Modul:languages/data2

Dari Wiktionary bahasa Indonesia, kamus bebas
Loncat ke navigasi Loncat ke pencarian

Dokumentasi untuk modul ini dapat dibuat di Modul:languages/data2/doc

local u = mw.ustring.char

-- UTF-8 encoded strings for some commonly-used diacritics
local GRAVE     = u(0x0300)
local ACUTE     = u(0x0301)
local CIRC      = u(0x0302)
local TILDE     = u(0x0303)
local MACRON    = u(0x0304)
local BREVE     = u(0x0306)
local DOTABOVE  = u(0x0307)
local DIAER     = u(0x0308)
local CARON     = u(0x030C)
local DGRAVE    = u(0x030F)
local INVBREVE  = u(0x0311)
local DOTBELOW  = u(0x0323)
local RINGBELOW = u(0x0325)
local CEDILLA   = u(0x0327)
local OGONEK    = u(0x0328)

-- Puncuation to be used for standardChars field
local PUNCTUATION = ' !#$%&*+,-./:;<=>?@^_`|~\'()'

local Cyrl = {"Cyrl"}
local Latn = {"Latn"}
local LatnArab = {"Latn", "Arab"}

local m = {}

m["aa"] = {
	"bahasa Afar",
	"Q27811",
	"cus",
	otherNames = {"Qafar"},
	scripts = Latn,
}

m["ab"] = {
	"bahasa Abkhaz",
	"Q5111",
	"cau-abz",
	otherNames = {"Abkhazian", "Abxazo"},
	scripts = {"Cyrl", "Geor", "Latn"},
	translit_module = "ab-translit",
	override_translit = true,
	entry_name = {
		from = {GRAVE, ACUTE},
		to   = {}} ,
}

m["ae"] = {
	"bahasa Avesta",
	"Q29572",
	"ira-cen",
	otherNames = {"Zend", "Old Bactrian"},
	scripts = {"Avst", "Gujr"},
	translit_module = "Avst-translit",
	wikipedia_article = "Avestan",
}

m["af"] = {
	"bahasa Afrikaans",
	"Q14196",
	"gmw",
	scripts = LatnArab,
	ancestors = {"nl"},
	sort_key = {
		from = {"[äáâà]", "[ëéêè]", "[ïíîì]", "[öóôò]", "[üúûù]", "[ÿýŷỳ]", "^-", "'"},
		to   = {"a"	 , "e"	, "i"	, "o"	, "u"  , "y" }} ,
}

m["ak"] = {
	"bahasa Akan",
	"Q28026",
	"alv-kwa",
	otherNames = {"Twi-Fante", "Twi", "Fante", "Fanti", "Asante", "Akuapem"},
	scripts = Latn,
}

m["am"] = {
	"bahasa Amhar",
	"Q28244",
	"sem-eth",
	scripts = {"Ethi"},
	translit_module = "Ethi-translit",
}

m["an"] = {
	"bahasa Aragon",
	"Q8765",
	"roa",
	scripts = Latn,
	ancestors = {"roa-oan"},
}

m["ar"] = {
	"bahasa Arab",
	"Q13955",
	"sem-arb",
	otherNames = {"Modern Standard Arabic", "Standard Arabic", "Literary Arabic", "Classical Arabic"},
	scripts = {"Arab", "Brai"},
	-- alif waṣl to alif, remove diacritics
	entry_name = {
		from = {u(0x0671), u(0x064B), u(0x064C), u(0x064D), u(0x064E), u(0x064F), u(0x0650), u(0x0651), u(0x0652), u(0x0670), u(0x0640)},
		to   = {u(0x0627)}},
	translit_module = "ar-translit",
	standardChars = "ء-غف-ْٰٱ" .. PUNCTUATION .. "٠-٩،؛؟٫٬ـ",
}

m["as"] = {
	"bahasa Assam",
	otherNames = {"Asamiya"},
	"Q29401",
	"inc",
	scripts = {"Beng"},
	ancestors = {"inc-mas"},
	translit_module = "as-translit",
}

m["av"] = {
	"bahasa Avar",
	"Q29561",
	"cau-nec",
	otherNames = {"Avaric"},
	scripts = Cyrl,
	ancestors = {"oav"},
	translit_module = "av-translit",
	override_translit = true,
	entry_name = {
		from = {GRAVE, ACUTE},
		to   = {}} ,
}

m["ay"] = {
	"bahasa Aymara",
	"Q4627",
	"sai-aym",
	otherNames = {"Southern Aymara", "Central Aymara"},
	scripts = Latn,
}

m["az"] = {
	"bahasa Azerbaijan",
	"Q9292",
	"trk-ogz",
	otherNames = {"Azeri", "Azari", "Azeri Turkic", "Azerbaijani Turkic", "North Azerbaijani", "South Azerbaijani", "Afshar", "Afshari", "Afshar Azerbaijani", "Afchar", "Qashqa'i", "Qashqai", "Kashkay", "Sonqor"},
	scripts = {"Latn", "Cyrl", "fa-Arab"},
	ancestors = {"trk-oat"},
}

m["ba"] = {
	"bahasa Bashkir",
	"Q13389",
	"trk-kip",
	scripts = Cyrl,
	translit_module = "ba-translit",
	override_translit = true,
}

m["be"] = {
	"bahasa Belarusia",
	"Q9091",
	"zle",
	otherNames = {"Belorussian", "Belarusan", "Bielorussian", "Byelorussian", "Belarussian", "White Russian"},
	scripts = Cyrl,
	ancestors = {"orv"},
	translit_module = "be-translit",
	sort_key = {
		from = {"Ё", "ё"},
		to   = {"Е" , "е"}},
	entry_name = {
		from = {"Ѐ", "ѐ", GRAVE, ACUTE},
		to   = {"Е", "е"}},
}

m["bg"] = {
	"bahasa Bulgaria",
	"Q7918",
	"zls",
	scripts = {"Cyrl"},
	ancestors = {"cu"},
	translit_module = "bg-translit",
	entry_name = {
		from = {"Ѐ", "ѐ", "Ѝ", "ѝ", GRAVE, ACUTE},
		to   = {"Е", "е", "И", "и"}},
}

m["bh"] = {
	"bahasa Bihari",
	"Q135305",
	"inc",
	scripts = {"Deva"},
	ancestors = {"inc-mgd"},
}

m["bi"] = {
	"bahasa Bislama",
	"Q35452",
	"crp",
	scripts = Latn,
	ancestors = {"en"},
}

m["bm"] = {
	"bahasa Bambara",
	"Q33243",
	"dmn",
	otherNames = {"Bamanankan"},
	scripts = Latn,
}

m["bn"] = {
	"bahasa Benggala",
	"Q9610",
	"inc",
	otherNames = {"Bangla"},
	scripts = {"Beng", "Newa"},
	ancestors = {"inc-mbn"},
	translit_module = "bn-translit",
}

m["bo"] = {
	"bahasa Tibet",
	"Q34271",
	"tbq",
	otherNames = {"Ü", "Dbus", "Lhasa", "Lhasa Tibetan", "Amdo Tibetan", "Amdo", "Panang", "Khams", "Khams Tibetan", "Khamba", "Tseku", "Dolpo", "Humla", "Limi", "Lhomi", "Shing Saapa", "Mugom", "Mugu", "Nubri", "Walungge", "Gola", "Thudam", "Lowa", "Loke", "Mustang", "Tichurong"}, -- and "Gyalsumdo", "Lower Manang"? "Kyirong"?
	scripts = {"Tibt"}, -- sometimes Deva?
	ancestors = {"xct"},
	translit_module = "bo-translit",
	override_translit = true,
}

m["br"] = {
	"bahasa Breton",
	"Q12107",
	"cel-bry",
	scripts = Latn,
	ancestors = {"xbm"},
}

m["ca"] = {
	"bahasa Katala",
	"Q7026",
	"roa",
	otherNames = {"Valencian"},
	scripts = Latn,
	ancestors = {"roa-oca"},
	sort_key = {
		from = {"à", "[èé]", "[íï]", "[òó]", "[úü]", "ç", "l·l"},
		to   = {"a", "e"   , "i"   , "o"   , "u"   , "c", "ll" }} ,
}

m["ce"] = {
	"bahasa Chechen",
	"Q33350",
	"cau-nkh",
	scripts = Cyrl,
	translit_module = "ce-translit",
	override_translit = true,
	entry_name = {
		from = {MACRON},
		to   = {}},
}

m["ch"] = {
	"bahasa Chamorro",
	"Q33262",
	"poz-sus",
	otherNames = {"Chamoru"},
	scripts = Latn,
}

m["co"] = {
	"bahasa Korsika",
	"Q33111",
	"roa",
	otherNames = {"Corsu"},
	scripts = Latn,
}

m["cr"] = {
	"bahasa Cree",
	"Q33390",
	"alg",
	scripts = {"Cans", "Latn"},
	translit_module = "cr-translit",
}

