mirror of
https://github.com/moparisthebest/SickRage
synced 2024-11-15 22:05:02 -05:00
536 lines
15 KiB
Python
536 lines
15 KiB
Python
|
# -*- coding: utf-8 -*-
|
||
|
# enzyme - Video metadata parser
|
||
|
# Copyright 2011-2012 Antoine Bertin <diaoulael@gmail.com>
|
||
|
# Copyright 2003-2006 Dirk Meyer <dischi@freevo.org>
|
||
|
#
|
||
|
# This file is part of enzyme.
|
||
|
#
|
||
|
# enzyme is free software; you can redistribute it and/or modify it under
|
||
|
# the terms of the GNU General Public License as published by
|
||
|
# the Free Software Foundation; either version 3 of the License, or
|
||
|
# (at your option) any later version.
|
||
|
#
|
||
|
# enzyme is distributed in the hope that it will be useful,
|
||
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||
|
# GNU General Public License for more details.
|
||
|
#
|
||
|
# You should have received a copy of the GNU General Public License
|
||
|
# along with enzyme. If not, see <http://www.gnu.org/licenses/>.
|
||
|
import re
|
||
|
|
||
|
__all__ = ['resolve']
|
||
|
|
||
|
|
||
|
def resolve(code):
|
||
|
"""
|
||
|
Transform the given (2- or 3-letter) language code to a human readable
|
||
|
language name. The return value is a 2-tuple containing the given
|
||
|
language code and the language name. If the language code cannot be
|
||
|
resolved, name will be 'Unknown (<code>)'.
|
||
|
"""
|
||
|
if not code:
|
||
|
return None, None
|
||
|
if not isinstance(code, basestring):
|
||
|
raise ValueError('Invalid language code specified by parser')
|
||
|
|
||
|
# Take up to 3 letters from the code.
|
||
|
code = re.split(r'[^a-z]', code.lower())[0][:3]
|
||
|
|
||
|
for spec in codes:
|
||
|
if code in spec[:-1]:
|
||
|
return code, spec[-1]
|
||
|
|
||
|
return code, u'Unknown (%r)' % code
|
||
|
|
||
|
|
||
|
# Parsed from http://www.loc.gov/standards/iso639-2/ISO-639-2_utf-8.txt
|
||
|
codes = (
|
||
|
('aar', 'aa', u'Afar'),
|
||
|
('abk', 'ab', u'Abkhazian'),
|
||
|
('ace', u'Achinese'),
|
||
|
('ach', u'Acoli'),
|
||
|
('ada', u'Adangme'),
|
||
|
('ady', u'Adyghe'),
|
||
|
('afa', u'Afro-Asiatic '),
|
||
|
('afh', u'Afrihili'),
|
||
|
('afr', 'af', u'Afrikaans'),
|
||
|
('ain', u'Ainu'),
|
||
|
('aka', 'ak', u'Akan'),
|
||
|
('akk', u'Akkadian'),
|
||
|
('alb', 'sq', u'Albanian'),
|
||
|
('ale', u'Aleut'),
|
||
|
('alg', u'Algonquian languages'),
|
||
|
('alt', u'Southern Altai'),
|
||
|
('amh', 'am', u'Amharic'),
|
||
|
('ang', u'English, Old '),
|
||
|
