Module:User:Vampyricon/txb-noun/data

local decl = {}

local lang = require("Module:languages").getByCode("txb") local m_links = require("Module:links") local m_utilities = require("Module:utilities")

local rsplit = mw.text.split local rfind = mw.ustring.find local rmatch = mw.ustring.match local rsubn = mw.ustring.gsub

-- 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

local function glossary_link(anchor, text) text = text or anchor return "" .. text .. "" end

local current_title = mw.title.getCurrentTitle.nsText local NAMESPACE = current_title.nsText local PAGENAME = current_title.text

local abl = "meṃ" local all = "ś(c)" local prl = "sa" local loc = "ne" local com = "mpa"

decl["IV"] = function(data, args) local stem = args[1]

-- normal IV -- need to check if ending is polysyllabic (i.e. multiple vowels) -- if so, check to see if stem has ā or a, and change to a and ä resp. data.forms["nom_sg"] = stem .. "cer" data.forms["gen_sg"] = stem .. "tri" data.forms["obl_sg"] = stem .. "tär" data.forms["abl_sg"] = data.forms["obl_sg"] .. abl data.forms["all_sg"] = data.forms["obl_sg"] .. all data.forms["prl_sg"] = data.forms["obl_sg"] .. prl data.forms["loc_sg"] = data.forms["obl_sg"] .. loc data.forms["com_sg"] = data.forms["obl_sg"] .. com

data.forms["nom_pl"] = stem .. "cera" data.forms["gen_pl"] = stem .. "tärnts" data.forms["obl_pl"] = stem .. "cera" data.forms["abl_pl"] = data.forms["obl_pl"] .. abl data.forms["all_pl"] = data.forms["obl_pl"] .. all data.forms["prl_pl"] = data.forms["obl_pl"] .. prl data.forms["loc_pl"] = data.forms["obl_pl"] .. loc data.forms["com_pl"] = data.forms["obl_pl"] .. com end

decl["2"] = function(data, args) local stem1 = args[1] local stem2 = args[2]

-- normal 2nd data.forms["nom_sg"] = stem1 .. "us" data.forms["gen_sg"] = stem1 .. "ī" data.forms["dat_sg"] = stem1 .. "ō" data.forms["acc_sg"] = stem1 .. "um" data.forms["abl_sg"] = stem1 .. "ō" data.forms["voc_sg"] = stem1 .. "e"

data.forms["nom_pl"] = stem1 .. "ī" data.forms["gen_pl"] = stem1 .. "ōrum" data.forms["dat_pl"] = stem1 .. "īs" data.forms["acc_pl"] = stem1 .. "ōs" data.forms["abl_pl"] = stem1 .. "īs" data.forms["voc_pl"] = stem1 .. "ī"

-- all neuter if data.types.N then table.insert(data.subtitles, "neuter")

data.forms["nom_sg"] = stem1 .. "um" data.forms["voc_sg"] = stem1 .. "um"

data.forms["nom_pl"] = stem1 .. "a" data.forms["acc_pl"] = stem1 .. "a" data.forms["voc_pl"] = stem1 .. "a"

-- neuter ium if data.types.ium then data.forms["nom_sg"] = stem1 .. "ium" data.forms["gen_sg"] = {stem1 .. "iī", stem1 .. "ī"} data.forms["dat_sg"] = stem1 .. "iō" data.forms["acc_sg"] = stem1 .. "ium" data.forms["abl_sg"] = stem1 .. "iō" data.forms["voc_sg"] = stem1 .. "ium"

data.forms["nom_pl"] = stem1 .. "ia" data.forms["gen_pl"] = stem1 .. "iōrum" data.forms["dat_pl"] = stem1 .. "iīs" data.forms["acc_pl"] = stem1 .. "ia" data.forms["abl_pl"] = stem1 .. "iīs" data.forms["voc_pl"] = stem1 .. "ia"

data.notes["gen_sg2"] = "Found in older Latin (until the Augustan Age)." -- neuter us, plural a		elseif data.types.a then table.insert(data.subtitles, "nominative/accusative/vocative plural in -a")

data.forms["nom_sg"] = stem1 .. "us" data.forms["acc_sg"] = stem1 .. "us" data.forms["voc_sg"] = stem1 .. "us"

data.forms["nom_pl"] = stem1 .. "a" data.forms["acc_pl"] = stem1 .. "a" data.forms["voc_pl"] = stem1 .. "a" -- -vom (aevom, etc.) elseif data.types.vom then table.insert(data.subtitles, "nominative singular in -om after v") data.forms["nom_sg"] = stem1 .. "om" data.forms["acc_sg"] = stem1 .. "om" data.forms["voc_sg"] = stem1 .. "om"

