Module:grc-pronunciation/data

local U = require("Module:string utilities").char local gmatch = mw.ustring.gmatch local m_utils_data = require("Module:grc-utilities/data") local diacritics = m_utils_data.diacritics

local nonsyllabic = U(0x32F) local high = U(0x341) -- combining acute tone mark local low = U(0x340) -- combining grave tone mark local midHigh = U(0x1DC4) -- mid–high pitch local midLow = U(0x1DC6) -- mid–low pitch local highMid = U(0x1DC7) -- high–mid pitch local rising = U(0x30C)		-- combining caron local falling = diacritics.Latin_circum	-- combining circumflex local voiceless = U(0x325) -- combining ring below local aspirated = 'ʰ' local stress_mark = 'ˈ' local long = 'ː' local macron = diacritics.spacing_macron local breve = diacritics.spacing_breve

local circumflex_on_long_vowel = falling local acute_on_long_vowel = rising local acute_on_short_vowel = high local grave_pitch_mark = low

local data = {}

local function get_pitch_marks(accent_type, long) if accent_type == 'acute' then if long then return acute_on_long_vowel else return acute_on_short_vowel end elseif accent_type == 'grave' then return grave_pitch_mark elseif accent_type == 'circum' then return circumflex_on_long_vowel end return '' end

local function alpha(breathing, accent, iota, isLong) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, isLong) local length = (isLong or iota or accent == 'circum') and long or '' local offglide = iota and ('i' .. nonsyllabic) or '' return { ['cla'] = breathing[1] .. 'a' .. pitch .. length .. offglide, ['koi1'] = breathing[2] .. stress .. 'a', ['koi2'] = stress .. 'a', ['byz1'] = stress .. 'a', ['byz2'] = stress .. 'a'		} end

local function iota(breathing, accent, isLong) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, isLong) local length = (isLong or accent == 'circum') and long or '' return { ['cla'] = breathing[1] .. 'i' .. pitch .. length, ['koi1'] = breathing[2] .. stress .. 'i', ['koi2'] = stress .. 'i', ['byz1'] = stress .. 'i', ['byz2'] = stress .. 'i'		} end

local function ypsilon(breathing, accent, isLong) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, isLong) local length = (isLong or accent == 'circum') and long or '' return { ['cla'] = breathing[1] .. 'y' .. pitch .. length, ['koi1'] = breathing[2] .. stress .. 'y', ['koi2'] = stress .. 'y', ['byz1'] = stress .. 'y', ['byz2'] = stress .. 'i'		} end

local function omicron(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, false) return { ['cla'] = breathing[1] .. 'o' .. pitch, ['koi1'] = breathing[2] .. stress .. 'o', ['koi2'] = stress .. 'o', ['byz1'] = stress .. 'o', ['byz2'] = stress .. 'o'		} end

local function epsilon(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, false) return { ['cla'] = breathing[1] .. 'e' .. pitch, ['koi1'] = breathing[2] .. stress .. 'e', ['koi2'] = stress .. 'e', ['byz1'] = stress .. 'e', ['byz2'] = stress .. 'e'		} end

local function eta(breathing, accent, iota) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) local offglide = iota and ('i' .. nonsyllabic) or '' return { ['cla'] = breathing[1] .. 'ɛ' .. pitch .. long .. offglide, ['koi1'] = breathing[2] .. stress .. 'e̝', ['koi2'] = stress .. 'i', ['byz1'] = stress .. 'i', ['byz2'] = stress .. 'i'		} end

local function omega(breathing, accent, iota) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) local offglide = iota and ('i' .. nonsyllabic) or '' return { ['cla'] = breathing[1] .. 'ɔ' .. pitch .. long .. offglide, ['koi1'] = breathing[2] .. stress .. 'o', ['koi2'] = stress .. 'o', ['byz1'] = stress .. 'o', ['byz2'] = stress .. 'o'		} end

