মডিউল:cmn-pron
এই মডিউলের জন্য মডিউল:cmn-pron/নথি-এ নথিপত্র তৈরি করা হয়ে থাকতে পারে
local export = {}
local find = mw.ustring.find
local gsub = mw.ustring.gsub
local len = mw.ustring.len
local match = mw.ustring.match
local sub = mw.ustring.sub
local split = mw.text.split
local gsplit = mw.text.gsplit
local m_zh = require("Module:zh")
local _m_zh_data = nil
local hom_data = mw.loadData("Module:zh/data/cmn-hom")
-- if not empty
local function ine(var)
if var == "" then
return nil
else
return var
end
end
local breve, hacek, circumflex = mw.ustring.char(0x306), mw.ustring.char(0x30C), mw.ustring.char(0x302)
local decompose = mw.ustring.toNFD
local function breve_error(text)
if type(text) ~= "string" then
return
end
text = decompose(text)
if text:find(breve) then
error('The pinyin text "' .. text .. '" contains a breve. Replace it with "' .. text:gsub(breve, hacek) .. '".', 2)
end
if find(text, '[ẑĉŝŋẐĈŜ]') or find(text, '[zcsZCS][' .. circumflex .. ']') then
error('The pinyin text "' .. text .. '" contains an uncommon shorthand. Replace it with "' .. gsub(text, '[ẑĉŝŋẐĈŜ]', {['ẑ'] = 'zh', ['ĉ'] = 'ch', ['ŝ'] = 'sh', ['ŋ'] = 'ng', ['Ẑ'] = 'Zh', ['Ĉ'] = 'Ch', ['Ŝ'] = 'Sh', [circumflex] = 'h'}) .. '".', 2)
end
end
local function m_zh_data()
if _m_zh_data == nil then _m_zh_data = mw.loadData("Module:zh/data/cmn-tag") end;
return _m_zh_data;
end
local py_detone = {
['ā'] = 'a', ['á'] = 'a', ['ǎ'] = 'a', ['à'] = 'a',
['ō'] = 'o', ['ó'] = 'o', ['ǒ'] = 'o', ['ò'] = 'o',
['ē'] = 'e', ['é'] = 'e', ['ě'] = 'e', ['è'] = 'e',
['ê̄'] = 'ê', ['ế'] = 'ê', ['ê̌'] = 'ê', ['ề'] = 'ê',
['ī'] = 'i', ['í'] = 'i', ['ǐ'] = 'i', ['ì'] = 'i',
['ū'] = 'u', ['ú'] = 'u', ['ǔ'] = 'u', ['ù'] = 'u',
['ǖ'] = 'ü', ['ǘ'] = 'ü', ['ǚ'] = 'ü', ['ǜ'] = 'ü',
['m̄'] = 'm', ['ḿ'] = 'm', ['m̌'] = 'm', ['m̀'] = 'm',
['n̄'] = 'n', ['ń'] = 'n', ['ň'] = 'n', ['ǹ'] = 'n',
}
local py_tone = {
['̄'] = '1',
['́'] = '2',
['̌'] = '3',
['̀'] = '4'
}
local tones = '[̄́̌̀]'
function export.py_detone(f)
local text = type(f) == 'table' and f.args[1] or f
return mw.ustring.toNFC(gsub(mw.ustring.toNFD(text), tones, ''))
end
function export.py_transf(f)
local text = type(f) == 'table' and f.args[1] or f
return export.py_detone(text) .. export.tone_determ(text)
end
function export.tone_determ(f)
local text = type(f) == 'table' and f.args[1] or f
text = mw.ustring.toNFD(text)
return py_tone[match(text, tones)] or '5'
end
function export.py_transform(text, detone, not_spaced)
if type(text) == 'table' then text, detone, not_spaced = text.args[1], text.args[2], text.args[3] end
if find(text, '') then
error("Pinyin contains the hidden character: (U+200B). Please remove that character from the text.")
end
detone = ine(detone)
not_spaced = ine(not_spaced)
text = gsub(gsub(mw.ustring.toNFD(text), mw.ustring.toNFD('ê'), 'ê'), mw.ustring.toNFD('ü'), 'ü')
if find(mw.ustring.lower(text), '[aeiouêü]' .. tones .. '[aeiou]?[aeiouêü]' .. tones .. '') and not not_spaced then
error(("Missing apostrophe before null-initial syllable - should be \"%s\" instead."):format(gsub(text, '([aeiouêü]' .. tones .. '[aeiou]?)([aeiouêü]' .. tones .. ')', "%1'%2"))) end
original_text = text
text = gsub(text,'([aoeAOE])([iou])(' .. tones .. ')', '%1%3%2')
text = gsub(text,'([iuü])(' .. tones .. ')([aeiou])', '%1%3%2')
if text ~= original_text then
error("Incorrect diacritic placement in Pinyin - should be \"".. text .. "\" instead.") end
text = mw.ustring.lower(text)
if not mw.ustring.find(text, tones) and text:find('[1-5]') then
return gsub(text, '(%d)(%l)', '%1 %2')
end
text = gsub(text, "#", " #")
if find(text, '[一不,.?]') then
text = gsub(text, '([一不])$', {['一'] = ' yī', ['不'] = ' bù'})
text = gsub(text, '([一不])', ' %1 ')
text = gsub(text, '([,.?])', ' %1 ')
text = gsub(text, ' +', ' ')
text = gsub(text, '^ ', '')
text = gsub(text, ' $', '')
text = gsub(text, '%. %. %.', '...')
end
text = gsub(text, "['%-]", ' ')
text = gsub(text, '([aeiouêümn]' .. tones .. '?n?g?r?)([bpmfdtnlgkhjqxzcsywr]h?)', '%1 %2')
text = gsub(text, ' ([grn])$', '%1')
text = gsub(text, ' ([grn]) ', '%1 ')
if detone then
text = mw.ustring.gsub(text, tones, py_tone)
text = gsub(text, '([1234])([^ ]*)', '%2%1')
text = mw.ustring.gsub(text, '([%lüê]) ', '%15 ')