-- neuter Greek in -os (cētos) elseif data.types.Greek and data.types.us then table.insert(data.subtitles, "Greek-type") table.insert(data.subtitles, "nominative/accusative/vocative in -os")

data.forms["nom_sg"] = stem1 .. "os" data.forms["acc_sg"] = stem1 .. "os" data.forms["voc_sg"] = stem1 .. "os"

data.forms["nom_pl"] = stem1 .. "ē" data.forms["gen_pl"] = stem1 .. "ōn" data.forms["acc_pl"] = stem1 .. "ē" data.forms["voc_pl"] = stem1 .. "ē"

-- neuter Greek elseif data.types.Greek then table.insert(data.subtitles, "Greek-type")

data.forms["nom_sg"] = stem1 .. "on" data.forms["acc_sg"] = stem1 .. "on" data.forms["voc_sg"] = stem1 .. "on"

-- neuter us		elseif data.types.us then table.insert(data.subtitles, "nominative/accusative/vocative in -us")

data.forms["nom_sg"] = stem1 .. "us" data.forms["acc_sg"] = stem1 .. "us" data.forms["voc_sg"] = stem1 .. "us"

data.forms["nom_pl"] = stem1 .. "ī" data.forms["acc_pl"] = stem1 .. "ōs" data.forms["voc_pl"] = stem1 .. "ī"

elseif data.types.not_Greek or data.types.not_us then table.insert(data.subtitles, "nominative/accusative/vocative in -um") end

-- er	elseif data.types.er then if mw.ustring.match(stem1, "[aiouy]r$") then table.insert(data.subtitles, "nominative singular in -r") else table.insert(data.subtitles, "nominative singular in -er") end

data.forms["nom_sg"] = stem1 data.forms["gen_sg"] = stem2 .. "ī" data.forms["dat_sg"] = stem2 .. "ō" data.forms["acc_sg"] = stem2 .. "um" data.forms["abl_sg"] = stem2 .. "ō" data.forms["voc_sg"] = stem1

data.forms["nom_pl"] = stem2 .. "ī" data.forms["gen_pl"] = stem2 .. "ōrum" data.forms["dat_pl"] = stem2 .. "īs" data.forms["acc_pl"] = stem2 .. "ōs" data.forms["abl_pl"] = stem2 .. "īs" data.forms["voc_pl"] = stem2 .. "ī"

-- ius elseif data.types.ius then data.forms["nom_sg"] = stem1 .. "ius" data.forms["gen_sg"] = {stem1 .. "iī", stem1 .. "ī"} data.forms["dat_sg"] = stem1 .. "iō" data.forms["acc_sg"] = stem1 .. "ium" data.forms["abl_sg"] = stem1 .. "iō" if data.types.voci then -- Only for proper names and fīlius, genius data.forms["voc_sg"] = stem1 .. "ī" else data.forms["voc_sg"] = stem1 .. "ie" end

data.forms["nom_pl"] = stem1 .. "iī" data.forms["gen_pl"] = stem1 .. "iōrum" data.forms["dat_pl"] = stem1 .. "iīs" data.forms["acc_pl"] = stem1 .. "iōs" data.forms["abl_pl"] = stem1 .. "iīs" data.forms["voc_pl"] = stem1 .. "iī"

data.notes["gen_sg2"] = "Found in older Latin (until the Augustan Age)."

-- -vos (servos, etc.) elseif data.types.vos then table.insert(data.subtitles, "nominative singular in -os after v") data.forms["nom_sg"] = stem1 .. "os" data.forms["acc_sg"] = stem1 .. "om"

-- Greek elseif data.types.Greek then table.insert(data.subtitles, "Greek-type")

data.forms["nom_sg"] = stem1 .. "os" data.forms["acc_sg"] = {stem1 .. "on"}

elseif data.types.not_Greek then table.insert(data.subtitles, "non-Greek-type") end