local function ai(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'a' .. pitch .. 'i' .. nonsyllabic, ['koi1'] = breathing[2] .. stress .. 'ɛ', ['koi2'] = stress .. 'ɛ', ['byz1'] = stress .. 'e', ['byz2'] = stress .. 'e'		} end

local function ei(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'e' .. pitch .. long, ['koi1'] = breathing[2] .. stress .. 'i', ['koi2'] = stress .. 'i', ['byz1'] = stress .. 'i', ['byz2'] = stress .. 'i'		} end

local function oi(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'o' .. pitch .. 'i' .. nonsyllabic, ['koi1'] = breathing[2] .. stress .. 'y', ['koi2'] = stress .. 'y', ['byz1'] = stress .. 'y', ['byz2'] = stress .. 'i'		} end

local function ui(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'y' .. pitch .. long, ['koi1'] = breathing[2] .. stress .. 'y', ['koi2'] = stress .. 'y', ['byz1'] = stress .. 'y', ['byz2'] = stress .. 'i'		} end

local function au(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'a' .. pitch .. 'u' .. nonsyllabic, ['koi1'] = { {'2=σ+3=μ', stress .. 'a' .. 'w'}, {'2.unvoiced', breathing[2] .. stress .. 'a' .. 'ʍ'}, breathing[2] .. stress .. 'a' .. 'w', },		['koi2'] = { {'2=σ+3=μ', stress .. 'aβ', }, { '2.unvoiced', stress .. 'aɸ', }, stress .. 'aβ', },		['byz1'] = { {'2=σ+3=μ', stress .. 'av', }, { '2.unvoiced', stress .. 'af', }, stress .. 'av', },		['byz2'] = { {'2=σ+3=μ', stress .. 'av', }, { '2.unvoiced', stress .. 'af', }, stress .. 'av', },	} end

local function eu(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'e' .. pitch .. 'u' .. nonsyllabic, ['koi1'] = { {'2=σ+3=μ', stress .. 'e' .. 'w'}, {'2.unvoiced+3=μ', breathing[2] .. stress .. 'e' .. 'ʍ'}, breathing[2] .. stress .. 'e' .. 'w', },		['koi2'] = { {'2=σ+3=μ', stress .. 'eβ', }, { '2.unvoiced', stress .. 'eɸ', }, stress .. 'eβ', },		['byz1'] = { {'2=σ+3=μ', stress .. 'ev', }, { '2.unvoiced', stress .. 'ef', }, stress .. 'ev', },		['byz2'] = { {'2=σ+3=μ', stress .. 'ev', }, { '2.unvoiced', stress .. 'ef', }, stress .. 'ev', },	} end

local function hu(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'ɛ' .. pitch .. 'ːu' .. nonsyllabic, ['koi1'] = { {'2=σ+3=μ', stress .. 'e̝' .. 'w'}, {'2.unvoiced', breathing[2] .. stress .. 'e̝' .. 'ʍ'}, breathing[2] .. stress .. 'e̝' .. 'w', },		['koi2'] = { {'2=σ+3=μ', stress .. 'iβ', }, { '2.unvoiced', stress .. 'iɸ', }, stress .. 'iβ', },		['byz1'] = { {'2=σ+3=μ', stress .. 'iv', }, { '2.unvoiced', stress .. 'if', }, stress .. 'iv', },		['byz2'] = { {'2=σ+3=μ', stress .. 'iv', }, { '2.unvoiced', stress .. 'if', }, stress .. 'iv', },	} end

local function ou(breathing, accent) local breathing = breathing == 'rough' and { 'h', '(h)' } or { ,  } local stress = accent and stress_mark or '' local pitch = get_pitch_marks(accent, true) return { ['cla'] = breathing[1] .. 'u' .. pitch .. long, ['koi1'] = breathing[2] .. stress .. 'u', ['koi2'] = stress .. 'u', ['byz1'] = stress .. 'u', ['byz2'] = stress .. 'u'		} end

data[' '] = { ['p'] = { ['cla'] = ' ', ['koi1'] = ' ', ['koi2'] = ' ', ['byz1'] = ' ', ['byz2'] = ' ', }, }

