Module:etymology languages/data

local m_lang = require("Module:languages") local m_langdata = require("Module:languages/data")

local u = require("Module:string utilities").char

local c = m_langdata.chars local p = m_langdata.puaChars local s = m_langdata.shared

local m = {}

-- Akan varieties

m["tw"] = { "Twi Akan", 36850,	"ak", aliases = {"Twi"}, }

m["tw-abr"] = { "Bono Twi", 34831,	"tw", aliases = {"Bono", "Abron", "Brong"}, }

m["tw-asa"] = { "Asante Twi", 19261685,	"tw", aliases = {"Asante", "Ashanti", "Ashante"}, }

m["tw-aku"] = { "Akuapem Twi", 31150449,	"tw", aliases = {"Akuapem", "Akuapim", "Akwapem Twi", "Akwapi"}, }

m["fat"] = { "Fante Akan", 35570,	"ak", aliases = {"Fante", "Fanti", "Fantse", "Mfantse"}, }

-- Albanian varieties

m["aln"] = { "Gheg Albanian", 181037,	"sq", aliases = {"Gheg"}, }

m["aae"] = { "Arbëresh Albanian", 1075302,	"als", aliases = {"Arbëreshë", "Arbëresh"}, }

m["aat"] = { "Arvanitika Albanian", 29347,	"als", aliases = {"Arvanitika"}, }

m["als"] = { "Tosk Albanian", 180937,	"sq", aliases = {"Tosk"}, }

-- Bantu varieties

m["bnt-cmn"] = { "Common Bantu", nil, "bnt-pro", }

-- Semitic varieties

-- Akkadian varieties

m["akk-old"] = { "Old Akkadian", nil, "akk", }

m["akk-obb"] = { "Old Babylonian", nil, "akk", }

m["akk-oas"] = { "Old Assyrian", nil, "akk", }

m["akk-mbb"] = { "Middle Babylonian", nil, "akk", }

m["akk-mas"] = { "Middle Assyrian", nil, "akk", }

m["akk-nbb"] = { "Neo-Babylonian", nil, "akk", }

m["akk-nas"] = { "Neo-Assyrian", nil, "akk", }

m["akk-lbb"] = { "Late Babylonian", nil, "akk", }

m["akk-stb"] = { "Standard Babylonian", nil, "akk", }

-- Arabic varieties

m["jrb"] = { "Judeo-Arabic", 37733,	"ar", }

-- Aramaic varieties

m["arc-bib"] = { "Biblical Aramaic", 843235,	"arc", family = "sem-are", }

m["arc-cpa"] = { "Christian Palestinian Aramaic", 60790119,	"arc", family = "sem-arw", aliases = {"Melkite Aramaic", "Palestinian Syriac", "Syropalestinian Aramaic"}, }

m["arc-imp"] = { "Imperial Aramaic", 7079491,	"arc", aliases = {"Official Aramaic"}, }

m["arc-hat"] = { "Hatran Aramaic", 3832926,	"arc", family = "sem-are", }

m["arc-jla"] = { "Jewish Literary Aramaic", 105952842,	"arc", }

m["arc-nab"] = { "Nabataean Aramaic", 36178,	"arc", }

m["arc-old"] = { "Old Aramaic", 3398392,	"arc", }

m["arc-pal"] = { "Palmyrene Aramaic", 1510113,	"arc", family = "sem-arw", }

m["tmr"] = { "Jewish Babylonian Aramaic", 33407,	"arc", family = "sem-ase", }

m["jpa"] = { "Jewish Palestinian Aramaic", 948909,	"arc", family = "sem-arw", aliases = {"Galilean Aramaic"}, }

-- Catalan varieties

m["ca-val"] = { "Valencian", 32641,	"ca", }

-- Central Nicobarese varieties

m["ncb-cam"] = { "Camorta", 5026908,	"ncb", aliases = {"Kamorta"}, }

m["ncb-kat"] = { "Katchal", 17064263,	"ncb", aliases = {"Tehnu"}, }

m["ncb-nan"] = { "Nancowry", 6962504,	"ncb", aliases = {"Nankwari"}, }

- --               Chinese varieties                -- -

- Old Chinese, Middle Chinese -

m["och-ear"] = { "Early Old Chinese", nil, "och", }

m["och-lat"] = { "Late Old Chinese", nil, "och", }

m["ltc-ear"] = { "Early Middle Chinese", nil, "ltc", }

m["ltc-lat"] = { "Late Middle Chinese", nil, "ltc", }

- Classical/Literary varieties -

-- FIXME: Temporary. m["lzh-cii"] = { "Ci", 1091366,	"lzh", }

-- FIXME: Temporary. m["lzh-yue"] = { "Classical Cantonese", nil, "lzh", }

-- FIXME: Temporary. m["lzh-cmn"] = { "Classical Mandarin", nil, "lzh", }

-- FIXME: Temporary. m["lzh-tai"] = { "Classical Taishanese", nil, "lzh", }

-- FIXME: Temporary. m["lzh-cmn-TW"] = { "Classical Taiwanese Mandarin", nil, "lzh-cmn", }

-- FIXME: Temporary. m["lzh-VI"] = { "Vietnamese Classical Chinese", 17034227,	"lzh", }

-- FIXME: Temporary. m["lzh-KO"] = { "Korean Classical Chinese", 10496257,	"lzh", ietf_subtag = "lzh-KR" -- KR = South Korea, as there is no code for Korea as a whole }

-- FIXME: Temporary. m["lzh-lit"] = { "Literary Chinese", nil, "lzh", }

-- FIXME: Temporary. FIXME: Do we need this? How does it differ from Old Chinese? m["lzh-pre"] = { "Pre-Classical Chinese", nil, "lzh", }

- Written Vernacular varieties -

-- FIXME: Temporary. m["cmn-wvc"] = { "Written vernacular Mandarin", 783605,	"cmn", }

-- FIXME: Temporary. FIXME: How does this differ from "Literary Cantonese"? m["yue-wvc"] = { "Written vernacular Cantonese", nil, "yue", }

-- FIXME: Temporary. m["zhx-tai-wvc"] = { "Written vernacular Taishanese", nil, "zhx-tai", }

- Mandarin varieties -

-- FIXME: Temporary. NOTE: The Linguist List assigns the "w:Beijing dialect" (Wikidata 1147606) the code "cmn-bej" and -- the larger "w:Beijing Mandarin (division of Mandarin)" dialect group (Wikidata 2169652; what we call "Beijingic -- Mandarin", after Glottolog) the code "cmn-bei". m["cmn-bei"] = { "Beijing Mandarin", 1147606,	"cmn-bec", }

-- FIXME: Temporary. m["cmn-bec"] = { "Beijingic Mandarin", 2169652,	"cmn", }

-- FIXME: Temporary. NOTE: The Linguist List uses the code cmn-zho. m["cmn-cep"] = { "Central Plains Mandarin", 3048775,	"cmn", aliases = {"Zhongyuan Mandarin"}, }

m["cmn-ear"] = { "Early Mandarin", 837169,	"cmn", ancestors = "ltc", }

-- FIXME: Temporary. m["cmn-gua"] = { "Guanzhong Mandarin", 3431648,	"cmn-cep", }

-- FIXME: Temporary. Appears to be a subdialect of Guiliu Mandarin, which in turn is a subdialect of Southwestern Mandarin. m["cmn-gui"] = { "Guilin Mandarin", 11111636,	"cmn", }

m["cmn-jhu"] = { "Jianghuai Mandarin", 2128953,	"cmn", aliases = {"Lower Yangtze Mandarin"}, }

-- FIXME: Temporary. m["cmn-lan"] = { "Lanyin Mandarin", 662754,	"cmn", }

-- FIXME: Temporary. m["cmn-MY"] = { "Malaysian Mandarin", 13646143,	"cmn", }

-- FIXME: Temporary. m["cmn-nan"] = { "Nanjing Mandarin", 2681098,	"cmn-jhu", }

-- FIXME: Temporary. m["cmn-noe"] = { "Northeastern Mandarin", 1064504,	"cmn", }

-- FIXME: Temporary. m["cmn-PH"] = { "Philippine Mandarin", 7185155,	"cmn", }

-- FIXME: Temporary. m["cmn-SG"] = { "Singapore Mandarin", 1048980,	"cmn", }

-- FIXME: Temporary. m["cmn-sow"] = { "Southwestern Mandarin", 2609239,	"cmn", }

-- FIXME: Temporary. Appears to be a subdialect of Jilu Mandarin. m["cmn-tia"] = { "Tianjin Mandarin", 7800220,	"cmn", }

