Module:Ml-translit

local export = {} local consonants = { ['ക']='k', ['ഖ']='kh', ['ഗ']='g', ['ഘ']='gh', ['ങ']='ṅ', ['ച']='c', ['ഛ']='ch', ['ജ']='j', ['ഝ']='jh', ['ഞ']='ñ', ['ട']='ṭ', ['ഠ']='ṭh', ['ഡ']='ḍ', ['ഢ']='ḍh', ['ണ']='ṇ', ['ത']='t', ['ഥ']='th', ['ദ']='d', ['ധ']='dh', ['ന']='n', ['പ']='p', ['ഫ']='ph', ['ബ']='b', ['ഭ']='bh', ['മ']='m', ['യ']='y', ['ര']='r', ['ല']='l', ['വ']='v', ['ശ']='ś', 	['ഷ']='ṣ', ['സ']='s', ['ഹ']='h', ['ള']='ḷ', ['ഴ']='ḻ', ['റ']='ṟ', ['ഩ']='ṉ' , ['ഺ']='ṯ' , }

local diacritics = { ['\224\181\129\224\181\141'] = 'ŭ', ['\224\180\190'] = 'ā' , ['\224\180\191'] = 'i' , ['\224\181\128'] = 'ī' , ['\224\181\129'] = 'u' , ['\224\181\130'] = 'ū' , ['\224\181\131'] = 'r̥' , ['\224\181\132'] = 'r̥̄' , ['\224\181\134'] = 'e' , ['\224\181\135'] = 'ē' , ['\224\181\136'] = 'ai', ['\224\181\138'] = 'o' , ['\224\181\139'] = 'ō' , ['\224\181\151'] = 'au', ['\224\181\162'] = 'l̥ ', ['\224\181\163'] = 'l̥̄' , --virama, supresses the inherent vowel "a" ['\224\181\141'] = '',	-- no diacritic [''] = 'a' }

local nonconsonants = { -- vowels ['അ']='a', ['ആ']='ā' , ['ഇ']='i' , ['ഈ']='ī' , ['ഉ']='u' , ['ഊ']='ū' , ['ഋ']='r̥', ['ൠ']='r̥̄' , ['ഌ']='l̥' , ['ൡ']='l̥̄', ['എ']='e' , ['ഏ']='ē' , ['ഐ']='ai', ['ഒ']='o' , ['ഓ']='ō' , ['ഔ']='au' , -- other symbols ['ം']='ṃ', -- anusvara ['ഃ']='ḥ', -- visarga ['ഽ']='’', -- praślēṣam -- chillus, consonants that never take vowels ['ൺ']='ṇ', ['ൻ']='n' , ['ർ']='r' , ['ൽ']='l' , ['ൾ']='ḷ' , ['ൿ']='k' , -- digits ['൦'] = '0', ['൧'] = '1', ['൨'] = '2', ['൩'] = '3', ['൪'] = '4',	['൫'] = '5', ['൬'] = '6', ['൭'] = '7', ['൮'] = '8', ['൯']= '9',	['൰']='10', ['൱']='100', ['൲']='1000', ['൳']='¼', ['൴']='½', ['൵']='¾', }

-- translit any words or phrases function export.tr(text, lang, sc) local VIRAMA = '്' -- final virama rules text = mw.ustring.gsub(text, VIRAMA .. "([%,%.%!%?%:%;]?)$", VIRAMA .. "ŭ%1") text = mw.ustring.gsub(text, VIRAMA .. "([%,%.%!%?%:%;]?) ", VIRAMA .. "ŭ%1 ") text = mw.ustring.gsub(		text,		'([കഖഗഘങചഛജഝഞടഠഡഢണതഥദധനപഫബഭമയരലവശഷസഹളഴറഩഺ])'..		'(\224\181\129?[\224\180\190\224\180\191\224\181\128\224\181\129\224\181\130\224\181\131\224\181\132\224\181\162\224\181\163\224\181\134\224\181\135\224\181\138\224\181\139\224\181\136\224\181\151\224\181\141]?)',		function(c, d)			return consonants[c] .. (diacritics[d] or d)		end) text = mw.ustring.gsub(text, '.', nonconsonants) -- anusvara text = mw.ustring.gsub(text, 'ṃ([kgṅ])', 'ṅ%1') text = mw.ustring.gsub(text, 'ṃ([cjñ])', 'ñ%1') text = mw.ustring.gsub(text, 'ṃ([ṭḍṇ])', 'ṇ%1') text = mw.ustring.gsub(text, 'ṃ([tdn])', 'n%1') text = mw.ustring.gsub(text, 'ṃ([pbm])', 'm%1') text = mw.ustring.gsub(text, "ŭ ([%,%.%!%?%:%;]?)([aāiīuūeo])", " %1%2") -- ŭ is elided before vowels return text end return export