('anp', u'Angika'),
|
||
|
('apa', u'Apache languages'),
|
||
|
('ara', 'ar', u'Arabic'),
|
||
|
('arc', u'Official Aramaic '),
|
||
|
('arg', 'an', u'Aragonese'),
|
||
|
('arm', 'hy', u'Armenian'),
|
||
|
('arn', u'Mapudungun'),
|
||
|
('arp', u'Arapaho'),
|
||
|
('art', u'Artificial '),
|
||
|
('arw', u'Arawak'),
|
||
|
('asm', 'as', u'Assamese'),
|
||
|
('ast', u'Asturian'),
|
||
|
('ath', u'Athapascan languages'),
|
||
|
('aus', u'Australian languages'),
|
||
|
('ava', 'av', u'Avaric'),
|
||
|
('ave', 'ae', u'Avestan'),
|
||
|
('awa', u'Awadhi'),
|
||
|
('aym', 'ay', u'Aymara'),
|
||
|
('aze', 'az', u'Azerbaijani'),
|
||
|
('bad', u'Banda languages'),
|
||
|
('bai', u'Bamileke languages'),
|
||
|
('bak', 'ba', u'Bashkir'),
|
||
|
('bal', u'Baluchi'),
|
||
|
('bam', 'bm', u'Bambara'),
|
||
|
('ban', u'Balinese'),
|
||
|
('baq', 'eu', u'Basque'),
|
||
|
('bas', u'Basa'),
|
||
|
('bat', u'Baltic '),
|
||
|
('bej', u'Beja'),
|
||
|
('bel', 'be', u'Belarusian'),
|
||
|
('bem', u'Bemba'),
|
||
|
('ben', 'bn', u'Bengali'),
|
||
|
('ber', u'Berber '),
|
||
|
('bho', u'Bhojpuri'),
|
||
|
('bih', 'bh', u'Bihari'),
|
||
|
('bik', u'Bikol'),
|
||
|
('bin', u'Bini'),
|
||
|
('bis', 'bi', u'Bislama'),
|
||
|
('bla', u'Siksika'),
|
||
|
('bnt', u'Bantu '),
|
||
|
('bos', 'bs', u'Bosnian'),
|
||
|
('bra', u'Braj'),
|
||
|
('bre', 'br', u'Breton'),
|
||
|
('btk', u'Batak languages'),
|
||
|
('bua', u'Buriat'),
|
||
|
('bug', u'Buginese'),
|
||
|
('bul', 'bg', u'Bulgarian'),
|
||
|
('bur', 'my', u'Burmese'),
|
||
|
('byn', u'Blin'),
|
||
|
('cad', u'Caddo'),
|
||
|
('cai', u'Central American Indian '),
|
||
|
('car', u'Galibi Carib'),
|
||
|
('cat', 'ca', u'Catalan'),
|
||
|
('cau', u'Caucasian '),
|
||
|
('ceb', u'Cebuano'),
|
||
|
('cel', u'Celtic '),
|
||
|
('cha', 'ch', u'Chamorro'),
|
||
|
('chb', u'Chibcha'),
|
||
|
('che', 'ce', u'Chechen'),
|
||
|
('chg', u'Chagatai'),
|
||
|
('chi', 'zh', u'Chinese'),
|
||
|
('chk', u'Chuukese'),
|
||
|
('chm', u'Mari'),
|
||
|
('chn', u'Chinook jargon'),
|
||
|
('cho', u'Choctaw'),
|
||
|
('chp', u'Chipewyan'),
|
||
|
('chr', u'Cherokee'),
|
||
|
('chu', 'cu', u'Church Slavic'),
|
||
|
('chv', 'cv', u'Chuvash'),
|
||
|
('chy', u'Cheyenne'),
|
||
|
('cmc', u'Chamic languages'),
|
||
|
('cop', u'Coptic'),
|
||
|
('cor', 'kw', u'Cornish'),
|
||
|
('cos', 'co', u'Corsican'),
|
||
|
('cpe', u'Creoles and pidgins, English based '),
|
||
|
('cpf', u'Creoles and pidgins, French-based '),
|
||
|
('cpp', u'Creoles and pidgins, Portuguese-based '),
