Modül:ja/k2r-old
< Modül:ja
Bu belgeleme Modül:ja/k2r-old/belge (düzenle | geçmiş) sayfasından yansıtılmaktadır. Arayüz düzenleyicilerinin deney yapabilmeleri için ayrıca Modül:ja/k2r-old/deneme tahtası sayfası kullanılabilir.
local export = {}
local data = mw.loadData("Modül:ja/veri")
function export.kana_to_romaji(text, nodiacr, keepdot)
if type(text) == "table" then -- assume a frame.
text, nodiacr = text.args[1], text.args[2]
end
-- remove %
text = mw.ustring.gsub(text, '%%', '')
text = mw.ustring.gsub(text, "%^'''", "'''^")
-- local kr_minus_period = data.kr
-- kr_minus_period["。"] = "。"
-- convert Japanese spaces to western spaces
text = mw.ustring.gsub(text, ' ', ' ')
-- if there is a は separated by halfwidth spaces, romanize it as " wa "
text = mw.ustring.gsub(text, ' は ', ' wa ')
-- also if it follows a space and is the last character, e.g. それでは
text = mw.ustring.gsub(text, ' は$', ' wa')
-- or " は、"
text = mw.ustring.gsub(text, ' は、', ' wa, ')
-- or " は。"
text = mw.ustring.gsub(text, ' は。', ' wa. ')
text = mw.ustring.gsub(text, ' は?', ' wa? ')
text = mw.ustring.gsub(text, ' は)', ' wa)')
-- or " '''は''' "
text = mw.ustring.gsub(text, " '''は''' ", " '''wa''' ")
-- romanize では as "dewa"
text = mw.ustring.gsub(text, ' では ', ' dewa ')
text = mw.ustring.gsub(text, ' では$', ' dewa')
text = mw.ustring.gsub(text, ' では、', ' dewa, ')
text = mw.ustring.gsub(text, ' では。', ' dewa. ')
text = mw.ustring.gsub(text, ' では?', ' dewa? ')
text = mw.ustring.gsub(text, ' では)', ' dewa)')
text = mw.ustring.gsub(text, "'''では'''", "'''dewa'''")
-- romanize で は (with space) as "de wa"
text = mw.ustring.gsub(text, "'''で は'''", "'''de wa'''")
-- same sort of thing for へ
text = mw.ustring.gsub(text, ' へ ', ' e ')
text = mw.ustring.gsub(text, " '''へ''' ", " '''e''' ")
text = mw.ustring.gsub(text, ' へ$', ' e')
text = mw.ustring.gsub(text, ' へ、', ' e, ')
text = mw.ustring.gsub(text, ' へ。', ' e. ')
text = mw.ustring.gsub(text, ' へ?', ' e?')
text = mw.ustring.gsub(text, ' へ)', ' e)')
-- dangling small tsu is romanized as nothing
text = mw.ustring.gsub(text, 'ッ。', '。')
text = mw.ustring.gsub(text, 'ッ!', '!')
text = mw.ustring.gsub(text, 'ッ」', '」')
text = mw.ustring.gsub(text, 'ッ、', '、')
-- ゝ means "repeat the previous character" and is used with hiragana, like 々 is for kanji
text = mw.ustring.gsub(text, '(.)ゝ', '%1%1')
text = mw.ustring.gsub(text, '(.)ゞ', function(char) return mw.ustring.toNFC(char .. char .. '゙') end)
-- romanising ヶ
text = mw.ustring.gsub(text, 'ヶげつ', 'kagetsu')
text = mw.ustring.gsub(text, 'ヶ', 'ga')
-- convert hiragana to katakana
text = require('Modül:ja').hira_to_kata(text)
-- replace katakana with romaji (?? not sure what the pattern below does ??)
-- this is hackish, but we're using the period to indicate morpheme boundaries to prevent macrons
-- from forming across them, so we'll remove the ASCII periods used for markup but not the Japanese periods
-- convert the Japanese periods at the end
--table.remove(kr_minus_period, "。")
text = mw.ustring.gsub(text, '.[ィェォャュョァヮゥ゜]?ェ?',
function (char)
if char == "。" then
return char
else
return data.kr_old[char] or char
end
end)
text = require('Modül:ja').fullwidth_to_halfwidth(text)
-- replace long vowel mark with the vowel that comes before
text = mw.ustring.gsub(text, '([aeiou])ー', '%1%1')
-- add vowels with diacritics
if not nodiacr then
text = mw.ustring.gsub(text, 'oo', 'ō')
text = mw.ustring.gsub(text, 'aa', 'ā')
text = mw.ustring.gsub(text, 'ee', 'ē')
text = mw.ustring.gsub(text, 'ou', 'ō')
text = mw.ustring.gsub(text, 'uu', 'ū')
text = mw.ustring.gsub(text, 'ii', 'ī')
end
if not keepdot then
-- if input had spaces, keep them
-- if the input string had periods, then remove them now
text = mw.ustring.gsub(text, '%.', '')
-- now that markup periods are gone, convert the Japanese periods to western periods
text = mw.ustring.gsub(text, "。", ". ")
end
-- romanize sokuon or geminate consonants
-- text = mw.ustring.gsub(text, '^ッ', '')
-- double the previous consonant letter if there is a small tsu
text = mw.ustring.gsub(text, 'ッ([kstpgdbjzrfh])', '%1%1')
text = mw.ustring.gsub(text, 'ッ\'\'\'([kstpgdbjzrfh])', '%1\'\'\'%1')
-- replace ッc with tc
text = mw.ustring.gsub(text, 'ッc', 'tc')
text = mw.ustring.gsub(text, 'ッ\'\'\'c', 't\'\'\'c')
-- if small tsu comes at the end, just throw it away
-- text = mw.ustring.gsub(text, 'ッ$', '')
-- the @ is used to determine when to insert an opostrophe after ん or ン
-- (all is kata at that point)
-- insert apostrophe when ン is followed by a vowel or
-- y, which corresponds to the cases んや (n'ya) んゆ (n'yu) and んよ (n'yo)
text = mw.ustring.gsub(text, "@([aeiouāēīōūy])", "'%1")
-- remove @
text = mw.ustring.gsub(text, "@", "")
-- capitalize any letter following a ^ symbol
text = mw.ustring.gsub(text, "%^%l", mw.ustring.upper)
-- remove ^
text = mw.ustring.gsub(text, "%^", "")
-- clean up spaces
text = mw.text.trim(text)
text = mw.ustring.gsub(text, ' +', ' ')
return text
end
return export