Module:User:Benwing2/ar-verb
- This module sandbox lacks a documentation subpage. You may create it.
- Useful links: root page • root page’s subpages • links • transclusions • testcases • user page • user talk page • userspace
This is a private module sandbox of Benwing2, for their own experimentation. Items in this module may be added and removed at Benwing2's discretion; do not rely on this module's stability.
local export = {}
--[=[
This module implements {{ar-conj}} and provides the underlying conjugation functions for {{ar-verb}}
(whose actual formatting is done in [[Module:ar-headword]]).
Author: User:Benwing, from an early version (2013-2014) by User:Atitarev, User:ZxxZxxZ.
]=]
--[=[
TERMINOLOGY:
-- "slot" = A particular combination of tense/mood/person/number/etc.
Example slot names for verbs are "past_1s" (past tense first-person singular), "juss_pass_3fp" (non-past jussive
passive third-person feminine plural) "ap" (active participle). Each slot is filled with zero or more forms.
-- "form" = The conjugated Arabic form representing the value of a given slot.
-- "lemma" = The dictionary form of a given Arabic term. For Arabic, normally the third person masculine singular past,
although other forms may be used if this form is missing (e.g. in passive-only verbs or verbs lacking the past).
]=]
--[=[
FIXME:
1. Finish unimplemented conjugation types. Only IX-final-weak left (extremely rare, possibly only one verb اِعْمَايَ
(according to Haywood and Nahmad p. 244, who are very specific about the irregular occurrence of alif + yā instead
of expected اِعْمَيَّ with doubled yā). Not in Hans Wehr. NOTE: Not true about this, cf. form IX اِرْعَوَى "to desist,
to repent, to see the light". Also note form XII اِخْضَوْضَرَ = form IX اِخْضَرَّ "to be or become green".
[DONE except for اِعْمَايَ]
2. Implement irregular verbs as special cases and recognize them, e.g.
-- laysa "to not be"; only exists in the past tense, no non-past, no imperative, no participles, no passive, no
verbal noun. Irregular alternation las-/lays-. [IMPLEMENTABLE USING OVERRIDES]
-- istaḥā yastaḥī "be ashamed of" -- this is complex according to Hans Wehr because there are two verbs, regular
istaḥyā yastaḥyī "to spare (someone)'s life" and irregular istaḥyā yastaḥyī "to be ashamed to face (someone)",
which is irregular because it has the alternate irregular form istaḥā yastaḥī which only applies to this meaning.
Currently we follow Haywood and Nahmad in saying that both varieties can be spelled istaḥyā/istaḥā/istaḥḥā, but we
should instead use a variant= param similar to حَيَّ to distinguish the two possibilities, and maybe not include
istaḥḥā.
-- ʿayya/ʿayiya yaʿayyu/yaʿyā "to not find the right way, be incapable of, stammer, falter, fall ill". This appears
to be a mixture of a geminate and final-weak verb. Unclear what the whole paradigm looks like. Do the
consonant-ending parts in the past follow the final-weak paradigm? Is it the same in the non-past? Or can you
conjugate the non-past fully as either geminate or final-weak?
-- اِنْمَحَى inmaḥā or يمَّحَى immaḥā "to be effaced, obliterated; to disappear, vanish" has irregular assimilation of inm-
to imm- as an alternative. inmalasa "to become smooth; to glide; to slip away; to escape" also has immalasa as an
alternative. The only other form VII verbs in Hans Wehr beginning with -m- are inmalaḵa "to be pulled out, torn
out, wrenched" and inmāʿa "to be melted, to melt, to dissolve", which are not listed with imm- alternatives, but
might have them; if so, we should handle this generally. [DONE]
-- يَرَعَ yaraʕa yariʕu "to be a coward, to be chickenhearted" as an alternative form of يَرِعَ yariʕa yayraʕu (as given in
Wehr). [IMPLEMENTABLE USING OVERRIDES]
3. Implement individual override parameters for each paradigm part. See Module:fro-verb for an example of how to do this
generally. Note that {{temp|ar-conj-I}} and other of the older templates already had such individual override params.
[DONE]
Irregular verbs already implemented:
-- [ḥayya/ḥayiya yaḥyā "live" -- behaves like a normal final-weak verb
(e.g. past first singular ḥayītu) except in the past-tense parts with
vowel-initial endings (all the third person except for the third feminine
plural). The normal singular and dual endings have -yiya- in them, which
compresses to -yya-, with the normal endings the less preferred ones.
