モジュール:ja-pron

出典: フリー多機能辞典『ウィクショナリー日本語版(Wiktionary)』
ナビゲーションに移動 検索に移動

このモジュールについての説明文ページを モジュール:ja-pron/doc に作成できます

local export = {}
local gsub = mw.ustring.gsub
local match = mw.ustring.match
local sub = mw.ustring.sub
local len = mw.ustring.len

local lang = require("モジュール:languages").getByCode("ja")
local m_ja = require("モジュール:ja")
local m_accent = require("モジュール:accent qualifier")
-- also [[モジュール:qualifier]]

local PAGENAME = mw.title.getCurrentTitle().text

local function quote(text)
	return  "“" .. text .. "”"
end

local ref_template_name_data = {
	['DJR'] = '書誌・大辞林',
	['DJR1'] = '書誌・大辞林|1',
	['DJR4'] = '書誌・大辞林|4',
	['NHK'] = '書誌・NHKアクセント新版',
	['SMK2'] = 'R:Shinmeikai2',
	['SMK5'] = 'R:Shinmeikai5',
	['ZAJ'] = 'R:Zenkoku Akusento Jiten',
	['JEL'] = 'R:Kenkyusha JEL Pocket',
}

local function generate_ref_tag(ref_name)
	local ref_template_name = ref_template_name_data[ref_name]
	
	if not ref_template_name then
		-- [[Special:WhatLinksHere/Template:tracking/ja-pron/unrecognized ref]]
		require("モジュール:debug").track("ja-pron/unrecognized ref")
		return nil
	end

	local ref_tag = mw.getCurrentFrame():extensionTag{
		name = 'ref',
		args = { name = ref_name },
		content = '{{' .. ref_template_name .. '}}',
	}

	return ref_tag
end

local function add_acc_refs(text)
	local output = {}

	if mw.ustring.match(text, 'ref') then
		table.insert(output, mw.getCurrentFrame():preprocess(text))
	else
		for ref_name in mw.text.gsplit(text, '%s*,%s*') do
			table.insert(output, generate_ref_tag(ref_name))
		end
	end

	return table.concat(output)
end

function export.show(frame)
	local params = {
		[1] = {default = PAGENAME, list = true},
		
		["accent"] = {list = true},
		["accent=_loc"] = {list = true},
		["accent=_ref"] = {list = true},
		["accent=_note"] = {list = true},
		
		["acc"] = {alias_of = "accent", list = true},
		["acc=_loc"] = {alias_of = "accent_loc", list = true},	-- using "accent=_loc" (etc.) doesn't work
		["acc=_ref"] = {alias_of = "accent_ref", list = true},
		["acc=_note"] = {alias_of = "accent_note", list = true},
		
		["dev"] = {},
		["dev2"] = {},
		["devm"] = {},
		
		["noipa"] = {},
		
		["a"] = {alias_of = "audio"},
		["audio"] = {},

		["homonym"] = {}
	}
	
	local args = require("モジュール:parameters").process(frame:getParent().args, params)
	
	local au = args.audio
	local dev = args.dev or args.devm
	local dev2 = args.dev2
	local homonym = args.homonym
	local maxindex = table.getn(args[1])
	local result = ""
	local text
	
	-- Deals with the accents
	local a, al, ar, an = args.accent, args.accent_loc, args.accent_ref, args.accent_note
	for i, position in ipairs(a) do
		text = args[1][math.min(maxindex,i)]
		result = result .. "\n*"
		if not al[i] then
			al[i] = "[[w:東京式アクセント|東京式]]"
		end
		result = result .. m_accent.show({al[i]}) .. " "
		
		result = result .. export.accent(text, position, dev, dev2)
		if ar[i] then
			result = result .. add_acc_refs(ar[i])
		end
		result = result .. (an[i] and (" " .. an[i]) or "")
	end
	
	-- Deals with the IPA
	if not noipa then
		local m_IPA = require("モジュール:IPA")
		for i, text in ipairs(args[1]) do
			local sortkey = m_ja.jsort(text)
			result = result .. "\n*"
			result = result .. m_IPA.format_IPA_full(lang, {{ pron = "[" .. export.ipa(text, dev, dev2) .. "]" }}, nil, nil, sortkey)
		end
	end
	