text = mw.ustring.gsub(text, '([%lüê])$', '%15')
end
if not_spaced then
text = gsub(text, ' ', '')
end
return mw.ustring.toNFC(text)
end
function export.py_ipa(text)
local ipa_initial = {
['b'] = 'p', ['p'] = 'pʰ', ['m'] = 'm', ['f'] = 'f',
['d'] = 't', ['t'] = 'tʰ', ['n'] = 'n', ['l'] = 'l',
['g'] = 'k', ['k'] = 'kʰ', ['h'] = 'x', ['ng'] = 'ŋ',
['j'] = 't͡ɕ', ['q'] = 't͡ɕʰ', ['x'] = 'ɕ',
['z'] = 't͡s', ['c'] = 't͡sʰ', ['s'] = 's', ['r'] = 'ʐ',
['zh'] = 'ʈ͡ʂ', ['ch'] = 'ʈ͡ʂʰ', ['sh'] = 'ʂ',
[''] = ''
}
local ipa_initial_tl = {
['p'] = 'b̥', ['t'] = 'd̥', ['k'] = 'g̊', ['t͡ɕ'] = 'd͡ʑ̥', ['t͡s'] = 'd͡z̥', ['ʈ͡ʂ'] = 'ɖ͡ʐ̥'
}
local ipa_final = {
['yuanr'] = 'ɥɑɻ', ['iangr'] = 'jɑ̃ɻ', ['yangr'] = 'jɑ̃ɻ', ['uangr'] = 'wɑ̃ɻ', ['wangr'] = 'wɑ̃ɻ', ['yingr'] = 'iɤ̯̃ɻ', ['wengr'] = 'ʊ̃ɻ', ['iongr'] = 'jʊ̃ɻ', ['yongr'] = 'jʊ̃ɻ',
['yuan'] = 'ɥɛn', ['iang'] = 'jɑŋ', ['yang'] = 'jɑŋ', ['uang'] = 'wɑŋ', ['wang'] = 'wɑŋ', ['ying'] = 'iŋ', ['weng'] = 'wəŋ', ['iong'] = 'jʊŋ', ['yong'] = 'jʊŋ', ['ianr'] = 'jɑɻ', ['yanr'] = 'jɑɻ', ['uair'] = 'wɑɻ', ['wair'] = 'wɑɻ', ['uanr'] = 'wɑɻ', ['wanr'] = 'wɑɻ', ['iaor'] = 'jaʊ̯ɻʷ', ['yaor'] = 'jaʊ̯ɻʷ', ['üanr'] = 'ɥɑɻ', ['vanr'] = 'ɥɑɻ', ['angr'] = 'ɑ̃ɻ', ['yuer'] = 'ɥɛɻ', ['weir'] = 'wəɻ', ['wenr'] = 'wəɻ', ['your'] = 'jɤʊ̯ɻʷ', ['yinr'] = 'iə̯ɻ', ['yunr'] = 'yə̯ɻ', ['engr'] = 'ɤ̃ɻ', ['ingr'] = 'iɤ̯̃ɻ', ['ongr'] = 'ʊ̃ɻ',
['uai'] = 'waɪ̯', ['wai'] = 'waɪ̯', ['yai'] = 'jaɪ̯', ['iao'] = 'jɑʊ̯', ['yao'] = 'jɑʊ̯', ['ian'] = 'jɛn', ['yan'] = 'jɛn', ['uan'] = 'wän', ['wan'] = 'wän', ['üan'] = 'ɥɛn', ['van'] = 'ɥɛn', ['ang'] = 'ɑŋ', ['yue'] = 'ɥɛ', ['wei'] = 'weɪ̯', ['you'] = 'joʊ̯', ['yin'] = 'in', ['wen'] = 'wən', ['yun'] = 'yn', ['eng'] = 'ɤŋ', ['ing'] = 'iŋ', ['ong'] = 'ʊŋ', ['air'] = 'ɑɻ', ['anr'] = 'ɑɻ', ['iar'] = 'jɑɻ', ['yar'] = 'jɑɻ', ['uar'] = 'wɑɻ', ['war'] = 'wɑɻ', ['aor'] = 'aʊ̯ɻʷ', ['ier'] = 'jɛɻ', ['yer'] = 'jɛɻ', ['uor'] = 'wɔɻ', ['wor'] = 'wɔɻ', ['üer'] = 'ɥɛɻ', ['ver'] = 'ɥɛɻ', ['eir'] = 'əɻ', ['enr'] = 'əɻ', ['uir'] = 'wəɻ', ['unr'] = 'wəɻ', ['our'] = 'ɤʊ̯ɻʷ', ['iur'] = 'jɤʊ̯ɻʷ', ['inr'] = 'iə̯ɻ', ['ünr'] = 'yə̯ɻ', ['vnr'] = 'yə̯ɻ', ['yir'] = 'iə̯ɻ', ['wur'] = 'uɻʷ', ['yur'] = 'yə̯ɻ', ['yor'] = 'jɔɻ',
['io'] = 'jɔ', ['yo'] = 'jɔ', ['ia'] = 'jä', ['ya'] = 'jä', ['ua'] = 'wä', ['wa'] = 'wä', ['ai'] = 'aɪ̯', ['ao'] = 'ɑʊ̯', ['an'] = 'än', ['ie'] = 'jɛ', ['ye'] = 'jɛ', ['uo'] = 'wɔ', ['wo'] = 'wɔ', ['ue'] = 'ɥɛ', ['üe'] = 'ɥɛ', ['ve'] = 'ɥɛ', ['ei'] = 'eɪ̯', ['ui'] = 'weɪ̯', ['ou'] = 'oʊ̯', ['iu'] = 'joʊ̯', ['en'] = 'ən', ['in'] = 'in', ['un'] = 'wən', ['ün'] = 'yn', ['vn'] = 'yn', ['yi'] = 'i', ['wu'] = 'u', ['yu'] = 'y', ['mˋ'] = 'm̩', ['ng'] = 'ŋ̍', ['ňg'] = 'ŋ̍', ['ńg'] = 'ŋ̍', ['ê̄'] = 'ɛ', ['ê̌'] = 'ɛ', ['ar'] = 'ɑɻ', ['er'] = 'ɤɻ', ['or'] = 'wɔɻ', ['ir'] = 'iə̯ɻ', ['ur'] = 'uɻʷ', ['ür'] = 'yə̯ɻ', ['vr'] = 'yə̯ɻ',
['a'] = 'ä', ['e'] = 'ɤ', ['o'] = 'wɔ', ['i'] = 'i', ['u'] = 'u', ['ü'] = 'y', ['v'] = 'y', ['m'] = 'm̩', ['ḿ'] = 'm̩', ['n'] = 'n̩', ['ń'] = 'n̩', ['ň'] = 'n̩', ['ê'] = 'ɛ'
}
local ipa_null = {
['a'] = true, ['o'] = true, ['e'] = true, ['ê'] = true,
['ai'] = true, ['ei'] = true, ['ao'] = true, ['ou'] = true,
['an'] = true, ['en'] = true, ['er'] = true,
['ang'] = true, ['ong'] = true, ['eng'] = true,
}
local ipa_tl_ts = {
['1'] = '²', ['2'] = '³', ['3'] = '⁴', ['4'] = '¹', ['5'] = '¹'
}
local ipa_third_t_ts = {
['1'] = '²¹⁴⁻²¹¹', ['3'] = '²¹⁴⁻³⁵', ['#3'] = '²¹⁴⁻²¹¹', ['5'] = '²¹⁴', ['2'] = '²¹⁴⁻²¹¹', ['1-2'] = '²¹⁴⁻²¹¹', ['4-2'] = '²¹⁴⁻²¹¹', ['4'] = '²¹⁴⁻²¹¹', ['1-4'] = '²¹⁴⁻²¹¹'
}
local ipa_t_values = {
['4'] = '⁵¹', ['1-4'] = '⁵⁵⁻⁵¹', ['1'] = '⁵⁵', ['2'] = '³⁵', ['1-2'] = '⁵⁵⁻³⁵', ['4-2'] = '⁵¹⁻³⁵'
}
local tone = {}
local tone_cat = {}
text = gsub(export.py_transform(text), '[,.]', '')
text = gsub(text, ' +', ' ')
local p = split(text, " ")
for i = 1, #p do
tone_cat[i] = export.tone_determ(p[i])
p[i] = gsub(p[i], '.[̄́̌̀]?', py_detone)
if p[i] == '一' then
tone_cat[i] = (export.tone_determ(p[i+1]) == '4' or p[i+1] == 'ge') and '1-2' or '1-4'
p[i] = 'yi'
elseif p[i] == '不' then
tone_cat[i] = (export.tone_determ(p[i+1]) == '4') and '4-2' or '4'
p[i] = 'bu'
end
end
tone_cat.length = #tone_cat
local function get_initial_and_final_IPA(a, b, c)
return a ..
(ipa_initial[b] or error(("Unrecognised initial: \"%s\""):format(b))) ..