-- with -um genitive plural if data.types.genplum then table.insert(data.subtitles, {"contracted", " genitive plural"}) data.notes["gen_pl2"] = "Contraction found in poetry." if data.types.ius or data.types.ium then data.forms["gen_pl"] = {stem2 .. "iōrum", stem2 .. "ium"} else data.forms["gen_pl"] = {stem2 .. "ōrum", stem2 .. "um"} end elseif data.types.not_genplum then table.insert(data.subtitles, {"normal", " genitive plural"}) end end

local acc_sg_i_stem_subtypes = { acc_im = { -- amussis, basis, buris, cucumis, gummis, mephitis, paraphrasis, poesis, ravis, sitis, tussis, (vis) [abl -ī]; -- cannabis, senapis, sinapis [abl -e, -ī] acc_sg = {"im"}, title = {"accusative singular in -im"}, },	acc_im_in = { -- cities, rivers, gods, e.g. Bilbilis, Syrtis, Tiberis, Anubis, Osiris [abl -ī]; -- Baetis, Tigris [acc -e, -ī] acc_sg = {"im", "in"}, title = {"accusative singular in -im or -in"}, },	acc_im_in_em = { -- e.g. tigris, river Līris acc_sg = {"im", "in", "em"}, title = {"accusative singular in -im, -in or -em"}, },	acc_im_em = { acc_sg = {"im", "em"}, title = {"accusative singular in -im or -em"}, },	acc_im_occ_em = { -- febris, pelvis, puppis, restis, securis, turris [abl -ī, -e] acc_sg = {"im", "em"}, title = {"accusative singular in -im or occasionally -em"}, },	acc_em_im = { -- aqualis, clavis, lens, navis [abl -e, -ī]; -- cutis, restis [abl -e] acc_sg = {"em", "im"}, title = {"accusative singular in -em or -im"}, }, }

local abl_sg_i_stem_subtypes = { abl_i = { -- amussis, basis, buris, cucumis, gummis, mephitis, paraphrasis, poesis, ravis, sitis, tussis, (vis) [acc -im]; -- cities, rivers, gods, e.g. Bilbilis, Syrtis, Tiberis, Anubis, Osiris [acc -im or -in]; -- canalis "water pipe", months in -is or -er, nouns originally i-stem adjectives such as aedilis, affinis, bipennis, familiaris, sodalis, volucris, etc. [acc -em] abl_sg = {"ī"}, title = {"ablative singular in -ī"}, },	abl_i_e = { -- febris, pelvis, puppis, restis, securis, turris [acc -im, -em] abl_sg = {"ī", "e"}, title = {"ablative singular in -ī or -e"}, },	abl_e_i = { -- cannabis, senapis, sinapis [acc -im]; -- Baetis, Tigris [acc -im, -in]; -- aqualis, clavis, lens, navis [acc -em, -im]; -- finis, mugilis, occiput, pugil, rus, supellex, vectis [acc -em] abl_sg = {"e", "ī"}, title = {"ablative singular in -e or -ī"}, },	abl_e_occ_i = { -- amnis, anguis, avis, civis, classis, fustis, ignis, imber, orbis, pars, postis, sors, unguis, vesper [acc -em] abl_sg = {"e", "ī"}, title = {"ablative singular in -e or occasionally -ī"}, }, }

local function extract_stem(form, ending) local base = rmatch(form, "^(.*)" .. ending .. "$") if not base then error("Form " .. form .. " should end in -" .. ending) end return base end

decl["3"] = function(data, args) local stem1 = args[1] local stem2 = args[2]

local function parisyllabic_type local stem1_vowels = rsub(			require("Module:la-utilities").strip_macrons(stem1),			"[^AEIOUYaeiouy]", ""		) local stem2_vowels = rsub(			require("Module:la-utilities").strip_macrons(stem2),			"[^AEIOUYaeiouy]", ""		) return #stem1_vowels > #stem2_vowels and "parisyllabic" or "imparisyllabic" end

local function non_i_stem_type return parisyllabic_type .. " non-i-stem" end

--normal 3rd data.forms["nom_sg"] = stem1 data.forms["gen_sg"] = stem2 .. "is" data.forms["dat_sg"] = stem2 .. "ī" data.forms["acc_sg"] = stem2 .. "em" data.forms["abl_sg"] = stem2 .. "e" data.forms["voc_sg"] = stem1

data.forms["nom_pl"] = stem2 .. "ēs" data.forms["gen_pl"] = stem2 .. "um" data.forms["dat_pl"] = stem2 .. "ibus" data.forms["acc_pl"] = stem2 .. "ēs" data.forms["abl_pl"] = stem2 .. "ibus" data.forms["voc_pl"] = stem2 .. "ēs"

local acc_sg_i_stem_subtype = false local not_acc_sg_i_stem_subtype = false for subtype, _ in pairs(data.types) do		if acc_sg_i_stem_subtypes[subtype] then acc_sg_i_stem_subtype = true break end end for acc_sg_subtype, _ in pairs(acc_sg_i_stem_subtypes) do if data.types["not_" .. acc_sg_subtype] then not_acc_sg_i_stem_subtype = true break end end local abl_sg_i_stem_subtype = false local not_abl_sg_i_stem_subtype = false for subtype, _ in pairs(data.types) do		if abl_sg_i_stem_subtypes[subtype] then abl_sg_i_stem_subtype = true break end end for abl_sg_subtype, _ in pairs(abl_sg_i_stem_subtypes) do if data.types["not_" .. abl_sg_subtype] then not_abl_sg_i_stem_subtype = true break end end