data['β'] = { ['clusters'] = { ['δ'] = true, ['λ'] = true, ['ρ'] = true, },	['p'] = { ['cla'] = 'b', ['koi1'] = 'b', ['koi2'] = { { '-1=μ', 'b', }, 'β' },		['byz1'] = { { '-1=μ', 'b', }, 'v'		}, ['byz2'] = { { '1=β', '', }, { '-1=μ', 'b', }, 'v', },	}, }

data['γ'] = { ['clusters'] = { ['λ'] = true, ['ν'] = true, ['ρ'] = true, },	['p'] = { ['cla'] = { { '1.dorsal/1=μ', 'ŋ', }, 'ɡ', },		['koi1'] = { { '1.dorsal', 'ŋ', }, 'ɡ', },		['koi2'] = { { '1.dorsal', {					{ '1~preFront', 'ɲ', }, 'ŋ', },			},			{ '0~preFront', {					{ '-1=γ', 'ɟ', }, 'ʝ', },			},			{ '-1=γ', 'ɡ', }, 'ɣ',		}, ['byz1'] = { { '1.dorsal', {					{ '1~preFront', 'ɲ', }, 'ŋ', },			},			{ '0~preFront', {					{ '-1=γ', 'ɟ', }, 'ʝ', },			},			{ '-1=γ', 'ɡ', }, 'ɣ', },		['byz2'] = { { '1.dorsal', { { '1~preFront', 'ɲ', }, 'ŋ', }, },			{ '0~preFront', {					{ '-1=γ', 'ɟ', }, 'ʝ', },			},			{ '-1=γ', 'ɡ', }, 'ɣ', },	}, }

data['δ'] = { ['clusters'] = { ['ρ'] = true, },	['p'] = { ['cla'] = 'd', ['koi1'] = 'd', ['koi2'] = { { '-1=ν', 'd', }, 'ð', },		['byz1'] = { { '-1=ν', 'd', }, 'ð', },		['byz2'] = { { '1=δ', '', }, { '-1=ν', 'd', }, 'ð', },	}, }

data['ζ'] = { ['clusters'] = { }, ['p'] = { ['cla'] = 'zd', ['koi1'] = 'z', ['koi2'] = 'z', ['byz1'] = 'z', ['byz2'] = { { '1=ζ', '', }, 'z', },	}, }

data['θ'] = { ['clusters'] = { ['ρ'] = true, },	['p'] = { ['cla'] = 'tʰ', ['koi1'] = 'tʰ', ['koi2'] = 'θ', ['byz1'] = 'θ', ['byz2'] = { { '1=θ', '', }, 'θ', },	}, }

data['κ'] = { ['clusters'] = { ['λ'] = true, ['ν'] = true, ['τ'] = true, ['ρ'] = true, },	['p'] = { ['cla'] = { { '1.voiced+1.stop', 'ɡ', }, { '1.aspirated', 'kʰ', }, 'k', },		['koi1'] = { { '1.voiced+1.stop', 'ɡ', }, 'k', },		['koi2'] = { { '1=κ', '', }, { '1.voiced+1.stop', 'ɡ', }, {				'-1=γ', {					{ '0~preFront', 'ɟ' }, 'ɡ', },			},			{ '0~preFront', 'c', }, 'k', },		['byz1'] = { { '1=κ', '', }, { '1.voiced+1.stop', 'ɡ', }, {				'-1=γ', {					{ '0~preFront', 'ɟ' }, 'ɡ', },			},			{ '0~preFront', 'c', }, 'k', },		['byz2'] = { { '1=κ', '', }, { '1.voiced+1.stop', 'ɡ', }, -- In what case does a voiced stop precede κ? {				'-1=γ', {					{ '0~preFront', 'ɟ' }, 'ɡ', },			},			{ '0~preFront', 'c', }, 'k', },	}, }

