Module:hi-IPA
Documentation for this module may be created at Module:hi-IPA/doc
local export = {}
local lang = require("Module:languages").getByCode("hi")
local sc = require("Module:scripts").getByCode("Deva")
local m_IPA = require("Module:IPA")
local m_a = require("Module:accent qualifier")
local gsub = mw.ustring.gsub
local gmatch = mw.ustring.gmatch
local find = mw.ustring.find
local correspondences = {
["ṅ"] = "ŋ", ["g"] = "ɡ",
["c"] = "t͡ʃ", ["j"] = "d͡ʒ",
["ṭ"] = "ʈ", ["ḍ"] = "ɖ", ["ṇ"] = "ɳ",
["t"] = "t̪", ["d"] = "d̪",
["y"] = "j", ["r"] = "ɾ", ["v"] = "ʋ", ["l"] = "l̪",
["ś"] = "ʃ", ["ṣ"] = "ʂ", ["ź"] = "ʒ", ["h"] = "ɦ",
["ṛ"] = "ɽ", ["ž"] = "ʒ", ["ḻ"] = "l̪", ["ġ"] = "ɣ", ["q"] = "q", ["x"] = "x", ["ṉ"] = "n̪", ["ṟ"] = "ɾ",
["a"] = "ə", ["ā"] = "ɑː", ["i"] = "ɪ",
["ī"] = "iː", ["o"] = "oː", ["e"] = "eː",
["u"] = "ʊ", ["ū"] = "uː", ["ŏ"] = "ɔ", ["ĕ"] = "æ",
["ẽ"] = "ẽː", ["ũ"] = "ʊ̃", ["õ"] = "õː", ["ã"] = "ə̃", ["ā̃"] = "ɑ̃ː", ["ĩ"] = "ɪ̃", ["ī̃"] = "ĩː",
["ॐ"] = "oːm", ["ḥ"] = "(ɦ)", ["'"] = "(ʔ)",
}
local perso_arabic = {
["x"] = "kh", ["ġ"] = "g", ["q"] = "k", ["ž"] = "z", ["z"] = "j", ["f"] = "ph", ["'"] = "",
}
local urdu = {
["ṣ"] = "ʃ", ["ṇ"] = "n",
}
local deccani = {
["q"] = "x",
}
local lengthen = {
["a"] = "ā", ["i"] = "ī", ["u"] = "ū",
}
local vowels = "aāiīuūoŏĕʊɪɔɔ̃ɛeæãā̃ẽĩī̃õũū̃ː"
local vowel = "[aāiīuūoŏĕʊɪɔɔ̃ɛeæãā̃ẽĩī̃õũū̃]ː?"
local weak_h = "([gjdḍbṛnm])h"
local aspirate = "([kctṭp])"
local syllabify_pattern = "([" .. vowels .. "]̃?)([^" .. vowels .. "%.%-]+)([" .. vowels .. "]̃?)"
local function find_consonants(text)
local current = ""
local cons = {}
for cc in mw.ustring.gcodepoint(text .. " ") do
local ch = mw.ustring.char(cc)
if find(current .. ch, "^[kgṅcjñṭḍṇtdnpbmyrlvśṣshqxġzžḻṛṟfθṉḥ]$") or find(current .. ch, "^[kgcjṭḍṇtdpbṛ]h$") then
current = current .. ch
else
table.insert(cons, current)
current = ch
end
end
return cons
end
local function syllabify(text)
for count = 1, 2 do
text = gsub(text, syllabify_pattern, function(a, b, c)
b_set = find_consonants(b)
table.insert(b_set, #b_set > 1 and 2 or 1, ".")
return a .. table.concat(b_set) .. c
end)
text = gsub(text, "(" .. vowel .. ")(?=" .. vowel .. ")", "%1.")