m["cs"] = {
	"bahasa Ceska",
	"Q9056",
	"zlw",
	scripts = Latn,
	ancestors = {"zlw-ocs"},
	sort_key = {
		from = {"á", "é", "í", "ó", "[úů]", "ý"},
		to   = {"a", "e", "i", "o", "u"   , "y"}} ,
}

m["cu"] = {
	"bahasa Gereja Slavonia Lama",
	"Q35499",
	"zls",
	otherNames = {"Old Church Slavic"},
	scripts = {"Cyrs", "Glag"},
	translit_module = "Cyrs-Glag-translit",
	entry_name = {
		from = {u(0x0484)}, -- kamora
		to   = {}},
	sort_key = {
		from = {"оу", "є"},
		to   = {"у" , "е"}} ,
}

m["cv"] = {
	"bahasa Chuvash",
	"Q33348",
	"trk-ogr",
	scripts = Cyrl,
	translit_module = "cv-translit",
	override_translit = true,
}

m["cy"] = {
	"bahasa Wales",
	"Q9309",
	"cel-bry",
	scripts = Latn,
	ancestors = {"wlm"},
	sort_key = {
		from = {"[âáàä]", "ch", "dd", "[êéèë]", "ff", "ngh", "[îíìï]", "ll", "[ôóòö]", "ph", "rh", "th", "[ûúùü]", "[ŵẃẁẅ]", "[ŷýỳÿ]", "'"},
		to   = {"a"	    , "c~", "d~", "e"	  , "f~", "g~h", "i"	  , "l~", "o"	  , "p~", "r~", "t~", "u"	  , "w"     , "y"	       }} ,
	standardChars = "A-IL-PR-UWYa-il-pr-uwy0-9ÂâÊêÎîÔôÛûŴŵŶŷ" .. PUNCTUATION,
}

m["da"] = {
	"bahasa Denmark",
	"Q9035",
	"gmq",
	scripts = Latn,
	ancestors = {"gmq-oda"},
}

m["de"] = {
	"bahasa Jerman",
	"Q188",
	"gmw",
	otherNames = {"High German", "New High German", "Deutsch"},
	scripts = {"Latn", "Latf"},
	ancestors = {"gmh"},
	sort_key = {
		from = {"[äàáâå]", "[ëèéê]", "[ïìíî]", "[öòóô]", "[üùúû]", "ß" },
		to   = {"a"	  , "e"	 , "i"	 , "o"	 , "u"	 , "ss"}} ,
	standardChars = "A-Za-z0-9ÄäÖöÜüß" .. PUNCTUATION,
}

m["dv"] = {
	"bahasa Dhivehi",
	"Q32656",
	"inc",
	otherNames = {"Divehi", "Mahal", "Mahl", "Maldivian"},
	scripts = {"Thaa"},
	ancestors = {"pmh"},	-- or Helu?
	translit_module = "dv-translit",
	override_translit = true,
}

m["dz"] = {
	"bahasa Dzongkha",
	"Q33081",
	"tbq",
	scripts = {"Tibt"},
	ancestors = {"xct"},
	translit_module = "bo-translit",
	override_translit = true,
}

m["ee"] = {
	"bahasa Ewe",
	"Q30005",
	"alv-von",
	scripts = Latn,
}

m["el"] = {
	"bahasa Yunani",
	"Q9129",
	"grk",
	otherNames = {"Modern Greek", "Neo-Hellenic"},
	scripts = {"Grek", "Brai"},
	ancestors = {"grc"},
	translit_module = "el-translit",
	override_translit = true,
	sort_key = {  -- Keep this synchronized with grc, cpg, pnt
		from = {"[ᾳάᾴὰᾲᾶᾷἀᾀἄᾄἂᾂἆᾆἁᾁἅᾅἃᾃἇᾇ]", "[έὲἐἔἒἑἕἓ]", "[ῃήῄὴῂῆῇἠᾐἤᾔἢᾒἦᾖἡᾑἥᾕἣᾓἧᾗ]", "[ίὶῖἰἴἲἶἱἵἳἷϊΐῒῗ]", "[όὸὀὄὂὁὅὃ]", "[ύὺῦὐὔὒὖὑὕὓὗϋΰῢῧ]", "[ῳώῴὼῲῶῷὠᾠὤᾤὢᾢὦᾦὡᾡὥᾥὣᾣὧᾧ]", "ῥ", "ς"},
		to   = {"α"						, "ε"		 , "η"						, "ι"				, "ο"		 , "υ"				, "ω"						, "ρ", "σ"}} ,
	standardChars = "ͺ;΄-ώϜϝ" .. PUNCTUATION .. "ἀ-῾",
}

m["en"] = {
	"bahasa Inggris",
	"Q1860",
	"gmw",
	otherNames = {"Modern English", "New English", "Hawaiian Creole English", "Hawai'ian Creole English", "Hawaiian Creole", "Hawai'ian Creole", "Polari", "Yinglish"}, -- all but the first three are names of subsumed dialects which once had codes
	scripts = {"Latn", "Brai", "Shaw", "Dsrt"}, -- entries in Shaw or Dsrt might require prior discussion
	ancestors = {"enm"},
	sort_key = {
		from = {"[äàáâåā]", "[ëèéêē]", "[ïìíîī]", "[öòóôō]", "[üùúûū]", "æ" , "œ" , "[çč]", "ñ", "'"},
		to   = {"a"       , "e"      , "i"      , "o"      , "u"      , "ae", "oe", "c"   , "n"}},
	wikimedia_codes = {"en", "simple"},
	standardChars = "A-Za-z0-9" .. PUNCTUATION .. u(0x2800) .. "-" .. u(0x28FF),
}

m["eo"] = {
	"bahasa Esperanto",
	"Q143",
	"art",
	scripts = Latn,
	sort_key = {
		from = {"[áà]", "[éè]", "[íì]", "[óò]", "[úù]", "[ĉ]", "[ĝ]", "[ĥ]", "[ĵ]", "[ŝ]", "[ŭ]"},
		to   = {"a"	   , "e"  , "i"  , "o"  , "u", "cĉ", "gĉ", "hĉ", "jĉ", "sĉ", "uĉ"}} ,
}

m["es"] = {
	"bahasa Spanyol",
	"Q1321",
	"roa",
	otherNames = {"Castilian", "Amazonian Spanish", "Amazonic Spanish", "Loreto-Ucayali Spanish"},
	scripts = {"Latn", "Brai"},
	ancestors = {"osp"},
	sort_key = {
		from = {"á", "é", "í", "ó", "[úü]", "ç", "ñ"},
		to   = {"a", "e", "i", "o", "u"   , "c", "n"}},
	standardChars = "A-VXYZa-vxyz0-9ÁáÉéÍíÓóÚúÑñ¿¡" .. PUNCTUATION,
}

m["et"] = {
	"bahasa Estonia",
	"Q9072",
	"fiu-fin",
	scripts = Latn,
}

m["eu"] = {
	"bahasa Basque",
	"Q8752",
	"euq",
	otherNames = {"Euskara"},
	scripts = Latn,
}

m["fa"] = {
	"bahasa Persia",
	"Q9168",
	"ira-swi",
	otherNames = {"Farsi", "New Persian", "Modern Persian", "Western Persian", "Iranian Persian", "Eastern Persian", "Dari", "Aimaq", "Aimak", "Aymaq", "Eimak"},
	scripts = {"fa-Arab"},
	ancestors = {"pal"}, -- "ira-mid"
	entry_name = {
		from = {u(0x064E), u(0x064F), u(0x0650), u(0x0651), u(0x0652)},
		to   = {}} ,
}

m["ff"] = {
	"bahasa Fula",
	"Q33454",
	"alv-sng",
	otherNames = {"Adamawa Fulfulde", "Bagirmi Fulfulde", "Borgu Fulfulde", "Central-Eastern Niger Fulfulde", "Fulani", "Fulfulde", "Maasina Fulfulde", "Nigerian Fulfulde", "Pular", "Pulaar", "Western Niger Fulfulde"}, -- Maasina, etc are dialects, subsumed into this code
	scripts = {"Latn", "Adlm"},
}

m["fi"] = {
	"bahasa Finlandia",
	"Q1412",
	"fiu-fin",
	otherNames = {"Suomi", "Botnian"},
	scripts = Latn,
	entry_name = {
		from = {"ˣ"},  -- Used to indicate gemination of the next consonant
		to   = {}},
	sort_key = {
		from = {"[áàâã]", "[éèêẽ]", "[íìîĩ]", "[óòôõ]", "[úùûũ]", "[ýỳŷüű]", "[øõő]", "æ" , "œ" , "[čç]", "š", "ž", "ß" , "[':]"},
		to   = {"a"	 , "e"	 , "i"	 , "o"	 , "u"	 ,  "y"	 , "ö"	, "ae", "oe", "c"   , "s", "z", "ss"}} ,
}

m["fj"] = {
	"bahasa Fiji",
	"Q33295",
	"poz-occ",
	scripts = Latn,
}

m["fo"] = {
	"bahasa Faroe",
	"Q25258",
	"gmq",
	scripts = Latn,
	ancestors = {"non"},
}

