Coverage and data model

Languages you can parse names in

extract_langcode, pronounce_lang, and get_lang_data all take a lang argument: the language the names are written in. The library ships wordlists for 21 of them:

Code	Language	Code	Language	Code	Language
an	Aragonese	de	German	it	Italian
ar	Arabic	en	English	kab	Kabyle
ast	Asturian	es	Spanish	nl	Dutch
bg	Bulgarian	eu	Basque	oc	Occitan
ca	Catalan	fr	French	pt	Portuguese
da	Danish	fy	Frisian	ro	Romanian
		gl	Galician	sk	Slovak
		hr	Croatian

The authoritative, runtime list is always:

import ovos_lang_parser
ovos_lang_parser.LANGS

A lang value does not have to be an exact member. It is matched to the closest wordlist, so regional variants like en-us or pt-pt resolve to en / pt. A lang too far from any bundled list raises ValueError.

What each wordlist covers

Each wordlist maps a few hundred target languages (keyed by ISO 639 code) to their names in that language. So the en wordlist knows the English names of hundreds of languages, the pt wordlist their Portuguese names, and so on. The target set is far broader than the 21 name languages above. You can resolve "Swahili", "Tibetan", or "Esperanto" even though the library cannot parse names written in those languages.

Approximate distinct target codes per wordlist:

lang	targets	lang	targets	lang	targets
an	242	es	153	nl	153
ar	134	eu	140	oc	156
ast	182	fr	153	pt	186
bg	148	fy	169	ro	153
ca	153	gl	153	sk	148
da	153	hr	148
de	184	it	153
en	153	kab	134

The data model

Wordlists live under ovos_lang_parser/res/<lang>/langs.json. Each file is a JSON object keyed by BCP-47 code:

{
  "de": "German",
  "pt": "Portuguese",
  "pt-br": "Brazilian Portuguese",
  "en-us": "American English"
}

A value may also be a template using (a|b) alternation, which the loader expands into every spelling. For example "Bislamá Bichlamar" and templated forms like "(Modern |)Greek" become multiple accepted names, all mapping to the same code. The first name listed for a code is treated as canonical. It is what pronounce_lang returns.

At load time (get_lang_data / the internal loader):

• codes are normalized to modern lowercase form (legacy iw → he, jw → jv, mo → ro), so duplicate aliases merge onto one code
• templates are expanded to individual names
• results are memoized so a wordlist is parsed only once.

Both full and base tags coexist: pt and pt-br are separate keys, which is what lets extract_langcode return pt-br for "Brazilian Portuguese" while pronounce_lang("pt-br", …) falls back to the pt name when a region-specific one is absent.

To add a language, see extending.md.

---

← API reference · Home · Extending →