-- FIXME: Temporary. NOTE: Wikidata also has Q4380827 "Taiwanese Mandarin", defined as "rare dialect of Standard Chinese -- (Mandarin) used in Taiwan, which is strongly influenced by Taiwanese Hokkien; mostly used by elderlies" and having no -- English Wikipedia article (but see w:zh:臺灣國語). m["cmn-TW"] = { "Taiwanese Mandarin", 262828,	"cmn", }

-- FIXME: Temporary. Appears to be a subdialect of Wu-Tian Mandarin, in turn a subdialect of Southwestern Mandarin. -- Given the code cmn-xwu in the Linguist List. m["cmn-wuh"] = { "Wuhan Mandarin", 11124731,	"cmn-sow", aliases = {"Wuhanese"}, }

-- FIXME: Temporary. Appears to be a subdialect of Lanyin Mandarin. m["cmn-xin"] = { "Xining Mandarin", nil, "cmn-lan", }

-- FIXME: Temporary. m["cmn-yan"] = { "Yangzhou Mandarin", nil, "cmn-jhu", }

- Cantonese varieties -

-- FIXME: Temporary. m["yue-gua"] = { "Guangzhou Cantonese", nil, "yue", }

-- FIXME: Temporary. Given the codes yue-yue or yue-can in the Linguist List. m["yue-HK"] = { "Hong Kong Cantonese", 5894342,	"yue", }

-- FIXME: Temporary. FIXME: How does this differ from "Written vernacular Cantonese"? m["yue-lit"] = { "Literary Cantonese", 2472605,	"yue", }

- Wu varieties -

m["wuu-han"] = { "Hangzhounese", 5648144,	"wuu", }

m["wuu-nin"] = { "Ningbonese", 3972199,	"wuu", }

-- FIXME: Temporary. m["wuu-nor"] = { "Northern Wu", 7675988,	"wuu", aliases = {"Taihu Wu"}, }

-- FIXME: Temporary? Subvariety of Taihu Wu. NOTE: "chm" stands for Chongming, the main dialect, to avoid a conflict -- with Shanghainese. m["wuu-chm"] = { "Shadi Wu", 6112340,	"wuu-nor", }

m["wuu-sha"] = { "Shanghainese", 36718,	"wuu-nor", }

m["wuu-suz"] = { "Suzhounese", 831744,	"wuu-nor", }

-- FIXME: Temporary. May be converted into a full language and/or split. m["wuu-wen"] = { "Wenzhounese", 710218,	"wuu", }

- Xiang varieties -

m["hsn-lou"] = { "Loudi Xiang", 10943823,	"hsn-old", }

m["hsn-hya"] = { "Hengyang Xiang", 20689035,	"hsn-hzh", }

m["hsn-hzh"] = { "Hengzhou Xiang", nil, "hsn", }

m["hsn-new"] = { "New Xiang", 7012696,	"hsn", aliases = {"Chang-Yi"}, }

m["hsn-old"] = { "Old Xiang", 7085453,	"hsn", aliases = {"Lou-Shao"}, }

- Hakka varieties -

-- FIXME: Temporary. m["hak-dab"] = { "Dabu Hakka", 19855566,	"hak", -- formerly hak-TW but seems to be spoken primary in Dabu County in Guangdong }

-- FIXME: Temporary. m["hak-eam"] = { "Early Modern Hakka", nil, "hak", }

-- FIXME: Temporary. m["hak-hai"] = { "Hailu Hakka", 17038519,	"hak", -- often considered a Taiwanese lect but also spoken in Shanwei, Guangdong }

-- FIXME: Temporary. m["hak-hui"] = { "Huiyang Hakka", 16873881,	"hak", }

-- FIXME: Temporary. m["hak-hui-MY"] = { "Malaysian Huiyang Hakka", nil, "hak-hui", }

-- FIXME: Temporary. Similar to and possibly the parent of Sixian Hakka in Taiwan. m["hak-mei"] = { "Meixian Hakka", 839295,	"hak", aliases = {"Moiyan Hakka", "Meizhou Hakka"}, }

-- FIXME: Temporary. m["hak-six"] = { "Sixian Hakka", 9668261,	"hak-TW", }

-- FIXME: Temporary. m["hak-TW"] = { "Taiwanese Hakka", 2391532,	"hak", }

-- FIXME: Temporary. m["hak-zha"] = { "Zhao'an Hakka", 6703311,	"hak", aliases = {"Zhangzhou Hakka"}, }

-- Southern Min varieties --

m["nan-anx"] = { "Anxi Hokkien", 97064149,	"nan-qua", }

m["nan-cha"] = { "Changtai Hokkien", nil, "nan-zha", }

m["nan-hou"] = { "Houlu Min", 19855492,	"nan-dat", }

m["nan-hui"] = { "Hui'an Hokkien", 16241797,	"nan-qua", }

m["nan-jin"] = { "Jinjiang Hokkien", 11089375,	"nan-qua", }

m["nan-kin"] = { "Kinmenese Hokkien", 56278342,	"nan-xia", aliases = {"Kinmen Hokkien"}, }

m["nan-med"] = { "Medan Hokkien", 6805114,	"nan-zha", }

m["nan-pen"] = { "Penang Hokkien", 11120689,	"nan-zha", }

m["nan-hbl-PH"] = { "Philippine Hokkien", 3236692,	"nan-qua", }

m["nan-qia"] = { "Qianlu Min", 19842517,	"nan-dat", }

m["nan-qua"] = { "Quanzhou Hokkien", 2251677,	"nan-hbl", aliases = {"Chinchew", "Choanchew"}, }

-- FIXME: Temporary? Derived from both Quanzhou and Zhangzhou Hokkien. m["nan-hbl-SG"] = { "Singapore Hokkien", 3846528,	"nan-hbl", }

m["nan-spm"] = { "Southern Malaysian Hokkien", 7570322,	"nan-qua", aliases = {"Southern Malaysia Hokkien", "Southern Peninsular Malaysian Hokkien", "Southern Peninsular Malaysia Hokkien"} }

m["nan-hbl-TW"] = { "Taiwanese Hokkien", 36778,	"nan-hbl", }

m["nan-ton"] = { "Tong'an Hokkien", nil, "nan-xia", }

m["nan-xia"] = { "Xiamen Hokkien", 68744,	"nan-hbl", aliases = {"Amoy", "Amoyese", "Amoynese", "Xiamenese"}, }

m["nan-yon"] = { "Yongchun Hokkien", 65118728,	"nan-qua", }

m["nan-zha"] = { "Zhangzhou Hokkien", 8070492,	"nan-hbl", aliases = {"Changchew", "Chiangchew", "Changchow"}, }

m["nan-zho"] = { "Zhao'an Hokkien", 65118728,	"nan-zha", aliases = {"Zhao'an", "Chawan", "Chawan Hokkien"}, }

m["nan-zhp"] = { "Zhangping Hokkien", 15937822,	"nan-zha", }

- Other Min varieties -

-- FIXME: Temporary. Affiliation within Min uncertain; some combination of Eastern and Southern. m["zhx-zho"] = { "Zhongshan Min", 8070958,	"zhx", }

- Other Chinese varieties -

-- FIXME: Temporary. Affiliation within Chinese uncertain; possibly Yue. m["zhx-dan"] = { "Danzhou Chinese", 2578935,	"zhx", }