m["fr"] = {
	"bahasa Prancis",
	"Q150",
	"roa-oil",
	otherNames = {"Modern French"},
	scripts = {"Latn", "Brai"},
	ancestors = {"frm"},
	sort_key = {
		from = {"[áàâä]", "[éèêë]", "[íìîï]", "[óòôö]", "[úùûü]", "[ýỳŷÿ]", "ç", "æ" , "œ" , "'"},
		to   = {"a"	 , "e"	 , "i"	 , "o"	 , "u"	 , "y"	 , "c", "ae", "oe"}},
	standardChars = "A-Za-z0-9ÀÂÇÉÈÊËÎÏÔŒÛÙÜàâçéèêëîïôœûùü«»" .. PUNCTUATION,
}

m["fy"] = {
	"bahasa Frisia Barat",
	"Q27175",
	"gmw-fri",
	otherNames = {"Western Frisian", "Frisian"},
	scripts = Latn,
	ancestors = {"ofs"},
	sort_key = {
		from = {"[àáâä]", "[èéêë]", "[ìíîïyỳýŷÿ]", "[òóôö]", "[ùúûü]", "æ", "[ /.-]"},
		to   = {"a"	 , "e"	, "i"	, "o"	, "u", "ae"}} ,
	standardChars = "A-PR-WYZa-pr-wyz0-9Ææâäàéêëèïìôöòúûüùỳ" .. PUNCTUATION,
}

m["ga"] = {
	"bahasa Irlandia",
	"Q9142",
	"cel-gae",
	otherNames = {"Irish Gaelic"},
	scripts = Latn,
	ancestors = {"mga"},
	sort_key = {
		from = {"á", "é", "í", "ó", "ú", "ý", "ḃ" , "ċ" , "ḋ" , "ḟ" , "ġ" , "ṁ" , "ṗ" , "ṡ" , "ṫ" },
		to   = {"a", "e", "i", "o", "u", "y", "bh", "ch", "dh", "fh", "gh", "mh", "ph", "sh", "th"}} ,
	standardChars = "A-IL-PR-Ua-il-pr-u0-9ÁáÉéÍíÓóÚú" .. PUNCTUATION,
}

m["gd"] = {
	"bahasa Gaelik Skotlandia",
	"Q9314",
	"cel-gae",
	otherNames = {"Gàidhlig", "Highland Gaelic", "Scots Gaelic", "Scottish"},
	scripts = Latn,
	ancestors = {"mga"},
	sort_key = {
		from = {"[áà]", "[éè]", "[íì]", "[óò]", "[úù]", "[ýỳ]"},
		to   = {"a"   , "e"   , "i"   , "o"   , "u"   , "y"   }} ,
	standardChars = "A-IL-PR-Ua-il-pr-u0-9ÀàÈèÌìÒòÙù" .. PUNCTUATION,
}

m["gl"] = {
	"bahasa Galisia",
	"Q9307",
	"roa",
	scripts = Latn,
	ancestors = {"roa-opt"},
	sort_key = {
		from = {"á", "é", "í", "ó", "ú"},
		to   = {"a", "e", "i", "o", "u"}} ,
}

m["gn"] = {
	"bahasa Guaraní",
	"Q35876",
	"tup-gua",
	scripts = Latn,
}

m["gu"] = {
	"bahasa Gujarat",
	"Q5137",
	"inc",
	scripts = {"Gujr"},
	ancestors = {"inc-mgu"},
	translit_module = "gu-translit",
}

m["gv"] = {
	"bahasa Manx",
	"Q12175",
	"cel-gae",
	otherNames = {"Manx Gaelic"},
	scripts = Latn,
	ancestors = {"mga"},
	sort_key = {
		from = {"ç", "-"},
		to   = {"c"}} ,
	standardChars = "A-WYÇa-wyç0-9" .. PUNCTUATION,
}

m["ha"] = {
	"bahasa Hausa",
	"Q56475",
	"cdc-wst",
	scripts = LatnArab,
    sort_key = {
		from = {"ɓ",   "ɗ",   "ƙ",  "'y", "ƴ",  "'" },
		to   = {"b~" , "d~"	, "k~", "y~", "y~", ""  }},
    entry_name = {
		from = {"R̃", "r̃", "À", "à", "È", "è", "Ì", "ì", "Ò", "ò", "Ù", "ù", "Â", "â", "Ê", "ê", "Î", "î", "Ô", "ô", "Û", "û", "Ā", "ā", "Ē", "ē", "Ī", "ī", "Ō", "ō", "Ū", "ū", "Á", "á", "É", "é", "Í", "í", "Ó", "ó", "Ú", "ú", "Ā̀", "ā̀", "Ḕ", "ḕ", "Ī̀", "ī̀", "Ṑ", "ṑ", "Ū̀", "ū̀", GRAVE, ACUTE},
		to   = {"R", "r", "A", "a", "E", "e", "I", "i", "O", "o", "U", "u", "A", "a", "E", "e", "I", "i", "O", "o", "U", "u", "A", "a", "E", "e", "I", "i", "O", "o", "U", "u", "A", "a", "E", "e", "I", "i", "O", "o", "U", "u", "A", "a", "E", "e", "I", "i", "O", "o", "U", "u"}},
}

m["he"] = {
	"bahasa Ibrani",
	"Q9288",
	"sem-can",
	otherNames = {"Ivrit"},
	scripts = {"Hebr", "Phnx", "Brai"},
	entry_name = {
		from = {"[" .. u(0x0591) .. "-" .. u(0x05BD) .. u(0x05BF) .. "-" .. u(0x05C5) .. u(0x05C7) .. "]"},
		to   = {}} ,
}

m["hi"] = {
	"bahasa Hindi",
	"Q1568",
	"inc",
	otherNames = {"Hindavi", "Khariboli", "Khari Boli", "Manak Hindi"},
	scripts = {"Deva", "Kthi", "Newa"},
	ancestors = {"inc-ohi"},
	translit_module = "hi-translit",
}

m["ho"] = {
	"bahasa Hiri Motu",
	"Q33617",
	"crp",
	otherNames = {"Pidgin Motu", "Police Motu"},
	scripts = Latn,
	ancestors = {"meu"},
}

m["ht"] = {
	"bahasa Kreol Haiti",
	"Q33491",
	"crp",
	otherNames = {"Creole", "Haitian", "Kreyòl"},
	scripts = Latn,
}

m["hu"] = {
	"bahasa Hungaria",
	"Q9067",
	"urj-ugr",
	otherNames = {"Magyar"},
	scripts = {"Latn", "Hung"},
	ancestors = {"ohu"},
	sort_key = {
		from = {"á", "é", "í", "ó", "ú", "ő", "ű"},
		to   = {"a", "e", "i", "o", "u", "ö", "ü"}} ,
}

m["hy"] = {
	"bahasa Armenian",
	"Q8785",
	"hyx",
	otherNames = {"Modern Armenian", "Eastern Armenian", "Western Armenian"},
	scripts = {"Armn", "Brai"},
	ancestors = {"axm"},
	translit_module = "Armn-translit",
	override_translit = true,
	sort_key = {
		from = {"ու", "և", "եւ"},
		to   = {"ւ", "եվ", "եվ"}},
	entry_name = {
		from = {"՞", "՜", "՛", "՟", "և", "<sup>յ</sup>", "<sup>ի</sup>"},
		to   = {"", "", "", "", "եւ", "յ", "ի"}} ,
}

m["hz"] = {
	"bahasa Herero",
	"Q33315",
	"bnt",
	scripts = Latn,
}

m["ia"] = {
	"bahasa Interlingua",
	"Q35934",
	"art",
	scripts = Latn,
}

m["id"] = {
	"bahasa Indonesia",
	"Q9240",
	"poz-mly",
	scripts = Latn,
	ancestors = {"ms"}
}

m["ie"] = {
	"bahasa Interlingue",
	"Q35850",
	"art",
	otherNames = {"Occidental"},
	scripts = Latn,
}

m["ig"] = {
	"bahasa Igbo",
	"Q33578",
	"nic-bco",
	scripts = Latn,
}

m["ii"] = {
	"bahasa Sichuan Yi",
	"Q34235",
	"tbq-lol",
	otherNames = {"Nuosu", "Nosu", "Northern Yi", "Liangshan Yi"},
	scripts = {"Yiii"},
	translit_module = "ii-translit",
}

m["ik"] = {
	"bahasa Inupiak",
	"Q27183",
	"esx-inu",
	otherNames = {"Inupiaq", "Iñupiaq", "Inupiatun"},
	scripts = Latn,
}

m["io"] = {
	"bahasa Ido",
	"Q35224",
	"art",
	scripts = Latn,
}

m["is"] = {
	"bahasa Islandia",
	"Q294",
	"gmq",
	scripts = Latn,
	ancestors = {"non"},
}

m["it"] = {
	"bahasa Italia",
	"Q652",
	"roa",
	scripts = Latn,
	sort_key = {
		from = {"[àáâäå]", "[èéêë]", "[ìíîï]", "[òóôö]", "[ùúûü]"},
		to   = {"a"	  , "e"	 , "i"	 , "o"	 , "u"	 }} ,
}