|
||
|
('cre', 'cr', u'Cree'),
|
||
|
('crh', u'Crimean Tatar'),
|
||
|
('crp', u'Creoles and pidgins '),
|
||
|
('csb', u'Kashubian'),
|
||
|
('cus', u'Cushitic '),
|
||
|
('cze', 'cs', u'Czech'),
|
||
|
('dak', u'Dakota'),
|
||
|
('dan', 'da', u'Danish'),
|
||
|
('dar', u'Dargwa'),
|
||
|
('day', u'Land Dayak languages'),
|
||
|
('del', u'Delaware'),
|
||
|
('den', u'Slave '),
|
||
|
('dgr', u'Dogrib'),
|
||
|
('din', u'Dinka'),
|
||
|
('div', 'dv', u'Divehi'),
|
||
|
('doi', u'Dogri'),
|
||
|
('dra', u'Dravidian '),
|
||
|
('dsb', u'Lower Sorbian'),
|
||
|
('dua', u'Duala'),
|
||
|
('dum', u'Dutch, Middle '),
|
||
|
('dut', 'nl', u'Dutch'),
|
||
|
('dyu', u'Dyula'),
|
||
|
('dzo', 'dz', u'Dzongkha'),
|
||
|
('efi', u'Efik'),
|
||
|
('egy', u'Egyptian '),
|
||
|
('eka', u'Ekajuk'),
|
||
|
('elx', u'Elamite'),
|
||
|
('eng', 'en', u'English'),
|
||
|
('enm', u'English, Middle '),
|
||
|
('epo', 'eo', u'Esperanto'),
|
||
|
('est', 'et', u'Estonian'),
|
||
|
('ewe', 'ee', u'Ewe'),
|
||
|
('ewo', u'Ewondo'),
|
||
|
('fan', u'Fang'),
|
||
|
('fao', 'fo', u'Faroese'),
|
||
|
('fat', u'Fanti'),
|
||
|
('fij', 'fj', u'Fijian'),
|
||
|
('fil', u'Filipino'),
|
||
|
('fin', 'fi', u'Finnish'),
|
||
|
('fiu', u'Finno-Ugrian '),
|
||
|
('fon', u'Fon'),
|
||
|
('fre', 'fr', u'French'),
|
||
|
('frm', u'French, Middle '),
|
||
|
('fro', u'French, Old '),
|
||
|
('frr', u'Northern Frisian'),
|
||
|
('frs', u'Eastern Frisian'),
|
||
|
('fry', 'fy', u'Western Frisian'),
|
||
|
('ful', 'ff', u'Fulah'),
|
||
|
('fur', u'Friulian'),
|
||
|
('gaa', u'Ga'),
|
||
|
('gay', u'Gayo'),
|
||
|
('gba', u'Gbaya'),
|
||
|
('gem', u'Germanic '),
|
||
|
('geo', 'ka', u'Georgian'),
|
||
|
('ger', 'de', u'German'),
|
||
|
('gez', u'Geez'),
|
||
|
('gil', u'Gilbertese'),
|
||
|
('gla', 'gd', u'Gaelic'),
|
||
|
('gle', 'ga', u'Irish'),
|
||
|
('glg', 'gl', u'Galician'),
|
||
|
('glv', 'gv', u'Manx'),
|
||
|
('gmh', u'German, Middle High '),
|
||
|
('goh', u'German, Old High '),
|
||
|
('gon', u'Gondi'),
|
||
|
('gor', u'Gorontalo'),
|
||
|
('got', u'Gothic'),
|
||
|
('grb', u'Grebo'),
|
||
|
('grc', u'Greek, Ancient '),
|
||
|
('gre', 'el', u'Greek, Modern '),
|
||
|
('grn', 'gn', u'Guarani'),
|
||
|
('gsw', u'Swiss German'),
|
||
|
('guj', 'gu', u'Gujarati'),
|
||
|
('gwi', u"Gwich'in"),
|
||
|
('hai', u'Haida'),
|
||
|
('hat', 'ht', u'Haitian'),
|
||
|
('hau', 'ha', u'Hausa'),
|
||
|
('haw', u'Hawaiian'),
|
||
|
('heb', 'he', u'Hebrew'),
|
||
|