- Chinese romanization varieties -

-- Information desk/2022/June -- Grease pit/2022/June

m["cmn-pinyin"] = { "Hanyu Pinyin", 42222,	"cmn", aliases = {"Pinyin"}, }

m["cmn-tongyong"] = { "Tongyong Pinyin", 700739,	"cmn", }

m["cmn-wadegiles"] = { "Wade–Giles", 208442,	"cmn", aliases = {"Wade-Giles", "Wade Giles"}, }

m["zh-postal"] = { "Postal Romanization", 151868,	"zh", }

-- Chinese cyrillization

m["cmn-palladius"] = { "Palladius", 1234239,	"cmn", aliases = {"Palladius system"}, }

- --                Coptic varieties                -- -

m["cop-akh"] = { "Akhmimic Coptic", nil, "cop", aliases = {"Akhmimic"}, }

m["cop-boh"] = { "Bohairic Coptic", 890733,	"cop", aliases = {"Bohairic", "Memphitic Coptic", "Memphitic"}, }

m["cop-ggg"] = { "Coptic Dialect G", nil, "cop", aliases = {"Dialect G", "Mansuric Coptic", "Mansuric"}, }

m["cop-jjj"] = { "Coptic Dialect J", nil, "cop", }

m["cop-kkk"] = { "Coptic Dialect K", nil, "cop", }

m["cop-ppp"] = { "Coptic Dialect P", nil, "cop", aliases = {"Proto-Theban Coptic", "Palaeo-Theban Coptic"}, }

m["cop-fay"] = { "Fayyumic Coptic", 1399115,	"cop", aliases = {"Fayyumic", "Faiyumic Coptic", "Faiyumic", "Fayumic Coptic", "Fayumic", "Bashmuric Coptic", "Bashmuric"}, }

m["cop-her"] = { "Hermopolitan Coptic", nil, "cop", aliases = {"Hermopolitan", "Coptic Dialect H", "Ashmuninic", "Ashmuninic Coptic"}, }

m["cop-lyc"] = { "Lycopolitan Coptic", nil, "cop", aliases = { "Lycopolitan", "Assiutic Coptic", "Asyutic Coptic", "Assiutic", "Asyutic", "Lyco-Diospolitan Coptic", "Lyco-Diospolitan", "Subakhmimic Coptic", "Subakhmimic" }, }

m["cop-old"] = { "Old Coptic", nil, "cop", }

m["cop-oxy"] = { "Oxyrhynchite Coptic", nil, "cop", aliases = {"Oxyrhynchite", "Mesokemic Coptic", "Mesokemic", "Middle Egyptian Coptic"}, }

m["cop-ply"] = { "Proto-Lycopolitan Coptic", nil, "cop", aliases = {"Coptic Dialect i", "Proto-Lyco-Diospolitan Coptic"}, }

m["cop-sah"] = { "Sahidic Coptic", 2645851,	"cop", aliases = {"Sahidic", "Saidic Coptic", "Saidic", "Thebaic Coptic", "Thebaic"}, }

- --                Dutch varieties                 -- -

m["nl-BE"] = { "Belgian Dutch", 34147,	"nl", aliases = { "Flemish", "Flemish Dutch", "Southern Dutch"}, }

- --              Low Saxon varieties               -- -

m["act"] = { "Achterhoeks", 153627,	"nds-nl", aliases = {"Achterhoek", "Achterhooks"}, }

m["drt"] = { "Drents", 2736709,	"nds-nl", aliases = {"Drèents", "Dreins", "Dreints", "Drints"}, }

m["frs"] = { "East Frisian Low German", 149208,	"nds-de", aliases = {"East Frisian", "East Frisian Low Saxon"}, }

m["gos"] = { "Gronings", 508854,	"nds-nl", aliases = {"Grunnegs", "Grönnegs"}, }

m["nds-lpr"] = { "Low Prussian", 33982,	"nds-de", }

m["sdz"] = { "Sallands", 3436668,	"nds-nl", aliases = {"Sallaans", "Sallaands"}, }

m["stl"] = { "Stellingwerfs", 506010,	"nds-nl", aliases = {"Stellingwarfs"}, }

m["twd"] = { "Twents", 497363,	"nds-nl", aliases = {"Tweants"}, }

m["vel"] = { "Veluws", 2484810,	"nds-nl", }

m["wep"] = { "Westphalian", 505655,	"nds-de", aliases = {"Westfalish", "Westphalien"}, }

- --              Egyptian varieties                -- -

m["egy-old"] = { "Old Egyptian", 447117,	"egy", }

m["egy-mid"] = { "Middle Egyptian", 657330,	"egy", aliases = {"Classical Egyptian"}, }

m["egy-nmi"] = { "Neo-Middle Egyptian", 123735278,	"egy", aliases = {"Égyptien de tradition", "Traditional Egyptian"}, }

m["egy-lat"] = { "Late Egyptian", 1852329,	"egy", }

- --               Elamite varieties                -- -

m["elx-old"] = { "Old Elamite", nil, "elx", }

m["elx-mid"] = { "Middle Elamite", nil, "elx", }

m["elx-neo"] = { "Neo-Elamite", nil, "elx", }

m["elx-ach"] = { "Achaemenid Elamite", nil, "elx", }

- --           English and Scots varieties          -- -

-- English varieties

m["en-AU"] = { "Australian English", 44679,	"en", }

m["en-GB"] = { "British English", 7979,	"en", }

m["en-GB-SCT"] = { "Scottish English", 44676,	"en-GB", }

m["en-GB-WLS"] = { "Welsh English", 44676,	"en-GB", }

m["en-IM"] = { "Manx English", 6753295,	"en-GB", }

m["en-aae"] = { "Australian Aboriginal English", 783347,	"en-AU", }

m["en-ear"] = { "Early Modern English", 1472196,	"en", ancestors = "enm", aliases = {"Early New English"}, }

m["en-geo"] = { "Geordie English", 653421,	"en", ancestors = "enm-nor", }

m["en-IE"] = { -- FIXME: "IE" doesn't cover Northern Ireland "Irish English", 665624,	"en", }

m["en-uls"] = { "Ulster English", 6840826,	"en-IE", }

m["en-GB-NIR"] = { "Northern Irish English", 6840826, -- actually the code for Ulster English "en-uls", }

m["en-NNN"] = { -- NA = Namibia; NNN is NATO 3-letter code for North America "North American English", 7053766,	"en", ietf_subtag = "en-021" -- 021 = UN M49 code for "Northern America" (i.e. North America wihout Central America or the Caribbean) }

m["en-US"] = { "American English", 7976,	"en-NNN", }

m["en-US-CA"] = { "California English", 1026812,	"en-US", }

m["en-CA"] = { "Canadian English", 44676,	"en-US", }

m["en-HK"] = { "Hong Kong English", 1068863,	"en", }

m["en-IN"] = { "Indian English", 1348800,	"en", }

m["pld"] = { "Polari", 1359130,	"en", }

-- Scots varieties

m["sco-osc"] = { "Early Scots", 5326738,	"enm", ancestors = "enm-nor", aliases = {"Old Scots"}, }

m["sco-smi"] = { "Middle Scots", 3327000,	"sco", ancestors = "sco-osc", }

m["sco-ins"] = { "Insular Scots", 16919205,	"sco", }

m["sco-uls"] = { "Ulster Scots", 201966,	"sco", }

m["sco-nor"] = { "Northern Scots", 16928150,	"sco", }

m["sco-sou"] = { "South Scots", 7570457,	"sco", aliases = {"Southern Scots", "Borders Scots"}, }

-- Middle English varieties

m["enm-nor"] = { "Northern Middle English", nil, "enm", ancestors = "ang-nor", aliases = {"Northumbrian Middle English"}, }