-- all Greek if data.types.Greek then table.insert(data.subtitles, "Greek-type")

-- Greek er		if data.types.er then table.insert(data.subtitles, "variant with nominative singular in -ēr") stem1 = extract_stem(stem1, "ēr")

data.forms["nom_sg"] = stem1 .. "ēr" data.forms["gen_sg"] = stem1 .. "eris" data.forms["dat_sg"] = stem1 .. "erī" data.forms["acc_sg"] = {stem1 .. "era", stem1 .. "erem"} data.forms["abl_sg"] = stem1 .. "ere" data.forms["voc_sg"] = stem1 .. "ēr"

data.forms["nom_pl"] = stem1 .. "erēs" data.forms["gen_pl"] = stem1 .. "erum" data.forms["dat_pl"] = stem1 .. "eribus" data.forms["acc_pl"] = stem1 .. "erēs" data.forms["abl_pl"] = stem1 .. "eribus" data.forms["voc_pl"] = stem1 .. "erēs"

-- Greek on		elseif data.types.on then table.insert(data.subtitles, "variant with nominative singular in -ōn") stem1 = extract_stem(stem1, "ōn")

data.forms["nom_sg"] = stem1 .. "ōn" data.forms["gen_sg"] = {stem1 .. "ontis", stem1 .. "ontos"} data.forms["dat_sg"] = stem1 .. "ontī" data.forms["acc_sg"] = stem1 .. "onta" data.forms["abl_sg"] = stem1 .. "onte" data.forms["voc_sg"] = stem1 .. "ōn"

data.forms["nom_pl"] = stem1 .. "ontēs" data.forms["gen_pl"] = {stem1 .. "ontum", stem1 .. "ontium"} data.forms["dat_pl"] = stem1 .. "ontibus" data.forms["acc_pl"] = {stem1 .. "ontēs", stem1 .. "ontās"} data.forms["abl_pl"] = stem1 .. "ontibus" data.forms["voc_pl"] = stem1 .. "ontēs"

-- Greek i-stem elseif data.types.I then table.insert(data.subtitles, "i-stem") data.forms["gen_sg"] = {stem2 .. "is", stem2 .. "eōs", stem2 .. "ios"} data.forms["acc_sg"] = {stem2 .. "im", stem2 .. "in", stem2 .. "em"} data.forms["abl_sg"] = {stem2 .. "ī", stem2 .. "e"} data.forms["voc_sg"] = {stem2 .. "is", stem2 .. "i"}

data.notes["acc_sg3"] = "Found sometimes in Medieval and New Latin." data.notes["abl_sg2"] = "Found sometimes in Medieval and New Latin."

data.forms["nom_pl"] = {stem2 .. "ēs", stem2 .. "eis"} data.forms["gen_pl"] = {stem2 .. "ium", stem2 .. "eōn"} data.forms["acc_pl"] = {stem2 .. "ēs", stem2 .. "eis"} data.forms["voc_pl"] = {stem2 .. "ēs", stem2 .. "eis"}

if data.types.poetic_esi then data.forms["dat_pl"] = {stem2 .. "ibus", stem2 .. "esi"} data.forms["abl_pl"] = {stem2 .. "ibus", stem2 .. "esi"} data.notes["dat_pl2"] = "Primarily in poetry." data.notes["abl_pl2"] = "Primarily in poetry." end

-- normal Greek else table.insert(data.subtitles, "normal variant") data.forms["gen_sg"] = stem2 .. "os" if stem2:find("y$") then data.forms["acc_sg"] = stem2 .. "n" else data.forms["acc_sg"] = stem2 .. "a" end

data.forms["nom_pl"] = stem2 .. "es" data.forms["acc_pl"] = stem2 .. "as" data.forms["voc_pl"] = stem2 .. "es" if rfind(stem1, "[iyï]s$") then -- Per Hiley, words in -is and -ys have a poetic vocative -- without the -s, but otherwise the vocative is the same -- as the nominative. data.forms["voc_sg"] = {stem1, rsub(stem1, "s", "")} data.notes["voc_sg2"] = "In poetry." end end elseif data.types.not_Greek then table.insert(data.subtitles, "non-Greek-type") end