m["iu"] = {
	"bahasa Inuktitut",
	"Q29921",
	"esx-inu",
	otherNames = {"Eastern Canadian Inuktitut", "Eastern Canadian Inuit", "Western Canadian Inuktitut", "Western Canadian Inuit", "Western Canadian Inuktun", "Inuinnaq", "Inuinnaqtun", "Inuvialuk", "Inuvialuktun", "Nunavimmiutit", "Nunatsiavummiut", "Aivilimmiut", "Natsilingmiut", "Kivallirmiut", "Siglit", "Siglitun"},
	scripts = {"Cans", "Latn"},
	translit_module = "iu-translit",
	override_translit = true,
}

m["ja"] = {
	"bahasa Jepang",
	"Q5287",
	"jpx",
	otherNames = {"Modern Japanese", "Nipponese", "Nihongo"},
	scripts = {"Jpan", "Latn", "Hira", "Brai"},
	ancestors = {"ojp"},
	--[=[
	-- Handled by jsort function in [[Module:ja]].
	sort_key = {
		from = {"[ぁァア]", "[ぃィイ]", "[ぅゔゥウヴ]", "[ぇェエ]", "[ぉォオ]", "[がゕカガヵ]", "[ぎキギ]", "[ぐクグㇰ]", "[げゖケゲヶ]", "[ごコゴ]", "[ざサザ]", "[じシジㇱ]", "[ずスズㇲ]", "[ぜセゼ]", "[ぞソゾ]", "[だタダ]", "[ぢチヂ]", "[っづッツヅ]", "[でテデ]", "[どトドㇳ]", "ナ", "ニ", "[ヌㇴ]", "ネ", "ノ", "[ばぱハバパㇵ]", "[びぴヒビピㇶ]", "[ぶぷフブプㇷ]", "[べぺヘベペㇸ]", "[ぼぽホボポㇹ]", "マ", "ミ", "[ムㇺ]", "メ", "モ", "[ゃャヤ]", "[ゅュユ]", "[ょョヨ]", "[ラㇻ]", "[リㇼ]", "[ルㇽ]", "[レㇾ]", "[ロㇿ]", "[ゎヮワヷ]", "[ヰヸ]", "[ヱヹ]", "[ヲヺ]", "ン", "[゙゚゛゜ゝゞ・ヽヾ]", "𛀀"},
		to   = {"あ", "い", "う", "え", "お", "か", "き", "く", "け", "こ", "さ", "し", "す", "せ", "そ", "た", "ち", "つ", "て", "と", "な", "に", "ぬ", "ね", "の", "は", "ひ", "ふ", "へ", "ほ", "ま", "み", "む", "め", "も", "や", "ゆ", "よ", "ら", "り", "る", "れ", "ろ", "わ", "ゐ", "ゑ", "を", "ん", "", "え"}},
	--]=]
}

m["jv"] = {
	"bahasa Jawa",
	"Q33549",
	"poz-sus",
	scripts = {"Latn", "Java"},
	translit_module = "jv-translit",
	ancestors = {"kaw"},
	link_tr = true,
}

m["ka"] = {
	"bahasa Georgia",
	"Q8108",
	"ccs-gzn",
	otherNames = {"Kartvelian", "Judeo-Georgian", "Kivruli", "Gruzinic"},
	scripts = {"Geor", "Geok", "Hebr"}, -- Hebr is used to write Judeo-Georgian
	ancestors = {"oge"},
	translit_module = "Geor-translit",
	override_translit = true,
	entry_name = {
		from = {"̂"},
		to   = {""}},
}

m["kg"] = {
	"bahasa Kongo",
	"Q33702",
	"bnt",
	otherNames = {"Kikongo", "Koongo", "Laari", "San Salvador Kongo", "Yombe"},
	scripts = Latn,
}

m["ki"] = {
	"bahasa Kikuyu",
	"Q33587",
	"bnt",
	otherNames = {"Gikuyu", "Gĩkũyũ"},
	scripts = Latn,
}

m["kj"] = {
	"bahasa Kwanyama",
	"Q1405077",
	"bnt",
	otherNames = {"Kuanyama", "Oshikwanyama"},
	scripts = Latn,
}

m["kk"] = {
	"bahasa Kazakh",
	"Q9252",
	"trk-kip",
	scripts = {"Cyrl", "Latn", "kk-Arab"},
	translit_module = "kk-translit",
	override_translit = true,
}

m["kl"] = {
	"bahasa Greenland",
	"Q25355",
	"esx-inu",
	otherNames = {"Kalaallisut"},
	scripts = Latn,
}

m["km"] = {
	"bahasa Khmer",
	"Q9205",
	"mkh",
	otherNames = {"Cambodian"},
	scripts = {"Khmr"},
	ancestors = {"mkh-mkm"},
	translit_module = "km-translit",
}

m["kn"] = {
	"bahasa Kannada",
	"Q33673",
	"dra",
	scripts = {"Knda"},
	ancestors = {"dra-mkn"},
	translit_module = "kn-translit",
}

m["ko"] = {
	"bahasa Korea",
	"Q9176",
	"qfa-kor",
	otherNames = {"Modern Korean"},
	scripts = {"Kore", "Hani", "Brai"},
	ancestors = {"okm"},
	translit_module = "ko-translit",
}

m["kr"] = {
	"bahasa Kanuri",
	"Q36094",
	"ssa-sah",
	otherNames = {"Kanembu", "Bilma Kanuri", "Central Kanuri", "Manga Kanuri", "Tumari Kanuri"},
	scripts = LatnArab,
	sort_key = {
		from = {"ny", "ǝ", "sh"},
		to   = {"n~", "e~", "s~"}} , -- the sortkey and entry_name are only for standard Kanuri; when dialectal entries get added, someone will have to work out how the dialects should be represented orthographically
	entry_name = {
		from = {"À", "à", "È", "è", "Ǝ̀", "ǝ̀", "Ì", "ì", "Ò", "ò", "Ù", "ù", "Â", "â", "Ê", "ê", "Ǝ̂", "ǝ̂", "Î", "î", "Ô", "ô", "Û", "û", "Ă", "ă", "Ĕ", "ĕ", "Ǝ̆", "ǝ̆", "Ĭ", "ĭ", "Ŏ", "ŏ", "Ŭ", "ŭ", "Á", "á", "É", "é", "Ǝ́", "ǝ́", "Í", "í", "Ó", "ó", "Ú", "ú", GRAVE, ACUTE},
		to   = {"A", "a", "E", "e", "Ǝ", "ǝ", "I", "i", "O", "o", "U", "u", "A", "a", "E", "e", "Ǝ", "ǝ", "I", "i", "O", "o", "U", "u", "A", "a", "E", "e", "Ǝ", "ǝ", "I", "i", "O", "o", "U", "u", "A", "a", "E", "e", "Ǝ", "ǝ", "I", "i", "O", "o", "U", "u"}},
}

m["ks"] = {
	"bahasa Kashmir",
	otherNames = {"Koshur"},
	"Q33552",
	"inc-dar",
	scripts = {"ks-Arab", "Deva", "Shrd", "Latn"},
	translit_module = "translit-redirect",
	ancestors = {"inc-dar-pro"},
}

m["ku"] = {
	"bahasa Kurdish",
	"Q36368",
	"ira-nwi",
	scripts = {"Latn", "ku-Arab", "Armn", "Cyrl"},
	translit_module = "translit-redirect",
}

-- "kv" IS TREATED AS "koi", "kpv", SEE WT:LT

m["kw"] = {
	"bahasa Kernowek",
	"Q25289",
	"cel-bry",
	scripts = Latn,
	ancestors = {"cnx"},
}

m["ky"] = {
	"bahasa Kyrgyz",
	"Q9255",
	"trk-kip",
	otherNames = {"Kirghiz", "Kirgiz"},
	scripts = {"Cyrl", "Latn", "Arab"},
	translit_module = "ky-translit",
	override_translit = true,
}

m["la"] = {
	"bahasa Latin",
	"Q397",
	"itc",
	scripts = Latn,
	ancestors = {"itc-ola"},
	entry_name = {
		from = {"[ĀĂ]", "[āă]", "[ĒĔ]", "[ēĕë]", "[ĪĬÏ]", "[īĭï]", "[ŌŎ]", "[ōŏ]", "[ŪŬÜ]", "[ūŭü]", "Ȳ", "ȳ", MACRON, BREVE, DIAER},
		to   = {"A", "a", "E", "e", "I", "i", "O", "o", "U", "u", "Y", "y"}},
	standardChars = "A-Za-zÆ挜Ā-ăĒ-ĕĪ-ĭŌ-ŏŪ-ŭȲȳ" .. MACRON .. BREVE .. PUNCTUATION,
}

m["lb"] = {
	"bahasa Luksemburg",
	"Q9051",
	"gmw",
	scripts = Latn,
	ancestors = {"gmh"},
}