-- Old English varieties

-- Includes both Mercian and Northumbrian. m["ang-ang"] = { "Anglian Old English", nil, "ang", }

m["ang-ken"] = { "Kentish Old English", 11687485,	"ang", }

m["ang-mer"] = { "Mercian Old English", 602072,	"ang-ang", }

m["ang-nor"] = { "Northumbrian Old English", 1798915,	"ang-ang", }

-- m["ang-wsx"] = {	"West Saxon Old English",	nil,	"ang", }

- --    French and French-based creole varieties    -- -

m["fro-nor"] = { "Old Northern French", 2044917,	"fro", aliases = {"Old Norman", "Old Norman French"}, }

m["fro-pic"] = { "Picard Old French", nil, "fro", }

m["xno"] = { "Anglo-Norman", 35214,	"fro-nor", }

m["xno-law"] = { "Law French", 2044323,	"xno", }

m["fr-CA"] = { "Canadian French", 1450506,	"fr", }

m["fr-CH"] = { "Swiss French", 1480152,	"fr", }

m["fr-aca"] = { "Acadian French", 415109,	"fr", }

m["fr-lou"] = { "Louisiana French", 3083213,	"fr", }

m["fr-mis"] = { "Missouri French", 3083210,	"fr", }

m["frc"] = { "Cajun French", 880301,	"fr-lou", }

m["ht-sdm"] = { "Saint Dominican Creole French", nil, "ht", ancestors = "fr", }

-- Norman varieties

m["nrf-grn"] = { "Guernsey Norman", 56428,	"nrf", aliases = {"Guernsey"}, }

m["nrf-jer"] = { "Jersey Norman", 56430,	"nrf", aliases = {"Jersey"}, }

- --               Brythonic varieties              -- -

m["bry-ear"] = { "Early Brythonic", nil, "cel-bry-pro", }

m["bry-lat"] = { "Late Brythonic", nil, "cel-bry-pro", }

- --                Gaulish varieties               -- -

m["xcg"] = { "Cisalpine Gaulish", 3832927,	"cel-gau", }

m["xtg"] = { "Transalpine Gaulish", 29977,	"cel-gau", }

- --               Portuguese varieties             -- -

m["pt-BR"] = { "Brazilian Portuguese", 750553,	"pt", }

m["pt-PT"] = { "European Portuguese", 922399,	"pt", }

- --                 Spanish varieties              -- -

m["es-ear"] = { "Early Modern Spanish", 5364419,	"es", ancestors = "osp", }

m["es-AR"] = { "Rioplatense Spanish", 509780,	"es", }

m["es-CO"] = { "Colombian Spanish", 1115875,	"es", }

m["es-CU"] = { "Cuban Spanish", 824909,	"es", }

m["es-MX"] = { "Mexican Spanish", 616620,	"es", }

m["es-US"] = { "United States Spanish", 2301077,	"es", aliases = {"US Spanish"}, } --use label "US Spanish" to put Spanish terms in this category

m["es-PR"] = { "Puerto Rican Spanish", 7258609,	"es", }

m["es-VE"] = { "Venezuelan Spanish", 840017,	"es", }

- --                  Fula varieties                -- -

m["fuc"] = { "Pulaar", 1420205,	"ff", }

m["fuf"] = { "Pular", 3915357,	"ff", }

m["ffm"] = { "Maasina Fulfulde", 3915322,	"ff", }

m["fue"] = { -- no enwiki entry as of yet but frwiki and pmswiki have one "Borgu Fulfulde", 12952426,	"ff", }

m["fuh"] = { -- no enwiki entry as of yet but frwiki and pmswiki have one "Western Niger Fulfulde", 12952430,	"ff", }

m["fuq"] = { -- no enwiki entry as of yet but frwiki, hrwiki and pmswiki have one "Central-Eastern Niger Fulfulde", 12628799,	"ff", }

m["fuv"] = { -- no enwiki entry as of yet but dewiki, frwiki, hrwiki, pmswiki and swwiki have one "Nigerian Fulfulde", 36129,	"ff", }

m["fub"] = { -- no enwiki entry as of yet but dewiki, frwiki, hrwiki, pmswiki, ptwiki, swwiki and yowiki have one "Adamawa Fulfulde", 34776,	"ff", }

m["fui"] = { -- no enwiki entry as of yet but pmswiki and swwiki have one "Bagirmi Fulfulde", 11003859,	"ff", }

- --              German(ic) varieties              -- -

-- (modern) German varieties

m["de-AT"] = { "Austrian German", 306626,	"de", }

m["de-AT-vie"] = { "Viennese German", 56474,	"de-AT", }

m["de-CH"] = { "Switzerland German", 1366643,	"de", aliases = {"Schweizer Hochdeutsch", "Swiss Standard German", "Swiss High German"}, }

m["ksh"] = { "Kölsch", 4624,	"gmw-cfr", }

m["pfl"] = { "Palatine German", 23014,	"gmw-rfr", aliases = {"Pfälzisch", "Pälzisch", "Palatinate German"}, }

m["sli"] = { "Silesian East Central German", 152965,	"gmw-ecg", aliases = {"Silesian"}, }

m["sxu"] = { "Upper Saxon German", 699284,	"gmw-ecg", }

-- Old High German varieties

m["lng"] = { "Lombardic", 35972,	"goh", }

-- Proto-West Germanic varieties

m["frk"] = { "Frankish", 10860505,	"gmw-pro", aliases = {"Old Frankish"}, }

-- Alemannic German varieties

m["gsw-low"] = { "Low Alemannic German", 503724,	"gsw", }

m["gsw-FR-als"] = { "Alsatian Alemannic German", 8786,	"gsw-low", }

m["gsw-hig"] = { "High Alemannic German", 503728,	"gsw", }

m["gsw-hst"] = { "Highest Alemannic German", 687538,	"gsw", }

m["wae"] = { "Walser German", 680517,	"gsw-hst", }

- --              Old Norse varieties               -- -

m["non-grn"] = { "Greenlandic Norse", 855236,	"non-own", }

m["non-oen"] = { "Old East Norse", 10498031,	"non", ancestors = "non", }

m["non-own"] = { "Old West Norse", 10498026,	"non", ancestors = "non", }

- --              Old Swedish varieties             -- -

m["gmq-osw-lat"] = { "Late Old Swedish", 10723594,	"gmq-osw", ancestors = "gmq-osw", }

- --                 Greek varieties                -- -

m["qsb-grc"] = { "Pre-Greek", 965052,	"und", family = "qfa-sub", }

m["grc-aeo"] = { "Aeolic Greek", 406373,	"grc", aliases = {"Lesbic Greek", "Lesbian Greek", "Aeolian Greek"}, }

m["grc-arc"] = { "Arcadian Greek", nil, "grc-arp", }

m["grc-arp"] = { "Arcadocypriot Greek", 499602,	"grc", }

m["grc-att"] = { "Attic Greek", 506588,	"grc", }

m["grc-boi"] = { "Boeotian Greek", 406373,	"grc-aeo", }

m["grc-dor"] = { "Doric Greek", 285494,	"grc", }

m["grc-ela"] = { "Elean Greek", nil, "grc", }

m["grc-epc"] = { "Epic Greek", 990062,	"grc", aliases = {"Homeric Greek"}, }

m["grc-ion"] = { "Ionic Greek", 504165,	"grc", }

m["grc-koi"] = { "Koine Greek", 107358,	"grc", ancestors = "grc-att", aliases = {"Hellenistic Greek"}, }

m["grc-kre"] = { "Cretan Ancient Greek", -- to distinguish from Cretan Greek below nil, "grc-dor", }

m["grc-opl"] = { "Opuntian Locrian", nil, "grc", }

m["grc-ozl"] = { "Ozolian Locrian", nil, "grc", }

m["grc-pam"] = { "Pamphylian Greek", 2271793,	"grc", }

m["grc-ths"] = { "Thessalian Greek", 406373,	"grc-aeo", }

m["gkm"] = { "Byzantine Greek", 36387,	"grc", ancestors = "grc-koi", aliases = {"Medieval Greek"}, }

m["el-cyp"] = { "Cypriot Greek", 245899,	"el", aliases = {"Cypriotic Greek"}, }

m["el-pap"] = { "Paphian Greek", nil, "el", }

m["el-crt"] = { "Cretan Greek", 588306,	"el", }

m["el-kth"] = { "Katharevousa", 35961,	"el", ancestors = "gkm", aliases = {"Katharevousa Greek"}, }