(ipa_final[c] or error(("Unrecognised final: \"%s\". Are you missing an apostrophe before the null-initial syllable, or using an invalid Pinyin final?"):format(c)))
end
for i, item in ipairs(p) do
if ipa_null[item] then item = 'ˀ' .. item end
item = gsub(item, '([jqx])u', '%1ü')
if item == 'ng' then
item = ipa_final['ng']
else
item = gsub(item, '^(#?ˀ?)([bcdfghjklmnpqrstxz]?h?)(.+)$',
get_initial_and_final_IPA)
end
item = gsub(item, '(ʈ?͡?[ʂʐ]ʰ?)ir', '%1ʐ̩ɻ')
item = gsub(item, '(ʈ?͡?[ʂʐ]ʰ?)i', '%1ʐ̩')
item = gsub(item, '(t?͡?sʰ?)ir', '%1z̩ɻ')
item = gsub(item, '(t?͡?sʰ?)i', '%1z̩')
item = gsub(item, 'ʐʐ̩', 'ʐ̩')
item = gsub(item, 'ˀwɔ', 'ˀɔ')
local curr_tone_cat, next_tone_cat = tone_cat[i], tone_cat[i+1]
if curr_tone_cat == '5' then
item = gsub(item, '^([ptk])([^͡ʰ])', function(a, b) return ipa_initial_tl[a] .. b end)
item = gsub(item, '^([tʈ]͡[sɕʂ])([^ʰ])', function(a, b) return ipa_initial_tl[a] .. b end)
item = gsub(item, 'ɤ$', 'ə')
tone[i] = ipa_tl_ts[tone_cat[i-1]] or ""
elseif curr_tone_cat == '3' then
if p[i+1] and match(p[i+1], "#") then next_tone_cat = "#3" end
if i == tone_cat.length then
if i == 1 then tone[i] = '²¹⁴' else tone[i] = '²¹⁴⁻²¹⁽⁴⁾' end
else
tone[i] = ipa_third_t_ts[next_tone_cat]
end
elseif curr_tone_cat == '4' and next_tone_cat == '4' then
tone[i] = '⁵¹⁻⁵³'
elseif curr_tone_cat == '4' and next_tone_cat == '1-4' then
tone[i] = '⁵¹⁻⁵³'
elseif curr_tone_cat == '1-4' and next_tone_cat == '4' then
tone[i] = '⁵⁵⁻⁵³'
else
tone[i] = ipa_t_values[curr_tone_cat]
end
p[i] = item .. tone[i]
p[i] = gsub(p[i], "#", "")
end
return table.concat(p, " ")
end
function export.py_number_to_mark(text)
local priority = { "a", "o", "e", "ê", "i", "u", "ü" }
local toneMark = { ["1"] = "̄", ["2"] = "́", ["3"] = "̌", ["4"] = "̀", ["5"] = "", ["0"] = "", [""] = "" }
local mark = toneMark[match(text, "[0-5]?$")]
local toneChars = "[̄́̌̀]"
text = gsub(text, "[0-5]?$", "")
for _, letter in ipairs(priority) do
text = gsub(text, letter, letter .. mark)
if find(text, toneChars) then break end
end
return mw.ustring.toNFC(gsub(text, "i("..toneChars..")u", "iu%1"))
end
function export.py_zhuyin(text)
local zhuyin_initial = {
['b'] = 'ㄅ', ['p'] = 'ㄆ', ['m'] = 'ㄇ', ['f'] = 'ㄈ',
['d'] = 'ㄉ', ['t'] = 'ㄊ', ['n'] = 'ㄋ', ['l'] = 'ㄌ',
['g'] = 'ㄍ', ['k'] = 'ㄎ', ['h'] = 'ㄏ',
['j'] = 'ㄐ', ['q'] = 'ㄑ', ['x'] = 'ㄒ',
['z'] = 'ㄗ', ['c'] = 'ㄘ', ['s'] = 'ㄙ', ['r'] = 'ㄖ',
['zh'] = 'ㄓ', ['ch'] = 'ㄔ', ['sh'] = 'ㄕ',
[''] = ''
}
local zhuyin_final = {
['yuan'] = 'ㄩㄢ', ['iang'] = 'ㄧㄤ', ['yang'] = 'ㄧㄤ', ['uang'] = 'ㄨㄤ', ['wang'] = 'ㄨㄤ', ['ying'] = 'ㄧㄥ', ['weng'] = 'ㄨㄥ', ['iong'] = 'ㄩㄥ', ['yong'] = 'ㄩㄥ',
['uai'] = 'ㄨㄞ', ['wai'] = 'ㄨㄞ', ['yai'] = 'ㄧㄞ', ['iao'] = 'ㄧㄠ', ['yao'] = 'ㄧㄠ', ['ian'] = 'ㄧㄢ', ['yan'] = 'ㄧㄢ', ['uan'] = 'ㄨㄢ', ['wan'] = 'ㄨㄢ', ['üan'] = 'ㄩㄢ', ['ang'] = 'ㄤ', ['yue'] = 'ㄩㄝ', ['wei'] = 'ㄨㄟ', ['you'] = 'ㄧㄡ', ['yin'] = 'ㄧㄣ', ['wen'] = 'ㄨㄣ', ['yun'] = 'ㄩㄣ', ['eng'] = 'ㄥ', ['ing'] = 'ㄧㄥ', ['ong'] = 'ㄨㄥ',
['io'] = 'ㄧㄛ', ['yo'] = 'ㄧㄛ', ['ia'] = 'ㄧㄚ', ['ya'] = 'ㄧㄚ', ['ua'] = 'ㄨㄚ', ['wa'] = 'ㄨㄚ', ['ai'] = 'ㄞ', ['ao'] = 'ㄠ', ['an'] = 'ㄢ', ['ie'] = 'ㄧㄝ', ['ye'] = 'ㄧㄝ', ['uo'] = 'ㄨㄛ', ['wo'] = 'ㄨㄛ', ['ue'] = 'ㄩㄝ', ['üe'] = 'ㄩㄝ', ['ei'] = 'ㄟ', ['ui'] = 'ㄨㄟ', ['ou'] = 'ㄡ', ['iu'] = 'ㄧㄡ', ['en'] = 'ㄣ', ['in'] = 'ㄧㄣ', ['un'] = 'ㄨㄣ', ['ün'] = 'ㄩㄣ', ['yi'] = 'ㄧ', ['wu'] = 'ㄨ', ['yu'] = 'ㄩ',
['a'] = 'ㄚ', ['e'] = 'ㄜ', ['o'] = 'ㄛ', ['i'] = 'ㄧ', ['u'] = 'ㄨ', ['ü'] = 'ㄩ', ['ê'] = 'ㄝ', [''] = ''
}
local zhuyin_er = {
['r'] = 'ㄦ', [''] = ''
}
local zhuyin_tone = {
['1'] = '', ['2'] = 'ˊ', ['3'] = 'ˇ', ['4'] = 'ˋ', ['5'] = '˙', ['0'] = '˙'
}
if type(text) == 'table' then
if text.args[1] == '' then
text = mw.title.getCurrentTitle().text
else
text = text.args[1]
end
end
breve_error(text)
text = gsub(text, "#", "")
text = export.py_transform(text, true)
text = gsub(text, '([jqx])u', '%1ü')
text = gsub(text, '([zcs]h?)i', '%1')
text = gsub(text, '([r])i', '%1')
local function add_tone(syllable, tone)
if tone == '5' then
return zhuyin_tone[tone] .. syllable
else
return syllable .. zhuyin_tone[tone]
end
end
local function fun1(a, b) return add_tone((({['ng'] = 'ㄫ', ['hm'] = 'ㄏㄇ'})[a] or a), b) end
local function fun2(number) return add_tone('ㄏㄫ', number) end
local function fun3(number) return add_tone('ㄦ', number) end
local function fun4(a, b, c, d) return add_tone(zhuyin_initial[a] .. zhuyin_final[b], d) .. zhuyin_er[c] end
local word = split(text, " ", true)
for i, syllable in ipairs(word) do
if find(syllable, '^[hn][mg][012345]$') then
syllable = gsub(syllable, '^([hn][mg])([012345])$', fun1)
elseif find(syllable, '^hng[012345]$') then
syllable = gsub(syllable, '^hng([012345])$', fun2)
elseif find(syllable, '^er[012345]$') then
syllable = gsub(syllable, '^er([012345])$', fun3)
else
syllable = gsub(syllable, '^([bpmfdtnlgkhjqxzcsr]?h?)([aeiouêüyw]?[aeioun]?[aeioung]?[ng]?)(r?)([012345])$',
fun4)
end
if find(syllable, '[%l%d]') then
error(("Zhuyin conversion unsuccessful: \"%s\". Are you using a valid Pinyin syllable? Is the text using a breve letter instead of a caron one?"):format(syllable))
end
word[i] = syllable
end
text = gsub(table.concat(word, " "), ' , ', ', ')
return text
end
function export.zhuyin_py(text)
local zhuyin_py_initial = {
["ㄅ"] = "b", ["ㄆ"] = "p", ["ㄇ"] = "m", ["ㄈ"] = "f",
["ㄉ"] = "d", ["ㄊ"] = "t", ["ㄋ"] = "n", ["ㄌ"] = "l",
["ㄍ"] = "g", ["ㄎ"] = "k", ["ㄏ"] = "h",
["ㄐ"] = "j", ["ㄑ"] = "q", ["ㄒ"] = "x",
["ㄓ"] = "zh", ["ㄔ"] = "ch", ["ㄕ"] = "sh", ["ㄖ"] = "r",
["ㄗ"] = "z", ["ㄘ"] = "c", ["ㄙ"] = "s",
[""] = ""
}
local zhuyin_py_final = {
['ㄚ'] = 'a', ['ㄛ'] = 'o', ['ㄜ'] = 'e', ['ㄝ'] = 'ê', ['ㄞ'] = 'ai', ['ㄟ'] = 'ei', ['ㄠ'] = 'ao', ['ㄡ'] = 'ou', ['ㄢ'] = 'an', ['ㄣ'] = 'en', ['ㄤ'] = 'ang', ['ㄥ'] = 'eng',
['ㄧ'] = 'i', ['ㄧㄚ'] = 'ia', ['ㄧㄛ'] = 'io', ['ㄧㄝ'] = 'ie', ['ㄧㄞ'] = 'iai', ['ㄧㄠ'] = 'iao', ['ㄧㄡ'] = 'iu', ['ㄧㄢ'] = 'ian', ['ㄧㄣ'] = 'in', ['ㄧㄤ'] = 'iang', ['ㄧㄥ'] = 'ing',
['ㄨ'] = 'u', ['ㄨㄚ'] = 'ua', ['ㄨㄛ'] = 'uo', ['ㄨㄞ'] = 'uai', ['ㄨㄟ'] = 'ui', ['ㄨㄢ'] = 'uan', ['ㄨㄣ'] = 'un', ['ㄨㄤ'] = 'uang', ['ㄨㄥ'] = 'ong',
['ㄩ'] = 'ü', ['ㄩㄝ'] = 'ue', ['ㄩㄝ'] = 'üe', ['ㄩㄢ'] = 'üan', ['ㄩㄣ'] = 'ün', ['ㄩㄥ'] = 'iong',
['ㄦ'] = 'er', ['ㄫ'] = 'ng', ['ㄇ'] = 'm', [''] = 'i'
}
local zhuyin_py_tone = {
["ˊ"] = "\204\129", ["ˇ"] = "\204\140", ["ˋ"] = "\204\128", ["˙"] = "", [""] = "\204\132"
}
if type(text) == "table" then text = text.args[1] end
local word = split(text, " ", true)
local function process_syllable(syllable)
syllable = gsub(syllable, '^([ㄓㄔㄕㄖㄗㄘㄙ])([ˊˇˋ˙]?)$', '%1ㄧ%2')
return gsub(syllable, '([ㄅㄆㄇㄈㄉㄊㄋㄌㄍㄎㄏㄐㄑㄒㄓㄔㄕㄖㄗㄘㄙ]?)([ㄧㄨㄩ]?[ㄚㄛㄜㄝㄞㄟㄠㄡㄢㄣㄤㄥㄦㄫㄧㄨㄩㄇ])([ˊˇˋ˙]?)(ㄦ?)', function(initial, final, tone, erhua)
initial = zhuyin_py_initial[initial]
final = zhuyin_py_final[final]