m["lg"] = {
	"bahasa Luganda",
	"Q33368",
	"bnt",
	otherNames = {"Ganda", "Oluganda"},
	scripts = Latn,
	entry_name = {
		from = {"á", "Á", "é", "É", "í", "Í", "ó", "Ó", "ú", "Ú", "ń", "Ń", "ḿ", "Ḿ", "â", "Â", "ê", "Ê", "î", "Î", "ô", "Ô", "û", "Û" },
		to   = {"a", "A", "e", "E", "i", "I", "o", "O", "u", "U", "n", "N", "m", "M", "a", "A", "e", "E", "i", "I", "o", "O", "u", "U",}},
	sort_key = {
		from = {"ŋ"},
		to   = {"n"}} ,
}

m["li"] = {
	"bahasa Limburg",
	"Q102172",
	"gmw",
	otherNames = {"Limburgan", "Limburgian", "Limburgic"},
	scripts = Latn,
	ancestors = {"dum"},
}

m["ln"] = {
	"bahasa Lingala",
	"Q36217",
	"bnt",
	otherNames = {"Ngala"},
	scripts = Latn,
}

m["lo"] = {
	"bahasa Laos",
	"Q9211",
	"tai-swe",
	otherNames = {"Laotian"},
	scripts = {"Laoo"},
	translit_module = "lo-translit",
	sort_key = {
		from = {"[່-ໍ]", "ຼ", "ຽ", "ໜ", "ໝ", "([ເແໂໃໄ])([ກ-ຮ])"},
		to   = {"", "ລ", "ຍ", "ຫນ", "ຫມ", "%2%1"}},
}

m["lt"] = {
	"bahasa Lithuania",
	"Q9083",
	"bat",
	scripts = Latn,
	ancestors = {"olt"},
	entry_name = {
		from = {"[áãà]", "[ÁÃÀ]", "[éẽè]", "[ÉẼÈ]", "[íĩì]", "[ÍĨÌ]", "[ýỹ]", "[ÝỸ]", "ñ", "[óõò]", "[ÓÕÒ]", "[úũù]", "[ÚŨÙ]", ACUTE, GRAVE, TILDE},
		to   = {"a",       "A",     "e",     "E",     "i",     "I",     "y",   "Y",   "n",   "o",    "O",     "u",      "U"}} ,
}

m["lu"] = {
	"bahasa Luba-Katanga",
	"Q36157",
	"bnt",
	scripts = Latn,
}

m["lv"] = {
	"bahasa Latvia",
	"Q9078",
	"bat",
	otherNames = {"Lettish", "Lett"},
	scripts = Latn,
	entry_name = {
		-- This attempts to convert vowels with tone marks to vowels either with
		-- or without macrons. Specifically, there should be no macrons if the
		-- vowel is part of a diphthong (including resonant diphthongs such
		-- pìrksts -> pirksts not #pīrksts). What we do is first convert the
		-- vowel + tone mark to a vowel + tilde in a decomposed fashion,
		-- then remove the tilde in diphthongs, then convert the remaining
		-- vowel + tilde sequences to macroned vowels, then delete any other
		-- tilde. We leave already-macroned vowels alone: Both e.g. ar and ār
		-- occur before consonants. FIXME: This still might not be sufficient.
		from = {"Ȩ", "ȩ", "[ÂÃÀ]", "[âãà]", "[ÊẼÈ]", "[êẽè]", "[ÎĨÌ]", "[îĩì]", "[ÔÕÒ]", "[ôõò]", "[ÛŨÙ]", "[ûũù]", "[ÑǸ]", "[ñǹ]", "[" .. CIRC .. TILDE ..GRAVE .."]", "([aAeEiIoOuU])" .. TILDE .."?([lrnmuiLRNMUI])" .. TILDE .. "?([^aAeEiIoOuUāĀēĒīĪūŪ])", "([aAeEiIoOuU])" .. TILDE .."?([lrnmuiLRNMUI])" .. TILDE .."?$", "([iI])" .. TILDE .. "?([eE])" .. TILDE .. "?", "A" .. TILDE, "a" .. TILDE, "E" .. TILDE, "e" .. TILDE, "I" .. TILDE, "i" .. TILDE, "U" .. TILDE, "u" .. TILDE, TILDE},
		to   = {"E", "e", "A" .. TILDE, "a" .. TILDE, "E" .. TILDE, "e" .. TILDE, "I" .. TILDE, "i" .. TILDE, "O", "o", "U" .. TILDE, "u" .. TILDE, "N", "n", TILDE, "%1%2%3", "%1%2", "%1%2", "Ā", "ā", "Ē", "ē", "Ī", "ī", "Ū", "ū", ""}},
}

m["mg"] = {
	"bahasa Malagasi",
	"Q7930",
	"poz-bre",
	otherNames = {"Betsimisaraka Malagasy", "Betsimisaraka", "Northern Betsimisaraka Malagasy", "Northern Betsimisaraka", "Southern Betsimisaraka Malagasy", "Southern Betsimisaraka", "Bara Malagasy", "Bara", "Masikoro Malagasy", "Masikoro", "Antankarana", "Antankarana Malagasy", "Plateau Malagasy", "Sakalava", "Tandroy Malagasy", "Tandroy", "Tanosy", "Tanosy Malagasy", "Tesaka", "Tsimihety", "Tsimihety Malagasy", "Bushi", "Shibushi", "Kibushi"},
	scripts = Latn,
}

m["mh"] = {
	"bahasa Marshall",
	"Q36280",
	"poz-mic",
	scripts = Latn,
	sort_key = {
		from = {"ā" , "ļ" , "m̧" , "ņ" , "n̄"  , "o̧" , "ō"  , "ū" },
		to   = {"a~", "l~", "m~", "n~", "n~~", "o~", "o~~", "u~"}} ,
}

m["mi"] = {
	"bahasa Maori",
	"Q36451",
	"poz-pep",
	otherNames = {"Māori"},
	scripts = Latn,
}

m["mk"] = {
	"bahasa Makedonia",
	"Q9296",
	"zls",
	scripts = Cyrl,
	translit_module = "mk-translit",
	entry_name = {
		from = {ACUTE},
		to   = {}},
}

m["ml"] = {
	"bahasa Malayalam",
	"Q36236",
	"dra",
	scripts = {"Mlym"},
	translit_module = "ml-translit",
	override_translit = true,
}

m["mn"] = {
	"bahasa Mongolia",
	"Q9246",
	"xgn",
	otherNames = {"Khalkha Mongolian"},
	scripts = {"Cyrl", "Mong", "Soyo", "Zanb"}, -- entries in Soyo or Zanb might require prior discussion
	ancestors = {"cmg"},
	translit_module = "mn-translit",
	override_translit = true,
}

-- "mo" IS TREATED AS "ro", SEE WT:LT

m["mr"] = {
	"bahasa Marathi",
	"Q1571",
	"inc",
	scripts = {"Deva", "Modi"},
	ancestors = {"omr"},
	translit_module = "mr-translit",
}

m["ms"] = {
	"bahasa Melayu",
	"Q9237",
	"poz-mly",
	otherNames = {"Malaysian", "Standard Malay"},
	scripts = {"Latn", "ms-Arab"},
	ancestors = {"msa-cls"}
}

m["mt"] = {
	"bahasa Malta",
	"Q9166",
	"sem-arb",
	scripts = Latn,
	ancestors = {"sqr"},
}

m["my"] = {
	"bahasa Myanmar",
	"Q9228",
	"tbq-brm",
	otherNames = {"Burma"},
	scripts = {"Mymr"},
	ancestors = {"obr"},
	translit_module = "my-translit",
	override_translit = true,
}

m["na"] = {
	"bahasa Nauru",
	"Q13307",
	"poz-mic",
	otherNames = {"Nauru"},
	scripts = Latn,
}

m["nb"] = {
	"bahasa Norwegia Bokmål",
	"Q25167",
	"gmq",
	otherNames = {"Bokmål"},
	scripts = Latn,
	ancestors = {"gmq-mno"},
	wikimedia_codes = {"no"},
}

m["nd"] = {
	"bahasa Ndebele Utara",
	"Q35613",
	"bnt-ngu",
	otherNames = {"North Ndebele"},
	scripts = Latn,
	entry_name = {
		from = {"[āàáâǎ]", "[ēèéêě]", "[īìíîǐ]", "[ōòóôǒ]", "[ūùúûǔ]", "ḿ", "[ǹńň]", MACRON, ACUTE, GRAVE, CIRC, CARON},
		to   = {"a"      , "e"      , "i"      , "o"      , "u"      , "m", "n"    }},
}

m["ne"] = {
	"bahasa Nepal",
	"Q33823",
	"inc-pah",
	otherNames = {"Nepalese"},
	scripts = {"Deva", "Newa"},
	translit_module = "ne-translit",
	ancestors = {"psu"},
}