end
for count = 1, 2 do
text = gsub(text, "(" .. vowel .. ")(" .. vowel .. ")", "%1.%2")
end
return text
end
local identical = "knlsfzθ"
for character in gmatch(identical, ".") do
correspondences[character] = character
end
local function transliterate(text)
return (lang:transliterate(text))
end
function export.link(term)
return require("Module:links").full_link{ term = term, lang = lang, sc = sc }
end
function export.narrow_IPA(ipa)
-- what /ɑ/ really is
ipa = gsub(ipa, 'ɑ', 'ä')
-- dentals
ipa = gsub(ipa, '([snl])', '%1̪')
-- nasals induce nasalization
ipa = gsub(ipa, '([əäɪiʊueɛoɔæ])(ː?)([nɳŋm])', '%1̃%2%3')
-- cc, jj
ipa = gsub(ipa, 't͡ʃ(%.?)t͡ʃ', 't̚%1t͡ʃ')
ipa = gsub(ipa, 'd͡ʒ(%.?)d͡ʒ', 'd̚%1d͡ʒ')
-- syllable boundary consonants
ipa = gsub(ipa, '([kg])%.([kg])', '%1̚.%2')
ipa = gsub(ipa, '([ʈɖ])%.([ʈɖ])', '%1̚.%2')
ipa = gsub(ipa, '([td]̪?)%.([tdn])', '%1̚.%2')
ipa = gsub(ipa, '([pb])%.([pb])', '%1̚.%2')
-- aspiration rules
ipa = gsub(ipa, 'əɦ%.([kgŋtdɲʈɖɳnpbmɾlzqfʂʃsʒɭɣɹʋj])', 'ɛɦ.%1')
ipa = gsub(ipa, 'ʊɦ%.([kgŋtdɲʈɖɳnpbmɾlzqfʂʃsʒɭɣɹʋj])', 'ɔɦ.%1')
ipa = gsub(ipa, 'ə%.ɦə', 'ɛ.ɦɛ')
ipa = gsub(ipa, 'ʊ%.ɦə', 'ɔ.ɦɔ')
ipa = gsub(ipa, 'ə%.ɦʊ', 'ɔ.ɦɔ')
-- v/w
ipa = gsub(ipa, '([kgŋtdɲʈɖɳnpbm]̪?%.?)ʋ', '%1w')
-- retroflex s rules
ipa = gsub(ipa, 'ʂ(%.?[^ʈɖ])', 'ʃ%1')
ipa = gsub(ipa, '([ŋn])%.([q])', 'ɴ.%2')
ipa = gsub(ipa, 'ʂ$', 'ʃ')
ipa = gsub(ipa, "ɪ%.j", "i.j")
return ipa
end
function export.toIPA(text, style)
text = gsub(text, '॰', '-')
local translit = text
if lang:findBestScript(text):isTransliterated() then
translit = transliterate(text)
end
if not translit then
error('The term "' .. text .. '" could not be transliterated.')
end
if style == "nonpersianized" then
translit = gsub(translit, "[xġqžzf']", perso_arabic)
end
if style == "dakhini" then
translit = gsub(translit, "[q]", deccani)
end
-- force final schwa for Hindi
translit = gsub(translit, "a~$", "ə")
if style == "desanskritanize" then
translit = gsub(translit, "(...)ə$", "%1ɑ(ː)")
translit = gsub(translit, "[ṣṇ]", urdu)
end
-- vowels
translit = gsub(translit, "͠", "̃")
translit = gsub(translit, 'a(̃?)i', 'ɛ%1ː')
translit = gsub(translit, 'a(̃?)u', 'ɔ%1ː')
translit = gsub(translit, "%-$", "")
translit = gsub(translit, "^%-", "")
translit = gsub(translit, "ŕ$", "r")
translit = gsub(translit, "ŕ(" .. vowel .. ")", "r%1")
translit = gsub(translit, "ŕ", "ri")
translit = gsub(translit, ",", "")
translit = gsub(translit, " ", "..")
translit = syllabify(translit)
translit = gsub(translit, "%.ː", "ː.")
translit = gsub(translit, "%.̃", "̃")
-- gy
translit = gsub(translit, 'jñ', 'gy')
translit = gsub(translit, aspirate .. "h", '%1ʰ')
translit = gsub(translit, weak_h, '%1ʱ')
local result = gsub(translit, ".", correspondences)
-- remove final schwa (Pandey, 2014)
-- actually weaken
result = gsub(result, "(...)ə$", "%1ᵊ")
result = gsub(result, "(...)ə ", "%1ᵊ ")
result = gsub(result, "(...)ə%.?%-", "%1ᵊ-")
result = gsub(result, "%.?%-", ".")
result = gsub(result, "%.%.", "‿")
-- formatting
result = gsub(result, "ː̃", "̃ː")
result = gsub(result, "ː%.̃", "̃ː.")
result = gsub(result, "%.$", "")
-- ñ
result = gsub(result, "ñ", "n")
-- i and u lengthening
result = gsub(result, "ʊ(̃?)(ʱ?)$", "u%1ː%2")
result = gsub(result, "ɪ(̃?)(ʱ?)$", "i%1ː%2")
result = gsub(result, "ɛː(%.?)j", function(a)
local res = "ə̯i"
res = res .. a .. "j"
return res
end)
result = gsub(result, "ɔː(%.?)ʋ", function(a)
local res = "ə̯u"
res = res .. a .. "ʋ"
return res
end)
return result
end
function export.make(frame)
local args = frame:getParent().args
local pagetitle = mw.title.getCurrentTitle().text
local p, results = {}, {}, {}
if args[1] then
for index, item in ipairs(args) do
table.insert(p, (item ~= "") and item or nil)
end
else
p = { pagetitle }
end
for _, Hindi in ipairs(p) do
local persianized = export.toIPA(Hindi, "persianized")
local nonpersianized = export.toIPA(Hindi, "nonpersianized")
table.insert(results, { pron = "/" .. persianized .. "/" })
local narrow = export.narrow_IPA(persianized)
if narrow ~= persianized then table.insert(results, { pron = "[" .. narrow .. "]" }) end
if persianized ~= nonpersianized then
table.insert(results, { pron = "/" .. nonpersianized .. "/" })
local narrow = export.narrow_IPA(nonpersianized)
if narrow ~= nonpersianized then table.insert(results, { pron = "[" .. narrow .. "]" }) end
end
end
return m_a.show({'Delhi'}) .. ' ' .. m_IPA.format_IPA_full(lang, results)
end
function export.make_ur(frame)
local args = frame:getParent().args
local pagetitle = mw.title.getCurrentTitle().text
local lang = require("Module:languages").getByCode("ur")
local sc = require("Module:scripts").getByCode("ur-Arab")
local p, results = {}, {}, {}
if args[1] then
for index, item in ipairs(args) do
table.insert(p, (item ~= "") and item or nil)
end
else
error("No transliterations given.")
end
for _, Urdu in ipairs(p) do
local desanskritanize = export.toIPA(Urdu, "desanskritanize")
table.insert(results, { pron = "/" .. desanskritanize .. "/" })
end
return m_a.show({'urd'}) .. ' ' .. m_IPA.format_IPA_full(lang, results)
end
function export.make_deccani(frame)
local args = frame:getParent().args
local pagetitle = mw.title.getCurrentTitle().text
local lang = require("Module:languages").getByCode("ur")
local sc = require("Module:scripts").getByCode("ur-Arab")
local p, results = {}, {}, {}
if args[1] then
for index, item in ipairs(args) do
table.insert(p, (item ~= "") and item or nil)
end
else
error("No transliterations given.")
end
for _, Urdu in ipairs(p) do
local dakhini = export.toIPA(Urdu, "dakhini")
table.insert(results, { pron = "/" .. dakhini .. "/" })
end
return m_a.show({'Deccani'}) .. ' ' .. m_IPA.format_IPA_full(lang, results)
end
return export