('her', 'hz', u'Herero'),
|
||
|
('hil', u'Hiligaynon'),
|
||
|
('him', u'Himachali'),
|
||
|
('hin', 'hi', u'Hindi'),
|
||
|
('hit', u'Hittite'),
|
||
|
('hmn', u'Hmong'),
|
||
|
('hmo', 'ho', u'Hiri Motu'),
|
||
|
('hsb', u'Upper Sorbian'),
|
||
|
('hun', 'hu', u'Hungarian'),
|
||
|
('hup', u'Hupa'),
|
||
|
('iba', u'Iban'),
|
||
|
('ibo', 'ig', u'Igbo'),
|
||
|
('ice', 'is', u'Icelandic'),
|
||
|
('ido', 'io', u'Ido'),
|
||
|
('iii', 'ii', u'Sichuan Yi'),
|
||
|
('ijo', u'Ijo languages'),
|
||
|
('iku', 'iu', u'Inuktitut'),
|
||
|
('ile', 'ie', u'Interlingue'),
|
||
|
('ilo', u'Iloko'),
|
||
|
('ina', 'ia', u'Interlingua '),
|
||
|
('inc', u'Indic '),
|
||
|
('ind', 'id', u'Indonesian'),
|
||
|
('ine', u'Indo-European '),
|
||
|
('inh', u'Ingush'),
|
||
|
('ipk', 'ik', u'Inupiaq'),
|
||
|
('ira', u'Iranian '),
|
||
|
('iro', u'Iroquoian languages'),
|
||
|
('ita', 'it', u'Italian'),
|
||
|
('jav', 'jv', u'Javanese'),
|
||
|
('jbo', u'Lojban'),
|
||
|
('jpn', 'ja', u'Japanese'),
|
||
|
('jpr', u'Judeo-Persian'),
|
||
|
('jrb', u'Judeo-Arabic'),
|
||
|
('kaa', u'Kara-Kalpak'),
|
||
|
('kab', u'Kabyle'),
|
||
|
('kac', u'Kachin'),
|
||
|
('kal', 'kl', u'Kalaallisut'),
|
||
|
('kam', u'Kamba'),
|
||
|
('kan', 'kn', u'Kannada'),
|
||
|
('kar', u'Karen languages'),
|
||
|
('kas', 'ks', u'Kashmiri'),
|
||
|
('kau', 'kr', u'Kanuri'),
|
||
|
('kaw', u'Kawi'),
|
||
|
('kaz', 'kk', u'Kazakh'),
|
||
|
('kbd', u'Kabardian'),
|
||
|
('kha', u'Khasi'),
|
||
|
('khi', u'Khoisan '),
|
||
|
('khm', 'km', u'Central Khmer'),
|
||
|
('kho', u'Khotanese'),
|
||
|
('kik', 'ki', u'Kikuyu'),
|
||
|
('kin', 'rw', u'Kinyarwanda'),
|
||
|
('kir', 'ky', u'Kirghiz'),
|
||
|
('kmb', u'Kimbundu'),
|
||
|
('kok', u'Konkani'),
|
||
|
('kom', 'kv', u'Komi'),
|
||
|
('kon', 'kg', u'Kongo'),
|
||
|
('kor', 'ko', u'Korean'),
|
||
|
('kos', u'Kosraean'),
|
||
|
('kpe', u'Kpelle'),
|
||
|
('krc', u'Karachay-Balkar'),
|
||
|
('krl', u'Karelian'),
|
||
|
('kro', u'Kru languages'),
|
||
|
('kru', u'Kurukh'),
|
||
|
('kua', 'kj', u'Kuanyama'),
|
||
|
('kum', u'Kumyk'),
|
||
|
('kur', 'ku', u'Kurdish'),
|
||
|
('kut', u'Kutenai'),
|
||
|
('lad', u'Ladino'),
|
||
|
('lah', u'Lahnda'),
|
||
|
('lam', u'Lamba'),
|
||
|
('lao', 'lo', u'Lao'),
|
||
|
('lat', 'la', u'Latin'),
|
||
|
('lav', 'lv', u'Latvian'),
|
||
|
('lez', u'Lezghian'),
|
||
|
('lim', 'li', u'Limburgan'),
|
||
|
('lin', 'ln', u'Lingala'),
|
||
|
('lit', 'lt', u'Lithuanian'),
|
||
|
('lol', u'Mongo'),
|
||
|
('loz', u'Lozi'),