m["ng"] = {
	"bahasa Ndonga",
	"Q33900",
	"bnt",
	scripts = Latn,
}

m["nl"] = {
	"bahasa Belanda",
	"Q7411",
	"gmw",
	otherNames = {"Netherlandic", "Flemish"},
	scripts = Latn,
	ancestors = {"dum"},
	sort_key = {
		from = {"[äáâå]", "[ëéê]", "[ïíî]", "[öóô]", "[üúû]", "ç", "ñ", "^-"},
		to   = {"a"	 , "e"	, "i"	, "o"	, "u"	, "c", "n"}} ,
	standardChars = "A-Za-z0-9" .. PUNCTUATION .. u(0x2800) .. "-" .. u(0x28FF),
}

m["nn"] = {
	"bahasa Norwegia Nynorsk",
	"Q25164",
	"gmq",
	otherNames = {"New Norwegian", "Nynorsk"},
	scripts = Latn,
	ancestors = {"gmq-mno"},
}

m["no"] = {
	"bahasa Norwegia",
	"Q9043",
	"gmq",
	scripts = Latn,
	ancestors = {"gmq-mno"},
}

m["nr"] = {
	"bahasa Ndebele Selatan",
	"Q36785",
	"bnt-ngu",
	otherNames = {"South Ndebele"},
	scripts = Latn,
	entry_name = {
		from = {"[āàáâǎ]", "[ēèéêě]", "[īìíîǐ]", "[ōòóôǒ]", "[ūùúûǔ]", "ḿ", "[ǹńň]", MACRON, ACUTE, GRAVE, CIRC, CARON},
		to   = {"a"      , "e"      , "i"      , "o"      , "u"      , "m", "n"    }},
}

m["nv"] = {
	"bahasa Navajo",
	"Q13310",
	"apa",
	otherNames = {"Navaho", "Diné bizaad"},
	scripts = {"nv-Latn"},
	sort_key = {
		from = {"[áą]", "[éę]", "[íį]", "[óǫ]", "ń", "^n([djlt])", "ł" , "[ʼ’']", ACUTE},
		to   = {"a"   , "e"   , "i"   , "o"   , "n", "ni%1"	  , "l~"}}, -- the tilde is used to guarantee that ł will always be sorted after all other words with l
}

m["ny"] = {
	"bahasa Chichewa",
	"Q33273",
	"bnt",
	otherNames = {"Chicheŵa", "Chinyanja", "Nyanja", "Chewa", "Cicewa", "Cewa", "Cinyanja"},
	scripts = Latn,
	entry_name = {
		from = {"ŵ", "Ŵ", "á", "Á", "é", "É", "í", "Í", "ó", "Ó", "ú", "Ú", "ń", "Ń", "ḿ", "Ḿ" },
		to   = {"w", "W", "a", "A", "e", "E", "i", "I", "o", "O", "u", "U", "n", "N", "m", "M"}},
	sort_key = {
		from = {"ng'"},
		to   = {"ng"}} ,
}

m["oc"] = {
	"bahasa Oksitan",
	"Q14185",
	"roa",
	otherNames = {"Provençal", "Auvergnat", "Auvernhat", "Gascon", "Languedocien", "Lengadocian", "Shuadit", "Chouhadite", "Chouhadit", "Chouadite", "Chouadit", "Shuhadit", "Judeo-Provençal", "Judeo-Provencal", "Judeo-Comtadin"},
	scripts = {"Latn", "Hebr"},
	ancestors = {"pro"},
	sort_key = {
		from = {"[àá]", "[èé]", "[íï]", "[òó]", "[úü]", "ç", "([lns])·h"},
		to   = {"a"   , "e"   , "i"   , "o"   , "u"   , "c", "%1h"	  }} ,
}

m["oj"] = {
	"bahasa Ojibwe",
	"Q33875",
	"alg",
	otherNames = {"Chippewa", "Ojibway", "Ojibwemowin", "Southwestern Ojibwa"},
	scripts = {"Cans", "Latn"},
}

m["om"] = {
	"bahasa Oromo",
	"Q33864",
	"cus",
	otherNames = {"Orma", "Borana-Arsi-Guji Oromo", "West Central Oromo"},
	scripts = {"Latn", "Ethi"},
}

m["or"] = {
	"bahasa Oriya",
	"Q33810",
	"inc",
	otherNames = {"Odia", "Oorya"},
	scripts = {"Orya"},
	ancestors = {"inc-mor"},
	translit_module = "or-translit",
}

m["os"] = {
	"bahasa Ossetia",
	"Q33968",
	"ira-scy",
	otherNames = {"Ossete", "Ossetic", "Digor", "Iron"},
	scripts = {"Cyrl", "Geor", "Latn"},
	ancestors = {"oos"},
	translit_module = "os-translit",
	override_translit = true,
	entry_name = {
		from = {GRAVE, ACUTE},
		to   = {}} ,
}

m["pa"] = {
	"bahasa Punjab",
	"Q58635",
	"inc",
	otherNames = {"Panjabi"},
	scripts = {"Guru", "pa-Arab", "Deva"},
	ancestors = {"psu"},
	translit_module = "translit-redirect",
	entry_name = {
		from = {u(0x064B), u(0x064C), u(0x064D), u(0x064E), u(0x064F), u(0x0650), u(0x0651), u(0x0652)},
		to   = {}} ,
}

m["pi"] = {
	"bahasa Pali",
	"Q36727",
	"inc-old",
	scripts = {"Latn", "Brah", "Deva", "Beng", "Sinh", "Mymr", "Thai", "Lana", "Khmr"}, -- holds Laoo, not enough glyphes to render
	ancestors = {"sa"},
	sort_key = {
		from = {"ā", "ī", "ū", "ḍ", "ḷ", "[ṁṃ]", "[ṇñṅ]", "ṭ", "([เแโใไ])([ก-ฮ])", "ᩔ", "ᩕ", "ᩖ", "ᩘ", "([ᨭ-ᨱ])ᩛ", "([ᨷ-ᨾ])ᩛ", "ᩤ", u(0xFE00)},
		to   = {"a", "i", "u", "d", "l", "m"   , "n"	, "t", "%2%1", "ᩈ᩠ᩈ", "᩠ᩁ", "᩠ᩃ", "ᨦ᩠", "%1᩠ᨮ", "%1᩠ᨻ", "ᩣ"}} ,
	entry_name = {
		from = {u(0xFE00)},
		to   = {}},
}

m["pl"] = {
	"bahasa Polandia",
	"Q809",
	"zlw",
	scripts = Latn,
	ancestors = {"zlw-opl"},
	sort_key = {
		from = {"[Ąą]", "[Ćć]", "[Ęę]", "[Łł]", "[Ńń]", "[Óó]", "[Śś]", "[Żż]", "[Źź]"},
		to   = {
			"a" .. u(0x10FFFF),
			"c" .. u(0x10FFFF),
			"e" .. u(0x10FFFF),
			"l" .. u(0x10FFFF),
			"n" .. u(0x10FFFF),
			"o" .. u(0x10FFFF),
			"s" .. u(0x10FFFF),
			"z" .. u(0x10FFFF),
			"z" .. u(0x10FFFE)}} ,
}

m["ps"] = {
	"bahasa Pashtun",
	"Q58680",
	"ira-sei",
	otherNames = {"Pashtun", "Pushto", "Pashtu", "Central Pashto", "Northern Pashto", "Southern Pashto", "Pukhto", "Pakhto", "Pakkhto", "Afghani"},
	scripts = {"ps-Arab"},
}

m["pt"] = {
	"bahasa Portugis",
	"Q5146",
	"roa",
	otherNames = {"Modern Portuguese"},
	scripts = {"Latn", "Brai"},
	ancestors = {"roa-opt"},
	sort_key = {
		from = {"[àãáâä]", "[èẽéêë]", "[ìĩíï]", "[òóôõö]", "[üúùũ]", "ç", "ñ"},
		to   = {"a"	  , "e"	  , "i"	 , "o"	  , "u"	 , "c", "n"}} ,
}

m["qu"] = {
	"bahasa Quechua",
	"Q5218",
	"qwe",
	scripts = Latn,
}

m["rm"] = {
	"bahasa Romansh",
	"Q13199",
	"roa-rhe",
	otherNames = {"Romansh", "Rumantsch", "Romanche"},
	scripts = Latn,
}

m["ro"] = {
	"bahasa Romania",
	"Q7913",
	"roa-eas",
	otherNames = {"Daco-Romanian", "Roumanian", "Rumanian"},
	scripts = {"Latn", "Cyrl"},
	sort_key = {
        from = {"ă" , "â"  , "î" , "ș" , "ț" },
        to   = {"a~", "a~~", "i~", "s~", "t~"}},
}

m["ru"] = {
	"bahasa Rusia",
	"Q7737",
	"zle",
	scripts = {"Cyrl", "Brai"},
	translit_module = "ru-translit",
	sort_key = {
		from = {"ё"},
		to   = {"е" .. mw.ustring.char(0x10FFFF)}},
	entry_name = {
		from = {"Ѐ", "ѐ", "Ѝ", "ѝ", GRAVE, ACUTE},
		to   = {"Е", "е", "И", "и"}},
	standardChars = "ЁІА-яёі0-9—" .. PUNCTUATION,
}