data['λ'] = { ['clusters'] = { }, ['p'] = { ['cla'] = 'l', ['koi1'] = 'l', ['koi2'] = 'l', ['byz1'] = 'l', ['byz2'] = { { '1=λ', '', }, 'l', },	}, }

data['μ'] = { ['clusters'] = { ['ν'] = true, },	['p'] = { ['cla'] = 'm', ['koi1'] = 'm', ['koi2'] = 'm', ['byz1'] = 'm', ['byz2'] = { { '1=μ', '', }, 'm', },	}, }

data['ν'] = { ['clusters'] = { }, ['p'] = { ['cla'] = 'n', ['koi1'] = 'n', ['koi2'] = 'n', ['byz1'] = 'n', ['byz2'] = { { '1=ν', '', }, 'n', },	}, }

data['ξ'] = { ['clusters'] = { }, ['p'] = { ['cla'] = 'ks', ['koi1'] = 'ks', ['koi2'] = 'ks', ['byz1'] = 'ks', ['byz2'] = 'ks', }, }

data['π'] = { ['clusters'] = { ['λ'] = true, ['ν'] = true, ['ρ'] = true, ['τ'] = true, },	['p'] = { ['cla'] = { { '1.aspirated', 'pʰ', }, 'p', },		['koi1'] = 'p', ['koi2'] = 'p', ['byz1'] = 'p', ['byz2'] = { { '-1=μ', 'b' }, { '1=π', '', }, 'p', },	}, }

data['ρ'] = { ['clusters'] = { }, ['p'] = { ['cla'] = { { '1=ρ/1=ῥ/-1=ρ', 'r̥', }, 'r', },		['koi1'] = 'r', ['koi2'] = 'r', ['byz1'] = 'r', ['byz2'] = { { '1=ρ', '', }, 'r', },	}, }

data['ῥ'] = { ['clusters'] = { }, ['p'] = { ['cla'] = 'r̥', ['koi1'] = 'r', ['koi2'] = 'r', ['byz1'] = 'r', ['byz2'] = { { '1=ρ', '', }, 'r', },	}, }

data['σ'] = { ['clusters'] = { ['β'] = true, ['θ'] = true, ['κ'] = true, ['μ'] = true, ['π'] = true, ['τ'] = true, ['φ'] = true, ['χ'] = true, },	['p'] = { ['cla'] = { { '1.voiced', 'z', }, 's', },		['koi1'] = { { '1.voiced', 'z', }, 's', },		['koi2'] = { { '1.voiced', 'z', }, 's', },		['byz1'] = { { '1.voiced', 'z', }, 's', },		['byz2'] = { { '1=σ', '', }, { '1.voiced', 'z', }, 's', },	}, }

data['τ'] = { ['clusters'] = { ['λ'] = true, ['μ'] = true, ['ρ'] = true, },	['p'] = { ['cla'] = { { '1.aspirated', 'tʰ', }, 't', },		['koi1'] = 't', ['koi2'] = 't', ['byz1'] = 't', ['byz2'] = { { '-1=ν', 'd' }, { '1=τ', '', }, 't', },	}, }

data['φ'] = { ['clusters'] = { ['θ'] = true, ['λ'] = true, ['ρ'] = true, },	['p'] = { ['cla'] = 'pʰ', ['koi1'] = 'pʰ', ['koi2'] = 'ɸ', ['byz1'] = 'f', ['byz2'] = { { '1=φ', '', }, 'f', },	}, }

data['χ'] = { ['clusters'] = { ['θ'] = true, ['λ'] = true, ['ρ'] = true, },	['p'] = { ['cla'] = 'kʰ', ['koi1'] = 'kʰ', ['koi2'] = { { '1=χ', '', }, { '0~preFront', 'ç', }, 'x', }, ['byz1'] = { { '1=χ', '', }, { '0~preFront', 'ç', }, 'x', }, ['byz2'] = { { '1=χ', '', }, { '0~preFront', 'ç', }, 'x', },	}, }

data['ψ'] = { ['clusters'] = { }, ['p'] = { ['cla'] = 'ps', ['koi1'] = 'ps', ['koi2'] = 'ps', ['byz1'] = 'ps', ['byz2'] = 'ps', }, }

data['ϝ'] = { ['clusters'] = { }, ['p'] = { ['cla'] = 'w', ['koi1'] = '', ['koi2'] = '', ['byz1'] = '', ['byz2'] = '', }, }

data['α'] = { ['pre'] = { { '0~isIDiphth/0~isUDiphth/0~hasMacronBreve', 1}, 0,	},	['p'] = { ['cla'] = 'a', ['koi1'] = 'a', ['koi2'] = 'a', ['byz1'] = 'a', ['byz2'] = 'a', }, }

data['ε'] = { ['pre'] = { { '0~isIDiphth/0~isUDiphth', 1}, 0,	},	['p'] = { ['cla'] = 'e', ['koi1'] = 'e', ['koi2'] = 'e', ['byz1'] = 'e', ['byz2'] = 'e', }, }

data['η'] = { ['pre'] = { { '0~isUDiphth', 1}, 0,	},	['p'] = { ['cla'] = 'ɛ', ['koi1'] = 'e̝', ['koi2'] = 'i', ['byz1'] = 'i', ['byz2'] = 'i', }, }

data['ι'] = { ['p'] = { ['cla'] = 'i', ['koi1'] = 'i', ['koi2'] = 'i', ['byz1'] = 'i', ['byz2'] = 'i', }, }

data['ο'] = { ['pre'] = { { '0~isIDiphth/0~isUDiphth', 1}, 0,	},	['p'] = { ['cla'] = 'o', ['koi1'] = 'o', ['koi2'] = 'o', ['byz1'] = 'o', ['byz2'] = 'o', }, }

data['υ'] = { ['pre'] = { { '0~isIDiphth/0~hasMacronBreve', 1}, 0,	},	['p'] = { ['cla'] = 'y', ['koi1'] = 'y', ['koi2'] = 'y', ['byz1'] = 'y', ['byz2'] = 'i', }, }

data['ω'] = { ['p'] = { ['cla'] = 'ɔ', ['koi1'] = 'o', ['koi2'] = 'o', ['byz1'] = 'o', ['byz2'] = 'o', }, }

local categories = { [1] = {		["stop"] = { "π", "τ", "κ", "β", "δ", "γ", "φ", "θ", "χ", "ψ", "ξ", }, ["dorsal"] = { "κ", "γ", "χ", "ξ", }, ["voiced"] = { "β", "δ", "γ", "ζ", "μ", "ν", "λ", "ρ", "ϝ", }, ["unvoiced"] = { "π", "ψ", "τ", "κ", "ξ", "φ", "θ", "χ", "σ", "ς", }, ["aspirated"] = { "φ", "θ", "χ", }, ["diaer"] = { "ϊ", "ϋ", "ΐ", "ΰ", "ῒ", "ῢ", "ῗ", "ῧ", }, ["subi"] = { "ᾳ", "ῃ", "ῳ", "ᾴ", "ῄ", "ῴ", "ᾲ", "ῂ", "ῲ", "ᾷ", "ῇ", "ῷ", "ᾀ", "ᾐ", "ᾠ", "ᾄ", "ᾔ", "ᾤ", "ᾂ", "ᾒ", "ᾢ", "ᾆ", "ᾖ", "ᾦ", "ᾁ", "ᾑ", "ᾡ", "ᾅ", "ᾕ", "ᾥ", "ᾃ", "ᾓ", "ᾣ", "ᾇ", "ᾗ", "ᾧ", }, },	["type"] = { ["vowel"] = { "α", "ε", "η", "ι", "ο", "ω", "υ", }, -- Not currently used; if it were, it might need to include all the accented vowel characters. ["consonant"] = { "β", "γ", "δ", "ζ", "θ", "κ", "λ", "μ", "ν", "ξ", "π", "ρ", "σ", "ς", "τ", "φ", "χ", "ψ", }, ["long"] = { "η", "ω", "ᾱ", "ῑ", "ῡ", }, ["short"] = { "ε", "ο", "ᾰ", "ῐ", "ῠ", }, ["either"] = { "α", "ι", "υ", }, ["diacritic"] = { diacritics.macron, diacritics.spacing_macron, diacritics.modifier_macron, diacritics.breve, diacritics.spacing_breve, diacritics.rough, diacritics.smooth, diacritics.diaeresis, diacritics.acute, diacritics.grave, diacritics.circum, diacritics.Latin_circum, diacritics.coronis, diacritics.subscript, }, },	["accent"] = { ["acute"] = { "ά", "έ", "ή", "ί", "ό", "ύ", "ώ", "ᾴ", "ῄ", "ῴ", "ἄ", "ἔ", "ἤ", "ἴ", "ὄ", "ὔ", "ὤ", "ᾄ", "ᾔ", "ᾤ", "ἅ", "ἕ", "ἥ", "ἵ", "ὅ", "ὕ", "ὥ", "ᾅ", "ᾕ", "ᾥ", "ΐ", "ΰ", }, ["grave"] = { "ὰ", "ὲ", "ὴ", "ὶ", "ὸ", "ὺ", "ὼ", "ᾲ", "ῂ", "ῲ", "ἂ", "ἒ", "ἢ", "ἲ", "ὂ", "ὒ", "ὢ", "ᾂ", "ᾒ", "ᾢ", "ἃ", "ἓ", "ἣ", "ἳ", "ὃ", "ὓ", "ὣ", "ᾃ", "ᾓ", "ᾣ", "ῒ", "ῢ", }, ["circum"] = { "ᾶ", "ῆ", "ῖ", "ῦ", "ῶ", "ᾷ", "ῇ", "ῷ", "ἆ", "ἦ", "ἶ", "ὖ", "ὦ", "ᾆ", "ᾖ", "ᾦ", "ἇ", "ἧ", "ἷ", "ὗ", "ὧ", "ᾇ", "ᾗ", "ᾧ", "ῗ", "ῧ", }, },	["breath"] = { ["rough"] = { "ἁ", "ἑ", "ἡ", "ἱ", "ὁ", "ὑ", "ὡ", "ᾁ", "ᾑ", "ᾡ", "ἅ", "ἕ", "ἥ", "ἵ", "ὅ", "ὕ", "ὥ", "ᾅ", "ᾕ", "ᾥ", "ἃ", "ἓ", "ἣ", "ἳ", "ὃ", "ὓ", "ὣ", "ᾃ", "ᾓ", "ᾣ", "ἇ", "ἧ", "ἷ", "ὗ", "ὧ", "ᾇ", "ᾗ", "ᾧ", }, ["smooth"] = { "ἀ", "ἐ", "ἠ", "ἰ", "ὀ", "ὐ", "ὠ", "ᾀ", "ᾐ", "ᾠ", "ῤ", "ἄ", "ἔ", "ἤ", "ἴ", "ὄ", "ὔ", "ὤ", "ᾄ", "ᾔ", "ᾤ", "ἂ", "ἒ", "ἢ", "ἲ", "ὂ", "ὒ", "ὢ", "ᾂ", "ᾒ", "ᾢ", "ἆ", "ἦ", "ἶ", "ὖ", "ὦ", "ᾆ", "ᾖ", "ᾦ", }, },	}

for key1, list in pairs(categories) do	for key2, letters in pairs(list) do		if type(key1) == "number" then for _, letter in ipairs(letters) do				if not data[letter] then data[letter] = {} end data[letter][key2] = true end elseif type(key1) == "string" then for _, letter in ipairs(letters) do				if not data[letter] then data[letter] = {} end data[letter][key1] = key2 end end end end

for letter in gmatch("εέὲἐἔἒἑἕἓ", ".") do	local l_data = data[letter] l_data.p = epsilon(l_data.breath, l_data.accent) end

for letter in gmatch("οόὸὀὄὂὁὅὃ", ".") do	local l_data = data[letter] l_data.p = omicron(l_data.breath, l_data.accent) end

for letter in gmatch("ηῃήῄὴῂῆῇἠᾐἤᾔἢᾒἦᾖἡᾑἥᾕἣᾓἧᾗ", ".") do	local l_data = data[letter] l_data.p = eta(l_data.breath, l_data.accent, l_data.subi) end

for letter in gmatch("ωῳώῴὼῲῶῷὠᾠὤᾤὢᾢὦᾦὡᾡὥᾥὣᾣὧᾧ", ".") do	local l_data = data[letter] l_data.p = omega(l_data.breath, l_data.accent, l_data.subi) end

for letter in gmatch("αᾳάᾴὰᾲᾶᾷἀᾀἄᾄἂᾂἆᾆἁᾁἅᾅἃᾃἇᾇ", ".") do	local l_data = data[letter] l_data.p = alpha(l_data.breath, l_data.accent, l_data.subi) if not l_data.subi and l_data.accent ~= 'circum' then if not l_data.pre then l_data.pre = { { '0~hasMacronBreve', 1}, 0, } end data[letter .. breve] = {p = alpha(l_data.breath, l_data.accent, false, false)} data[letter .. macron] = {p = alpha(l_data.breath, l_data.accent, false, true)} end end

for letter in gmatch("ιίὶῖἰἴἲἶἱἵἳἷϊΐῒῗ", ".") do	local l_data = data[letter] l_data.p = iota(l_data.breath, l_data.accent) if l_data.accent ~= 'circum' then l_data.pre = { { '0~hasMacronBreve', 1}, 0, } data[letter .. breve] = {p = iota(l_data.breath, l_data.accent, false)} data[letter .. macron] = {p = iota(l_data.breath, l_data.accent, true)} end if not l_data.diar then data['α' .. letter] = {p = ai(l_data.breath, l_data.accent)} data['ε' .. letter] = {p = ei(l_data.breath, l_data.accent)} data['ο' .. letter] = {p = oi(l_data.breath, l_data.accent)} data['υ' .. letter] = {p = ui(l_data.breath, l_data.accent)} end end

for letter in gmatch("υύὺῦὐὔὒὖὑὕὓὗϋΰῢῧ", ".") do	local l_data = data[letter] l_data.p = ypsilon(l_data.breath, l_data.accent) if l_data.accent ~= 'circum' then if letter ~= 'υ' then l_data.pre = { { '0~hasMacronBreve', 1}, 0, } end data[letter .. breve] = {p = ypsilon(l_data.breath, l_data.accent, false)} data[letter .. macron] = {p = ypsilon(l_data.breath, l_data.accent, true)} end if not l_data.diar then data['α' .. letter] = {p = au(l_data.breath, l_data.accent)} data['η' .. letter] = {p = hu(l_data.breath, l_data.accent)} data['ε' .. letter] = {p = eu(l_data.breath, l_data.accent)} data['ο' .. letter] = {p = ou(l_data.breath, l_data.accent)} end end

data['chars'] = { ['cons'] = 'bɡŋdzklmnprstβðɣɸθxfvɟʝcçwʍj', ['vowel'] = "aeiouyɛɔ", ['diacritic'] = high .. low .. midHigh .. midLow .. highMid .. long .. aspirated .. voiceless .. nonsyllabic .. rising .. falling, ['liquid'] = "rln", ['obst'] = "bɡdkptβðɣɸθxfv", ['frontDiphth'] = "[αο]ι", ['frontVowel'] = "ιηευ", ['iDiaer'] = "ϊΐῒῗ", ['long'] = "ηω", ['short'] = "εο", ['ambig'] = "αιυ", ['uDiphth'] = 'αεηο', -- first members for diphthongs ending in 'υ' ['iDiphth'] = 'αεου', -- first members for diphthongs ending in 'ι' ['Greekdiacritic'] = m_utils_data.all, ['Greekconsonant'] = m_utils_data.consonants } data.chars.frontDiphthong = data.chars.frontDiphth

return data