m["el-kal"] = { "Kaliarda", nil, "el", }

- --                Hebrew varieties                -- -

m["hbo"] = { "Biblical Hebrew", 1982248,	"he", aliases = {"Classical Hebrew"}, }

m["he-mis"] = { "Mishnaic Hebrew", 1649362,	"he", ancestors = "hbo", }

m["he-med"] = { "Medieval Hebrew", 2712572,	"he", ancestors = "he-mis", }

m["he-IL"] = { "Israeli Hebrew", 8141,	"he", }

m["bsh-kat"] = { "Kativiri", 2605045,	"bsh", aliases = {"Katə́viri"}, }

m["xvi"] = { "Kamviri", 1193495,	"bsh", aliases = {"Kamvíri"}, }

m["bsh-mum"] = { "Mumviri", nil, aliases = {"Mumvíri"}, "bsh" }

- --                Inuit varieties                 -- -

m["esi"] = { "North Alaskan Inupiatun", nil, "ik" }

m["esk"] = { "Northwest Alaskan Inupiatun", 25559714,	"ik" }

- --                Iranian varieties               -- -

m["qsb-bma"] = { "the BMAC substrate", 1054850,	"und", family = "qfa-sub", }

-- Historical and current Iranian dialects

m["ae-old"] = { "Old Avestan", 29572,	"ae", aliases = {"Gathic Avestan"}, }

m["ae-yng"] = { "Younger Avestan", 29572,	"ae-old", aliases = {"Young Avestan"}, }

m["bcc"] = { "Southern Balochi", 33049,	"bal", aliases = {"Southern Baluchi"}, }

m["bgp"] = { "Eastern Balochi", 33049,	"bal", aliases = {"Eastern Baluchi"}, }

m["bgn"] = { "Western Balochi", 33049,	"bal", aliases = {"Western Baluchi"}, }

m["bsg-ban"] = { "Bandari", nil, "bsg", }

m["bsg-hor"] = { "Hormozi", nil, "bsg", }

m["bsg-min"] = { "Minabi", nil, "bsg", }

m["kho-old"] = { "Old Khotanese", nil, "kho", }

m["kho-lat"] = { "Late Khotanese", nil, "kho-old", }

m["peo-ear"] = { "Early Old Persian", nil, "peo", }

m["peo-lat"] = { "Late Old Persian", nil, "peo", }

m["pal-ear"] = { "Early Middle Persian", nil, "pal", }

m["pal-lat"] = { "Late Middle Persian", nil, "pal", ancestors = "pal-ear", }

m["ps-nwe"] = { "Northwestern Pashto", nil, "ps", }

m["ps-cgi"] = { "Central Ghilzay", nil, "ps-nwe", }

m["ps-mah"] = { "Mahsudi", nil, "ps-nwe", }

m["ps-nea"] = { "Northeastern Pashto", nil, "ps", }

m["ps-afr"] = { "Afridi", nil, "ps-nea", }

m["ps-bng"] = { "Bangash", nil, "ps-nea", }

m["ps-xat"] = { "Khatak", nil, "ps-nea", }

m["ps-pes"] = { "Peshawari", nil, "ps-nea", }

m["ps-sea"] = { "Southeastern Pashto", nil, "ps", }

m["ps-ban"] = { "Bannu", nil, "ps-sea", }

m["ps-kak"] = { "Kakari", nil, "ps-sea", }

m["ps-ser"] = { "Sher", nil, "ps-sea", }

m["ps-waz"] = { "Waziri", 12274473,	"ps-sea", }

m["ps-swe"] = { "Southwestern Pashto", nil, "ps", }

m["ps-kan"] = { "Kandahari", nil, "ps-swe", }

m["ps-jad"] = { "Jadrani", nil, "ps", ancestors = "ira-pat-pro" }

m["xme-azr"] = { "Old Azari", nil, "xme-ott", aliases = {"Old Azeri", "Azari", "Azeri", "Āḏarī", "Adari", "Adhari"}, }

m["xme-ttc-cen"] = { "Central Tati", nil, "xme-ott", }

m["xme-ttc-eas"] = { "Eastern Tati", nil, "xme-ott", }

m["xme-ttc-nor"] = { "Northern Tati", nil, "xme-ott", }

m["xme-ttc-sou"] = { "Southern Tati", nil, "xme-ott", }

m["xme-ttc-wes"] = { "Western Tati", nil, "xme-ott", }

m["xmn"] = { "Manichaean Middle Persian", nil, "pal-lat", }

m["fa-ira"] = { "Iranian Persian", 3513637,	"fa", aliases = {"Modern Persian", "Western Persian"}, translit = "fa-ira-translit", }

m["fa-cls"] = { "Classical Persian", 9168,	"fa", ancestors = "pal-lat", translit = "fa-cls-translit", }

m["prs"] = { "Dari", 178440,	"fa", aliases = {"Dari Persian", "Central Persian", "Eastern Persian", "Afghan Persian"}, translit = "fa-cls-translit", }

m["haz"] = { "Hazaragi", 33398,	"prs", translit = "fa-cls-translit", }

m["os-dig"] = { "Digor Ossetian", 3027861,	"os", aliases = {"Digoron", "Digor"}, }

m["os-iro"] = { "Iron Ossetian", nil, "os", aliases = {"Iron"}, }

m["sog-ear"] = { "Early Sogdian", nil, "sog", }

m["sog-lat"] = { "Late Sogdian", nil, "sog-ear", }

m["oru-kan"] = { "Kaniguram", 6363164,	"oru", }

m["oru-log"] = { "Logar", nil, "oru", }

m["oos-ear"] = { "Early Old Ossetic", nil, "oos", }

m["oos-lat"] = { "Late Old Ossetic", nil, "oos", }

m["xln"] = { "Alanic", 3658580,	"oos", }

m["rdb-jir"] = { "Jirofti", nil, "rdb", }

m["rdb-kah"] = { "Kahnuji", nil, "rdb", }

-- Southwestern Fars lects

m["fay-bur"] = { "Burenjani", nil, "fay", }

m["fay-bsh"] = { "Bushehri", nil, "fay", }

m["fay-dsh"] = { "Dashtaki", nil, "fay", }

m["fay-dav"] = { "Davani", 5228140,	"fay", }

m["fay-eze"] = { "Emamzada Esma’ili", nil, "fay", }

m["fay-gav"] = { "Gavkoshaki", nil, "fay", }

m["fay-kho"] = { "Khollari", nil, "fay", }

m["fay-kon"] = { "Kondazi", nil, "fay", }

m["fay-kzo"] = { "Old Kazeruni", nil, "fay", }

m["fay-mas"] = { "Masarami", nil, "fay", }

m["fay-pap"] = { "Papuni", nil, "fay", }

m["fay-sam"] = { "Samghani", nil, "fay", }

m["fay-shr"] = { "Shirazi", nil, "fay", }

m["fay-sho"] = { "Old Shirazi", nil, "fay", }

m["fay-sam"] = { "Samghani", nil, "fay", }

m["fay-kar"] = { "Khargi", nil, "fay", }

m["fay-sor"] = { "Sorkhi", nil, "fay", }

-- Talysh lects

m["tly-cen"] = { "Central Talysh", nil, "tly", }

m["tly-asa"] = { "Asalemi", nil, "tly-cen", }

m["tly-kar"] = { "Karganrudi", nil, "tly-cen", }

m["tly-tul"] = { "Tularudi", nil, "tly-cen", }

m["tly-tal"] = { "Taleshdulabi", nil, "tly-cen", }

m["tly-nor"] = { "Northern Talysh", nil, "tly", }

m["tly-aze"] = { "Azerbaijani Talysh", nil, "tly-nor", }

m["tly-anb"] = { "Anbarani", nil, "tly-nor", }

m["tly-sou"] = { "Southern Talysh", nil, "tly", }

m["tly-fum"] = { "Fumani", nil, "tly-sou", }

m["tly-msu"] = { "Masulei", nil, "tly-sou", }

m["tly-msa"] = { "Masali", nil, "tly-sou", }

m["tly-san"] = { "Shandarmani", nil, "tly-sou", }

-- Tafreshi lects

m["xme-amo"] = { "Amorehi", nil, "xme-taf", }

m["atn"] = { "Ashtiani", 3436590,	"xme-taf", wikipedia_article = "Ashtiani language", }

m["xme-bor"] = { "Borujerdi", nil, "xme-taf", }

m["xme-ham"] = { "Hamadani", 6302426,	"xme-taf", }

m["xme-kah"] = { "Kahaki", nil, "xme-taf", }

m["vaf"] = { "Vafsi", 32611,	"xme-taf", }

-- Kermanic lects

m["kfm"] = { "Khunsari", 6403030,	"xme-ker", wikipedia_article = "Khunsari language", }

m["xme-mah"] = { "Mahallati", nil, "xme-ker", }

m["xme-von"] = { "Vonishuni", nil, "xme-ker", }

m["xme-bdr"] = { "Badrudi", nil, "xme-ker", }

m["xme-del"] = { "Delijani", nil, "xme-ker", }

m["xme-kas"] = { "Kashani", nil, "xme-ker", }

m["xme-kes"] = { "Kesehi", nil, "xme-ker", }

m["xme-mey"] = { "Meymehi", nil, "xme-ker", }

m["ntz"] = { "Natanzi", 6968399,	"xme-ker", wikipedia_article = "Natanzi language", }

m["xme-abz"] = { "Abuzeydabadi", nil, "xme-ker", }

m["xme-aby"] = { "Abyanehi", nil, "xme-ker", }

m["xme-far"] = { "Farizandi", nil, "xme-ker", }

m["xme-jow"] = { "Jowshaqani", nil, "xme-ker", }

m["xme-nas"] = { "Nashalji", nil, "xme-ker", }

m["xme-qoh"] = { "Qohrudi", nil, "xme-ker", }

m["xme-yar"] = { "Yarandi", nil, "xme-ker", }

m["soj"] = { "Soi", 7930463,	"xme-ker", aliases = {"Sohi"}, wikipedia_article = "Soi language", }

m["xme-tar"] = { "Tari", nil, "xme-ker", }

m["gzi"] = { "Gazi", 5529130,	"xme-ker", wikipedia_article = "Gazi language", }

m["xme-sed"] = { "Sedehi", nil, "xme-ker", }

m["xme-ard"] = { "Ardestani", nil, "xme-ker", }

m["xme-zef"] = { "Zefrehi", nil, "xme-ker", }

m["xme-isf"] = { "Isfahani", nil, "xme-ker", }

m["xme-kaf"] = { "Kafroni", nil, "xme-ker", }

m["xme-vrz"] = { "Varzenehi", nil, "xme-ker", }

m["xme-xur"] = { "Khuri", nil, "xme-ker", }

m["nyq"] = { "Nayini", 6983146,	"xme-ker", wikipedia_article = "Nayini language", }

m["xme-ana"] = { "Anaraki", nil, "xme-ker", }

m["gbz"] = { "Zoroastrian Dari", 32389,	"xme-ker", aliases = {"Behdināni", "Gabri", "Gavrŭni", "Gabrōni"}, wikipedia_article = "Zoroastrian Dari language", }

m["xme-krm"] = { "Kermani", nil, "xme-ker", }

m["xme-yaz"] = { "Yazdi", nil, "xme-ker", }

m["xme-bid"] = { "Bidhandi", nil, "xme-ker", }

m["xme-bij"] = { "Bijagani", nil, "xme-ker", }

m["xme-cim"] = { "Chimehi", nil, "xme-ker", }

m["xme-han"] = { "Hanjani", nil, "xme-ker", }

m["xme-kom"] = { "Komjani", nil, "xme-ker", }

m["xme-nar"] = { "Naraqi", nil, "xme-ker", }

m["xme-nus"] = { "Nushabadi", nil, "xme-ker", }

m["xme-qal"] = { "Qalhari", nil, "xme-ker", }

m["xme-trh"] = { "Tarehi", nil, "xme-ker", }

m["xme-val"] = { "Valujerdi", nil, "xme-ker", }

m["xme-var"] = { "Varani", nil, "xme-ker", }

m["xme-zor"] = { "Zori", nil, "xme-ker", }

-- Ramandi lects

m["tks-ebr"] = { "Ebrahimabadi", nil, "tks", }

m["tks-sag"] = { "Sagzabadi", nil, "tks", }

m["tks-esf"] = { "Esfarvarini", nil, "tks", }

m["tks-tak"] = { "Takestani", nil, "tks", }

m["tks-cal"] = { "Chali Tati", nil, "tks", aliases = {"Chāli"}, wikipedia_article = "Tati language (Iran)", }

m["tks-dan"] = { "Danesfani", nil, "tks", }

m["tks-xia"] = { "Khiaraji", nil, "tks", }

m["tks-xoz"] = { "Khoznini", nil, "tks", }

-- Shughni dialects

m["sgh-bro"] = { "Bartangi-Oroshori", nil, "sgh", }

m["sgh-bar"] = { "Bartangi", nil, "sgh-bro", }

m["sgh-oro"] = { "Oroshori", nil, "sgh-bro", aliases = {"Roshorvi"}, }

m["sgh-rsx"] = { "Roshani-Khufi", nil, "sgh", }

m["sgh-xuf"] = { "Khufi", 2562249,	"sgh-rsx", aliases = {"Xufi", "Xūfī"}, wikipedia_article = "Khufi language", }

m["sgh-ros"] = { "Roshani", 2597566,	"sgh-rsx", aliases = {"Rushani", "Rōšāni"}, wikipedia_article = "Rushani language", }

m["sgh-xgb"] = { "Khughni-Bajui", nil, "sgh", }

m["sgh-xug"] = { "Khughni", nil, "sgh-xgb", }

m["sgh-baj"] = { "Bajui", nil, "sgh-xgb", }

-- Indo-Aryan varieties

m["inc-mit"] = { "Mitanni", 1986700,	"inc-pro", }

m["awa-old"] = { "Old Awadhi", nil, "awa", }

m["bra-old"] = { "Old Braj", nil, "bra", }

m["gu-kat"] = { "Kathiyawadi", nil, "gu", aliases = {"Kathiyawadi Gujarati", "Kathiawadi"}, }

m["gu-lda"] = { "Lisan ud-Dawat Gujarati", nil, "gu", aliases = {"Lisan ud-Dawat", "LDA"}, }

m["hi-mum"] = { "Bombay Hindi", 3543151,	"hi", aliases = {"Mumbai Hindi", "Bambaiyya Hindi"}, }

m["hi-mid"] = { "Middle Hindi", nil, "inc-ohi", ancestors = "inc-ohi", }

m["sa-bhs"] = { "Buddhist Hybrid Sanskrit", 248758,	"sa", }

m["sa-bra"] = { "Brahmanic Sanskrit", 36858,	"sa", }

m["sa-cls"] = { "Classical Sanskrit", 11059,	"sa", }

m["sa-neo"] = { "New Sanskrit", 11059,	"sa", }

m["sa-ved"] = { "Vedic Sanskrit", 36858,	"sa", ietf_subtag = "vsn" -- should we make this the actual code? }

m["si-med"] = { "Medieval Sinhalese", nil, "si", aliases = {"Medieval Sinhala"}, }

m["kok-mid"] = { "Middle Konkani", nil, "kok", aliases = {"Medieval Konkani"}, }

m["kok-old"] = { "Old Konkani", nil, "kok", aliases = {"Early Konkani"}, }

-- Indian subcontinent languages

-- Dhivehi varieties

m["dv-mul"] = { "Mulaku Dhivehi", nil, "dv", aliases = {"Mulaku Divehi", "Mulaku Bas"}, }

m["dv-huv"] = { "Huvadhu Dhivehi", nil, "dv", aliases = {"Huvadhu Divehi", "Huvadhu Bas"}, }

m["dv-add"] = { "Addu Dhivehi", nil, "dv", aliases = {"Addu Divehi", "Addu Bas"}, }

-- Dravidian varieties

m["ta-mid"] = { "Middle Tamil", 20987434,	"ta", }

m["kn-hav"] = { "Havigannada", 24276369,	"kn", }

m["kn-kun"] = { "Kundagannada", 6444255,	"kn", }

-- Prakrits

m["pra-ard"] = { "Ardhamagadhi Prakrit", 35217,	"inc-pra", aliases = {"Ardhamagadhi"}, }

m["pra-hel"] = { "Helu Prakrit", 15080869,	"inc-pra", aliases = {"Elu", "Elu Prakrit", "Helu"}, }

m["pra-kha"] = { "Khasa Prakrit", nil, "inc-pra", aliases = {"Khasa"}, }

m["pra-mag"] = { "Magadhi Prakrit", 2652214,	"inc-pra", aliases = {"Magadhi"}, }

m["pra-mah"] = { "Maharastri Prakrit", 2586773,	"inc-pra", aliases = {"Maharashtri Prakrit", "Maharastri", "Maharashtri"}, }

m["pra-pai"] = { "Paisaci Prakrit", 2995607,	"pra-sau", aliases = {"Paisaci", "Paisachi"}, ancestors = "pra-sau" }

m["pra-sau"] = { "Sauraseni Prakrit", 2452885,	"inc-pra", aliases = {"Sauraseni", "Shauraseni"}, }

m["pra-ava"] = { "Avanti", nil, "inc-pra", aliases = {"Avanti Prakrit"}, }

m["pra-pra"] = { "Pracya", nil, "inc-pra", aliases = {"Pracya Prakrit"}, }

m["pra-bah"] = { "Bahliki", nil, "inc-pra", aliases = {"Bahliki Prakrit"}, }

m["pra-dak"] = { "Daksinatya", nil, "inc-pra", aliases = {"Daksinatya Prakrit"}, }

m["pra-sak"] = { "Sakari", nil, "inc-pra", aliases = {"Sakari Prakrit"}, }

m["pra-can"] = { "Candali", nil, "inc-pra", aliases = {"Candali Prakrit"}, }

m["pra-sab"] = { "Sabari", nil, "inc-pra", aliases = {"Sabari Prakrit"}, }

m["pra-abh"] = { "Abhiri", nil, "inc-pra", aliases = {"Abhiri Prakrit"}, }

m["pra-dra"] = { "Dramili", nil, "inc-pra", aliases = {"Dramili Prakrit"}, }

m["pra-odr"] = { "Odri", nil, "inc-pra", aliases = {"Odri Prakrit"}, }

-- Italian, Latin and other Italic varieties

m["roa-oit"] = { "Old Italian", 652,	"it", }

m["it-CH"] = { "Switzerland Italian", 672147,	"it", }

-- Latin varieties by period

m["itc-ola"] = { "Old Latin", 12289,	"la", }

m["la-cla"] = { "Classical Latin", 253854,	"la", }

m["la-lat"] = { "Late Latin", 1503113,	"la", ancestors = "la-cla", }

m["la-vul"] = { "Vulgar Latin", 37560,	"la", ancestors = "la-cla", }

m["la-med"] = { "Medieval Latin", 1163234,	"la", ancestors = "la-lat", }

m["la-eme"] = { "Early Medieval Latin", nil, "la-med", wikipedia_article = "Medieval Latin", }

m["la-ecc"] = { "Ecclesiastical Latin", 1247932,	"la", aliases = {"Church Latin"}, ancestors = "la-lat", }

m["la-ren"] = { "Renaissance Latin", 499083,	"la", ancestors = "la-med", }

m["la-new"] = { "New Latin", 1248221,	"la", aliases = {"Modern Latin"}, ancestors = "la-ren", }

m["la-con"] = { "Contemporary Latin", 1246397,	"la-new", }

-- other Italic lects

m["osc-luc"] = { "Lucanian", nil, "osc", }

m["osc-sam"] = { "Samnite", nil, "osc", }

m["xum-her"] = { "Hernician", nil, "xum", }

-- Malay and related varieties

m["ms-old"] = { "Old Malay", nil, "ms", }

m["ms-cla"] = { "Classical Malay", nil, "ms", ancestors = "ms-old", }

m["pse-bsm"] = { "Besemah", nil, "pse", }

m["bew-kot"] = { "Betawi Kota", nil, "bew", }

m["bew-ora"] = { "Betawi Ora", nil, "bew", }

m["bew-udi"] = { "Betawi Udik", nil, "bew", }

-- Mongolic lects

m["xng-ear"] = { "Early Middle Mongol", nil, "xng", }

m["xng-lat"] = { "Late Middle Mongol", nil, "xng", ancestors = "xng-ear", }

m["mn-kha"] = { "Khalkha Mongolian", 6399808,	"mn", aliases = {"Khalkha"}, }

m["mn-ord"] = { "Ordos Mongolian", 716904,	"mn", aliases = {"Ordos"}, }

m["mn-cha"] = { "Chakhar Mongolian", 907425,	"mn", aliases = {"Chakhar"}, }

m["mn-khr"] = { "Khorchin Mongolian", 3196210,	"mn", aliases = {"Khorchin"}, }

-- Japanese varieties

m["ja-mid"] = { "Middle Japanese", 6841474,	"ojp", ancestors = "ojp", }

m["ja-mid-ear"] = { "Early Middle Japanese", 182695,	"ja-mid", }

m["ja-mid-lat"] = { "Late Middle Japanese", 1816184,	"ja-mid", ancestors = "ja-mid-ear", }

m["ja-ear"] = { "Early Modern Japanese", 5326692,	"ja", ancestors = "ja-mid-lat", }

m["ojp-eas"] = { "Eastern Old Japanese", 65247957,	"ojp", }

-- Kartvelian varieties

m["ka-mid"] = { "Middle Georgian", nil, "ka", ancestors = "oge", }

-- Korean varieties

m["oko-lat"] = { "Late Old Korean", nil, "oko", }

m["okm-ear"] = { "Early Middle Korean", nil, "okm", }

m["ko-cen"] = { "Central Korean", nil, "ko", }

m["ko-gyg"] = { "Gyeonggi Korean", 485492,	"ko-cen", aliases = {"Seoul Korean"}, }

m["ko-chu"] = { "Chungcheong Korean", 625800,	"ko-cen", aliases = {"Hoseo Korean"}, }

m["ko-hwa"] = { "Hwanghae Korean", 16183706,	"ko-cen", }

m["ko-gan"] = { "Gangwon Korean", 11260444,	"ko-cen", aliases = {"Yeongdong Korean"}, }

m["ko-gys"] = { "Gyeongsang Korean", 488002,	"ko", aliases = {"Southeastern Korean"}, }

m["ko-jeo"] = { "Jeolla Korean", 11250166,	"ko", aliases = {"Southwestern Korean"}, }

m["ko-pyo"] = { "Pyongan Korean", 7263142,	"ko", aliases = {"Northwestern Korean"}, }

m["ko-ham"] = { "Hamgyong Korean", 860702,	"ko", aliases = {"Northeastern Korean"}, }

m["ko-yuk"] = { "Yukjin Korean", 16171275,	"ko", aliases = {"Yukchin Korean", "Ryukjin Korean", "Ryukchin Korean"}, }

-- Occitan varieties

m["oc-auv"] = { "Auvergnat", 35359,	"oc", aliases = {"Auvernhat", "Auvergnese"}, }

m["oc-gas"] = { "Gascon", 35735,	"oc", }

-- standardized dialect of Gascon m["oc-ara"] = { "Aranese", 10196,	"oc-gas", }

m["oc-lan"] = { "Languedocien", 942602,	"oc", aliases = {"Lengadocian"}, }

m["oc-lim"] = { "Limousin", 427614,	"oc", }

m["oc-pro"] = { "Provençal", 241243,	"oc", aliases = {"Provencal"}, }

m["oc-pro-old"] = { "Old Provençal", 2779185,	"pro", }

m["oc-viv"] = { "Vivaro-Alpine", 1649613,	"oc", }

m["oc-jud"] = { "Shuadit", 56472,	"oc", aliases = { "Chouhadite", "Chouhadit", "Chouadite", "Chouadit", "Shuhadit", "Judeo-Occitan", "Judæo-Occitan", "Judaeo-Occitan", "Judeo-Provençal", "Judæo-Provençal", "Judaeo-Provençal", "Judeo-Provencal", "Judaeo-Provencal", "Judeo-Comtadin", "Judæo-Comtadin", "Judaeo-Comtadin", }, }