-- polis if data.types.polis then stem1 = extract_stem(stem1, "polis") table.insert(data.subtitles, "i-stem, partially Greek-type")

data.forms["nom_sg"] = stem1 .. "polis" data.forms["gen_sg"] = stem1 .. "polis" data.forms["dat_sg"] = stem1 .. "polī" data.forms["acc_sg"] = {stem1 .. "polim", stem1 .. "polin"} data.forms["abl_sg"] = stem1 .. "polī" data.forms["voc_sg"] = {stem1 .. "polis", stem1 .. "polī"} elseif data.types.not_polis then table.insert(data.subtitles, non_i_stem_type) end

-- all neuter if data.types.N then table.insert(data.subtitles, "neuter")

data.forms["acc_sg"] = stem1

-- neuter I stem if data.types.I then

-- pure variety if data.types.pure then table.insert(data.subtitles, "“pure” i-stem") data.forms["abl_sg"] = stem2 .. "ī"

data.forms["nom_pl"] = stem2 .. "ia" data.forms["gen_pl"] = stem2 .. "ium" data.forms["acc_pl"] = stem2 .. "ia" data.forms["voc_pl"] = stem2 .. "ia"

-- non-pure variety (rare) else table.insert(data.subtitles, "i-stem") data.forms["nom_pl"] = stem2 .. "a" data.forms["gen_pl"] = {stem2 .. "ium", stem2 .. "um"} data.forms["acc_pl"] = stem2 .. "a" data.forms["voc_pl"] = stem2 .. "a" end

-- normal neuter else table.insert(data.subtitles, non_i_stem_type) data.forms["nom_pl"] = stem2 .. "a" data.forms["acc_pl"] = stem2 .. "a" data.forms["voc_pl"] = stem2 .. "a" end

-- I stem elseif data.types.I or acc_sg_i_stem_subtype or abl_sg_i_stem_subtype then if data.types.not_N then table.insert(data.subtitles, "non-neuter i-stem") else table.insert(data.subtitles, "i-stem") end

data.forms["gen_pl"] = stem2 .. "ium" -- Per Allen and Greenough, Hiley and others, the acc_pl in -īs -- applied originally to all i-stem nouns, and was current as an -- alternative form up through Caesar. data.forms["acc_pl"] = {stem2 .. "ēs", stem2 .. "īs"}

for subtype, _ in pairs(data.types) do			local acc_sg_i_stem_props = acc_sg_i_stem_subtypes[subtype] if acc_sg_i_stem_props then data.forms["acc_sg"] = {} for _, ending in ipairs(acc_sg_i_stem_props.acc_sg) do table.insert(data.forms["acc_sg"], stem2 .. ending) end if data.num ~= "pl" then for _, t in ipairs(acc_sg_i_stem_props.title) do						table.insert(data.subtitles, t)					end end break end end

for subtype, _ in pairs(data.types) do			local abl_sg_i_stem_props = abl_sg_i_stem_subtypes[subtype] if abl_sg_i_stem_props then data.forms["abl_sg"] = {} for _, ending in ipairs(abl_sg_i_stem_props.abl_sg) do table.insert(data.forms["abl_sg"], stem2 .. ending) end if data.num ~= "pl" then for _, t in ipairs(abl_sg_i_stem_props.title) do						table.insert(data.subtitles, t)					end end break end end elseif data.types.not_N and data.types.not_I then table.insert(data.subtitles, "non-neuter " .. non_i_stem_type) elseif data.types.not_N then table.insert(data.subtitles, "non-neuter") elseif data.types.not_I then table.insert(data.subtitles, non_i_stem_type) end end

decl["4"] = function(data, args) local stem = args[1]

-- normal 4th data.forms["nom_sg"] = stem .. "us" data.forms["gen_sg"] = stem .. "ūs" data.forms["dat_sg"] = stem .. "uī" data.forms["acc_sg"] = stem .. "um" data.forms["abl_sg"] = stem .. "ū" data.forms["voc_sg"] = stem .. "us"

data.forms["nom_pl"] = stem .. "ūs" data.forms["gen_pl"] = stem .. "uum" data.forms["dat_pl"] = stem .. "ibus" data.forms["acc_pl"] = stem .. "ūs" data.forms["abl_pl"] = stem .. "ibus" data.forms["voc_pl"] = stem .. "ūs"

if data.types.echo then table.insert(data.subtitles, "nominative/vocative singular in -ō") data.forms["nom_sg"] = stem .. "ō" data.forms["voc_sg"] = stem .. "ō" elseif data.types.argo then table.insert(data.subtitles, "nominative/accusative/vocative singular in -ō, ablative singular in -uī") data.forms["nom_sg"] = stem .. "ō" data.forms["acc_sg"] = stem .. "ō" data.forms["abl_sg"] = stem .. "uī" data.forms["voc_sg"] = stem .. "ō" elseif data.types.Callisto then table.insert(data.subtitles, "all cases except the genitive singular in -ō") data.forms["nom_sg"] = stem .. "ō" data.forms["dat_sg"] = stem .. "ō" data.forms["acc_sg"] = stem .. "ō" data.forms["abl_sg"] = stem .. "ō" data.forms["voc_sg"] = stem .. "ō" end

