Modül:evn-alfabeçeviri

Vikisözlük sitesinden
Gezinti kısmına atla Arama kısmına atla
[Düzenle] [Tazele] Template-info.svg Modül belgelemesi


local u = mw.ustring.char

local MACRON    = u(0x0304)
local DOTABOVE  = u(0x0307)
local DOTBELOW  = u(0x0323)

local str_gsub, ugsub = string.gsub, mw.ustring.gsub
local UTF8char = '[\1-\127\194-\244][\128-\191]*'

local export = {}

local tab = {
	["А"]='A', ["а"]='a', ["В"]='W', ["в"]='w', ["Е"]='E', ["е"]='e',
	["Ё"]='Jo', ["ё"]='jo',	["Г"]='G', ["г"]='g', ["Д"]='D', ["д"]='d',
	["И"]='I', ["и"]='i',  ["Ӣ"]='Ī', ["ӣ"]='ī', ["Й"]='J', ["й"]='j',
	["К"]='K', ["к"]='k', ["Л"]='L', ["л"]='l', ["М"]='M', ["м"]='m',
	["Н"]='N', ["н"]='n', ["Ӈ"]='Ŋ', ["ӈ"]='ŋ', ["О"]='O', ["о"]='o',
	["П"]='P', ["п"]='p', ["Р"]='R', ["р"]='r', ["С"]='S', ["с"]='s', 
	["Т"]='T', ["т"]='t', ["У"]='U', ["у"]='u', ["Ӯ"]='Ū', ["ӯ"]='ū',
	["Ф"]='F', ["ф"]='f', ["Х"]='H', ["х"]='h', ["Ч"]='Ç', ["ч"]='ç',
	["Ы"]='I', ["ы"]='i', ["Э"]='Ə', ["э"]='ə', ["Ю"]='Ju', ["ю"]='ju',
	["Я"]='Ja', ["я"]='ja',
	-- non-native letters
	["Б"]='B', ["б"]='b', ["Ж"]='Z', ["ж"]='z',	["З"]='Z', ["з"]='z',
	 ["Ц"]='C', ["ц"]='c',  ["Ш"]='Ş', ["ш"]='ş',	["Щ"]='Ş', ["щ"]='ş', --in literary language ш is only found in Russian words and was originally represented with s, however some dialects have ш in native words
	 ['Ъ']='ʺ', ['ъ']='ʺ', ["Ь"]="’", ["ь"]="’"
}

local other = {
	{ 'Я', 'Ja' },
	{ 'я', 'ja' },
	{ 'Ё', 'Jo' },
	{ 'ё', 'jo' },
	{ 'Ю', 'Ju' },
	{ 'ю', 'ju' },

-- Unfortunately the Cyrillic alphabet doesn't distinguish between ʒe and ʒə
	{ 'Де', 'Ʒe' },
	{ 'де', 'ʒe' },
	{ 'Не', 'Ņe' },
	{ 'не', 'ņe' },

	{ 'Ди', 'Ʒi' },
	{ 'ди', 'ʒi' },
	{ 'Ни', 'Ņi' },
	{ 'ни', 'ņi' },

	{ 'Дӣ', 'Ʒī' },
	{ 'дӣ', 'ʒī' },
	{ 'Нӣ', 'Ņī' },
	{ 'нӣ', 'ņī' },

	{ 'Дj', 'Ʒ' },
	{ 'дj', 'ʒ' },
	{ 'Нj', 'Ņ' },
	{ 'нj', 'ņ' },
}
 
function export.tr(text, lang, sc)
	for i, replacement in ipairs(other) do
		text = str_gsub(text, unpack(replacement))
	end

    -- е after a vowel or at the beginning of a word becomes ye
    -- Again, the Cyrillic alphabet doesn't distinguish between je and jə
	text = ugsub(text,
		"([АОУЫЕИӢЪЬаӣиоуыэеъьaeiou]["..MACRON..DOTABOVE..DOTBELOW.."]?)е",
		"%1je")
    text = ugsub(text,
    	"([АОУЫЕИӢЪЬаӣиоуыэеъьaeiou]["..MACRON..DOTABOVE..DOTBELOW.."]?)и",
    	"%1ji")
    text = ugsub(text,
    	"([АОУЫЕИӢЪЬаӣиоуыэеъьaeiou]["..MACRON..DOTABOVE..DOTBELOW.."]?)ӣ",
    	"%1jī")
    text = str_gsub(text, "^Е","Je")
    text = str_gsub(text, "^е","je")
    text = ugsub(text, "([^Ѐ-ӿ])Е","%1Je")
    text = ugsub(text, "([^Ѐ-ӿ])е","%1je")
 
    return (str_gsub(text, UTF8char, tab))
end

return export