-- Oromo varieties

m["hae"] = { "Harar Oromo", 5330355,	"om", aliases = {"Eastern Oromo"}, }

m["gax"] = { "Borana", 2910610,	"om", aliases = {"Southern Oromo"}, }

m["orc"] = { "Orma", 2919128,	"om", }

m["ssn"] = { "Waata", 3501553,	"om", }

-- Phillipine varieties

m["tl-old"] = { "Old Tagalog", 12967437,	"tl", }

m["tl-cls"] = { "Classical Tagalog", nil, "tl", }

-- Pre-Roman substrates

m["qsb-ibe"] = { "a pre-Roman substrate of Iberia", 530799,	"und", family = "qfa-sub", }

m["qsb-bal"] = { "Paleo-Balkan", 1815070,	"und", family = "qfa-sub", }

-- Sardinian varieties

m["sc-src"] = { "Logudorese", 777974,	"sc", aliases = {"Logudorese Sardinian"}, }

m["sc-nuo"] = { "Nuorese", nil, "sc-src", aliases = {"Nuorese Sardinian"}, }

m["sc-sro"] = { "Campidanese", 35348,	"sc", aliases = {"Campidanese Sardinian"}, }

-- Rwanda-Rundi varieties

m["rw-kin"] = { "Kinyarwanda", 33573,	"rw", aliases = {"Rwanda"}, }

m["rw-run"] = { "Kirundi", 33583,	"rw", aliases = {"Rundi"}, }

-- Slavic varieties

m["cs-ear"] = { "Early Modern Czech", nil, "cs", ancestors = "zlw-ocs" }

m["cu-bgm"] = { "Middle Bulgarian", 12294897,	"cu", ancestors = "cu" }

m["zle-mru"] = { "Middle Russian", 35228,	"ru", "Cyrs", ancestors = "orv", translit = "ru-translit", }

m["zle-obe"] = { "Old Belarusian", 13211,	"zle-ort", }

m["zle-ouk"] = { "Old Ukrainian", 13211,	"zle-ort", }

m["zlw-mpl"] = { "Middle Polish", 402878,	"pl", ancestors = "zlw-opl", entry_name = { remove_diacritics = c.acute, remove_exceptions = {"Ć", "ć", "Ń", "ń", "Ó", "ó", "Ś", "ś", "Ź", "ź"}, }, }

m["zle-ops"] = { "Old Pskovian", 4167885,	"zle-ono", }

-- Serbo-Croatian varieties

m["ckm"] = { "Chakavian Serbo-Croatian", 337565,	"sh", aliases = {"Čakavian"}, }

m["kjv"] = { "Kajkavian Serbo-Croatian", 838165,	"sh", }

m["sh-tor"] = { -- Linguist code srp-tor "Torlakian Serbo-Croatian", 1078803,	"sh", aliases = {"Torlak"}, }

-- Tibetic lects

m["adx"] = { "Amdo Tibetan", 56509,	"bo", }

m["kbg"] = { "Khamba", 12952626,	"bo", }

m["khg"] = { "Khams Tibetan", 56601,	"bo", }

m["tsk"] = { "Tseku", 11159532,	"bo", }

-- Tuareg lects

m["thv"] = { "Tamahaq", 56703,	"tmh", }

m["ttq"] = { "Tawellemmet", 56390,	"tmh", }

m["taq"] = { "Tamasheq", 4670066,	"tmh", }

m["thz"] = { "Tayert", 56388,	"tmh", }

m["tmh-ght"] = { "Ghat", 47012900,	"tmh", wikipedia_article = "Tamahaq language", }

-- Turkic lects

m["trk-cmn"] = { "Common Turkic", 1126028,	"trk-pro", }

m["trk-ogz-pro"] = { "Proto-Oghuz", 494600,	"trk-pro", family = "trk-ogz", aliases = {"Southwestern Common Turkic"}, }

m["crh-dbj"] = { "Dobrujan Tatar", 12811566,	"crh", aliases = {"Romanian Tatar"}, }

m["cv-ana"] = { "Anatri", nil, "cv", aliases = {"Anatri Chuvash"}, }

m["cv-mid"] = { "Middle Chuvash", nil, "cv", ancestors = "cv-old", }

m["cv-old"] = { "Old Chuvash", nil, "cv", ancestors = "xbo", }

m["cv-vir"] = { "Viryal", 4278332,	"cv", aliases = {"Viryal Chuvash"}, }

m["kjh-fyu"] = { "Fuyu Kyrgyz", 2598963,	"kjh", aliases = {"Fuyu Kirgiz", "Fuyu Kirghiz", "Manchurian Kyrgyz", "Manchurian Kirgiz", "Manchurian Kirghiz"}, }

m["klj-arg"] = { "Arghu", 33455,	"klj", }

m["otk-kir"] = { "Old Kirghiz", 83142,	"otk", }

m["qwm-arm"] = { "Armeno-Kipchak", 2027503,	"qwm", }

m["qwm-mam"] = { "Mamluk-Kipchak", 4279942,	"qwm", }

m["az-cls"] = { "Classical Azerbaijani", nil, "az", aliases = {"Classical Azeri"}, }

m["qxq"] = { "Qashqai", 13192,	"az", aliases = {"Qaşqay", "Qashqayi", "Kashkai", "Kashkay"}, }

m["tr-CY"] = { "Cypriot Turkish", 7917392,	"tr", }

-- Uralic lects

m["mns-eas"] = { "Eastern Mansi", 30311755,	"mns-cen", }

m["mns-wes"] = { "Western Mansi", 30311756,	"mns-cen", }

-- Other lects

m["alv-kro"] = { "Kromanti", 1093206,	"crp-mar", }

m["bat-pro"] = { "Proto-Baltic", 1703347,	"ine-bsl-pro", }

m["es-lun"] = { "Lunfardo", 1401612,	"es", }

m["fiu-pro"] = { "Proto-Finno-Ugric", 79890,	"urj-pro", }

m["gem-sue"] = { "Suevic", 155085,	"gmw-pro", aliases = {"Suebian"}, }

m["iro-ohu"] = { "Old Wendat", nil, "wdt", wikipedia_article = "Huron language", }

m["iro-omo"] = { "Old Mohawk", nil, "moh", }

m["iro-oon"] = { "Old Onondaga", nil, "ono", }

m["mis-idn"] = { "Idiom Neutral", 35847,	"und", -- or "vo" wikipedia_article = "Idiom Neutral", }

m["mis-tdl"] = { "Turduli", nil, "und", wikipedia_article = "Turduli", }

m["mis-tdt"] = { "Turdetani", nil, "und", wikipedia_article = "Turdetani", }

m["mis-xnu"] = { "Xiongnu", 10901674,	"und", wikipedia_article = "Xiongnu", }

m["okz-ang"] = { "Angkorian Old Khmer", 9205,	"okz", wikipedia_article = "Khmer language#Historical periods", }

m["okz-pre"] = { "Pre-Angkorian Old Khmer", 9205,	"okz", wikipedia_article = "Khmer language#Historical periods", }

m["mul-tax"] = { "taxonomic name", nil, "mul", }

m["qsb-pyg"] = { "a substrate language originally spoken by the Pygmies", nil, "und", family = "qfa-sub", wikipedia_article = "Classification of Pygmy languages#Original Pygmy language(s)", }

m["tai-shz"] = { "Shangsi Zhuang", 13216,	"za", }

m["tbq-pro"] = { "Proto-Tibeto-Burman", 7251864,	"sit-pro", }

m["urj-fpr-pro"] = { "Proto-Finno-Permic", nil, "urj-pro", }

m["woy"] = { "Weyto", 3915918,	"und", }

m["th-new"] = { "Hacked Thai", -- temporary for testing new translit/display methods nil, "th", translit = "User:Benwing2/th-scraping-translit", display_text = "User:Benwing2/th-scraping-translit", entry_name = "User:Benwing2/th-scraping-translit", preprocess_links = "User:Benwing2/th-scraping-translit", }

return m_lang.finalizeLanguageData(m_lang.finalizeEtymologyData(m_lang.addDefaultTypes(m, false, "etymology-only")))