-- neuter if data.types.N then table.insert(data.subtitles, "neuter")

data.forms["nom_sg"] = stem .. "ū" data.forms["dat_sg"] = stem .. "ū" data.forms["acc_sg"] = stem .. "ū" data.forms["voc_sg"] = stem .. "ū"

data.forms["nom_pl"] = stem .. "ua" data.forms["acc_pl"] = stem .. "ua" data.forms["voc_pl"] = stem .. "ua" end

-- ubus if data.types.ubus then table.insert(data.subtitles, "dative/ablative plural in -ubus")

data.forms["dat_pl"] = stem .. "ubus" data.forms["abl_pl"] = stem .. "ubus" elseif data.types.not_ubus then table.insert(data.subtitles, "-ibus") end end

decl["5"] = function(data, args) local stem = args[1]

-- ies if data.types.i then stem = stem .. "i" end

data.forms["nom_sg"] = stem .. "ēs" data.forms["gen_sg"] = stem .. "eī" data.forms["dat_sg"] = stem .. "eī" data.forms["acc_sg"] = stem .. "em" data.forms["abl_sg"] = stem .. "ē" data.forms["voc_sg"] = stem .. "ēs"

data.forms["nom_pl"] = stem .. "ēs" data.forms["gen_pl"] = stem .. "ērum" data.forms["dat_pl"] = stem .. "ēbus" data.forms["acc_pl"] = stem .. "ēs" data.forms["abl_pl"] = stem .. "ēbus" data.forms["voc_pl"] = stem .. "ēs"

-- ies if data.types.i then data.forms["gen_sg"] = stem .. "ēī" data.forms["dat_sg"] = stem .. "ēī" end end

decl["0"] = function(data, args) local stem = args[1]

data.forms["nom_sg"] = stem data.forms["gen_sg"] = stem data.forms["dat_sg"] = stem data.forms["acc_sg"] = stem data.forms["abl_sg"] = stem data.forms["voc_sg"] = stem

data.forms["nom_pl"] = stem data.forms["gen_pl"] = stem data.forms["dat_pl"] = stem data.forms["acc_pl"] = stem data.forms["abl_pl"] = stem data.forms["voc_pl"] = stem end

decl["indecl"] = function(data, args) local title = {} data.title = "Not declined; used only in the nominative and accusative singular"

local stem = args[1]

data.forms["nom_sg"] = "-" data.forms["gen_sg"] = "-" data.forms["dat_sg"] = "-" data.forms["acc_sg"] = "-" data.forms["abl_sg"] = "-" data.forms["voc_sg"] = "-"

data.forms["nom_pl"] = "-" data.forms["gen_pl"] = "-" data.forms["dat_pl"] = "-" data.forms["acc_pl"] = "-" data.forms["abl_pl"] = "-" data.forms["voc_pl"] = "-"

data.forms["nom_sg"] = stem data.forms["acc_sg"] = stem data.num = "sg" end

decl["irreg"] = function(data, args) local stem = args[1]

data.forms["nom_sg"] = "-" data.forms["gen_sg"] = "-" data.forms["dat_sg"] = "-" data.forms["acc_sg"] = "-" data.forms["abl_sg"] = "-" data.forms["voc_sg"] = "-"

data.forms["nom_pl"] = "-" data.forms["gen_pl"] = "-" data.forms["dat_pl"] = "-" data.forms["acc_pl"] = "-" data.forms["abl_pl"] = "-" data.forms["voc_pl"] = "-"

if stem == "bōs" then data.forms["nom_sg"] = "bōs" data.forms["gen_sg"] = "bovis" data.forms["dat_sg"] = "bovī" data.forms["acc_sg"] = "bovem" data.forms["abl_sg"] = "bove" data.forms["voc_sg"] = "bōs"

data.forms["nom_pl"] = "bovēs" data.forms["gen_pl"] = "boum" data.forms["dat_pl"] = {"bōbus", "būbus"} data.forms["acc_pl"] = "bovēs" data.forms["abl_pl"] = {"bōbus", "būbus"} data.forms["voc_pl"] = "bovēs"