m["rw"] = {
	"bahasa Rwanda-Rundi",
	"Q33573",
	"bnt",
	otherNames = {"Rwanda", "Kinyarwanda", "Rundi", "Kirundi", "Ha", "Giha", "Hangaza", "Vinza", "Shubi", "Subi"},
	scripts = Latn,
}

m["sa"] = {
	"bahasa Sanskerta",
	"Q11059",
	"inc",
	scripts = {"Deva", "Bali", "as-Beng", "Beng", "Bhks", "Brah", "Gran", "Gujr", "Guru", "Java", "Khar", "Khmr", "Knda", "Lana", "Mlym", "Mymr", "Newa", "Orya", "Saur", "Shrd", "Sidd", "Sinh", "Taml", "Telu", "Thai", "Tibt", "Tirh"}, -- holds Laoo, not enough glyphes to render
	sort_key = {
		from = {"ā", "ī", "ū", "ḍ", "ḷ", "[ṁṃ]", "[ṇñṅ]", "ṭ", "([เแโใไ])([ก-ฮ])", "ᩔ", "ᩕ", "ᩖ", "ᩘ", "([ᨭ-ᨱ])ᩛ", "([ᨷ-ᨾ])ᩛ", "ᩤ", u(0xFE00)},
		to   = {"a", "i", "u", "d", "l", "m"   , "n"	, "t", "%2%1", "ᩈ᩠ᩈ", "᩠ᩁ", "᩠ᩃ", "ᨦ᩠", "%1᩠ᨮ", "%1᩠ᨻ", "ᩣ"}} ,
	entry_name = {
		from = {u(0xFE00)},
		to   = {}},
	translit_module = "translit-redirect",
}

m["sc"] = {
	"bahasa Sardinia",
	"Q33976",
	"roa",
	otherNames = {"Campidanese", "Campidanese Sardinian", "Logudorese", "Logudorese Sardinian", "Nuorese", "Nuorese Sardinian"},
	scripts = Latn,
}

m["sd"] = {
	"bahasa Sindhi",
	"Q33997",
	"inc",
	scripts = {"sd-Arab", "Deva", "Sind", "Khoj"},
	entry_name = {
		from = {u(0x0671), u(0x064B), u(0x064C), u(0x064D), u(0x064E), u(0x064F), u(0x0650), u(0x0651), u(0x0652), u(0x0670), u(0x0640)},
		to   = {u(0x0627)}},
	ancestors = {"psu"},
}

m["se"] = {
	"bahasa Sami Utara",
	"Q33947",
	"smi",
	otherNames = {"North Sami", "Northern Saami", "North Saami"},
	scripts = Latn,
	entry_name = {
		from = {"ạ", "[ēẹ]", "ī", "[ōọ]", "ū", "ˈ"},
		to   = {"a", "e"   , "i", "o"   , "u"} },
	sort_key = {
		from = {"á" , "č" , "đ" , "ŋ" , "š" , "ŧ" , "ž" },
		to   = {"a²", "c²", "d²", "n²", "s²", "t²", "z²"} },
	standardChars = "A-PR-VZa-pr-vz0-9ÁáČčĐđŊŋŠšŦŧŽž" .. PUNCTUATION,
}

m["sg"] = {
	"bahasa Sango",
	"Q33954",
	"crp",
	scripts = Latn,
}

m["sh"] = {
	"bahasa Serbo-Kroasia",
	"Q9301",
	"zls",
	otherNames = {"BCS", "Croato-Serbian", "Serbocroatian", "Bosnian", "Croatian", "Montenegrin", "Serbian"},
	scripts = {"Latn", "Cyrl"},
	entry_name = {
		from = {"[ȀÀȂÁĀÃ]", "[ȁàȃáāã]", "[ȄÈȆÉĒẼ]", "[ȅèȇéēẽ]", "[ȈÌȊÍĪĨ]", "[ȉìȋíīĩ]", "[ȌÒȎÓŌÕ]", "[ȍòȏóōõ]", "[ȐȒŔ]", "[ȑȓŕ]", "[ȔÙȖÚŪŨ]", "[ȕùȗúūũ]", "Ѐ", "ѐ", "[ӢЍ]", "[ӣѝ]", "[Ӯ]", "[ӯ]", GRAVE, ACUTE, DGRAVE, INVBREVE, MACRON, TILDE},
		to   = {"A"	  , "a"	  , "E"	  , "e"	  , "I"	  , "i"	  , "O"	  , "o"	  , "R"	, "r"	, "U"	  , "u"	  , "Е", "е", "И"   , "и", "У", "у"   }},
	wikimedia_codes = {"sh", "bs", "hr", "sr"},
}

m["si"] = {
	"bahasa Sinhala",
	"Q13267",
	"inc",
	otherNames = {"Singhalese", "Sinhala"},
	scripts = {"Sinh"},
	ancestors = {"elu-prk"},
	translit_module = "si-translit",
	override_translit = true,
}

m["sk"] = {
	"bahasa Slowakia",
	"Q9058",
	"zlw",
	scripts = Latn,
	sort_key = {
		from = {"[áä]", "é", "í", "[óô]", "ú", "ý", "ŕ", "ĺ", "[" .. DIAER .. ACUTE .. CIRC .. "]"},
		to   = {"a"   , "e", "i", "o"   , "u", "y", "r", "l", ""}} ,
}

m["sl"] = {
	"bahasa Slovenia",
	"Q9063",
	"zls",
	otherNames = {"Slovenian"},
	scripts = Latn,
	entry_name = {
		from = {"[ÁÀÂȂȀ]", "[áàâȃȁ]", "[ÉÈÊȆȄỆẸ]", "[éèêȇȅệẹə]", "[ÍÌÎȊȈ]", "[íìîȋȉ]", "[ÓÒÔȎȌỘỌ]", "[óòôȏȍộọ]", "[ŔȒȐ]", "[ŕȓȑ]", "[ÚÙÛȖȔ]", "[úùûȗȕ]", "ł", GRAVE, ACUTE, DGRAVE, INVBREVE, CIRC, DOTBELOW},
		to   = {"A"	  , "a"	  , "E"		, "e"		 , "I"	  , "i"	  , "O"		, "o"		, "R"	, "r"	, "U"	  , "u"	  , "l"}} ,
}

m["sm"] = {
	"bahasa Samoa",
	"Q34011",
	"poz-pnp",
	scripts = Latn,
}

m["sn"] = {
	"bahasa Shona",
	"Q34004",
	"bnt",
	scripts = Latn,
	entry_name = {
        from = {ACUTE},
        to = {}} ,
}

m["so"] = {
	"bahasa Somalia",
	"Q13275",
	"cus",
	scripts = {"Latn", "Arab", "Osma"},
	entry_name = {
		from = {"[ÁÀÂ]", "[áàâ]", "[ÉÈÊ]", "[éèê]", "[ÍÌÎ]", "[íìî]", "[ÓÒÔ]", "[óòô]", "[ÚÙÛ]", "[úùû]", "[ÝỲ]", "[ýỳ]"},
		to   = {"A"	  , "a"	  , "E"	, "e" , "I"	  , "i"	  , "O"	, "o"	, "U"  , "u", "Y", "y"}} ,
}

m["sq"] = {
	"bahasa Albania",
	"Q8748",
	"sqj",
	otherNames = {"Tosk", "Gheg", "Arvanitika", "Arbëreshë", "Arbëresh"},
	scripts = {"Latn", "Grek", "Elba"},
	sort_key = {
		from = { '[âãä]', '[ÂÃÄ]', '[êẽë]', '[ÊẼË]', 'ĩ', 'Ĩ', 'õ', 'Õ', 'ũ', 'Ũ', 'ỹ', 'Ỹ', 'ç', 'Ç' },
		to   = {     'a',     'A',     'e',     'E', 'i', 'I', 'o', 'O', 'u', 'U', 'y', 'Y', 'c', 'C' } } ,
}

m["ss"] = {
	"bahasa Swati",
	"Q34014",
	"bnt-ngu",
	otherNames = {"Swazi"},
	scripts = Latn,
	entry_name = {
		from = {"[āàáâǎ]", "[ēèéêě]", "[īìíîǐ]", "[ōòóôǒ]", "[ūùúûǔ]", "ḿ", "[ǹńň]", MACRON, ACUTE, GRAVE, CIRC, CARON},
		to   = {"a"      , "e"      , "i"      , "o"      , "u"      , "m", "n"    }},
}

m["st"] = {
	"bahasa Sotho",
	"Q34340",
	"bnt",
	otherNames = {"Sesotho", "Southern Sesotho", "Southern Sotho"},
	scripts = Latn,
}

m["su"] = {
	"bahasa Sunda",
	"Q34002",
	"poz-msa",
	scripts = {"Latn", "Sund"},
	translit_module = "su-translit",
}

m["sv"] = {
	"bahasa Swedia",
	"Q9027",
	"gmq",
	scripts = Latn,
	ancestors = {"gmq-osw"},
}

m["sw"] = {
	"bahasa Swahili",
	"Q7838",
	"bnt",
	otherNames = {"Settler Swahili", "KiSetla", "KiSettla", "Setla", "Settla", "Kitchen Swahili", "Kihindi", "Indian Swahili", "KiShamba", "Kishamba", "Field Swahili", "Kibabu", "Asian Swahili", "Kimanga", "Arab Swahili", "Kitvita", "Army Swahili"},
	scripts = LatnArab,
	sort_key = {
		from = {"ng'", "^-"},
		to   = {"ngz"}} ,
}

m["ta"] = {
	"bahasa Tamil",
	"Q5885",
	"dra",
	scripts = {"Taml"},
	ancestors = {"oty"},
	translit_module = "ta-translit",
	override_translit = true,
}

m["te"] = {
	"bahasa Telugu",
	"Q8097",
	"dra",
	scripts = {"Telu"},
	translit_module = "te-translit",
	override_translit = true,
}

m["tg"] = {
	"bahasa Tajik",
	"Q9260",
	"ira-swi",
	otherNames = {"Tadjik", "Tadzhik", "Tajiki", "Tajik Persian", "Tajiki Persian"},
	scripts = {"Cyrl", "fa-Arab", "Latn"},
	ancestors = {"pal"}, -- same as "fa", see WT:T:AFA
	translit_module = "tg-translit",
	override_translit = true,
	sort_key = {
		from = {"Ё", "ё"},
		to   = {"Е" , "е"}} ,
	entry_name = {
		from = {ACUTE},
		to   = {}} ,
}

m["th"] = {
	"bahasa Thai",
	"Q9217",
	"tai-swe",
	otherNames = {"Siamese", "Central Thai"},
	scripts = {"Thai", "Brai"},
	translit_module = "th-translit",
	sort_key = {
		from = {"%p", "[็-๎]", "([เแโใไ])([ก-ฮ])"},
		to   = {"", "", "%2%1"}},
}

m["ti"] = {
	"bahasa Tigrinya",
	"Q34124",
	"sem-eth",
	otherNames = {"Tigrigna"},
	scripts = {"Ethi"},
	translit_module = "Ethi-translit",
}

m["tk"] = {
	"bahasa Turkmen",
	"Q9267",
	"trk-ogz",
	scripts = {"Latn", "Cyrl"},
	entry_name = {
		from = {"ā", "ē", "ī", "ō", "ū", "ȳ", "ȫ", "ǖ", MACRON},
		to   = {"a", "e", "i", "o", "u", "y", "ö", "ü", ""}},
}

m["tl"] = {
	"bahasa Tagalog",
	"Q34057",
	"phi",
	scripts = {"Latn", "Tglg"},
	entry_name = {
		from = {"[áàâ]", "[éèê]", "[íìî]", "[óòô]", "[úùû]", ACUTE, GRAVE, CIRC},
		to   = {"a"    , "e"    , "i"    , "o"    , "u"    }},
}

m["tn"] = {
	"bahasa Tswana",
	"Q34137",
	"bnt",
	otherNames = {"Setswana"},
	scripts = Latn,
}

m["to"] = {
	"bahasa Tonga",
	"Q34094",
	"poz-pol",
	scripts = Latn,
}

m["tr"] = {
	"bahasa Turki",
	"Q256",
	"trk-ogz",
	scripts = Latn,
	ancestors = {"ota"},
}

m["ts"] = {
	"bahasa Tsonga",
	"Q34327",
	"bnt",
	scripts = Latn,
}

m["tt"] = {
	"bahasa Tatar",
	"Q25285",
	"trk-kip",
	scripts = {"Cyrl", "Latn", "tt-Arab"},
	translit_module = "tt-translit",
	override_translit = true,
}

-- "tw" IS TREATED AS "ak", SEE WT:LT

m["ty"] = {
	"bahasa Tahiti",
	"Q34128",
	"poz-pep",
	scripts = Latn,
}

m["ug"] = {
	"bahasa Uyghur",
	"Q13263",
	"trk",
	otherNames = {"Uigur", "Uighur", "Uygur"},
	scripts = {"ug-Arab", "Latn", "Cyrl"},
	ancestors = {"chg"},
	translit_module = "ug-translit",
	override_translit = true,
}

m["uk"] = {
	"bahasa Ukraina",
	"Q8798",
	"zle",
	scripts = Cyrl,
	ancestors = {"orv"},
	translit_module = "uk-translit",
	entry_name = {
		from = {"Ѐ", "ѐ", "Ѝ", "ѝ", GRAVE, ACUTE},
		to   = {"Е", "е", "И", "и"}},
	standardChars = "ЄІЇА-ЩЫЬЮ-щыьюяєії" .. PUNCTUATION,
} 
m["ur"] = {
	"bahasa Urdu",
	"Q1617",
	"inc",
	scripts = {"ur-Arab"},
	ancestors = {"inc-sap"},
	entry_name = {
		from = {u(0x064B), u(0x064C), u(0x064D), u(0x064E), u(0x064F), u(0x0650), u(0x0651), u(0x0652)},
		to   = {}} ,
}

m["uz"] = {
	"bahasa Uzbek",
	"Q9264",
	"trk",
	otherNames = {"Northern Uzbek", "Southern Uzbek"},
	scripts = {"Latn", "Cyrl", "fa-Arab"},
	ancestors = {"chg"},
}

m["ve"] = {
	"bahasa Venda",
	"Q32704",
	"bnt",
	scripts = Latn,
}

m["vi"] = {
	"bahasa Vietnam",
	"Q9199",
	"mkh-vie",
	otherNames = {"Annamese", "Annamite"},
	scripts = {"Latn", "Hani"},
	ancestors = {"mkh-mvi"},
	sort_key = "vi-sortkey",
}

m["vo"] = {
	"bahasa Volapük",
	"Q36986",
	"art",
	scripts = Latn,
}

m["wa"] = {
	"bahasa Walloon",
	"Q34219",
	"roa-oil",
	otherNames = {"Liégeois", "Namurois", "Wallo-Picard", "Wallo-Lorrain"},
	scripts = Latn,
	ancestors = {"fro"},
	sort_key = {
		from = {"[áàâäå]", "[éèêë]", "[íìîï]", "[óòôö]", "[úùûü]", "[ýỳŷÿ]", "ç", "'"},
		to   = {"a"	  , "e"	 , "i"	 , "o"	 , "u"	 , "y"	 , "c"}} ,
}

m["wo"] = {
	"bahasa Wolof",
	"Q34257",
	"alv-sng",
	otherNames = {"Gambian Wolof"}, -- the subsumed dialect 'wof'
	scripts = LatnArab,
}

m["xh"] = {
	"bahasa Xhosa",
	"Q13218",
	"bnt-ngu",
	scripts = Latn,
	entry_name = {
		from = {"[āàáâǎ]", "[ēèéêě]", "[īìíîǐ]", "[ōòóôǒ]", "[ūùúûǔ]", "ḿ", "[ǹńň]", MACRON, ACUTE, GRAVE, CIRC, CARON},
		to   = {"a"      , "e"      , "i"      , "o"      , "u"      , "m", "n"    }},
}

m["yi"] = {
	"bahasa Yiddi",
	"Q8641",
	"gmw",
	scripts = {"Hebr"},
	ancestors = {"gmh"},
	sort_key = {
		from = {"[אַאָ]", "בּ", "[וֹוּ]", "יִ", "ײַ", "פֿ"},
		to = {"א", "ב", "ו", "י",	"יי", "פ"}} ,
	translit_module = "yi-translit",
}

m["yo"] = {
	"bahasa Yoruba",
	"Q34311",
	"alv-von",
	scripts = Latn,
}

m["za"] = {
	"bahasa Zhuang",
	"Q13216",
	"tai",
	otherNames = {"Standard Zhuang", "Dai Zhuang", "Wenma Zhuang", "Wenma Thu", "Wenma", "Nong Zhuang", "Youjiang Zhuang", "Yongbei Zhuang", "Yang Zhuang", "Yongnan Zhuang", "Zuojiang Zhuang", "Central Hongshuihe Zhuang", "Eastern Hongshuihe Zhuang", "Guibei Zhuang", "Minz Zhuang", "Guibian Zhuang", "Liujiang Zhuang", "Lianshan Zhuang", "Liuqian Zhuang", "Qiubei Zhuang", "Chongzuo Zhuang", "Shangsi Zhuang"},
	scripts = {"Latn", "Hani"},
}

m["zh"] = {
	"bahasa Tionghoa",
	"Q7850",
	"sit",
	scripts = {"Hani", "Brai", "Nshu"},
	ancestors = {"ltc"},
	sort_key = "zh-sortkey",
}

m["zu"] = {
	"bahasa Zulu",
	"Q10179",
	"bnt-ngu",
	otherNames = {"isiZulu"},
	scripts = Latn,
	entry_name = {
                from = {"[āàáâǎ]", "[ēèéêě]", "[īìíîǐ]", "[ōòóôǒ]", "[ūùúûǔ]", "ḿ", "[ǹńň]", MACRON, ACUTE, GRAVE, CIRC, CARON},
                to   = {"a"      , "e"      , "i"      , "o"      , "u"      , "m", "n"    }},
}

return m