if erhua ~= '' then
final = final .. 'r'
end
if initial == '' then
final = final:gsub('^([iu])(n?g?)$', function(a, b) return a:gsub('[iu]', {['i'] = 'yi', ['u'] = 'wu'}) .. b end)
final = final:gsub('^(w?u)([in])$', 'ue%2')
final = final:gsub('^iu$', 'iou')
final = final:gsub('^([iu])', {['i'] = 'y', ['u'] = 'w'})
final = final:gsub('^ong', 'weng')
final = gsub(final, '^ü', 'yu')
end
if initial:find('[jqx]') then
final = gsub(final, '^ü', 'u')
end
local tone = zhuyin_py_tone[tone]
if final:find('[ae]') then
final = final:gsub("([ae])", "%1" .. tone)
elseif final:find('i[ou]') then
final = final:gsub("(i[ou])", "%1" .. tone)
elseif final:find('[io]') then
final = final:gsub("([io])", "%1" .. tone)
else
final = gsub(final, "^([wy]?)(.)", "%1" .. "%2" .. tone)
end
return initial .. final
end)
end
for i, syllable in ipairs(word) do
word[i] = process_syllable(syllable)
end
return mw.ustring.toNFC(table.concat(word, " "))
end
function export.py_wg(text)
local py_wg_initial = {
["b"] = "p", ["p"] = "pʻ",
["d"] = "t", ["t"] = "tʻ",
["g"] = "k", ["k"] = "kʻ",
["j"] = "ch", ["q"] = "chʻ", ["x"] = "hs",
["z"] = "ts", ["c"] = "tsʻ", ["r"] = "j",
["zh"] = "ch", ["ch"] = "chʻ",
}
local py_wg_final = {
["^([yw]?)e([^ih])"] = "%1ê%2",
["^e$"] = "ê",
["([iy])an$"] = "%1en",
["(i?)ong"] = "%1ung",
["([iy])e$"] = "%1eh",
["[uü]e"] = "üeh",
["r$"] = "rh",
["^ê$"] = "eh",
["^i$"] = "i",
["yi$"] = "i",
}
local py_wg_syl = {
["(t?sʻ?)uo"] = "%1o",
["^([tnlcj]h?ʻ?)uo"] = "%1o",
["shi"] = "shih", ["ji"] = "jih",
["tsi"] = "tzŭ", ["tsʻi"] = "tzʻŭ", ["^si$"] = "ssŭ",
["^([kh]?ʻ?)ê$"] = "%1o",
["yên"] = "yen",
["you"] = "yu", ["^ih"] = "i",
["k(ʻ?)ui"] = "k%1uei"
}
if type(text) == 'table' then text = text.args[1] end
local function process_syllable(initial, final, tone)
for text, replace in pairs(py_wg_final) do
final = gsub(final, text, replace)
end
if (initial == "zh" or initial == "ch") and final == "i" then
final = "ih"
end
local untoned = (py_wg_initial[initial] or initial) .. final
for text, replace in pairs(py_wg_syl) do
untoned = gsub(untoned, text, replace)
end
return untoned .. '<sup>' .. tone .. '</sup>'
end
local function process_word(word)
local uppercase
if word:find('^%u') then uppercase = true else uppercase = false end
local text = gsub(export.py_transform(word, true), '[,%.]', '')
text = gsub(gsub(text, ' +', ' '), '[一不]', {['一'] = 'yi1', ['不'] = 'bu4'})
text = gsub(text, '([jqxy])u', '%1ü')
local p = split(text, " ", true)
for i = 1, #p do
p[i] = gsub(p[i], '^([bcdfghjklmnpqrstxz]?h?)(.+)([1-5])$', process_syllable)
end
text = table.concat(p, "-")
if uppercase then
text = gsub(text, '^%l', mw.ustring.upper)
end
return text
end
local words = split(text, " ")
for i, word in ipairs(words) do
words[i] = process_word(word)
end
return table.concat(words, " ")
end
local function temp_bg(text, bg)
if bg == 'y' then
return '<' .. text .. '>'
end
return text
end
local function make_bg(text, bg)
if bg == 'y' then
return '<span style="background-color:#F5DEB3">' .. text .. '</span>'
else
return text
end
end
function export.py_gwoyeu(text, original_text)
local initials = {
['b'] = 'b', ['p'] = 'p', ['m'] = 'm', ['f'] = 'f',
['d'] = 'd', ['t'] = 't', ['n'] = 'n', ['l'] = 'l',
['g'] = 'g', ['k'] = 'k', ['h'] = 'h',
['j'] = 'j', ['q'] = 'ch', ['x'] = 'sh',
['zh'] = 'j', ['ch'] = 'ch', ['sh'] = 'sh', ['r'] = 'r',
['z'] = 'tz', ['c'] = 'ts', ['s'] = 's',
['y'] = 'i', ['w'] = 'u',
[''] = ''
}
local finals = {
['a'] = 'a', ['ai'] = 'ai', ['ao'] = 'au', ['an'] = 'an', ['ang'] = 'ang', ['e'] = 'e', ['ei'] = 'ei', ['ou'] = 'ou', ['en'] = 'en', ['eng'] = 'eng', ['o'] = 'o',
['ia'] = 'ia', ['iao'] = 'iau', ['ian'] = 'ian', ['iang'] = 'iang', ['ie'] = 'ie', ['iu'] = 'iou', ['in'] = 'in', ['ing'] = 'ing', ['i'] = 'i',
['ua'] = 'ua', ['uai'] = 'uai', ['uan'] = 'uan', ['uang'] = 'uang', ['uo'] = 'uo', ['ui'] = 'uei', ['un'] = 'uen', ['ong'] = 'ong', ['u'] = 'u',
['ɨ'] = 'y', ['üan'] = 'iuan', ['üe'] = 'iue', ['ün'] = 'iun', ['iong'] = 'iong', ['ü'] = 'iu',
['io'] = 'io', ['ê'] = 'è',
--erhua
['ar'] = 'al', ['air'] = 'al', ['aor'] = 'aul', ['anr'] = 'al', ['angr'] = 'angl', ['er'] = "e'l", ['eir'] = 'el', ['our'] = 'oul', ['enr'] = 'el', ['engr'] = 'engl', ['or'] = 'ol',
['iar'] = 'ial', ['iaor'] = 'iaul', ['ianr'] = 'ial', ['iangr'] = 'iangl', ['ier'] = "ie'l", ['iur'] = 'ioul', ['inr'] = 'iel', ['ingr'] = 'iengl', ['ir'] = 'iel',
['uar'] = 'ual', ['uair'] = 'ual', ['uanr'] = 'ual', ['uangr'] = 'uangl', ['uor'] = 'uol', ['uir'] = 'uel', ['unr'] = 'uel', ['ongr'] = 'ongl', ['ur'] = 'ul',
['ɨr'] = 'el', ['üanr'] = 'iual', ['üer'] = "iue'l", ['ünr'] = 'iul', ['iongr'] = 'iongl', ['ür'] = 'iuel',
}
if type(text) == 'table' then text = text.args[1] end
if text:find('^%s') or text:find('%s$') then error('invalid spacing') end
local words = split(text, " ")
local count = 0
for i, word in ipairs(words) do
local uppercase
if word:find('^%u') then uppercase = true else uppercase = false end
word = export.py_transform(word, true, true)
word = gsub(word, "([1-5])", "%1 ")
word = gsub(word, " $", "")
word = gsub(word, '([!-/:-@%[-`{|}~!-/:-@[-`{-・])', ' %1 ')
word = gsub(word, ' +', ' ')
word = gsub(word, ' $', '')
word = gsub(word, '^ ', '')
local syllables = split(word, " ")
for j, syllable in ipairs(syllables) do
count = count + 1
if not find(syllable, '^[!-/:-@%[-`{|}~!-/:-@[-`{-・]+$') then
local current = sub(mw.title.getCurrentTitle().text, count, count)
if find(current, '^[一七八不]$') then
local exceptions = {['一'] = 'i', ['七'] = 'chi', ['八'] = 'ba', ['不'] = 'bu'}
syllables[j] = exceptions[current]
else
local initial, final, tone = '', '', ''
syllable = gsub(syllable, '([jqxy])u', '%1ü')
syllable = gsub(syllable, '^([zcsr]h?)i(r?[1-5])$', '%1ɨ%2')
if mw.ustring.find(syllable, '([bpmfdtnlgkhjqxzcsryw]?h?)([iuü]?[aoeiɨuüê][ioun]?g?r?)([1-5])') then
syllable = gsub(syllable, '([bpmfdtnlgkhjqxzcsryw]?h?)([iuü]?[aoeiɨuüê][ioun]?g?r?)([1-5])', function(a, b, c)
initial = initials[a] or error('Unrecognised initial:' .. a); final = finals[b] or error('Unrecognised final:' .. b); tone = c
return (initial .. final .. tone) end)
elseif not find(mw.title.getCurrentTitle().text, "[噷嗯哦呸哼唔呣姆嘸們欸誒ㄝM]") then
error('Unrecognised syllable:' .. syllable)
end
local original = initial..final..tone
if initial:find('^[iu]$') then
final = initial .. final
initial = ''
end
if initial .. final == "e'l" then
final = "el"
end
final = gsub(final, '([iu])%1', '%1')
local len = len(initial) + len(final)
local detone = initial..final
local replace = detone
local fullstop = false
if tone == 5 or tone == '5' then
fullstop = true
if original_text then