elseif stem == "cherub" then data.title = "mostly " .. glossary_link("indeclinable") table.insert(data.subtitles, "with a distinct plural")

data.forms["nom_sg"] = "cherub" data.forms["gen_sg"] = "cherub" data.forms["dat_sg"] = "cherub" data.forms["acc_sg"] = "cherub" data.forms["abl_sg"] = "cherub" data.forms["voc_sg"] = "cherub"

data.forms["nom_pl"] = {"cherubim", "cherubin"} data.forms["gen_pl"] = {"cherubim", "cherubin"} data.forms["dat_pl"] = {"cherubim", "cherubin"} data.forms["acc_pl"] = {"cherubim", "cherubin"} data.forms["abl_pl"] = {"cherubim", "cherubin"} data.forms["voc_pl"] = {"cherubim", "cherubin"}

elseif stem == "deus" then data.forms["nom_sg"] = "deus" data.forms["gen_sg"] = "deī" data.forms["dat_sg"] = "deō" data.forms["acc_sg"] = "deum" data.forms["abl_sg"] = "deō" data.forms["voc_sg"] = {"dee", "deus"}

data.forms["nom_pl"] = {"dī", "diī", "deī"} data.forms["gen_pl"] = {"deōrum", "deûm", "divom"} data.forms["dat_pl"] = {"dīs", "diīs", "deīs"} data.forms["acc_pl"] = "deōs" data.forms["abl_pl"] = {"dīs", "diīs", "deīs"} data.forms["voc_pl"] = {"dī", "diī", "deī"}

elseif stem == "Deus" then data.forms["nom_sg"] = "Deus" data.forms["gen_sg"] = "Deī" data.forms["dat_sg"] = "Deō" data.forms["acc_sg"] = "Deum" data.forms["abl_sg"] = "Deō" data.forms["voc_sg"] = "Deus" data.num = "sg"

elseif stem == "domus" then data.title = "fourth/second-declension noun" data.forms["nom_sg"] = "domus" data.forms["gen_sg"] = {"domūs", "domī"} data.forms["dat_sg"] = {"domuī", "domō", "domū"} data.forms["acc_sg"] = "domum" data.forms["abl_sg"] = {"domū", "domō"} data.forms["voc_sg"] = "domus" data.forms["loc_sg"] = "domī"

data.forms["nom_pl"] = "domūs" data.forms["gen_pl"] = {"domuum", "domōrum"} data.forms["dat_pl"] = "domibus" data.forms["acc_pl"] = {"domūs", "domōs"} data.forms["abl_pl"] = "domibus" data.forms["voc_pl"] = "domūs" data.forms["loc_pl"] = "domibus"

data.loc = true

elseif stem == "Iēsus" or stem == "Iēsūs" then table.insert(data.subtitles, "highly " .. glossary_link("irregular")) ij = mw.ustring.sub(stem, 1, 1)

data.forms["nom_sg"] = stem data.forms["gen_sg"] = ij .. "ēsū" data.forms["dat_sg"] = ij .. "ēsū" data.forms["acc_sg"] = ij .. "ēsum" data.forms["abl_sg"] = ij .. "ēsū" data.forms["voc_sg"] = ij .. "ēsū" data.num = "sg" elseif stem == "Jesus" then table.insert(data.subtitles, "highly " .. glossary_link("irregular")) ij = mw.ustring.sub(stem, 1, 1)

data.forms["nom_sg"] = stem data.forms["gen_sg"] = ij .. "esu" data.forms["dat_sg"] = ij .. "esu" data.forms["acc_sg"] = ij .. "esum" data.forms["abl_sg"] = ij .. "esu" data.forms["voc_sg"] = ij .. "esu" data.num = "sg" elseif stem == "Jésus" then table.insert(data.subtitles, "highly " .. glossary_link("irregular")) ij = mw.ustring.sub(stem, 1, 1)

data.forms["nom_sg"] = stem data.forms["gen_sg"] = ij .. "ésu" data.forms["dat_sg"] = ij .. "ésu" data.forms["acc_sg"] = ij .. "ésum" data.forms["abl_sg"] = ij .. "ésu" data.forms["voc_sg"] = ij .. "ésu" data.num = "sg" elseif stem == "iūgerum" or stem == "jūgerum" then ij = stem == "iūgerum" and "i" or "j" data.title = "second–third-declension hybrid noun" table.insert(data.subtitles, "neuter")