|
||
|
('ltz', 'lb', u'Luxembourgish'),
|
||
|
('lua', u'Luba-Lulua'),
|
||
|
('lub', 'lu', u'Luba-Katanga'),
|
||
|
('lug', 'lg', u'Ganda'),
|
||
|
('lui', u'Luiseno'),
|
||
|
('lun', u'Lunda'),
|
||
|
('luo', u'Luo '),
|
||
|
('lus', u'Lushai'),
|
||
|
('mac', 'mk', u'Macedonian'),
|
||
|
('mad', u'Madurese'),
|
||
|
('mag', u'Magahi'),
|
||
|
('mah', 'mh', u'Marshallese'),
|
||
|
('mai', u'Maithili'),
|
||
|
('mak', u'Makasar'),
|
||
|
('mal', 'ml', u'Malayalam'),
|
||
|
('man', u'Mandingo'),
|
||
|
('mao', 'mi', u'Maori'),
|
||
|
('map', u'Austronesian '),
|
||
|
('mar', 'mr', u'Marathi'),
|
||
|
('mas', u'Masai'),
|
||
|
('may', 'ms', u'Malay'),
|
||
|
('mdf', u'Moksha'),
|
||
|
('mdr', u'Mandar'),
|
||
|
('men', u'Mende'),
|
||
|
('mga', u'Irish, Middle '),
|
||
|
('mic', u"Mi'kmaq"),
|
||
|
('min', u'Minangkabau'),
|
||
|
('mis', u'Uncoded languages'),
|
||
|
('mkh', u'Mon-Khmer '),
|
||
|
('mlg', 'mg', u'Malagasy'),
|
||
|
('mlt', 'mt', u'Maltese'),
|
||
|
('mnc', u'Manchu'),
|
||
|
('mni', u'Manipuri'),
|
||
|
('mno', u'Manobo languages'),
|
||
|
('moh', u'Mohawk'),
|
||
|
('mol', 'mo', u'Moldavian'),
|
||
|
('mon', 'mn', u'Mongolian'),
|
||
|
('mos', u'Mossi'),
|
||
|
('mul', u'Multiple languages'),
|
||
|
('mun', u'Munda languages'),
|
||
|
('mus', u'Creek'),
|
||
|
('mwl', u'Mirandese'),
|
||
|
('mwr', u'Marwari'),
|
||
|
('myn', u'Mayan languages'),
|
||
|
('myv', u'Erzya'),
|
||
|
('nah', u'Nahuatl languages'),
|
||
|
('nai', u'North American Indian'),
|
||
|
('nap', u'Neapolitan'),
|
||
|
('nau', 'na', u'Nauru'),
|
||
|
('nav', 'nv', u'Navajo'),
|
||
|
('nbl', 'nr', u'Ndebele, South'),
|
||
|
('nde', 'nd', u'Ndebele, North'),
|
||
|
('ndo', 'ng', u'Ndonga'),
|
||
|
('nds', u'Low German'),
|
||
|
('nep', 'ne', u'Nepali'),
|
||
|
('new', u'Nepal Bhasa'),
|
||
|
('nia', u'Nias'),
|
||
|
('nic', u'Niger-Kordofanian '),
|
||
|
('niu', u'Niuean'),
|
||
|
('nno', 'nn', u'Norwegian Nynorsk'),
|
||
|
('nob', 'nb', u'Bokm\xe5l, Norwegian'),
|
||
|
('nog', u'Nogai'),
|
||
|
('non', u'Norse, Old'),
|
||
|
('nor', 'no', u'Norwegian'),
|
||
|
('nqo', u"N'Ko"),
|
||
|
('nso', u'Pedi'),
|
||
|
('nub', u'Nubian languages'),
|
||
|
('nwc', u'Classical Newari'),
|
||
|
('nya', 'ny', u'Chichewa'),
|
||
|
('nym', u'Nyamwezi'),
|
||
|
('nyn', u'Nyankole'),
|
||
|
('nyo', u'Nyoro'),
|
||
|
('nzi', u'Nzima'),
|
||
|
('oci', 'oc', u'Occitan '),
|
||
|
('oji', 'oj', u'Ojibwa'),
|
||
|
('ori', 'or', u'Oriya'),
|
||
|
('orm', 'om', u'Oromo'),
|
||
|