	-- Deals with the audio
	if au then
		sortkey = m_ja.jsort(args[1][1])
		result = result .. '\n* <table class="audiotable" style="vertical-align: top; display:inline-block; list-style:none;line-height: 1em;"><tr><td class="unicode audiolink">音声</td><td class="audiofile">[[Image:' .. au .. '|noicon|175px]]</td><td class="audiometa" style="font-size: 80%;">([[:Image:' .. au ..'|file]])</td></tr></table>[[Category:日本語 音声リンクがある語句|' .. sortkey .. ']]'
	end

	local m_links = require("モジュール:links")
	if homonym and homonym ~= "" then
		homonym_links = {}
		if homonym == "y" then
			for i, text in ipairs(args[1]) do
				table.insert(homonym_links, m_links.language_link(text, nil, lang, nil, false))
			end
		else
			for text in mw.text.gsplit(homonym, ',') do
				m_links.getLinkPage(m_links.remove_links(i), lang)
				table.insert(homonym_links, m_links.language_link(text, nil, lang, nil, false))
			end
		end

		result = result .. "\n*"
		result = result .. "同音異義語: " .. table.concat(homonym_links, ", ") .. "を参照。"
	end
	
	return result
end

function export.ipa(text, dev, dev2)
	if type(text) == "table" then
		text, dev, dev2 = text.args[1], text.args["dev"], text.args["dev2"] end
	dev = dev or ""
	dev2 = dev2 or ""
	
	if dev2 ~= "" then error('Please remove parameter dev2 and change parameter dev to \"dev=' .. dev .. ',' .. dev2 .. '"') end
	
	local position_kana = {}
	local position_mora = {}
	
	for i=1,mw.ustring.len(text) do
		if sub(text,i,i) ~= ' ' then
			table.insert(position_kana, i)
			if not mw.ustring.match(sub(text,i,i), "[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ]") then
				table.insert(position_mora, i)
			end
		end
	end
	
	if dev ~= "" then
		for position in mw.text.gsplit(dev,",") do
			position = tonumber(position)
			if #position_mora == position then
				text = text .. "̥"
			else
				position = position_mora[tonumber(position)+1]
				text = sub(text, 1, position-1) .. "̥" .. sub(text, position, -1)
			end
			for i=position+1,table.getn(position_mora) do
				position_mora[i] = position_mora[i] + 1
			end
		end
	end
	
	text = m_ja.kana_to_romaji(text, { keep_period = true })
	
	text = gsub(text, "&#39;", "ʔ")
	
	text = gsub(text, "[ptkbjgzsdr][ptckbjgzsdr][hs]?", {
		["pp"] = "p̚p", ["tch"] = "t̚ch", ["kk"] = "k̚k", ["bb"] = "b̚b̥", ["jj"] = "d̚j",
		["dd"] = "d̚d̥", ["gg"] = "g̚g̊", ["zz"] = "d̚z", ["tt"] = "t̚t", ["tts"] = "t̚ts",
		["rr"] = "r̚r", ["ssh"] = "ɕː" })
	
	text = gsub(text, "ei", "ē")
	text = gsub(text, "[āēīōūfvjryz]", {
		["ā"] = "aː", ["ē"] = "eː", ["ī"] = "iː", ["ō"] = "oː", ["ū"] = "uː", 
		["f"] = "ɸ", ["v"] = "b", ["j"] = "d͡ʑ", ["r"] = "ɾ", ["y"] = "j", ["z"] = "d͡z" })
	
	text = gsub(text, "[sct][hs]", {
		["sh"] = "ɕ", 
		["ch"] = "t͡ɕ", 
		["ts"] = "t͡s" })
	
	text = gsub(text, "([aeiouː̥])d͡([zʑ])", "%1%2")
	text = gsub(text, "([pbtdkgnmɸszɾ][̥̊]?)i", "%1ʲi")
	text = gsub(text, "([pbtdkgnmɸszɾ][̥̊]?)j", "%1ʲ")
	text = gsub(text, "nʲ", "ɲ̟")
	
	text = gsub(text, "([^ ː])(ː?)n([^aeou])", "%1̃%2n%3")
	for i, args in pairs{
			{ "(ː?)n$", "̃%1ɴ" },
			{ "n( ?)([pbm])", "m%1%2" },
			{ "n( ?)(.͡[ɕʑ])", "ɲ̟%1%2" },
			{ "n( ?)ɲ̟", "ɲ̟%1ɲ̟" },
			{ "n( ?)([kg])(ʲ?)", "ŋ%1%3%2%3" },
			{ "n( ?)([ɸszɕhjw])", "ɰ̃%1%2" },
			{ "n'", "ɰ̃" },
			{ "n ([aeiou])", "ɰ̃ %1" },
		} do
		
		text = gsub(text, args[1], args[2])
	end
	
	text = gsub(text, "h[iju]", {
		["hi"] = "çi", ["hj"] = "ç", 
		["hu"] = "ɸu" })
	
	text = gsub(text, "h([çɸ])", "%1%1")
	text = gsub(text, "([snhçɸmɾjw])%1", "%1ː")
	text = gsub(text, "ːʲ", "ʲː")
	text = gsub(text, "̚(.[̥̊]?)ʲ", "̚ʲ%1ʲ")
	text = gsub(text, "[aeiouw]", {
		["a"] = "a̠", 
		["e"] = "e̞", 
		["o"] = "o̞", 
		["u"] = "ɯ̟ᵝ", 
		["w"] = "ɰᵝ"})
	
	text = gsub(text, "([szɕʑɲçʲ])ɯ̟", "%1ɨ")
	
	text = gsub(text, "ᵝ̥", "̥ᵝ")
	text = gsub(text, "ᵝ̃", "̃ᵝ")
	text = gsub(text, "̠[̥̃][̥̃]", "̥̃˗")
	text = gsub(text, "̞[̥̃][̥̃]", "̥̃˕")
	text = gsub(text, "̟[̥̃][̥̃]", "̥̃˖")
	text = gsub(text, "([̠̞̟])̥", "%1̊")
	text = gsub(text, "%.", "")
	text = gsub(text, "'", ".")
	text = gsub(text, "g", "ɡ")
	
	return text
end

function export.rise_and_fall(word, rftype)
	word = gsub(word, "([おこごそぞとどのほぼぽもよろぉょオコゴソゾトドノホボポモヨロォョ])([うウ])", "%1.%2")
	word = gsub(word, "([えけげせぜてでねへべぺめれゑぇエケゲセゼテデネヘベペメレェ])([いイ])", "%1.%2")
	word = m_ja.kana_to_romaji(word)

	if rftype == "rise" then
		word = gsub(word, ".", {
			["a"] = "á", ["e"] = "é", ["i"] = "í", ["o"] = "ó", ["u"] = "ú", 
			["ā"] = "áá", ["ē"] = "éé", ["ī"] = "íí", ["ō"] = "óó", ["ū"] = "úú" })
	
		word = gsub(gsub(word, "n([bcdfghjkmnprstvw%'z ])", "ń%1"), "n$", "ń")
		
	elseif rftype == "fall" then
		word = gsub(word, ".", {
			["a"] = "à", ["e"] = "è", ["i"] = "ì", ["o"] = "ò", ["u"] = "ù", 
			["ā"] = "àà", ["ē"] = "èè", ["ī"] = "ìì", ["ō"] = "òò", ["ū"] = "ùù" })
		
		word = gsub(gsub(word, "n([bcdfghjkmnprstvw%'z ])", "ǹ%1"), "n$", "ǹ")
		
	else
		return error("Type not recognised.")
	end
	
	return word
end

-- [[モジュール:ja-ojad]] and [[モジュール:ja-infl-demo]] rely on the output format of this function
function export.accent(text, class, dev, dev2)
	if(type(text)) == "table" then text, class, dev, dev2 = text.args[1], text.args[2], text.args["dev"], text.args["dev2"] end
	text = gsub(text, "([おこごそぞとどのほぼぽもよろぉょオコゴソゾトドノホボポモヨロォョ])[うウ]", "%1ー")
	text = gsub(text, "([えけげせぜてでねへべぺめれゑぇエケゲセゼテデネヘベペメレェ])[いイ]", "%1ー")
	text = gsub(text, "%.", "")
	if dev == "" then dev = false end
	if dev2 == "" then dev2 = false end
	
	local down_first = "<span style=\"border-top:1px solid black;position:relative;padding:1px;\">"
	local down_last = "<span style=\"position:absolute;top:0;bottom:67%;right:0%;border-right:1px solid black;\">&#8203;</span></span>"
	local high_first = "<span style=\"border-top:1px solid black\">"
	local start = "<span lang=\"ja\" class=\"Jpan\">"
	local romaji_start = " <span class=\"Latn\"><samp>["
	local romaji_last = "]</samp></span> "
	local last = "</span>"
	
	local position_kana = {}       --position of each kana (ぁ counted), text without space
	local position_mora = {}       --position of each mora (ぁ not counted), text without space
	local position_mora_space = {} --position of each mora (ぁ not counted), text with space
	
	for i=1,mw.ustring.len(text) do
		if not mw.ustring.match(sub(text,i,i), "[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ ]") then
			local extra = mw.ustring.len(mw.ustring.match(sub(text,i+1), "^[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ]*"))
			table.insert(position_mora_space, i+extra)
		end
	end
	local space_removed = mw.ustring.gsub(text," ","")
	for i=1,mw.ustring.len(space_removed) do
		table.insert(position_kana, i)
		if not mw.ustring.match(sub(space_removed,i,i), "[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ]") then
			local extra = mw.ustring.len(mw.ustring.match(sub(space_removed,i+1), "^[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ]*"))
			table.insert(position_mora, i+extra)
		end
	end
	
	if match(class, "^[h0]$") then
		acc_type, acc_number = "h", 0
		
	elseif match(class, "^[a1]$") then
		acc_type, acc_number = "a", 1
	
	elseif match(class, "^o$") then
		acc_type = "o"
	end
	
	if match(class, "^[0-9]+$") and not match(class,"^[01]$") then
		class = gsub(class, "[on]", "")
		acc_number = tonumber(class)
		
		morae_count = len(gsub(text, "[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ ]", ""))
			
		if morae_count == acc_number then
			acc_type = "o"
		elseif morae_count < acc_number then
			return error(("Mora count (%d) is smaller than position of downstep mora (%d).")
				:format(morae_count, acc_number))
		else
			acc_type = "n"
		end
	elseif not acc_number then
		acc_number = class
	end
	
	local start_index = 1
	while match(sub(text, start_index+1, start_index+1), "[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ]") do
		start_index = start_index + 1
	end
	
	local kanas = {}
	local single_mora
	for i=1,mw.ustring.len(text) do
		if not mw.ustring.match(sub(text,i,i), "[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ ]") then
			single_mora = gsub(sub(text, i, -1), "^(.[ァィゥェォャュョヮぁぃぅぇぉゃゅょゎ]*).*", "%1")
			table.insert(kanas, single_mora)
		end
	end
	
	local function kana_devoice(text)
		return '<span style="border:1px dotted gray; border-radius:50%;">' .. text .. "</span>"
	end
	
	if dev then
		for position in mw.text.gsplit(dev, ",") do
			position = tonumber(position)
			kanas[position] = kana_devoice(kanas[position])
		end
	end
	
	romaji_text = gsub(text, "([おこそとのほもよろをごぞどぼぽょぉオコソトノホモヨロヲゴゾドボポョォ])ー", "%1お")
	romaji_text = gsub(romaji_text, "([えけせてねへめれゑげぜでべぺぇエケセテネヘメレヱゲゼデベペェ])ー", "%1え")
	romaji_text = gsub(romaji_text, "([うくすつぬふむゆるぐずづぶぷゅうウクスツヌフムユルグズヅブプュゥゔヴ])ー", "%1う")
	romaji_text = gsub(romaji_text, "([いきしちにひみりゐぎじぢびぴぃイキシチニヒミリヰギジヂビピィ])ー", "%1い")
	romaji_text = gsub(romaji_text, "([あかさたなはまやらわんがざだばぱゃぁアカサタナハマヤラワンガザダバパャァ])ー", "%1あ")
	
	local romajis = mw.text.split(romaji_text, "")
	
	local function count_nspaces(text, index)
		local i, sample, nspaces = 0, "", 0
		while len(sample) < index do
			i = i + 1
			sample, nspaces = gsub(sub(text, 1, i), " ", "")
		end
		return nspaces
	end
	
	local function romaji_devoice(text)
		return text .. "̥"
	end
	
	if dev then
		for position in mw.text.gsplit(dev,",") do
			position = position_mora_space[tonumber(position)]
			romajis[position] = romaji_devoice(romajis[position])
		end
	end
	
	if acc_type == "n" then
		r_start_index = start_index + count_nspaces(romaji_text, start_index)
		local r_index = position_mora_space[acc_number]
		local k_index = acc_number
		
		r_parts = {
			[1] = table.concat(romajis, "", 1, r_start_index),
			[2] = table.concat(romajis, "", r_start_index + 1, r_index),
			[3] = table.concat(romajis, "", r_index + 1, #romajis)
		}
				
		k_parts = {
			[1] = table.concat(kanas, "", 1, 1),
			[2] = table.concat(kanas, "", 2, k_index),
			[3] = table.concat(kanas, "", k_index + 1, #kanas)
		}
		
		result = start .. 
			k_parts[1] .. 
			down_first .. 
			k_parts[2] .. 
			down_last .. 
			k_parts[3] .. 
			last .. 
			romaji_start .. 
			export.rise_and_fall(r_parts[1], "fall") .. 
			export.rise_and_fall(r_parts[2], "rise") .. 
			"ꜜ" .. 
			export.rise_and_fall(r_parts[3], "fall") .. 
			romaji_last .. 
			"([[中高型]] – [" .. acc_number .. "])"
		
	else
		r_start_index = start_index + count_nspaces(romaji_text, start_index)
		r_parts = {
			[1] = table.concat(romajis, "", 1, r_start_index),
			[2] = table.concat(romajis, "", r_start_index + 1, #romajis)
		}
		
		k_parts = {
			[1] = table.concat(kanas, "", 1, 1),
			[2] = table.concat(kanas, "", 2, #kanas)
		}
		
		if acc_type == "h" then
			result = start .. 
				k_parts[1] .. 
				high_first .. 
				k_parts[2] .. 
				last .. 
				last .. 
				romaji_start .. 
				export.rise_and_fall(r_parts[1], "fall") .. 
				export.rise_and_fall(r_parts[2], "rise") .. 
				romaji_last .. 
				"([[平板型]] – [" .. acc_number .. "])"
				
 		elseif acc_type == "a" then
			result = start .. 
				down_first .. 
				k_parts[1] .. 
				down_last .. 
				k_parts[2] .. 
				last .. 
				romaji_start .. 
				export.rise_and_fall(r_parts[1], "rise") .. 
				"ꜜ" .. 
				export.rise_and_fall(r_parts[2], "fall") .. 
				romaji_last .. 
				"([[頭高型]] – [" .. acc_number .. "])"
				
		elseif acc_type == "o" then
			result = start .. 
				k_parts[1] .. 
				down_first .. 
				k_parts[2] .. 
				down_last .. 
				last .. 
				romaji_start .. 
				export.rise_and_fall(r_parts[1], "fall") .. 
				export.rise_and_fall(r_parts[2], "rise") .. 
				"ꜜ" .. 
				romaji_last .. 
				"([[尾高型]] – [" .. acc_number .. "])"
				
		else
			return error("Accent type not recognised.")
		end
		
	end
	
	result = gsub(result, "(.)̥", "<del>%1</del>")
	
	return result
end

return export