data.forms["nom_sg"] = ij .. "ūgerum" data.forms["gen_sg"] = ij .. "ūgerī" data.forms["dat_sg"] = ij .. "ūgerō" data.forms["acc_sg"] = ij .. "ūgerum" data.forms["abl_sg"] = ij .. "ūgerō" data.forms["voc_sg"] = ij .. "ūgerum" data.forms["nom_pl"] = ij .. "ūgera" data.forms["gen_pl"] = ij .. "ūgerum" data.forms["dat_pl"] = ij .. "ūgeribus" data.forms["acc_pl"] = ij .. "ūgera" data.forms["abl_pl"] = {ij .. "ūgeribus", ij .. "ūgerīs"} data.forms["voc_pl"] = ij .. "ūgera"

data.notes["abl_pl2"] = "Once only, in: M. Terentius Varro, Res Rusticae, bk I, ch. x"

elseif stem == "sūs" then data.forms["nom_sg"] = "sūs" data.forms["gen_sg"] = "suis" data.forms["dat_sg"] = "suī" data.forms["acc_sg"] = "suem" data.forms["abl_sg"] = "sue" data.forms["voc_sg"] = "sūs"

data.forms["nom_pl"] = "suēs" data.forms["gen_pl"] = "suum" data.forms["dat_pl"] = {"suibus", "sūbus", "subus"} data.forms["acc_pl"] = "suēs" data.forms["abl_pl"] = {"suibus", "sūbus", "subus"} data.forms["voc_pl"] = "suēs"

elseif stem == "ēthos" then table.insert(data.subtitles, glossary_link("irregular")) table.insert(data.subtitles, "Greek-type")

data.forms["nom_sg"] = "ēthos" data.forms["gen_sg"] = "ētheos" data.forms["acc_sg"] = "ēthos" data.forms["voc_sg"] = "ēthos"

data.forms["nom_pl"] = {"ēthea", "ēthē"} data.forms["dat_pl"] = {"ēthesi", "ēthesin"} data.forms["acc_pl"] = {"ēthea", "ēthē"} data.forms["abl_pl"] = {"ēthesi", "ēthesin"} data.forms["voc_pl"] = {"ēthea", "ēthē"}

elseif stem == "lexis" then table.insert(data.subtitles, glossary_link("irregular")) table.insert(data.subtitles, "Greek-type")

data.forms["nom_sg"] = "lexis" data.forms["gen_sg"] = "lexeōs" data.forms["acc_pl"] = "lexeis"

elseif stem == "Athōs" then table.insert(data.subtitles, "highly " .. glossary_link("irregular")) table.insert(data.subtitles, "Greek-type")

data.forms["nom_sg"] = "Athōs" data.forms["gen_sg"] = "Athō" data.forms["dat_sg"] = "Athō" data.forms["acc_sg"] = {"Athō", "Athōn"} data.forms["abl_sg"] = "Athō" data.forms["voc_sg"] = "Athōs" data.num = "sg" elseif stem == "Pammenēs" then data.title = "Variably declined either according to Latin’s third declension or in a manner adapted from Ancient Greek’s first declension" data.forms["nom_sg"] = "Pammenēs" data.forms["gen_sg"] = {"Pammenis", "Pammenī"} data.forms["dat_sg"] = {"Pammenī", "Pammenē"} data.forms["acc_sg"] = {"Pammenem", "Pammenēn"} data.forms["abl_sg"] = "Pammenē̆" data.forms["voc_sg"] = {"Pammenēs", "Pammenē"} data.num = "sg"

elseif stem == "vēnum" then data.title = "fourth/second-declension noun" table.insert(data.subtitles, glossary_link("defective"))

data.forms["dat_sg"] = {"vēnuī", "vēnō"} data.forms["acc_sg"] = "vēnum" data.num = "sg"

elseif stem == "vīs" then table.insert(data.subtitles, glossary_link("irregular")) table.insert(data.subtitles, glossary_link("defective"))

data.forms["nom_sg"] = "vīs" data.forms["gen_sg"] = "*vīs" data.forms["dat_sg"] = "*vī" data.forms["acc_sg"] = "vim" data.forms["abl_sg"] = "vī" data.forms["voc_sg"] = "vīs"

data.forms["nom_pl"] = "vīrēs" data.forms["gen_pl"] = "vīrium" data.forms["dat_pl"] = "vīribus" data.forms["acc_pl"] = {"vīrēs", "vīrīs"} data.forms["abl_pl"] = "vīribus" data.forms["voc_pl"] = "vīrēs"

else error("Stem " .. stem .. " not recognized.") end end

return decl

-- For Vim, so we get 4-space tabs -- vim: set ts=4 sw=4 noet: