Module:la-pronunc

local export = {}

local m_a = require("Module:accent qualifier") local m_IPA = require("Module:IPA") local m_str_utils = require("Module:string utilities") local lang = require("Module:languages").getByCode("la")

local u = m_str_utils.char local rfind = m_str_utils.find local rsubn = m_str_utils.gsub local rmatch = m_str_utils.match local rsplit = m_str_utils.split local ulower = m_str_utils.lower local usub = m_str_utils.sub local ulen = m_str_utils.len

local BREVE = u(0x0306) -- breve = ̆ local TILDE = u(0x0303) -- ̃ local HALF_LONG = "ˑ" local LONG = "ː"

local letters_ipa = { ["a"] = "a",["e"] = "e",["i"] = "i",["o"] = "o",["u"] = "u",["y"] = "y", ["ā"] = "aː",["ē"] = "eː",["ī"] = "iː",["ō"] = "oː",["ū"] = "uː",["ȳ"] = "yː", ["ae"] = "ae̯",["oe"] = "oe̯",["ei"] = "ei̯",["au"] = "au̯",["eu"] = "eu̯", ["b"] = "b",["d"] = "d",["f"] = "f", ["c"] = "k",["g"] = "ɡ",["v"] = "w",["x"] = "ks", ["ph"] = "pʰ",["th"] = "tʰ",["ch"] = "kʰ",["rh"] = "r",["qv"] = "kʷ",["gv"] = "ɡʷ", ["'"] = "ˈ",["ˈ"] = "ˈ", }

local letters_ipa_eccl = { ["a"] = "a",["e"] = "e",["i"] = "i",["o"] = "o",["u"] = "u",["y"] = "i", ["ā"] = "aː",["ē"] = "eː",["ī"] = "iː",["ō"] = "oː",["ū"] = "uː",["ȳ"] = "iː", ["ae"] = "eː",["oe"] = "eː",["ei"] = "ei̯",["au"] = "au̯",["eu"] = "eu̯", ["b"] = "b",["d"] = "d",["f"] = "f", ["k"] = "q", -- dirty hack to make sure k isn't palatalized ["c"] = "k", ["g"] = "ɡ",["v"] = "v",["x"] = "ks", ["ph"] = "f",["th"] = "tʰ",["ch"] = "kʰ",["rh"] = "r",["qv"] = "kw",["gv"] = "ɡw", ["sv"] = "sw", --"sw" is needed to avoid [zv] in words like suavium ["h"] = "", ["'"] = "ˈ",["ˈ"] = "ˈ", }

local lax_vowel = { ["e"] = "ɛ", ["i"] = "ɪ", ["o"] = "ɔ", ["u"] = "ʊ", ["y"] = "ʏ", }

local tense_vowel = { ["ɛ"] = "e", ["ɪ"] = "i", ["ɔ"] = "o", ["ʊ"] = "u", ["ʏ"] = "y", }

local voicing = { ["p"] = "b", ["t"] = "d", ["k"] = "ɡ", }

local devoicing = { ["b"] = "p", ["d"] = "t", ["ɡ"] = "k", }

local classical_vowel_letters = "aeɛiɪoɔuʊyʏ" local classical_vowel = "[" .. classical_vowel_letters .. "]"

local phonetic_rules = {

-- Bibliography included at the end

-- Assimilation of [g] to [ŋ] before a following /n/. {"ɡ([.ˈ]?)n", "ŋ%1n"}, -- Per Allen (1978: 23), although note the reservations expressed on the next page.

-- Assimilation of word-internal /n/ and /m/ to following consonants. Exception: /m/ does not assimilate to a following /n/. {"n([.ˈ]?)([mpb])", "m%1%2"}, {"n([.ˈ]?)([kɡ])", "ŋ%1%2"}, {"m([.ˈ]?)([td])", "n%1%2"}, {"m([.ˈ]?)([kɡ])", "ŋ%1%2"}, -- Per George M. Lane: “Nasals changed their place of articulation to that of the following consonant. Thus, dental n before the labials p and b became the labial m... labial m before the gutturals c and g became guttural n...labial m before the dentals t, d, s became dental n…” (§164.3); “One nasal, n, is assimilated to another, m...but an m before n is never assimilated..." (§166.5).		-- Per Lloyd (1987: 84): “The opposition between nasals was neutralized in syllable-final position, with the realization of the nasality being assimilated to the point of articulation of the following consonant, e.g., [m] is found only before labials, [n] only before dentals or alveolars, and [ŋ] only before velars and /n/." -- Potential addition: assimilation of final /m/ and /n/ across word boundaries, per e.g. Allen (1987: 28, 31). -- No additional labialization before high back vowels {"ʷ%f[uʊ]", ""}, -- Tensing of short vowels before another vowel {		"([ɛɪʏɔʊ])([.ˈ][h]?)%f[aeɛiɪoɔuʊyʏ]", function (vowel, following) return (tense_vowel[vowel] or vowel) .. following end, },

-- But not before consonantal glides {"ei̯", "ɛi̯"}, {"eu̯", "ɛu̯"},

-- Nasal vowels {		"(" .. classical_vowel .. ")m$", function (vowel) return (lax_vowel[vowel] or vowel) .. TILDE .. HALF_LONG end, },	{		"(" .. classical_vowel .. ")[nm]([.ˈ]?[sf])", function (vowel, following) return (tense_vowel[vowel] or vowel) .. TILDE .. LONG .. following end, },

-- Dissimilation after homorganic glides (the tuom volgus-type) --{"([wu])([.ˈ]?)([h]?)ʊ", "%1%2%3o"}, --{"([ji])([.ˈ]?)([h]?)ɪ", "%1%2%3e"}, ---Disabled per 19 September 2021 discussion at Template_talk:la-IPA#Transcription_of_syllable-initial_semivowels -- Realization of /r/ as a tap -- Pultrová (2013) argues for Latin /r/ being an alveolar tap. -- Lloyd (1987: 81) agrees: “The /r/ was doubtlessly an alveolar flap."		-- Allen (1978: 33) expresses doubt: “By the classical period there is no reason to think that the sound had not strengthened to the trill described by later writers.”       -- Unconditional [r] transcription is preferable to unconditional [ɾ] per 18 September 2021 discussion at Module_talk:la-pronunc#Transcription_of_Latin's_rhotic_consonant        -- No consensus yet on how to implement conditional allophony of [r] vs. [ɾ]	-- Voicing and loss of intervocalic /h/.	{"([^ˈ].)h", "%1(ɦ)"},	-- Per Allen (1978: 43–45).

-- Phonetic (as opposed to lexical/phonemic) assimilations -- Place -- First because this accounts for 'atque' seemingly escaping total assimilation (and 'adque' presumably not) {"[d]([.ˈ]?)s", "s%1s"},  -- leave [t] out since etsi has [ts], not [sː] {"s[^ː]([.ˈ]?)s%f[ptk]", "s(ː)%1"}, {"st([.ˈ])([^aeɛiɪoɔuʊyʏe̯u̯])", "s(t)%1%2"},

{"d([.ˈ])([pkɡln])", "%2%1%2"}, --leave [r] out since dr does not assimilate, even when heterosyllabic (e.g. quadrans), except in prefixed words {"b([.ˈ])([mf])", "%2%1%2"}, {"s([.ˈ])f", "f%1f"},

-- Regressive voicing assimilation in consonant clusters {		"([bdɡ])([.ˈ]?)%f[ptksf]", function (consonant, following) return (devoicing[consonant] or consonant) .. following end, },	{		"([ptk])([.ˈ]?)%f[bdɡ]", function (consonant, following) return (voicing[consonant] or consonant) .. following end, },

-- Allophones of /l/ {"l", "ɫ̪"}, -- “Pinguis”. Dark/velarized. -- Per Weiss (2009: 117): “…pinguis (velar). l is exīlis before i and when geminate, otherwise l is pinguis.” -- Page 82: “…l is pinguis even before e, e.g. Herculēs < Hercolēs … < Hercelēs …” -- Per Sihler (1995: 174): “l exilis was found before the vowels -i- and -ī-, and before another -l-; l pinguis occurred before any other vowel; before any consonant except l; and in word-final position […] l pinguis actually had two degrees of avoirdupois, being fatter before a consonant than before a vowel…” -- Page 41: “…velarized l (that is, ‘l pinguis’)…” -- Sen (2015: §2) states that /l/ was velarized in word-final position or before consonants–other than another /l/–and that it had varying degrees of “dark resonance (velarization in articulatory terms)” (p. 23) before e, a, o, and u (p. 33). -- Both Sen and Sihler indicate different degrees of velarization, depending on the environment. IPA lacks a way to represent these gradations, unfortunately. {"ɫ̪([.ˈ]?)ɫ̪", "l%1lʲ"}, {"ɫ̪([.ˈ]?[iɪyʏ])", "lʲ%1"}, -- “Exīlis”. Not dark/velarized. Possibly palatalized. -- Per Sen (2015: 29): It is plausible […] that simple onset /l/ was palatalized before /i/, thus [lʲ] […] it seems likely that geminate /ll/ was also palatalized, given the similar behaviour of the two…” -- Per Weiss (2009: 82): “In Latin, l developed…a non-velar (possibly palatal) allophone called exīlis before i and when geminate…” -- Per Sihler (1995: 174): “l exilis was found before the vowels 􏰹-i-􏰹 and -ī-, and before another -l-.” -- Per Sihler (2000: §133.1): "It is less clear whether the 'thin' lateral [i.e. L exilis] was specifically palatal, or palatalized, or only neutral." -- Giannini and Marotta apparently argue that it was not palatalized (https://i.imgur.com/ytM1QDn.png). I do not have access to the book in question.

-- Retracted /s/ {"s", "s̠"}, -- Lloyd (1987: 80–81) expresses some uncertainty about this, but appears to overall be in favour of it: “…the evidence that the apico-alveolar pronunciation was ancient in Latin and inherited from Indo-European is quite strong.” -- Per Zampaulo (2019: 93), “…in many instances, Latin s was likely pronounced as an apical segment [s̺] (rather than laminal [s])."		-- Per Widdison (1987: 64), "In all, it would be fair to state that the apico-alveolar [ś] articulation represented the main allophonic variant of Latin and possibly IE /s/..."

-- dental Z	{"z([aeɛiɪoɔuʊyʏ])", "d͡z%1"},      --See discussion {"z([.ˈ])z", "z%1(d͡)z"}, {"z", "z̪"},

-- Dental articulations {"t", "t̪"}, {"d", "d̪"}, {"n([.ˈ]?)([td])", "n̪%1%2"},      --it's not as clear as for the stops

--Allophones of A	{"a", "ä"},

-- Works cited -- Allen, William Sidney. 1978. Vox Latina: A Guide to the pronunciation of Classical Latin. -- Lane, George M. A Latin grammar for schools and colleges. -- Lloyd, Paul M. 1987. From Latin to Spanish. -- Pultrová, Lucie. 2013. On the phonetic nature of the Latin R. -- Sen, Ranjan. 2015. Syllable and segment in Latin. -- Sihler, Andrew L. 1995. New comparative grammar of Greek and Latin. -- Sihler, Andrew L. 2000. Language history: An introduction. -- Weiss, Michael. 2009. Outline of the historical and comparative grammar of Latin. -- Widdison, Kirk A. 16th century Spanish sibilant reordering: Reasons for divergence. -- Zampaulo, André. 2019. Palatal Sound Change in the Romance languages: Diachronic and Synchronic Perspectives. }

local phonetic_rules_eccl = { -- Specifically the Roman Ecclesiastical for singing from the Liber Usualis

{"([aɛeiɔou][ː.ˈ]*)s([.ˈ]*)%f[aɛeiɔou]", "%1s̬%2"},      --partial voicing of s between vowels {"s([.ˈ]*)%f[bdgmnlv]", "z%1"},      --full voicing of s before voiced consonants {"ek([.ˈ]*)s([aɛeiɔoubdgmnlv])", "eɡ%1z%2"},      --voicing of the prefix ex- {"kz", "ɡz"},      --i give up, without this /ksˈl/ gives [kzˈl]

-- Tapped R intervocalically and in complex onset -- ^ Citation needed for this being the case in Ecclesiastical pronunciation -- {"([aɛeiɔou]ː?[.ˈ])r([aɛeiɔou]?)", "%1ɾ%2"}, -- {"([fbdgptk])r", "%1ɾ"}, {"a", "ä"}, --a is open and central per 17 September 2021 discussion at Template_talk:la-IPA#Ecclesiastical_a -- /e/ and /o/ realization is phonetic but handled in convert_word below as it is sensitive to stress

-- Dental articulations {"n([.ˈ]?)([td])([^͡])", "n̪%1%2%3"}, --assimilation of n to dentality. {"l([.ˈ]?)([td])([^͡])", "l̪%1%2%3"}, --Note that the quality of n might not be dental otherwise--it may be alveolar in most contexts in Italian, according to Wikipedia. {"t([^͡])", "t̪%1"},      --t is dental, except as the first element of a palatal affricate {"d([^͡])", "d̪%1"},      --d is dental, except as the first element of a palatal affricate {"t͡s", "t̪͡s̪"},      -- dental affricates {"d͡z", "d̪͡z̪"},      --dental affricates {"t̪([.ˈ]?)t͡ʃ", "t%1t͡ʃ"}, {"d̪([.ˈ]?)d͡ʒ", "d%1d͡ʒ"},

--end of words {"lt$", "l̪t̪"}, {"nt$", "n̪t̪"}, {"t$", "t̪"}, {"d$", "d̪"},

--Partial assimilation of l and n before palatal affricates, as in Italian {"l([.ˈ]?)t͡ʃ", "l̠ʲ%1t͡ʃ"}, {"l([.ˈ]?)d͡ʒ", "l̠ʲ%1d͡ʒ"}, {"l([.ˈ]?)ʃ", "l̠ʲ%1ʃ"}, {"n([.ˈ]?)t͡ʃ", "n̠ʲ%1t͡ʃ"}, {"n([.ˈ]?)d͡ʒ", "n̠ʲ%1d͡ʒ"}, {"n([.ˈ]?)ʃ", "n̠ʲ%1ʃ"},

-- other coda nasal assimilation, full and partial. Per Canepari, only applies to /n/ and not to /m/ {"n([.ˈ]?)([kɡ])", "ŋ%1%2"}, {"n([.ˈ]?)([fv])", "ɱ%1%2"},

}

local lenition = { ["ɡ"] = "ɣ", ["d"] = "ð", }

local lengthen_vowel = { ["a"] = "aː", ["aː"] = "aː", ["ɛ"] = "ɛː", ["ɛː"] = "ɛː", ["e"] = "eː", ["eː"] = "eː", ["i"] = "iː", ["iː"] = "iː", ["ɔ"] = "ɔː", ["ɔː"] = "ɔː", ["o"] = "oː", ["oː"] = "oː", ["u"] = "uː", ["uː"] = "uː", ["au̯"] = "aːu̯", ["ɛu̯"] = "ɛːu̯", ["eu̯"] = "eːu̯", }

local vowels = { "a", "ɛ", "e", "ɪ", "i", "ɔ", "o", "ʊ", "u", "y", "aː", "ɛː", "eː", "iː", "ɔː", "oː", "uː", "yː", "ae̯", "oe̯", "ei̯", "au̯", "eu̯", }

local onsets = { "b", "p", "pʰ", "d", "t", "tʰ", "β", "ɡ", "k", "kʰ", "kʷ", "ɡʷ", "kw", "ɡw", "t͡s", "t͡ʃ", "d͡ʒ", "ʃ", "f", "s", "z", "d͡z", "h", "l", "m", "n", "ɲ", "r", "j", "v", "w", "bl", "pl", "pʰl", "br", "pr", "pʰr", "dr", "tr", "tʰr", "ɡl", "kl", "kʰl", "ɡr", "kr", "kʰr", "fl", "fr", "sp", "st", "sk", "skʷ", "sw", "spr", "str", "skr", "spl", "skl", }

local codas = { "b", "p", "pʰ", "d", "t", "tʰ", "ɡ", "k", "kʰ", "β", "f", "s", "z", "l", "m", "n", "ɲ", "r", "j", "ʃ", "sp", "st", "sk", "spʰ", "stʰ", "skʰ", "lp", "lt", "lk", "lb", "ld", "lɡ", "lpʰ", "ltʰ", "lkʰ", "lf", "rp", "rt", "rk", "rb", "rd", "rɡ", "rpʰ", "rtʰ", "rkʰ", "rf", "mp", "nt", "nk", "mb", "nd", "nɡ", "mpʰ", "ntʰ", "nkʰ", "lm", "rl", "rm", "rn", "ps", "ts", "ks", "ls", "ns", "rs", "lks", "nks", "rks", "rps", "mps", "lms", "rls", "rms", "rns", }

-- Prefixes that end in a consonant; can be patterns. Occurrences of such -- prefixes + i + vowel cause the i to convert to j (to suppress this, add a -- dot, i.e. syllable boundary, after the i). local cons_ending_prefixes = { "a[bd]", "circum", "con", "dis", "ex", "in", "inter", "ob", "per", "sub", "subter", "super", "tr[aā]ns" }

local remove_macrons = { ["ā"] = "a", ["ē"] = "e", ["ī"] = "i", ["ō"] = "o", ["ū"] = "u", ["ȳ"] = "y", }

local macrons_to_breves = { ["ā"] = "ă", ["ē"] = "ĕ", ["ī"] = "ĭ", ["ō"] = "ŏ", ["ū"] = "ŭ", -- Unicode doesn't have breve-y ["ȳ"] = "y" .. BREVE, }

local remove_breves = { ["ă"] = "a", ["ĕ"] = "e", ["ĭ"] = "i", ["ŏ"] = "o", ["ŭ"] = "u", -- Unicode doesn't have breve-y }

local remove_ligatures = { ["æ"] = "ae", ["œ"] = "oe", }

for i, val in ipairs(vowels) do	vowels[val] = true end

for i, val in ipairs(onsets) do	onsets[val] = true end

for i, val in ipairs(codas) do	codas[val] = true end

-- NOTE: Everything is lowercased very early on, so we don't have to worry -- about capitalized letters. local short_vowels_string = "aeiouyăĕĭŏŭäëïöüÿ" -- no breve-y in Unicode local long_vowels_string = "āēīōūȳ" local vowels_string = short_vowels_string .. long_vowels_string local vowels_c = "[" .. vowels_string .. "]" local non_vowels_c = "[^" .. vowels_string .. "]"

local function track(page) require("Module:debug/track")("la-pronunc/" .. page) return true end

-- version of rsubn that discards all but the first return value local function rsub(term, foo, bar) local retval = rsubn(term, foo, bar) return retval end

-- version of rsubn that returns a 2nd argument boolean indicating whether -- a substitution was made. local function rsubb(term, foo, bar) local retval, nsubs = rsubn(term, foo, bar) return retval, nsubs > 0 end

local function letters_to_ipa(word,phonetic,eccl,vul) local phonemes = {} local dictionary = eccl and letters_ipa_eccl or (vul and letters_ipa_vul or letters_ipa) while ulen(word) > 0 do		local longestmatch = "" for letter, ipa in pairs(dictionary) do			if ulen(letter) > ulen(longestmatch) and usub(word, 1, ulen(letter)) == letter then longestmatch = letter end end if ulen(longestmatch) > 0 then if dictionary[longestmatch] == "ks" then table.insert(phonemes, "k") table.insert(phonemes, "s") else table.insert(phonemes, dictionary[longestmatch]) end word = usub(word, ulen(longestmatch) + 1) else table.insert(phonemes, usub(word, 1, 1)) word = usub(word, 2) end end if eccl then for i=1,#phonemes do		local prev, cur, next = phonemes[i-1], phonemes[i], phonemes[i+1] if next and (cur == "k" or cur == "ɡ") and rfind(next, "^[eɛi]ː?$") then if cur == "k" then if prev == "s" then --and ((not phonemes[i-2]) or phonemes[i-2] ~= "k") prev = "ʃ" cur = "ʃ" else cur = "t͡ʃ" if prev == "k" then prev = "t" end end else cur = "d͡ʒ" if prev == "ɡ" then prev = "d" end end end -- dirty hack to make sure k isn't palatalized if cur == "q" then cur = "k" end if cur == "t" and next == "i" and not (prev == "s" or prev == "t") and vowels[phonemes[i+2]] then cur = "t͡s" end if cur == "z" then if next == "z" then cur = "d" next = "d͡z" else cur = "d͡z" end end if cur == "kʰ" then cur = "k" end if cur == "tʰ" then cur = "t" end if cur == "ɡ" and next == "n" then cur = "ɲ" next = "ɲ" end phonemes[i-1], phonemes[i], phonemes[i+1] = prev, cur, next end end return phonemes end

local function get_onset(syll) local consonants = {} for i = 1, #syll do		if vowels[syll[i]] then break end if syll[i] ~= "ˈ" then table.insert(consonants, syll[i]) end end return table.concat(consonants) end

local function get_coda(syll) local consonants = {} for i = #syll, 1, -1 do		if vowels[syll[i]] then break end table.insert(consonants, 1, syll[i]) end return table.concat(consonants) end

local function get_vowel(syll) for i = 1,#syll do		if vowels[syll[i]] then return syll[i] end end end

-- Split the word into syllables of CV shape local function split_syllables(remainder) local syllables = {} local syll = {} for _, phoneme in ipairs(remainder) do if phoneme == "." then if #syll > 0 then table.insert(syllables, syll) syll = {} end -- Insert a special syllable consisting only of a period. -- We remove it later but it forces no movement of consonants across -- the period. table.insert(syllables, {"."}) elseif phoneme == "ˈ" then if #syll > 0 then table.insert(syllables,syll) end syll = {"ˈ"} elseif vowels[phoneme] then table.insert(syll, phoneme) table.insert(syllables, syll) syll = {} else table.insert(syll, phoneme) end end -- If there are phonemes left, then the word ends in a consonant. -- Add another syllable for them, which will get joined the preceding -- syllable down below. if #syll > 0 then table.insert(syllables, syll) end -- Split consonant clusters between syllables for i, current in ipairs(syllables) do if #current == 1 and current[1] == "." then -- If the current syllable is just a period (explicit syllable			-- break), remove it. The loop will then skip the next syllable, -- which will prevent movement of consonants across the syllable -- break (since movement of consonants happens from the current			-- syllable to the previous one). table.remove(syllables, i)		elseif i > 1 then local previous = syllables[i-1] local onset = get_onset(current) -- Shift over consonants until the syllable onset is valid while not (onset == "" or onsets[onset]) do				table.insert(previous, table.remove(current, 1)) onset = get_onset(current) end -- If the preceding syllable still ends with a vowel, -- and the current one begins with s + another consonant, then shift it over. if get_coda(previous) == "" and (current[1] == "s" and not vowels[current[2]]) then table.insert(previous, table.remove(current, 1)) end -- Check if there is no vowel at all in this syllable. That -- generally happens either (1) with an explicit syllable division -- specified, like 'cap.ra', which will get divided into the syllables -- [ca], [p], [.], [ra]; or (2) at the end of a word that ends with -- one or more consonants. We move the consonants onto the preceding -- syllable, then remove the resulting empty syllable. If the -- new current syllable is [.], remove it, too. The loop will then -- skip the next syllable, which will prevent movement of consonants -- across the syllable break (since movement of consonants happens			-- from the current syllable to the previous one). if not get_vowel(current) then for j=1,#current do					table.insert(previous, table.remove(current, 1)) end table.remove(syllables, i) if syllables[i] and #syllables[i] == 1 and syllables[i][1] == "." then table.remove(syllables, i)				end end end end for i, syll in ipairs(syllables) do		local onset = get_onset(syll) local coda = get_coda(syll) if not (onset == "" or onsets[onset]) then track("bad onset") --error("onset error:[" .. onset .. "]") end if not (coda == "" or codas[coda]) then track("bad coda") --error("coda error:[" .. coda .. "]") end end return syllables end

local function phoneme_is_short_vowel(phoneme) return rfind(phoneme, "^[aɛeiɔouy]$") end

local function detect_accent(syllables, is_prefix, is_suffix) -- Manual override for i=1,#syllables do		for j=1,#syllables[i] do			if syllables[i][j] == "ˈ" then table.remove(syllables[i],j) return i			end end end -- Prefixes have no accent. if is_prefix then return -1 end -- Suffixes have an accent only if the stress would be on the suffix when the -- suffix is part of a word. Don't get tripped up by the first syllable being -- nonsyllabic (e.g. in -rnus). if is_suffix then local syllables_with_vowel = #syllables - (get_vowel(syllables[1]) and 0 or 1) if syllables_with_vowel < 2 then return -1 end if syllables_with_vowel == 2 then local penult = syllables[#syllables - 1] if phoneme_is_short_vowel(penult[#penult]) then return -1 end end end -- Detect accent placement if #syllables > 2 then -- Does the penultimate syllable end in a single vowel? local penult = syllables[#syllables - 1] if phoneme_is_short_vowel(penult[#penult]) then return #syllables - 2 else return #syllables - 1 end elseif #syllables == 2 then return #syllables - 1 elseif #syllables == 1 then return #syllables       --mark stress on monosyllables so that stress-conditioned sound rules work correctly. Then, delete it prior to display end end

local function convert_word(word, phonetic, eccl, vul) -- Normalize i/j/u/v; do this before removing breves, so we keep the -- ŭ in langŭī (perfect of languēscō) as a vowel. word = rsub(word, "w", "v") word = rsub(word, "(" .. vowels_c .. ")v(" .. non_vowels_c .. ")", "%1u%2") word = rsub(word, "qu", "qv") word = rsub(word, "ngu(" .. vowels_c .. ")", "ngv%1") word = rsub(word, "^i(" .. vowels_c .. ")", "j%1") word = rsub(word, "^u(" .. vowels_c .. ")", "v%1") -- Per the August 31 2019 recommendation by User:Brutal Russian in	-- Module talk:la-pronunc, we convert i/j between vowels to jj if the -- preceding vowel is short but to single j if the preceding vowel is long. word = rsub(		word,		"(" .. vowels_c .. ")([iju])",		function (vowel, potential_consonant, pos)			if vowels_string:find(usub(word, pos, pos)) then				if potential_consonant == "u" then					return vowel .. "v"				else					if long_vowels_string:find(vowel) then						return vowel .. "j"					else						return vowel .. "jj"					end				end			end		end)

--Convert v to u syllable-finally word = rsub(word, "v%.", "u.") word = rsub(word, "v$", "u")

-- Convert i to j before vowel and after any prefix that ends in a consonant, -- per the August 23 2019 discussion in Module talk:la-pronunc. for _, pref in ipairs(cons_ending_prefixes) do word = rsub(word, "^(" .. pref .. ")i(" .. vowels_c .. ")", "%1j%2") end

-- Ecclesiastical has neither geminate j.j, nor geminate w.w in Greek words if eccl then word = rsub(word, "(" .. vowels_c .. ")u([.ˈ]?)v(" .. vowels_c .. ")", "%1%2v%3") word = rsub(word, "(" .. vowels_c .. ")j([.ˈ]?)j(" .. vowels_c .. ")", "%1%2j%3") end

-- Convert z to zz between vowels so that the syllable weight and stress assignment will be correct. word = rsub(word, "(" .. vowels_c .. ")z(" .. vowels_c .. ")", "%1zz%2")

if eccl then word = rsub(word, "(" .. vowels_c .. ")ti(" .. vowels_c .. ")", "%1tt͡si%2") end

-- Now remove breves. word = rsub(word, "([ăĕĭŏŭ])", remove_breves) -- BREVE sits uncombined in y+breve and vowel-macron + breve word = rsub(word, BREVE, "") -- Normalize aë, oë; do this after removing breves but before any -- other normalizations involving e.	word = rsub(word, "([ao])ë", "%1.e")

-- Eu and ei diphthongs word = rsub(word, "e(u[ms])$", "e.%1") word = rsub(word, "ei", "e.i") word = rsub(word, "_", "") -- Vowel length before nasal + fricative is allophonic word = rsub(word, "([āēīōūȳ])([mn][fs])",		function(vowel, nasalfric)			return remove_macrons[vowel] .. nasalfric		end	)

local vowel_before_yod = { ["a"] = "āj", ["e"] = "ēj", ["o"] = "ōj", ["u"] = "ūj", ["y"] = "ȳ", }   if eccl then word = rsub(word, "([aeiouy])([j])", vowel_before_yod) end -- Apply some basic phoneme-level assimilations for Ecclesiastical, which reads as written; in living varieties the assimilations were phonetic -- Italian (and therefore, by implication, Ecclesiastical Latin) does not show assimilation in clusters like /bk/ -- Source: "How can Italian phonology lack voice assimilation?", by Bálint Huszthy (2019): https://www.academia.edu/39347303/How_can_Italian_phonology_lack_voice_assimilation word = rsub(word, "xs", "x")

-- Per May 10 2019 discussion in Module talk:la-pronunc, we syllabify -- prefixes ab-, ad-, ob-, sub- separately from following l or r.	word = rsub(word, "^a([bd])([lr])", "a%1.%2") word = rsub(word, "^ob([lr])", "ob.%1") word = rsub(word, "^sub([lr])", "sub.%1")

-- Remove hyphens indicating prefixes or suffixes; do this after the above, -- some of which are sensitive to beginning or end of word and shouldn't	-- apply to end of prefix or beginning of suffix. local is_prefix, is_suffix word, is_prefix = rsubb(word, "%-$", "") word, is_suffix = rsubb(word, "^%-", "")

-- Convert word to IPA local phonemes = letters_to_ipa(word,phonetic,eccl,vul) -- Split into syllables local syllables = split_syllables(phonemes) -- Add accent local accent = detect_accent(syllables, is_prefix, is_suffix) -- poetic meter shows that a consonant before "h" was syllabified as an onset, not as a coda. -- Based on outcome of talk page discussion, this will be indicated by the omission of /h/ [h] in this context. word = rsub(word, "([^aeɛiɪoɔuʊyʏe̯u̯ptk])([.ˈ]?)h", "%1")

for i, syll in ipairs(syllables) do		for j, phoneme in ipairs(syll) do			if eccl or vul then syll[j] = rsub(syll[j], "ː", "") elseif phonetic then syll[j] = lax_vowel[syll[j]] or syll[j] end end end for i, syll in ipairs(syllables) do		if (eccl or vul) and i == accent and phonetic and vowels[syll[#syll]] then syll[#syll] = lengthen_vowel[syll[#syll]] or syll[#syll] end for j=1, #syll-1 do			if syll[j]==syll[j+1] then syll[j+1] = "" end end end

-- Atonic /ɔ/ and /ɛ/ merge with /o/ and /e/ respectively

for i, syll in ipairs(syllables) do		syll = table.concat(syll) if vul and i ~= accent then syll = rsub(syll, "ɔ", "o") syll = rsub(syll, "ɛ", "e") end if eccl and phonetic and i == accent then syll = rsub(syll, "o", "ɔ") syll = rsub(syll, "e", "ɛ") end syllables[i] = (i == accent and "ˈ" or "") .. syll end

word = (rsub(table.concat(syllables, "."), "%.ˈ", "ˈ")) if #syllables == 1 then word = rsub(word, "^ˈ", "")  --remove word-initial accent marks in monosyllables end

if eccl then word = rsub(word, "([^aeɛioɔu])ʃ([.ˈ]?)ʃ", "%1%2ʃ")    -- replace ʃ.ʃ or ʃˈʃ with .ʃ or ˈʃ after any consonant end

if not eccl then word = rsub(word, "j", "i̯")      -- normalize glide spelling word = rsub(word, "w", "u̯") end if phonetic then local rules = eccl and phonetic_rules_eccl or (vul and phonetic_rules_vul or phonetic_rules) for i, rule in ipairs(rules) do			word = rsub(word, rule[1], rule[2]) end

word = rsub(word, "[.]", "")      --remove the dots! >_<	end

if not eccl then word = rsub(word, "j", "i̯")      -- normalize glide spelling word = rsub(word, "w", "u̯") end

if phonetic then word = rsub(word, "(%a([̪̠̯]?))%1", "%1" .. LONG)      --convert double consonants into long ones word = rsub(word, "ːː", "ː") end

return word end

local function initial_canonicalize_text(text) -- Call ulower even though it's also called in phoneticize, -- in case convert_words is called externally. text = ulower(text) text = rsub(text, '[,?!:;"]', '')	text = rsub(text, '[æœ]', remove_ligatures)	return text end

function export.convert_words(text, phonetic, eccl, vul) text = initial_canonicalize_text(text) local disallowed = rsub(text, '[a-z%-āēīōūȳăĕĭŏŭë,.?!:;\'"_ ' .. BREVE .. ']', '')	if ulen(disallowed) > 0 then		if ulen(disallowed) == 1 then			error('The character "' .. disallowed .. '" is not allowed.')		else			error('The characters "' .. disallowed .. '" are not allowed.')		end		end	local result = {}	for word in mw.text.gsplit(text, " ") do		table.insert(result, convert_word(word, phonetic, eccl, vul))	end	return table.concat(result, " ") end

-- Phoneticize Latin TEXT. Return a list of one or more phoneticizations, -- each of which is a two-element list {PHONEMIC, PHONETIC}. If ECCL, use -- Ecclesiastical pronunciation. If VUL, use Vulgar Latin pronunciation. -- Otherwise, use Classical pronunciation. function export.phoneticize(text, eccl, vul) local function do_phoneticize(text, eccl, vul) return { export.convert_words(text, false, eccl, vul), export.convert_words(text, true, eccl, vul), }	end

text = ulower(text) -- If we have a macron-breve sequence, generate two pronunciations, one for -- the long vowel and one for the short. if rfind(text, "[āēīōūȳ]" .. BREVE) then local longvar = rsub(text, "([āēīōūȳ])" .. BREVE, "%1") local shortvar = rsub(text, "([āēīōūȳ])" .. BREVE, macrons_to_breves) local longipa = do_phoneticize(longvar, eccl, vul) local shortipa = do_phoneticize(shortvar, eccl, vul) -- Make sure long and short variants are actually different (they won't		-- be in Ecclesiastical pronunciation). if not require("Module:table").deepEquals(longipa, shortipa) then return {longipa, shortipa} else return {longipa} end elseif rfind(text, ";") then local tautosyllabicvar = rsub(text, ";", "") local heterosyllabicvar = rsub(text, ";", ".") local tautosyllabicipa = do_phoneticize(tautosyllabicvar, eccl, vul) local heterosyllabicipa = do_phoneticize(heterosyllabicvar, eccl, vul) if not require("Module:table").deepEquals(tautosyllabicipa, heterosyllabicipa) then return {tautosyllabicipa, heterosyllabicipa} else return {tautosyllabicipa} end else return {do_phoneticize(text, eccl, vul)} end end

local function make_row(phoneticizations, dials) local full_pronuns = {} for _, phoneticization in ipairs(phoneticizations) do		local phonemic = phoneticization[1] local phonetic = phoneticization[2] local IPA_args = table.insert(IPA_args, {pron = '[' .. phonetic .. ']'}) table.insert(full_pronuns, m_IPA.format_IPA_full { lang = lang, items = IPA_args }) end return m_a.format_qualifiers(lang, dials) .. ' ' .. table.concat(full_pronuns, ' or ') end

function export.show_full(frame) local params = { [1] = {default = mw.title.getCurrentTitle.nsText == 'Template' and 'īnspīrāre' or mw.title.getCurrentTitle.text}, classical = {type = "boolean", default = true}, cl = {type = "boolean", alias_of = "classical", default = true}, ecclesiastical = {type = "boolean", default = true}, eccl = {type = "boolean", alias_of = "ecclesiastical", default = true}, vul = {type = "boolean", default = false}, ann = {}, accent = {list = true}, indent = {} }	local parent_args = frame:getParent.args local function unrecognized_boolean(val) return val and val ~= "" and val ~= "1" and val ~= "0" and val ~= "yes" and val ~= "no" and val ~= "true" and val ~= "false" and val ~= "y" and val ~= "n" and val ~= "on" and val ~= "+" and val ~= "-" end

-- temporary tracking for strange boolean values for _, arg in ipairs {"classical", "cl", "ecclesiastical", "eccl", "vul"} do		if unrecognized_boolean(parent_args[arg]) then track("unrecognized-boolean") track("unrecognized-boolean/" .. arg) end end

local args = require("Module:parameters").process(parent_args, params) local text = args[1] local categories = {} local accent = args.accent

local indent = (args.indent or "*") .. " "	local out = '' if args.indent then out = indent end if args.classical then out = out .. make_row(export.phoneticize(text, false, false), #accent > 0 and accent or {'Classical'}) elseif not args.vul then table.insert(categories, lang:getCanonicalName .. ' terms with Ecclesiastical IPA pronunciation only') end local anntext = (		args.ann == "1" and "" .. rsub(text, "[.'_]", "") .. ":&#32;" or		args.ann and "" .. args.ann .. ":&#32;" or		"")

out = anntext .. out if args.ecclesiastical then if args.classical or args.vul then out = out .. '\n' .. indent .. anntext end out = out .. make_row(			export.phoneticize(text, true, false),			#accent > 0 and accent or {'Ecclesiastical'}		) end return out .. require("Module:utilities").format_categories(categories) end

function export.show(text, phonetic, eccl, vul) if type(text) == "table" then -- assume a frame eccl = text.args["eccl"] vul = text.args["vul"] text = text.args[1] or mw.title.getCurrentTitle.text end if vul then phonetic = true end return export.convert_words(text, phonetic, eccl, vul) end

function export.allophone(word, eccl, vul) return export.show(word, true, eccl, vul) end

return export