tone = split(export.py_transform(original_text, true), ' ')[count]:match('[1-5]')
elseif initial:find('^[lmnr]$') then
tone = 2
else tone = 1 end
if tone == 5 or tone == '5' then
tone = export.tone_determ(m_zh.py(current))
end
end
if tone == 1 or tone == '1' then
if initial:find('^[lmnr]$') then
replace = initial .. 'h' .. sub(detone, 2, len)
else
replace = detone
end
elseif tone == 2 or tone == '2' then
if not initial:find('^[lmnr]$') then
if final:find('^[iu]') then
replace = gsub(detone, '[iu]', {['i'] = 'y', ['u'] = 'w'}, 1)
replace = gsub(replace, '[yw]l?$', {['y'] = 'yi', ['w'] = 'wu', ['wl'] = 'wul',})
else
replace = gsub(detone, '([aiueoyè]+)', '%1r')
end
else
replace = detone
end
elseif tone == 3 or tone == '3' then
if final:find("^iu?e'l$") then
detone = detone:gsub("'", '')
end
detone = detone:gsub('^[iu]', {['i'] = 'yi', ['u'] = 'wu'})
if final:find('[aeiou][aeiou]') and (not final:find('^[ie][ie]') or initial..final=="ie") and (not final:find('^[uo][uo]') or initial..final=="uo") then
replace = detone:gsub('[iu]', {['i'] = 'e', ['u'] = 'o'}, 1)
elseif final:find('[aoeiuyè]') then
replace = detone:gsub('([iuyw]?)([aoeiuy\195]\168?)', '%1%2%2', 1) --\195\168 is the utf8 for è
else
error('Unrecognised final:'..final)
end
elseif tone == 4 or tone == '4' then
if final:find("^iu?e'l$") then
detone = detone:gsub("'", '')
end
detone = detone:gsub('^[iu]', {['i'] = 'yi', ['u'] = 'wu'})
if detone:find('[aeo][iu]l?$') then
replace = detone:gsub("[iu]l?$", {['i'] = 'y', ['u'] = 'w', ['ul'] = 'wl'})
elseif detone:find('[ngl]$') then
replace = detone:gsub("[ng'l]l?$", {['n'] = 'nn', ['g'] = 'q', ['l'] = 'll', ['gl'] = 'ql', ["'l"] = 'hl'})
else
replace = detone .. 'h'
end
replace = replace:gsub('yi([aeiou])', 'y%1')
replace = replace:gsub('wu([aeiou])', 'w%1')
end
if fullstop then replace = '.' .. replace end
syllables[j] = syllable:gsub(original, replace)
end
end
end
words[i] = table.concat(syllables, "")
if uppercase then
words[i] = gsub(words[i], '^%l', mw.ustring.upper)
end
end
return table.concat(words, " ")
end
-- Converts Hanyu Pinyin into Tongyong Pinyin.
function export.py_tongyong(text)
if type(text) == 'table' then text = text.args[1] end
local ty_tone = {
["1"] = "", ["2"] = "\204\129", ["3"] = "\204\140", ["4"] = "\204\128", ["5"] = "\204\138"
}
local function num_to_mark(syllable, tone)
tone = ty_tone[tone]
if tone ~= "" then
if syllable:find('[aeê]') then
syllable = syllable:gsub("([aeê])", "%1" .. tone)
elseif syllable:find('o') then
syllable = syllable:gsub("(o)", "%1" .. tone)
elseif syllable:find('[iu]') then
syllable = syllable:gsub("([iu])", "%1" .. tone)
end
end
return syllable
end
local words = {}
for word in gsplit(text, " ") do
local cap = word:find("^[A-Z]")
word = export.py_transform(word, true)
local syllables = {}
for syllable in gsplit(word, " ") do
syllable = syllable:gsub("([zcs]h?)i", "%1ih")
syllable = syllable:gsub("ü", "yu")
syllable = syllable:gsub("([jqx])u", "%1yu")
syllable = syllable:gsub("iu", "iou")
syllable = syllable:gsub("ui", "uei")
syllable = syllable:gsub("([wf])eng", "%1ong")
syllable = syllable:gsub("wen", "wun")
syllable = syllable:gsub("iong", "yong")
syllable = syllable:gsub("^zh", "jh")
syllable = syllable:gsub("^q", "c")
syllable = syllable:gsub("^x", "s")
syllable = #syllables ~= 0 and syllable:gsub("^([aeo])", "'%1") or syllable
syllable = syllable:gsub("^([^1-5]+)([1-5])$", num_to_mark)
table.insert(syllables, syllable)
end
word = table.concat(syllables, "")
word = cap and word:gsub("^.", string.upper) or word
table.insert(words, word)
end
return mw.ustring.toNFC(table.concat(words, " "))
end
function export.py_format(text, cap, bg, simple, nolink)
if cap == false then cap = nil end
if bg == false then bg = 'n' else bg = 'y' end
if simple == false then simple = nil end
if nolink == false then nolink = nil end
text = mw.ustring.toNFD(text)
local phon = text
local title = mw.title.getCurrentTitle().text
local cat = ''
local spaced = mw.ustring.toNFD(export.py_transform(text))
local space_count
spaced, space_count = gsub(spaced, ' ', '@')
local consec_third_count
for _ = 1, space_count do
spaced, consec_third_count = gsub(spaced, "([^@]+)̌([^#@]*)@([^#@]+̌)", function(a, b, c)
return temp_bg(a..'́'..b, bg)..'@'..c end, 1)
if consec_third_count > 0 then
phon = gsub(spaced, '@', '')
end
end
text = gsub(text, "#", "")
phon = gsub(phon, "#", "")
if title:find('一') and not text:find('一') and not simple then
cat = cat .. '[[Category:Mandarin words containing 一 not undergoing tone sandhi]]'
end
if text:find('[一不]') and not simple then
text = gsub(text, '[一不]$', {['一'] = 'yī', ['不'] = 'bù'})
phon = gsub(phon, '[一不]$', {['一'] = 'yī', ['不'] = 'bù'})
if find(text, '一') then
if find(text, '一[^̄́̌̀]*[̄́̌]') then
cat = cat .. '[[Category:Mandarin words containing 一 undergoing tone sandhi to the fourth tone]]'
phon = gsub(phon, '一([^̄́̌̀]*[̄́̌])', function(a) return temp_bg('yì', bg) .. a end)
text = gsub(text, '一([^̄́̌̀]*[̄́̌])', 'yī%1')
end
if find(text, '一[^̄́̌̀]*̀') or find(text, '一ge$') or find(text, '一ge[^nr]') then
cat = cat .. '[[Category:Mandarin words containing 一 undergoing tone sandhi to the second tone]]'
phon = gsub(phon, '一([^̄́̌̀]*̀)', function(a) return temp_bg('yí', bg) .. a end)
phon = gsub(phon, '一ge', temp_bg('yí', bg) .. 'ge')
text = gsub(text, '一([^̄́̌̀]*[̄́̌])', 'yī%1')
end
end
if find(text, '不 ?[bpmfdtnlgkhjqxzcsrwy]?h?[aeiou]*̀') then
cat = cat .. '[[Category:Mandarin words containing 不 undergoing tone sandhi|2]]'
phon = gsub(phon, '不( ?[bpmfdtnlgkhjqxzcsrwy]?h?[aeiou]*̀)', function(a) return temp_bg('bú', bg) .. a end)
end
end
text = gsub(text, '[一不]', {['一'] = 'yī', ['不'] = 'bù'})
text = gsub(text, '兒', function() return make_bg('r', bg) end) -- character is deleted
phon = gsub(phon, '<([^>]+)>', '<span style="background-color:#F5DEB3">%1</span>')
if not simple then
if cap then
text = gsub(text, '^%l', string.upper)
phon = gsub(phon, '^%l', string.upper)
end
if not nolink then
text = '[[' .. text .. ']]'
end
if '[[' .. gsub(phon, '[一不]', {['一'] = 'yī', ['不'] = 'bù'}) .. ']]' ~= text then
phon = gsub(phon, '[一不]', {['一'] = 'yī', ['不'] = 'bù'})
text = text .. ' [Phonetic: ' .. phon .. ']'
end
if mw.title.getCurrentTitle().nsText ~= 'টেমপ্লেট' and not nolink then
text = text .. cat
end
end
return mw.ustring.toNFC(text)
end
function export.make_tl(original_text, tl_pos, bg, cap)
if bg == false then bg = 'n' else bg = 'y' end
local _, countoriginal = gsub(original_text, " ", " ")
local spaced = export.py_transform(original_text)
if sub(spaced, -1, -1) == ' ' then spaced = sub(spaced, 1, -2) end
local _, count = gsub(spaced, " ", " ")
local index = {}