('osa', u'Osage'),
|
||
|
('oss', 'os', u'Ossetian'),
|
||
|
('ota', u'Turkish, Ottoman '),
|
||
|
('oto', u'Otomian languages'),
|
||
|
('paa', u'Papuan '),
|
||
|
('pag', u'Pangasinan'),
|
||
|
('pal', u'Pahlavi'),
|
||
|
('pam', u'Pampanga'),
|
||
|
('pan', 'pa', u'Panjabi'),
|
||
|
('pap', u'Papiamento'),
|
||
|
('pau', u'Palauan'),
|
||
|
('peo', u'Persian, Old '),
|
||
|
('per', 'fa', u'Persian'),
|
||
|
('phi', u'Philippine '),
|
||
|
('phn', u'Phoenician'),
|
||
|
('pli', 'pi', u'Pali'),
|
||
|
('pol', 'pl', u'Polish'),
|
||
|
('pon', u'Pohnpeian'),
|
||
|
('por', 'pt', u'Portuguese'),
|
||
|
('pra', u'Prakrit languages'),
|
||
|
('pro', u'Proven\xe7al, Old '),
|
||
|
('pus', 'ps', u'Pushto'),
|
||
|
('qaa-qtz', u'Reserved for local use'),
|
||
|
('que', 'qu', u'Quechua'),
|
||
|
('raj', u'Rajasthani'),
|
||
|
('rap', u'Rapanui'),
|
||
|
('rar', u'Rarotongan'),
|
||
|
('roa', u'Romance '),
|
||
|
('roh', 'rm', u'Romansh'),
|
||
|
('rom', u'Romany'),
|
||
|
('rum', 'ro', u'Romanian'),
|
||
|
('run', 'rn', u'Rundi'),
|
||
|
('rup', u'Aromanian'),
|
||
|
('rus', 'ru', u'Russian'),
|
||
|
('sad', u'Sandawe'),
|
||
|
('sag', 'sg', u'Sango'),
|
||
|
('sah', u'Yakut'),
|
||
|
('sai', u'South American Indian '),
|
||
|
('sal', u'Salishan languages'),
|
||
|
('sam', u'Samaritan Aramaic'),
|
||
|
('san', 'sa', u'Sanskrit'),
|
||
|
('sas', u'Sasak'),
|
||
|
('sat', u'Santali'),
|
||
|
('scc', 'sr', u'Serbian'),
|
||
|
('scn', u'Sicilian'),
|
||
|
('sco', u'Scots'),
|
||
|
('scr', 'hr', u'Croatian'),
|
||
|
('sel', u'Selkup'),
|
||
|
('sem', u'Semitic '),
|
||
|
('sga', u'Irish, Old '),
|
||
|
('sgn', u'Sign Languages'),
|
||
|
('shn', u'Shan'),
|
||
|
('sid', u'Sidamo'),
|
||
|
('sin', 'si', u'Sinhala'),
|
||
|
('sio', u'Siouan languages'),
|
||
|
('sit', u'Sino-Tibetan '),
|
||
|
('sla', u'Slavic '),
|
||
|
('slo', 'sk', u'Slovak'),
|
||
|
('slv', 'sl', u'Slovenian'),
|
||
|
('sma', u'Southern Sami'),
|
||
|
('sme', 'se', u'Northern Sami'),
|
||
|
('smi', u'Sami languages '),
|
||
|
('smj', u'Lule Sami'),
|
||
|
('smn', u'Inari Sami'),
|
||
|
('smo', 'sm', u'Samoan'),
|
||
|
('sms', u'Skolt Sami'),
|
||
|
('sna', 'sn', u'Shona'),
|
||
|
('snd', 'sd', u'Sindhi'),
|
||
|
('snk', u'Soninke'),
|
||
|
('sog', u'Sogdian'),
|
||
|
('som', 'so', u'Somali'),
|
||
|
('son', u'Songhai languages'),
|
||
|
('sot', 'st', u'Sotho, Southern'),
|
||
|
('spa', 'es', u'Spanish'),
|
||
|
('srd', 'sc', u'Sardinian'),
|
||
|
('srn', u'Sranan Tongo'),
|
||
|
('srr', u'Serer'),
|
||
|
('ssa', u'Nilo-Saharan '),
|
||
|
('ssw', 'ss', u'Swati'),
|
||
|
('suk', u'Sukuma'),
|
||
|
('sun', 'su', u'Sundanese'),
|
||
|
('sus', u'Susu'),
|
||
|
('sux', u'Sumerian'),
|
||
|
('swa', 'sw', u'Swahili'),
|
||
|
('swe', 'sv', u'Swedish'),
|
||
|
('syc', u'Classical Syriac'),
|
||
|
('syr', u'Syriac'),
|
||
|
('tah', 'ty', u'Tahitian'),
|
||
|
('tai', u'Tai '),
|
||
|
('tam', 'ta', u'Tamil'),
|
||
|
('tat', 'tt', u'Tatar'),
|
||
|
('tel', 'te', u'Telugu'),
|
||
|
('tem', u'Timne'),
|
||
|
('ter', u'Tereno'),
|
||
|
('tet', u'Tetum'),
|
||
|
('tgk', 'tg', u'Tajik'),
|
||
|
('tgl', 'tl', u'Tagalog'),
|
||
|
('tha', 'th', u'Thai'),
|
||
|
('tib', 'bo', u'Tibetan'),
|
||
|
('tig', u'Tigre'),
|
||
|
('tir', 'ti', u'Tigrinya'),
|
||
|
('tiv', u'Tiv'),
|
||
|
('tkl', u'Tokelau'),
|
||
|
('tlh', u'Klingon'),
|
||
|
('tli', u'Tlingit'),
|
||
|
('tmh', u'Tamashek'),
|
||
|
('tog', u'Tonga '),
|
||
|
('ton', 'to', u'Tonga '),
|
||
|
('tpi', u'Tok Pisin'),
|
||
|
('tsi', u'Tsimshian'),
|
||
|
('tsn', 'tn', u'Tswana'),
|
||
|
('tso', 'ts', u'Tsonga'),
|
||
|
('tuk', 'tk', u'Turkmen'),
|
||
|
('tum', u'Tumbuka'),
|
||
|
('tup', u'Tupi languages'),
|
||
|
('tur', 'tr', u'Turkish'),
|
||
|
('tut', u'Altaic '),
|
||
|
('tvl', u'Tuvalu'),
|
||
|
('twi', 'tw', u'Twi'),
|
||
|
('tyv', u'Tuvinian'),
|
||
|
('udm', u'Udmurt'),
|
||
|
('uga', u'Ugaritic'),
|
||
|
('uig', 'ug', u'Uighur'),
|
||
|
('ukr', 'uk', u'Ukrainian'),
|
||
|
('umb', u'Umbundu'),
|
||
|
('und', u'Undetermined'),
|
||
|
('urd', 'ur', u'Urdu'),
|
||
|
('uzb', 'uz', u'Uzbek'),
|
||
|
('vai', u'Vai'),
|
||
|
('ven', 've', u'Venda'),
|
||
|
('vie', 'vi', u'Vietnamese'),
|
||
|
('vol', 'vo', u'Volap\xfck'),
|
||
|
('vot', u'Votic'),
|
||
|
('wak', u'Wakashan languages'),
|
||
|
('wal', u'Walamo'),
|
||
|
('war', u'Waray'),
|
||
|
('was', u'Washo'),
|
||
|
('wel', 'cy', u'Welsh'),
|
||
|
('wen', u'Sorbian languages'),
|
||
|
('wln', 'wa', u'Walloon'),
|
||
|
('wol', 'wo', u'Wolof'),
|
||
|
('xal', u'Kalmyk'),
|
||
|
('xho', 'xh', u'Xhosa'),
|
||
|
('yao', u'Yao'),
|
||
|
('yap', u'Yapese'),
|
||
|
('yid', 'yi', u'Yiddish'),
|
||
|
('yor', 'yo', u'Yoruba'),
|
||
|
('ypk', u'Yupik languages'),
|
||
|
('zap', u'Zapotec'),
|
||
|
('zbl', u'Blissymbols'),
|
||
|
('zen', u'Zenaga'),
|
||
|
('zha', 'za', u'Zhuang'),
|
||
|
('znd', u'Zande languages'),
|
||
|
('zul', 'zu', u'Zulu'),
|
||
|
('zun', u'Zuni'),
|
||
|
('zxx', u'No linguistic content'),
|
||
|
('zza', u'Zaza'),
|
||
|
)
|