as tab-delimited text files, including a separate table of retirements with the reason (split, merge, duplicate, non-existent) and the date. Running a collection’s distinct language values against both tables once a year catches retired codes before they drift into search facets as orphan entries.

Example in RDF

<rdf:Description rdf:about="https://example.org/recordings/song-cycle-03.wav"
    xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
    xmlns:dcterms="http://purl.org/dc/terms/">
  <dcterms:title>Work song, third recording</dcterms:title>
  <dcterms:language rdf:datatype="http://purl.org/dc/terms/ISO639-3">gla</dcterms:language>
</rdf:Description>

Questions about dcterms:ISO639-3

What is ISO 639-3?

ISO 639-3 is the part of the ISO 639 family that aims to give every known human language its own three-letter code, living, extinct, ancient and constructed. SIL International acts as its registration authority and publishes the code tables.

How is ISO639-3 different from ISO639-2?

ISO 639-2 has around five hundred codes and includes collective codes for language groups. ISO 639-3 has several thousand and covers individual languages. For the major languages the codes are shared, using the terminology form of 639-2 where it has two.

Can I use a two-letter code like 'en' with dcterms:ISO639-3?

No. The scheme covers three-letter codes only. 'en' is an ISO 639-1 code. For English, ISO 639-3 uses 'eng'. A two-letter value typed as ISO639-3 will fail any validation against the code table.

Which property does dcterms:ISO639-3 go with?

It is normally used with dcterms:language, whose range is dcterms:LinguisticSystem. The scheme types the literal so software knows the string is an ISO 639-3 code and can look up its name.

What code should I use for content with no language?

ISO 639-2 and ISO 639-3 both define 'zxx' for no linguistic content, such as instrumental music or an image without text. 'und' means undetermined and 'mul' means multiple languages; the three are often confused.

More from DCMI classes and encoding schemes

All 10 in this group