local start, finish
local pos = 1
for i = 1, count, 1 do
if i ~= 1 then pos = (index[i-1] + 1) end
index[i] = mw.ustring.find(spaced, ' ', pos)
end
if tl_pos == 2 then
start = index[count-1] - count + countoriginal + 2
finish = index[count] - count + countoriginal
elseif tl_pos == 3 then
start = index[count-2] - count + countoriginal + 3
finish = index[count-1] - count + countoriginal + 1
else
start = count == 0 and 1 or (index[count] - count + countoriginal + 1)
finish = -1
end
local text = (sub(original_text, 1, start-1) .. make_bg(gsub(sub(original_text, start, finish), '.', py_detone), bg))
if finish ~= -1 then text = (text .. sub(original_text, finish+1, -1)) end
if cap == true then text = gsub(text, '^%l', string.upper) end
return text
end
function export.tag(first, second, third, fourth, fifth)
local text = "(''"
local tag = {}
local tagg = first or "Standard Chinese"
tag[1] = (second ~= '') and second or "Standard Chinese"
tag[2] = (third ~= '') and third or nil
tag[3] = (fourth ~= '') and fourth or nil
tag[4] = (fifth ~= '') and fifth or nil
text = text .. ((tagg == '') and table.concat(tag, ", ") or tagg) .. "'')"
text = gsub(text, 'Standard Chinese', "[[w:Standard Chinese|Standard Chinese]]")
text = gsub(text, 'Mainland', "[[w:Putonghua|Mainland]]")
text = gsub(text, 'Taiwan', "[[w:Taiwanese Mandarin|Taiwan]]")
text = gsub(text, 'Beijing', "[[w:Beijing dialect|Beijing]]")
text = gsub(text, 'erhua', "[[w:erhua|erhua]]")
text = gsub(text, 'Min Nan', "[[w:Min Nan|Min Nan]]")
text = gsub(text, 'shangkouzi', "''[[上口字|shangkouzi]]''")
return text
end
function export.straitdiff(text, pron_ind, tag)
local conv_text = text
for i = 1, #text do
if m_zh_data().MT[sub(text, i, i)] then conv_text = 'y' end
end
if tag == 'tag' then
conv_text = (conv_text == 'y') and m_zh_data().MT_tag[match(text, '[丁-丌与-龯㐀-䶵]')][pron_ind] or ''
elseif pron_ind == 1 or pron_ind == 2 or pron_ind == 3 or pron_ind == 4 or pron_ind == 5 then
local reading = {}
for a, b in pairs(m_zh_data().MT) do
reading[a] = b[pron_ind]
if reading[a] then reading[a] = gsub(reading[a], "^([āōēáóéǎǒěàòèaoe])", "'%1") end
end
conv_text = gsub(text, '.', reading)
text = gsub(text, "^'", "")
text = gsub(text, " '", " ")
if conv_text == text and tag == 'exist' then return nil end
end
conv_text = gsub(conv_text, "^'", "")
return conv_text
end
function export.str_analysis(text, conv_type, other_m_vars)
if type(text) == 'table' then text, conv_type = text.args[1], (text.args[2] or "") end
local MT = m_zh_data().MT
text = gsub(text, '=', '—')
text = gsub(text, ',', '隔')
text = gsub(text, '隔 ', ', ')
if conv_type == 'head' or conv_type == 'link' then
if find(text, '隔cap—') then
text = gsub(text, '[一不]', {['一'] = 'Yī', ['不'] = 'Bù'})
end
text = gsub(text, '[一不]', {['一'] = 'yī', ['不'] = 'bù'})
end
local comp = split(text, '隔', true)
local reading = {}
local alternative_reading = {}
local zhuyin = {}
--[[
-- not used
local param = {
'1n', '1na', '1nb', '1nc', '1nd', 'py', 'cap', 'tl', 'tl2', 'tl3', 'a', 'audio', 'er', 'ertl', 'ertl2', 'ertl3', 'era', 'eraudio',
'2n', '2na', '2nb', '2nc', '2nd', '2py', '2cap', '2tl', '2tl2', '2tl3', '2a', '2audio', '2er', '2ertl', '2ertl2', '2ertl3', '2era', '2eraudio',
'3n', '3na', '3nb', '3nc', '3nd', '3py', '3cap', '3tl', '3tl2', '3tl3', '3a', '3audio', '3er', '3ertl', '3ertl2', '3ertl3', '3era', '3eraudio',
'4n', '4na', '4nb', '4nc', '4nd', '4py', '4cap', '4tl', '4tl2', '4tl3', '4a', '4audio', '4er', '4ertl', '4ertl2', '4ertl3', '4era', '4eraudio',
'5n', '5na', '5nb', '5nc', '5nd', '5py', '5cap', '5tl', '5tl2', '5tl3', '5a', '5audio', '5er', '5ertl', '5ertl2', '5ertl3', '5era', '5eraudio'
}
--]]
if conv_type == '' then
return comp[1]
elseif conv_type == 'head' or conv_type == 'link' then
for i, item in ipairs(comp) do
if not find(item, '—') then
if find(item, '[一-龯㐀-䶵]') then
local M, T, t = {}, {}, {}
for a, b in pairs(MT) do
M[a] = b[1]; T[a] = b[2]; t[a] = b[3];
M[a] = gsub(M[a], "^([āōēáóéǎǒěàòèaoe])", "'%1")
T[a] = gsub(T[a], "^([āōēáóéǎǒěàòèaoe])", "'%1")
if t[a] then t[a] = gsub(t[a], "^([āōēáóéǎǒěàòèaoe])", "'%1") end
end
local mandarin = gsub(item, '.', M)
local taiwan = gsub(item, '.', T)
mandarin = gsub(mandarin, "^'", "")
mandarin = gsub(mandarin, " '", " ")
if conv_type == 'link' then return mandarin end
taiwan = gsub(taiwan, "^'", "")
taiwan = gsub(taiwan, " '", " ")
local tt = gsub(item, '.', t)
if find(text, 'cap—') then
mandarin = gsub(mandarin, '^%l', mw.ustring.upper)
taiwan = gsub(taiwan, '^%l', mw.ustring.upper)
tt = gsub(tt, '^%l', mw.ustring.upper)
end
if tt == item then
zhuyin[i] = export.py_zhuyin(mandarin, true) .. ', ' .. export.py_zhuyin(taiwan, true)
reading[i] = mandarin .. ']], [[' .. taiwan
else
tt = gsub(tt, "^'", "")
tt = gsub(tt, " '", " ")
zhuyin[i] = export.py_zhuyin(mandarin, true) .. ', ' .. export.py_zhuyin(taiwan, true) .. ', ' .. export.py_zhuyin(tt, true)
reading[i] = mandarin .. ']], [[' .. taiwan .. ']], [[' .. tt
end
else
if conv_type == 'link' then return item end
zhuyin[i] = export.py_zhuyin(item, true)
reading[i] = item
if len(mw.title.getCurrentTitle().text) == 1 and #mw.text.split(export.py_transform(item), " ") == 1 then
alternative_reading[i] = "[[" .. export.py_transf(reading[i]) .. "|" .. mw.ustring.gsub(export.py_transf(reading[i]), '([1-5])', '<sup>%1</sup>') .. "]]"
end
end
if reading[i] ~= '' then reading[i] = '[[' .. reading[i] .. ']]' end
reading[i] = gsub(reading[i], "#", "")
end
comp[i] = item
if conv_type == 'link' then return comp[1] end
end
local id = m_zh.ts_determ(mw.title.getCurrentTitle().text)
local accel
if id == 'trad' then
accel = '<span class="form-of pinyin-t-form-of transliteration-' .. m_zh.ts(mw.title.getCurrentTitle().text)
elseif id == 'simp' then
accel = '<span class="form-of pinyin-s-form-of transliteration-' .. m_zh.st(mw.title.getCurrentTitle().text)
elseif id == 'both' then
accel = '<span class="form-of pinyin-ts-form-of'
end
accel = accel .. '" lang="cmn" style="font-family: Consolas, monospace;">'
local result = other_m_vars and "*: <small>(''[[w:Standard Chinese|Standard]]'')</small>\n*::" or "*:"
result = result .. "<small>(''[[w:Pinyin|Pinyin]]'')</small>: " .. accel .. gsub(table.concat(reading, ", "), ", ,", ",")
if alternative_reading[1] then
result = result .. " (" .. table.concat(alternative_reading, ", ") .. ")"
end
result = result .. (other_m_vars and "</span>\n*::" or "</span>\n*:")
result = result .. "<small>(''[[w:Zhuyin|Zhuyin]]'')</small>: " .. '<span lang="zh-Bopo" class="Bopo">' .. gsub(table.concat(zhuyin, ", "), ", ,", ",") .. "</span>"
return result
elseif conv_type == '2' or conv_type == '3' or conv_type == '4' or conv_type == '5' then
if not find(text, '隔') or (comp[tonumber(conv_type)] and find(comp[tonumber(conv_type)], '—')) then
return ''
else
return comp[tonumber(conv_type)]
end
else
for i = 1, #comp, 1 do
local target = '^' .. conv_type .. '—'
if find(comp[i], target) then
text = gsub(comp[i], target, '')
return text
end
end
text = ''
end
return text
end
function export.homophones(pinyin)
local text = ''
if mw.title.getCurrentTitle().nsText == '' then
local args = hom_data.list[pinyin]
text = '<div style="visibility:hidden; float:left"><sup><span style="color:#FFF">edit</span></sup></div>'
for i, term in ipairs(args) do
if i > 1 then
text = text .. "<br>"
end
if mw.title.new(term).exists and term ~= mw.title.getCurrentTitle().text then
local forms = { term }
local content = mw.title.new(term):getContent()
local template = match(content, "{{zh%-forms[^}]*}}")
if template then
local simp = match(template, "|s=([^|}])+")
if simp then
table.insert(forms, simp)
end
for tradVar in mw.ustring.gmatch(template, "|t[0-9]=([^|}])+") do
table.insert(forms, tradVar)
end
for simpVar in mw.ustring.gmatch(template, "|s[0-9]=([^|}])+") do
table.insert(forms, simpVar)
end
term = table.concat(forms, "/")
end
end
text = text .. mw.getCurrentFrame():expandTemplate{ title = "Template:zh-l", args = { term, tr = "-" } }
end
text = text .. '[[Category:Mandarin terms with homophones]]'
end
return text
end
local function erhua(word, erhua_pos, pagename)
local title = split(pagename, '')
local linked_title = ''
local syllables = split(export.py_transform(word), ' ')
local count = #syllables
erhua_pos = find(erhua_pos, '[1-9]') and split(erhua_pos, ';') or { count }
for _, pos in ipairs(erhua_pos) do
pos = tonumber(pos)
title[pos] = title[pos] .. '兒'
syllables[pos] = syllables[pos] .. 'r'
end
local title = table.concat(title)
if mw.title.new(title).exists then
linked_title = ' (' .. require("Module:zh/link").link(nil, nil, {title, tr='-'}) .. ')'
end
for i, syllable in pairs(syllables) do
if i ~= 1 and mw.ustring.toNFD(syllable):find('^[aeiou]') then
syllables[i] = "'" .. syllable
end
end
word = table.concat(syllables, '')
return (export.tag('', '', 'erhua-ed') .. linked_title), word
end
export.erhua = erhua
function export.make(frame)
local args = frame:getParent().args
return export.make_args(args)
end
function export.make_args(args)
local pagename = mw.title.getCurrentTitle().text
local text = {}
local reading = {args[1] or '', args[2] or '', args[3] or '', args[4] or '', args[5] or ''}
args["1nb"] = ine(args["1nb"])
if reading[1] ~= '' then
local title = export.tag((args["1n"] or ''), (args["1na"] or ''), (args["1nb"] or export.straitdiff(args[1], 1, 'tag')), (args["1nc"] or ''), (args["1nd"] or ''))
local pinyin = export.straitdiff(reading[1], 1, '')
table.insert(text, export.make_table(title, pinyin, (args["py"] or ''), (args["cap"] or ''), (args["tl"] or ''), (args["tl2"] or ''), (args["tl3"] or ''), (args["a"] or args["audio"] or '')))
if args["er"] and args["er"] ~= '' then
title, pinyin = erhua(pinyin, args["er"], pagename)
table.insert(text, export.make_table(title, pinyin, '', (args["cap"] or ''), (args["ertl"] or ''), (args["ertl2"] or ''), (args["ertl3"] or ''), (args["era"] or args["eraudio"] or ''), true))
end
end
if reading[2] ~= '' or export.straitdiff(reading[1], 2, 'exist') then
if args["2nb"] and args["2nb"] ~= '' then tagb = args["2nb"] else tagb = export.straitdiff(args[1], 2, 'tag') end
title = export.tag((args["2n"] or ''), (args["2na"] or ''), tagb, (args["2nc"] or ''), (args["2nd"] or ''))
pinyin = (reading[2] ~= '') and reading[2] or export.straitdiff(reading[1], 2, '')
table.insert(text, export.make_table(title, pinyin, (args["2py"] or ''), (args["2cap"] or ''), (args["2tl"] or ''), (args["2tl2"] or ''), (args["2tl3"] or ''), (args["2a"] or args["2audio"] or ''), true))
table.insert(text, '[[Category:Mandarin terms with multiple pronunciations|' .. (export.straitdiff(args[1], 1, '') or args[1]) .. ']]')
if args["2er"] and args["2er"] ~= '' then
title, pinyin = erhua(pinyin, args["2er"], pagename)
table.insert(text, export.make_table(title, pinyin, '', (args["2cap"] or ''), (args["2ertl"] or ''), (args["2ertl2"] or ''), (args["2ertl3"] or ''), (args["2era"] or args["2eraudio"] or ''), true))
end
if reading[3] ~= '' or export.straitdiff(reading[1], 3, 'exist') then
if args["3nb"] and args["3nb"] ~= '' then tagb = args["3nb"] else tagb = export.straitdiff(args[1], 3, 'tag') end
title = export.tag((args["3n"] or ''), (args["3na"] or ''), tagb, (args["3nc"] or ''), (args["3nd"] or ''))
if reading[3] ~= '' then pinyin = reading[3] else pinyin = export.straitdiff(reading[1], 3, '') end
table.insert(text, export.make_table(title, pinyin, (args["3py"] or ''), (args["3cap"] or ''), (args["3tl"] or ''), (args["3tl2"] or ''), (args["3tl3"] or ''), (args["3a"] or args["3audio"] or ''), true))
if args["3er"] and args["3er"] ~= '' then
title, pinyin = erhua(pinyin, args["3er"], pagename)
table.insert(text, export.make_table(title, pinyin, '', (args["3cap"] or ''), (args["3ertl"] or ''), (args["3ertl2"] or ''), (args["3ertl3"] or ''), (args["3era"] or args["3eraudio"] or ''), true))
end
if reading[4] ~= '' or export.straitdiff(reading[1], 4, 'exist') then
if args["4nb"] and args["4nb"] ~= '' then tagb = args["4nb"] else tagb = export.straitdiff(args[1], 4, 'tag') end
title = export.tag((args["4n"] or ''), (args["4na"] or ''), tagb, (args["4nc"] or ''), (args["4nd"] or ''))
if reading[4] ~= '' then pinyin = reading[4] else pinyin = export.straitdiff(reading[1], 4, '') end
table.insert(text, export.make_table(title, pinyin, (args["4py"] or ''), (args["4cap"] or ''), (args["4tl"] or ''), (args["4tl2"] or ''), (args["4tl3"] or ''), (args["4a"] or args["4audio"] or ''), true))
if args["4er"] and args["4er"] ~= '' then
title, pinyin = erhua(pinyin, args["4er"], pagename)
table.insert(text, export.make_table(title, pinyin, '', (args["4cap"] or ''), (args["4ertl"] or ''), (args["4ertl2"] or ''), (args["4ertl3"] or ''), (args["4era"] or args["4eraudio"] or ''), true))
end
if reading[5] ~= '' or export.straitdiff(reading[1], 5, 'exist') then
if args["5nb"] and args["5nb"] ~= '' then tagb = args["5nb"] else tagb = export.straitdiff(args[1], 5, 'tag') end
title = export.tag((args["5n"] or ''), (args["5na"] or ''), tagb, (args["5nc"] or ''), (args["5nd"] or ''))
if reading[5] ~= '' then pinyin = reading[5] else pinyin = export.straitdiff(reading[1], 5, '') end
table.insert(text, export.make_table(title, pinyin, (args["5py"] or ''), (args["5cap"] or ''), (args["5tl"] or ''), (args["5tl2"] or ''), (args["5tl3"] or ''), (args["5a"] or args["5audio"] or ''), true))
if args["5er"] and args["5er"] ~= '' then
title, pinyin = erhua(pinyin, args["5er"], pagename)
table.insert(text, export.make_table(title, pinyin, '', (args["5cap"] or ''), (args["5ertl"] or ''), (args["5ertl2"] or ''), (args["5ertl3"] or ''), (args["5era"] or args["5eraudio"] or ''), true))
end
end
end
end
end
if (args["tl"] or '') .. (args["tl2"] or '') .. (args["tl3"] or '') .. (args["2tl"] or '') .. (args["2tl2"] or '') .. (args["2tl3"] or '') ~= '' then
table.insert(text, '[[Category:Mandarin words containing toneless variants|' .. export.straitdiff(args[1], 1, '') .. ']]')
end
return table.concat(text)
end
function export.make_audio(args)
local text, reading, pinyin = {}, {}, ""
local audio = {
args["a"] or args["audio"] or '',
args["2a"] or args["2audio"] or '',
args["3a"] or args["3audio"] or '',
args["4a"] or args["4audio"] or '',
args["5a"] or args["5audio"] or '',
}
for i=1,5 do
reading[i] = args[i] or ''
if i == 1 then
pinyin = export.straitdiff(reading[1], 1, '')
else
pinyin = (reading ~= '') and reading[i] or export.straitdiff(reading[1], i, '')
end
pinyin = export.py_format(pinyin, false, false, true)
add_audio(text, audio[i], pinyin)
end
return table.concat(text)
end
function add_audio(text, audio, pinyin)
if audio and audio ~= "" then
if audio == "y" then audio = string.format('zh-%s.ogg', pinyin) end
table.insert(text, '\n*:: [[File:')
table.insert(text, audio)
table.insert(text, ']]')
table.insert(text, '[[Category:Mandarin terms with audio links]]')
end
end
function export.make_table(title, pinyin, py, cap, tl, tl2, tl3, a, novariety)
py = ine(py);cap = ine(cap);tl = ine(tl);tl2 = ine(tl2);tl3 = ine(tl3);a = ine(a);novariety = ine(novariety)
local text = {}
local pinyin_simple_fmt = export.py_format(pinyin, false, false, true)
local pinyin_simple_fmt_nolink = export.py_format(pinyin, false, false, true, true)
if not novariety then
table.insert(text, '* [[w:Mandarin Chinese|Mandarin]]')
else
table.insert(text, '<br>')
end
table.insert(text, '\n** <small>' .. title .. '</small>')
local hom_found
if hom_data.list[mw.ustring.lower(pinyin_simple_fmt)] then
hom_found = true
else
hom_found = false
table.insert(text, '<sup><small><abbr title="Add Mandarin homophones"><span class="plainlinks">[' .. tostring(mw.uri.fullUrl("Module:zh/data/cmn-hom",{["action"]="edit"})) .. ' +]</span></abbr></small></sup>')
end
table.insert(text, "\n*** <small>''[[w:Pinyin|Pinyin]]''</small>: ")
local id = m_zh.ts_determ(mw.title.getCurrentTitle().text)
if id == 'trad' then
table.insert(text, '<span class="form-of pinyin-t-form-of transliteration-')
table.insert(text, m_zh.ts(mw.title.getCurrentTitle().text))
elseif id == 'simp' then
table.insert(text, '<span class="form-of pinyin-s-form-of transliteration-')
table.insert(text, m_zh.st(mw.title.getCurrentTitle().text))
else -- both
table.insert(text, '<span class="form-of pinyin-ts-form-of')
end
table.insert(text, '" lang="cmn" style="font-family: Consolas, monospace;">')
if py then
table.insert(text, py)
else
if cap then
table.insert(text, export.py_format(pinyin, true, true))
else
table.insert(text, export.py_format(pinyin, false, true))
end
if tl or tl2 or tl3 then
table.insert(text, ' → ')
if tl then tl_pos = 1 elseif tl2 then tl_pos = 2 elseif tl3 then tl_pos = 3 end
if cap then
table.insert(text, export.make_tl(export.py_format(pinyin, true, false, true, true), tl_pos, true, true))
else
table.insert(text, export.make_tl(pinyin_simple_fmt_nolink, tl_pos, true))
end
end
if tl then table.insert(text, ' <small>(toneless final syllable variant)</small>')
elseif tl2 or tl3 then table.insert(text, ' <small>(toneless variant)</small>') end
end
table.insert(text, "</span>\n*** <small>''[[w:Zhuyin|Zhuyin]]''</small>: ")
table.insert(text, '<span lang="zh-Bopo" class="Bopo">')
table.insert(text, export.py_zhuyin(pinyin_simple_fmt, true))
if tl or tl2 or tl3 then
table.insert(text, ' → ')
table.insert(text, export.py_zhuyin(export.make_tl(pinyin_simple_fmt_nolink, tl_pos, false), true))
end
table.insert(text, '</span>')
if tl then table.insert(text, ' <small>(toneless final syllable variant)</small>')
elseif tl2 or tl3 then table.insert(text, ' <small>(toneless variant)</small>') end
table.insert(text, "\n*** <small>''[[w:Gwoyeu Romatzyh|Gwoyeu Romatzyh]]''</small>: <code>")
if tl or tl2 or tl3 then
table.insert(text, export.py_gwoyeu(export.make_tl(pinyin_simple_fmt_nolink, tl_pos, false), pinyin_simple_fmt))
else
table.insert(text, export.py_gwoyeu(pinyin_simple_fmt))
end
table.insert(text, '</code>')
table.insert(text, "\n*** <small>''[[w:Tongyong Pinyin|Tongyong Pinyin]]''</small>: <code>")
if tl or tl2 or tl3 then
table.insert(text, export.py_tongyong(export.make_tl(pinyin_simple_fmt_nolink, tl_pos, false), pinyin_simple_fmt))
else
table.insert(text, export.py_tongyong(pinyin_simple_fmt))
end
table.insert(text, '</code>')
table.insert(text, "\n*** <small>''[[w:Wade–Giles|Wade–Giles]]''</small>: <code>")
if tl or tl2 or tl3 then
table.insert(text, export.py_wg(export.make_tl(pinyin_simple_fmt_nolink, tl_pos, false), pinyin_simple_fmt))
else
table.insert(text, export.py_wg(pinyin_simple_fmt))
end
table.insert(text, '</code>')
table.insert(text, '\n*** <small>Sinological [[Wiktionary:International Phonetic Alphabet|IPA]] <sup>([[Appendix:Mandarin pronunciation|key]])</sup></small>: <span class="IPA">/')
table.insert(text, export.py_ipa(pinyin))
if tl or tl2 or tl3 then
table.insert(text, '/ → /')
table.insert(text, export.py_ipa(export.make_tl(pinyin_simple_fmt_nolink, tl_pos, false)))
end
table.insert(text, '/</span>')
-- if a then
-- if a == 'y' then a = 'zh-' .. pinyin_simple_fmt .. '.ogg' end
-- table.insert(text, '\n*** <div style="display:inline-block; position:relative; top:0.5em;">[[File:')
-- table.insert(text, a)
-- table.insert(text, ']]</div>[[Category:Mandarin terms with audio links]]')
-- end
if hom_found then
table.insert(text, "\n*** <small>Homophones</small>: " ..
'<table class="wikitable" style="width:15em;margin:0; position:left; text-align:center">' ..
'<tr><th class="mw-customtoggle-cmnhom" style="color:#3366bb">[Show/Hide]</th></tr>' ..
'<tr class="mw-collapsible mw-collapsed" id="mw-customcollapsible-cmnhom">' ..
'<td><sup><div style="float: right; clear: right;"><span class="plainlinks">[')
table.insert(text, tostring(mw.uri.fullUrl("Module:zh/data/cmn-hom", {["action"]="edit"})))
table.insert(text, ' edit]</span></div></sup>')
table.insert(text, export.homophones(mw.ustring.lower(pinyin_simple_fmt)))
table.insert(text, '</td></tr></table>')
end
return table.concat(text)
end
return export