In masculine third plural, expected ḥayū is replaced by ḥayyū by
analogy to the -yy- parts, and the regular form is not given as an
alternant in John Mace. Barron's 201 verbs appears to have the regular
ḥayū as the part, however. Note also that final -yā appears with tall
alif. This appears to be a spelling convention of Arabic, also applying
in ḥayyā (form II, "to keep (someone) alive") and 'aḥyā (form IV,
"to animate, revive, give birth to, give new life to").] -- implemented
-- [ittaxadha yattaxidhu "take"] -- implemented
-- [sa'ala yas'alu "ask" with alternative jussive/imperative yasal/sal] -- implemented
-- [ra'ā yarā "see"] -- implemented
-- ['arā yurī "show"] -- implemented
-- ['akala ya'kulu "eat" with imperative kul] -- implemented
-- ['axadha ya'xudhu "take" with imperative xudh] -- implemented
-- ['amara ya'muru "order" with imperative mur] -- implemented
--]=]
local force_cat = false -- set to true for debugging
-- if true, always maintain manual translit during processing, and compare against full translit at the end
local debug_translit = false
local lang = require("Module:languages").getByCode("ar")
local m_links = require("Module:links")
local m_string_utilities = require("Module:string utilities")
local m_table = require("Module:table")
local ar_utilities = require("Module:ar-utilities")
local ar_nominals = require("Module:ar-nominals")
local iut = require("Module:inflection utilities")
local parse_utilities_module = "Module:parse utilities"
local pron_qualifier_module = "Module:pron qualifier"
local rfind = m_string_utilities.find
local rsubn = m_string_utilities.gsub
local rmatch = m_string_utilities.match
local rsplit = m_string_utilities.split
local usub = m_string_utilities.sub
local ulen = m_string_utilities.len
local u = m_string_utilities.char
local dump = mw.dumpObject
-- Within this module, conjugations are the functions that do the actual
-- conjugating by creating the parts of a basic verb.
-- They are defined further down.
local conjugations = {}
-- hamza variants
local HAMZA = u(0x0621) -- hamza on the line (stand-alone hamza) = ء
local HAMZA_ON_ALIF = u(0x0623)
local HAMZA_ON_W = u(0x0624)
local HAMZA_UNDER_ALIF = u(0x0625)
local HAMZA_ON_Y = u(0x0626)
local HAMZA_ANY = "[" .. HAMZA .. HAMZA_ON_ALIF .. HAMZA_UNDER_ALIF .. HAMZA_ON_W .. HAMZA_ON_Y .. "]"
local HAMZA_PH = u(0xFFF0) -- hamza placeholder
local BAD = u(0xFFF1)
local BORDER = u(0xFFF2)
-- diacritics
local A = u(0x064E) -- fatḥa
local AN = u(0x064B) -- fatḥatān (fatḥa tanwīn)
local U = u(0x064F) -- ḍamma
local UN = u(0x064C) -- ḍammatān (ḍamma tanwīn)
local I = u(0x0650) -- kasra
local IN = u(0x064D) -- kasratān (kasra tanwīn)
local SK = u(0x0652) -- sukūn = no vowel
local SH = u(0x0651) -- šadda = gemination of consonants
local DAGGER_ALIF = u(0x0670)
local DIACRITIC_ANY_BUT_SH = "[" .. A .. I .. U .. AN .. IN .. UN .. SK .. DAGGER_ALIF .. "]"
-- Pattern matching short vowels
local AIU = "[" .. A .. I .. U .. "]"
-- Pattern matching short vowels or sukūn
local AIUSK = "[" .. A .. I .. U .. SK .. "]"
-- Pattern matching any diacritics that may be on a consonant
local DIACRITIC = SH .. "?" .. DIACRITIC_ANY_BUT_SH
-- translit_patterns
local vowels = "aeiouāēīōū"
local NV = "[^" .. vowels .. "]"
local dia = {a = A, i = I, u = U}
local undia = {[A] = "a", [I] = "i", [U] = "u", ["-"] = "-"}
-- various letters and signs
local ALIF = u(0x0627) -- ʾalif = ا
local AMAQ = u(0x0649) -- ʾalif maqṣūra = ى
local AMAD = u(0x0622) -- ʾalif madda = آ
local TAM = u(0x0629) -- tāʾ marbūṭa = ة
local T = u(0x062A) -- tāʾ = ت
local HYPHEN = u(0x0640)
local N = u(0x0646) -- nūn = ن
local W = u(0x0648) -- wāw = و
local Y = u(0x064A) -- yāʾ = ي
local S = "س"
local M = "م"
local LRM = u(0x200e) -- left-to-right mark
-- common combinations
local AH = A .. TAM
local AT = A .. T
local AA = A .. ALIF
local AAMAQ = A .. AMAQ
local AAH = AA .. TAM
local AAT = AA .. T
local II = I .. Y
local UU = U .. W
local AY = A .. Y
local AW = A .. W
local AYSK = AY .. SK
local AWSK = AW .. SK
local NA = N .. A
local NI = N .. I
local AAN = AA .. N
local AANI = AA .. NI
local AYNI = AYSK .. NI
local AWNA = AWSK .. NA
local AYNA = AYSK .. NA
local AYAAT = AY .. AAT
local UNU = "[" .. UN .. U .. "]"
local MA = M .. A
local MU = M .. U
local TA = T .. A
local TU = T .. U
local _I = ALIF .. I
local _U = ALIF .. U
local translit_cache = {
-- hamza variants
[HAMZA] = "ʔ",
[HAMZA_ON_ALIF] = "ʔ",
[HAMZA_ON_W] = "ʔ",
[HAMZA_UNDER_ALIF] = "ʔ",
[HAMZA_ON_Y] = "ʔ",
[HAMZA_PH] = "ʔ",
-- diacritics
[A] = "a",
[AN] = "an",
[U] = "u",
[UN] = "un",
[I] = "i",
[IN] = "in",
[SK] = "",
[SH] = "*", -- handled specially
[DAGGER_ALIF] = "ā",
-- various letters and signs
[""] = "",
[ALIF] = BAD, -- we should never be transliterating ALIF by itself, as its translit in isolation is ambiguous
[AMAQ] = BAD,
[AMAD] = "ʔā",
[TAM] = "",
[T] = "t",
[N] = "n",
[W] = "w",
[Y] = "y",
[S] = "s",
[M] = "m",
[LRM] = "",
-- common combinations
[AH] = "a",
[AT] = "at",
[AA] = "ā",
[AAMAQ] = "ā",
[AAH] = "āh",
[AAT] = "āt",
[II] = "ī",
[UU] = "ū",
[AY] = "ay",
[AW] = "aw",
[AYSK] = "ay",
[AWSK] = "aw",
[NA] = "na",
[NI] = "ni",
[AAN] = "ān",
[AANI] = "āni",
[AYNI] = "ayni",
[AWNA] = "awna",
[AYNA] = "ayna",
[AYAAT] = "ayāt",
[MA] = "ma",
[MU] = "mu",
[TA] = "ta",
[TU] = "tu",
[_I] = "i",
[_U] = "u",
}
local function transliterate(text)
local cached = translit_cache[text]
if cached then
if cached == BAD then
error(("Internal error: Unable to transliterate %s because explicitly marked as BAD"):format(text))
end
return cached
end
local tr = (lang:transliterate(text))
if not tr then
error(("Internal error: Unable to transliterate: %s"):format(text))
end
translit_cache[text] = tr
return tr
end
local all_person_number_list = {
"1s",
"2ms",
"2fs",
"3ms",
"3fs",
"2d",
"3md",
"3fd",
"1p",
"2mp",
"2fp",
"3mp",
"3fp"
}
local function make_person_number_slot_accel_list(list)
local slot_accel_list = {}
return slot_accel_list
end
local imp_person_number_list = {}
for _, pn in ipairs(all_person_number_list) do
if pn:find("^2") then
table.insert(imp_person_number_list, pn)
end
end
local passive_types = m_table.listToSet {
"pass", -- verb has both active and passive
"ipass", -- verb is active with impersonal passive
"nopass", -- verb is active-only
"onlypass", -- verb is passive-only
"onlypass-impers", -- verb itself is impersonal, meaning passive-only with impersonal passive
}
local indicator_flags = m_table.listToSet {
"nopast", "no_nonpast", "noimp",
"nocat", -- don't categorize or include annotations about this; useful in suppletive parts of verbs
"reduced", -- verb has assimilation/reduction of initial coronals
"altgem", -- form X with alternative past geminate forms with final-weak endings
}
export.potential_lemma_slots = {"past_3ms", "past_pass_3ms", "ind_3ms", "ind_pass_3ms", "imp_2ms"}
export.unsettable_slots = {}
for _, potential_lemma_slot in ipairs(export.potential_lemma_slots) do
table.insert(export.unsettable_slots, potential_lemma_slot .. "_linked")
end
table.insert(export.unsettable_slots, "ap2") -- secondary default فَعِيل for form I active participles (stative I)
table.insert(export.unsettable_slots, "ap3") -- secondary default فَعِل for form I active participles (stative II)
table.insert(export.unsettable_slots, "apcd") -- secondary default أَفْعَل for form I active participles (color/defect)
table.insert(export.unsettable_slots, "apan") -- secondary default فَعْلَان for form I active participles (in -ān)
table.insert(export.unsettable_slots, "pp2") -- secondary default فَعِيل for form I passive participles (same as ap2)
table.insert(export.unsettable_slots, "vn2") -- secondary default فِعَال for form III verbal nouns
export.unsettable_slots_set = m_table.listToSet(export.unsettable_slots)
local default_indicator_to_active_participle_slot = {
["++"] = "ap2",
["+++"] = "ap3",
["+cd"] = "apcd",
["+an"] = "apan",
}
-- Initialize all the slots for which we generate forms.
local function add_slots(alternant_multiword_spec)
alternant_multiword_spec.verb_slots = {
{"ap", "act|part"},
{"pp", "pass|part"},
{"vn", "vnoun"},
}
for _, unsettable_slot in ipairs(export.unsettable_slots) do
table.insert(alternant_multiword_spec.verb_slots, {unsettable_slot, "-"})
end
-- Add entries for a slot with person/number variants.
-- `slot_prefix` is the prefix of the slot, typically specifying the tense/aspect.
-- `tag_suffix` is a string listing the set of inflection tags to add after the person/number tags.
-- `person_number_list` is a list of the person/number slot suffixes to add to `slot_prefix`.
local function add_personal_slot(slot_prefix, tag_suffix, person_number_list)
for _, persnum in ipairs(person_number_list) do
local slot = slot_prefix .. "_" .. persnum
local accel = persnum:gsub("(.)", "%1|") .. tag_suffix
table.insert(alternant_multiword_spec.verb_slots, {slot, accel})
end
end
local tenses = {
{"past", "past|%s"},
{"ind", "non-past|%s|ind"},
{"sub", "non-past|%s|sub"},
{"juss", "non-past|%s|juss"},
}
for _, slot_accel in ipairs(tenses) do
local slot, accel = unpack(slot_accel)
for _, voice in ipairs {"act", "pass"} do
add_personal_slot(voice == "act" and slot or slot .. "_pass", accel:format(voice),
all_person_number_list)
end
end
add_personal_slot("imp", "imp", imp_person_number_list)
alternant_multiword_spec.verb_slots_map = {}
for _, slot_accel in ipairs(alternant_multiword_spec.verb_slots) do
local slot, accel = unpack(slot_accel)
alternant_multiword_spec.verb_slots_map[slot] = accel
end
end
local overridable_stems = {}
local slot_override_param_mods = {
footnote = {
item_dest = "footnotes",
store = "insert",
},
alt = {},
t = {
-- [[Module:links]] expects the gloss in "gloss".
item_dest = "gloss",
},
gloss = {},
g = {
-- [[Module:links]] expects the genders in "g". `sublist = true` automatically splits on comma (optionally
-- with surrounding whitespace).
item_dest = "genders",
sublist = true,
},
pos = {},
lit = {},
id = {},
-- Qualifiers and labels
q = {
type = "qualifier",
},
qq = {
type = "qualifier",
},
l = {
type = "labels",
},
ll = {
type = "labels",
},
}
local function generate_obj(formval, parse_err)
local val, uncertain = formval:match("^(.*)(%?)$")
val = val or formval
uncertain = not not uncertain
local ar, translit = val:match("^(.*)//(.*)$")
if not ar then
ar = val
end
return {form = ar, translit = translit, uncertain = uncertain}
end
local function parse_inline_modifiers(comma_separated_group, parse_err)
return require(parse_utilities_module).parse_inline_modifiers_from_segments {
group = comma_separated_group,
props = {
param_mods = slot_override_param_mods,
parse_err = parse_err,
generate_obj = generate_obj,
pre_normalize_modifiers = function(data)
local modtext = data.modtext
modtext = modtext:match("^(%[.*%])$")
if modtext then
return ("<footnote:%s>"):format(modtext)
end
return data.modtext
end,
},
}
end
local function allow_multiple_values_for_override(comma_separated_groups, data, is_slot_override)
local retvals = {}
for _, comma_separated_group in ipairs(comma_separated_groups) do
local retval
if is_slot_override then
retval = parse_inline_modifiers(comma_separated_group, data.parse_err)
else
retval = generate_obj(comma_separated_group[1], data.parse_err)
retval.footnotes = data.fetch_footnotes(comma_separated_group)
end
table.insert(retvals, retval)
end
for _, form in ipairs(retvals) do
if form.form == "+" or default_indicator_to_active_participle_slot[form.form] then
if default_indicator_to_active_participle_slot[form.form] and not is_slot_override then
error(("Stem override '%s' cannot use %s to request a secondary default"):format(
data.prefix, form.form))
end
data.base.slot_override_uses_default[data.prefix] = true
end
end
for _, form in ipairs(retvals) do
if form.form == "-" then
data.base.slot_explicitly_missing[data.prefix] = true
break
end
end
if data.base.slot_explicitly_missing[data.prefix] then
for _, form in ipairs(retvals) do
if form.form ~= "-" then
data.parse_err(("For slot or stem '%s', saw both - and a value other than -, which isn't allowed"):
format(data.prefix))
end
end
return nil
end
return retvals
end
local function simple_choice(choices)
return function(separated_groups, data)
if #separated_groups > 1 then
data.parse_err("For spec '" .. data.prefix .. ":', only one value currently allowed")
end
if #separated_groups[1] > 1 then
data.parse_err("For spec '" .. data.prefix .. ":', no footnotes currently allowed")
end
local choice = separated_groups[1][1]
if not m_table.contains(choices, choice) then
data.parse_err("For spec '" .. data.prefix .. ":', saw value '" .. choice .. "' but expected one of '" ..
table.concat(choices, ",") .. "'")
end
return choice
end
end
for _, overridable_stem in ipairs {
"past",
"past_v",
"past_c",
"past_pass",
"past_pass_v",
"past_pass_c",
"nonpast",
"nonpast_v",
"nonpast_c",
"nonpast_pass",
"nonpast_pass_v",
"nonpast_pass_c",
"imp",
"imp_v",
"imp_c",
} do
overridable_stems[overridable_stem] = allow_multiple_values_for_override
end
overridable_stems.past_final_weak_vowel = simple_choice { "ay", "aw", "ī", "ū" }
overridable_stems.past_pass_final_weak_vowel = simple_choice { "ay", "aw", "ī", "ū" }
overridable_stems.nonpast_final_weak_vowel = simple_choice { "ā", "ī", "ū" }
overridable_stems.nonpast_pass_final_weak_vowel = simple_choice { "ā", "ī", "ū" }
-------------------------------------------------------------------------------
-- Utility functions --
-------------------------------------------------------------------------------
-- version of rsubn() that discards all but the first return value
local function rsub(term, foo, bar)
return (rsubn(term, foo, bar))
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
-- Concatenate one or more strings or form objects.
local function q(...)
local not_all_strings = debug_translit
local has_manual_translit = debug_translit
for i = 1, select("#", ...) do
local argt = select(i, ...)
if not argt then
error(("Internal error: Saw nil at index %s: %s"):format(i, dump({...})))
end
if type(argt) ~= "string" then
not_all_strings = true
if argt.translit then
has_manual_translit = true
break
end
end
end
if not not_all_strings then
-- just strings, concatenate directly
return table.concat({...})
end
local formvals = {}
local translit = has_manual_translit and {} or nil
local footnotes
for i = 1, select("#", ...) do
local argt = select(i, ...)
if type(argt) == "string" then
formvals[i] = argt
if has_manual_translit then
translit[i] = transliterate(argt)
end
else
formvals[i] = argt.form
if has_manual_translit then
translit[i] = argt.translit or transliterate(argt.form)
end
footnotes = iut.combine_footnotes(footnotes, argt.footnotes)
end
end
-- FIXME: Do we want to support other properties?
return {
form = table.concat(formvals),
translit = has_manual_translit and table.concat(translit) or nil,
footnotes = footnotes,
}
end
-- Return the formval associated with `rad` (a radical or past/non-past vowel, either a string or form object).
local function rget(rad)
if type(rad) == "string" then
return rad
elseif type(rad) == "table" then
return rad.form
else
error(("Internal error: Unexpected type for radical or past/non-past vowel: %s"):format(dump(rad)))
end
end
export.rget = rget -- for use in [[Module:ar-headword]]
-- Return the footnotes associated with `rad` (a radical or past/non-past vowel, either a string or form object).
local function rget_footnotes(rad)
if type(rad) == "string" then
return nil
elseif type(rad) == "table" then
return rad.footnotes
else
error(("Internal error: Unexpected type for radical or past/non-past vowel: %s"):format(dump(rad)))
end
end
-- Return true if the formval associated with `rad` (a radical or past/non-past vowel, either a string or form object)
-- is `val`.
local function req(rad, val)
return rget(rad) == val
end
-- Map `vow` (a past/non-past vowel, either a string or form object without translit) by passing the formval through
-- `fn`. Don't call this on radicals because they may have manual translit and it isn't clear how to handle that.
local function map_vowel(vow, fn)
if type(vow) == "string" then
return fn(vow)
elseif type(vow) == "table" then
return {form = fn(vow.form), footnotes = vow.footnotes}
else
error(("Internal error: Unexpected type for past/non-past vowel: %s"):format(dump(vow)))
end
end
local function get_radicals_3(vowel_spec)
return vowel_spec.rad1, vowel_spec.rad2, vowel_spec.rad3, vowel_spec.past, vowel_spec.nonpast
end
local function get_radicals_4(vowel_spec)
return vowel_spec.rad1, vowel_spec.rad2, vowel_spec.rad3, vowel_spec.rad4
end
local function is_final_weak(base, vowel_spec)
return vowel_spec.weakness == "final-weak" or base.form == "XV"
end
local function link_term(text, face, id)
return m_links.full_link({lang = lang, term = text, tr = "-", id = id}, face)
end
local function tag_text(text, tag, class)
return m_links.full_link({lang = lang, alt = text, tr = "-"})
end
local function track(page)
require("Module:debug/track")("ar-verb/" .. page)
return true
end
local function reorder_shadda(word)
-- shadda+short-vowel (including tanwīn vowels, i.e. -an -in -un) gets
-- replaced with short-vowel+shadda during NFC normalisation, which
-- MediaWiki does for all Unicode strings; however, it makes various
-- processes inconvenient, so undo it.
word = rsub(word, "(" .. DIACRITIC_ANY_BUT_SH .. ")" .. SH, SH .. "%1")
return word
end
-------------------------------------------------------------------------------
-- Basic functions to inflect tenses --
-------------------------------------------------------------------------------
local function skip_slot(base, slot, allow_overrides)
if base.slot_explicitly_missing[slot] then
return true
end
if not allow_overrides and base.slot_overrides[slot] and not base.slot_override_uses_default[slot] then
-- Skip any slots for which there are overrides, except those that request the default value using +, ++, etc.
return true
end
if base.passive == "nopass" and (slot == "pp" or slot:find("_pass")) then
return true
elseif base.passive == "onlypass" and slot ~= "pp" and slot ~= "vn" and not slot:find("_pass") then
return true
elseif base.passive == "ipass" and slot:find("_pass") and not slot:find("3ms") then
return true
elseif base.passive == "onlypass-impers" and slot ~= "pp" and slot ~= "vn" and (not slot:find("_pass") or
slot:find("_pass") and not slot:find("3ms")) then
return true
end
if base.nopast and slot:find("^past_") then
return true
end
if base.noimp and slot:find("^imp_") then
return true
end
if base.no_nonpast and (slot:find("^ind_") or slot:find("^sub_") or slot:find("^juss")) then
return true
end
return false
end
local function basic_combine_stem_ending(stem, ending)
return stem .. ending
end
local function basic_combine_stem_ending_tr(stem, ending)
return stem .. ending
end
-- Concatenate `prefixes`, `stems` and `endings` (any of which may be an abbreviate form list, i.e. strings, form
-- objects or lists of strings or form objects) and store into `slot`. If a user-supplied override exists for the slot,
-- nothing will happen unless `allow_overrides` is provided.
local function add3(base, slot, prefixes, stems, endings, allow_overrides)
if skip_slot(base, slot, allow_overrides) then
return
end
-- Optimization since the prefixes are almost always single strings.
if type(prefixes) == "string" then
local function do_combine_stem_ending(stem, ending)
return prefixes .. stem .. ending
end
local function do_combine_stem_ending_tr(stem, ending)
return transliterate(prefixes) .. stem .. ending
end
iut.add_forms(base.forms, slot, stems, endings, do_combine_stem_ending, transliterate,
do_combine_stem_ending_tr, base.form_footnotes)
else
iut.add_multiple_forms(base.forms, slot, {prefixes, stems, endings}, basic_combine_stem_ending, transliterate,
basic_combine_stem_ending_tr, base.form_footnotes)
end
end
-- Insert one or more forms in `form_or_forms` into `slot`. `form_or_forms` is an abbreviated form list (see comment at
-- top of [[Module:inflection utilities]]). If a user-supplied override exists for the slot, nothing will happen unless
-- `allow_overrides` is provided. BEWARE: One form object should never occur in two different slots, or twice in a given
-- slot; if taking a form object from an existing slot, make sure to shallowcopy() it.
local function insert_form_or_forms(base, slot, form_or_forms, allow_overrides, uncertain)
if not skip_slot(base, slot, allow_overrides) then
-- Some optimizations of the most common case of inserting a single string.
if type(form_or_forms) == "string" and not base.form_footnotes then
form_or_forms = {form = form_or_forms, uncertain = uncertain}
iut.insert_form(base.forms, slot, form_or_forms)
else
local list = iut.convert_to_general_list_form(form_or_forms, base.form_footnotes)
if uncertain then
for _, formobj in ipairs(list) do
formobj.uncertain = true
end
end
iut.insert_forms(base.forms, slot, list)
end
end
end
-- Insert `string_or_form` into both the ap2 and pp2 slots, shallowcopying a form object to make sure no form objects
-- occur in two slots.
local function insert_ap2_pp2(base, string_or_form)
insert_form_or_forms(base, "ap2", string_or_form)
if type(string_or_form) == "table" then
string_or_form = m_table.shallowcopy(string_or_form)
end
insert_form_or_forms(base, "pp2", string_or_form)
end
-- Convert `stemforms` (a string, a form object, or a list of strings and/or form objects) into "general form" (a list
-- of form objects) and map `fn` over the list of objects. `fn` is passed two arguments (form value and translit) and
-- should likewise return the new form value and translit. Footnotes will be preserved. FIXME: Preserve other metadata.
local function map_general(stemforms, fn)
return iut.map_forms(iut.convert_to_general_list_form(stemforms), fn)
end
-- Similar to map_general() except that `fn` should return a single value (one or more strings or form objects), instead
-- of two values (form value and translit), and the resulting value(s) from all calls to `fn` will be flattened to
-- construct the overall return value. Footnotes will be preserved. FIXME: Preserve other metadata.
local function flatmap_general(stemforms, fn)
return iut.flatmap_forms(iut.convert_to_general_list_form(stemforms), fn)
end
-- Given user-supplied stem overrides in `base`, construct any derived stem overrides (e.g. vowel-specific or
-- consonant-specific variants), and truncate initial y-/ي- in any non-past overrides.
local function construct_stems(base)
local stems = base.stem_overrides
stems.past_v = stems.past_v or stems.past
stems.past_c = stems.past_c or stems.past
stems.past_pass_v = stems.past_pass_v or stems.past_pass
stems.past_pass_c = stems.past_pass_c or stems.past_pass
stems.nonpast_v = stems.nonpast_v or stems.nonpast
stems.nonpast_c = stems.nonpast_c or stems.nonpast
stems.nonpast_pass_v = stems.nonpast_pass_v or stems.nonpast_pass
stems.nonpast_pass_c = stems.nonpast_pass_c or stems.nonpast_pass
stems.imp_v = stems.imp_v or stems.imp
stems.imp_c = stems.imp_c or stems.imp
local function truncate_nonpast_initial_cons(stem_type, form, translit)
if form == "+" then
return form, translit
end
if not form:find("^" .. Y) then
error(("Form value %s for stem type '%s' should begin with ي"):format(form, stem_type))
end
form = form:gsub("^" .. Y, "")
if translit then
if not translit:find("^y") then
error(("Translit value %s for stem type '%s' should begin with y"):format(translit, stem_type))
end
translit = translit:gsub("^y", "")
end
return form, translit
end
for _, nonpast_stem_type in ipairs { "nonpast_v", "nonpast_c", "nonpast_pass_v", "nonpast_pass_c" } do
if stems[nonpast_stem_type] then
stems[nonpast_stem_type] = map_general(stems[nonpast_stem_type], function(form, translit)
return truncate_nonpast_initial_cons(nonpast_stem_type, form, translit)
end)
end
end
end
-- Given user-specified overrides for stem `stemname`, return overrides with occurrences of + replaced by
-- `default_stem`. If no overrides, return `default_stem`, or {} if no default.
local function override_stem_if_needed(base, stemname, default_stem)
local overrides = base.stem_overrides[stemname]
if not overrides then
return default_stem or {}
end
return map_general(overrides, function(form, translit)
if default_indicator_to_active_participle_slot[form] then
error(("Stem overrides cannot use secondary default indicators but saw %s in stem override '%s'"):format(
form, stemname))
end
if form == "+" then
if translit then
error(("Cannot supply manual translit along with + for stem override '%s'"):format(stemname))
end
if not default_stem then
error(("Cannot use + for stem override '%s' because no default is available"):format(stemname))
end
if type(default_stem) ~= "string" then
error(("Internal error: Default stem for '%s' is not a string: %s"):format(stemname, dump(default_stem)))
end
return default_stem
end
return form, translit
end)
end
-------------------------------------------------------------------------------
-- Properties of different verbal forms --
-------------------------------------------------------------------------------
local allowed_vforms = {"I", "II", "III", "IV", "V", "VI", "VII", "VIII", "IX",
"X", "XI", "XII", "XIII", "XIV", "XV", "Iq", "IIq", "IIIq", "IVq"}
local allowed_vforms_set = m_table.listToSet(allowed_vforms)
local allowed_vforms_with_weakness = m_table.shallowcopy(allowed_vforms)
-- The user needs to be able to explicitly specify that a form-I verb (specifically one whose initial radical is و) is
-- sound. Cf. wajiʕa yawjaʕu (not #yajaʕu) "to ache, to hurt". In general, i~a and u~u verbs whose initial radical is و
-- seem to not assimilate the first radical; cf. وقح "to be shameless", variously waqaḥa~yaqiḥu, waquḥa~yawquḥu and
-- waqiḥa~yawqaḥu, whereas a~i verbs (wafaḍa~yafiḍu "to rush"), i~i verbs (wafiqa~yafiqu "to be proper, to be suitable")
-- and a~a verbs (waḍaʕa~yaḍaʕu "to set down, to place") do assimilate. But there are naturally exceptions, e.g.
-- waṭiʔa~yaṭaʔu "to tread, to trample"; wasiʕa~yasaʕu "to be spacious; to be well-off"; waṯiʔa~yaṯaʔu "to get bruised,
-- to be sprained". Also beware of waniya~yawnā "to be faint; to languish", which is sound in the first radical and
-- final-weak in the last radical. Nonetheless, the regularity of the patterns mentioned above suggest we should provide
-- them as defaults.
-- Note that there are other cases of unexpectedly sound verbs, e.g. izdawaja~yazdawiju "to be in pairs", layisa~yalyasu
-- "to be valiant, to be brave", ʔaḥwaja~yuḥwiju "to need", istahwana~yastahwinu "to consider easy", sawisa~yaswasu "to
-- be or become moth-eaten or worm-eaten" (vs. sāsa~yasūsu "to govern, to rule" from the same radicals), ʕawira~yaʕwaru
-- "to be one-eyed", istajwaba~yastajwibu "to interrogate", etc. But in these cases there is no need for explicit user
-- specification as the lemma itself specifies the unexpected soundness.
for _, form_with_weakness in ipairs { "I-sound", "I-assimilated", "none-sound", "none-hollow", "none-geminate",
"none-final-weak" } do
table.insert(allowed_vforms_with_weakness, form_with_weakness)
end
local allowed_vforms_with_weakness_set = m_table.listToSet(allowed_vforms_with_weakness)
local function vform_supports_final_weak(vform)
return vform ~= "XI" and vform ~= "XV" and vform ~= "IVq"
end
local function vform_supports_geminate(vform)
return vform == "I" or vform == "III" or vform == "IV" or
vform == "VI" or vform == "VII" or vform == "VIII" or vform == "X"
end
local function vform_supports_hollow(vform)
return vform == "I" or vform == "IV" or vform == "VII" or vform == "VIII" or
vform == "X"
end
local function vform_probably_impersonal_passive(vform)
return vform == "VI"
end
local function vform_probably_no_passive(vform, weakness, past_vowel, nonpast_vowel)
return vform == "I" and weakness ~= "hollow" and req(past_vowel, U) or
vform == "VII" or vform == "IX" or vform == "XI" or vform == "XII" or
vform == "XIII" or vform == "XIV" or vform == "XV" or vform == "IIq" or
vform == "IIIq" or vform == "IVq"
end
-- Active vforms II, III, IV, Iq use non-past prefixes in -u- instead of -a-.
local function prefix_vowel_from_vform(vform)
if vform == "II" or vform == "III" or vform == "IV" or vform == "Iq" then
return "u"
else
return "a"
end
end
-- True if the active non-past takes a-vocalization rather than i-vocalization in its last syllable.
local function vform_nonpast_a_vowel(vform)
return vform == "V" or vform == "VI" or vform == "XV" or vform == "IIq"
end
-- True if the `passive` spec indicates a passive-only verb.
local function is_passive_only(passive)
return passive == "onlypass" or passive == "onlypass-impers"
end
export.is_passive_only = is_passive_only -- for use in [[Module:ar-headword]]
-------------------------------------------------------------------------------
-- Properties of specific sounds --
-------------------------------------------------------------------------------
-- Is radical wāw (و) or yāʾ (ي)?
local function is_waw_ya(rad)
return req(rad, W) or req(rad, Y)
end
-- Check that radical is wāw (و) or yāʾ (ي), error if not
local function check_waw_ya(rad)
if not is_waw_ya(rad) then
error("Expecting weak radical: '" .. rget(rad) .. "' should be " .. W .. " or " .. Y)
end
end
-- Form-I verb حيّ or حيي and form-X verb استحيا or استحى
local function hayy_radicals(rad1, rad2, rad3)
return req(rad1, "ح") and req(rad2, Y) and is_waw_ya(rad3)
end
-- FUCK ME HARD. "Lua error at line 1514: main function has more than 200 local variables".
local function create_conjugations()
-------------------------------------------------------------------------------
-- Radicals associated with various irregular verbs --
-------------------------------------------------------------------------------
-- Form-I verb أخذ or form-VIII verb اتخذ
local function axadh_radicals(rad1, rad2, rad3)
return req(rad1, HAMZA) and req(rad2, "خ") and req(rad3, "ذ")
end
-- Form-I verb whose imperative has a reduced form: أكل and أخذ and أمر. Return "shortonly" if only
-- short-form imperatives exist (أكل and أخذ) or "shortlong" if long-form imperatives also exist (أمر);
-- they are used after a clitic like فَ and وَ.
local function reduced_imperative_verb(rad1, rad2, rad3)
return axadh_radicals(rad1, rad2, rad3) and "shortonly" or
req(rad1, HAMZA) and req(rad2, "ك") and req(rad3, "ل") and "shortonly" or
req(rad1, HAMZA) and req(rad2, "م") and req(rad3, "ر") and "shortlong"
end
-- Form-I verb رأى and form-IV verb أرى
local function raa_radicals(rad1, rad2, rad3)
return req(rad1, "ر") and req(rad2, HAMZA) and is_waw_ya(rad3)
end
-- Form-I verb سأل
local function saal_radicals(rad1, rad2, rad3)
return req(rad1, "س") and req(rad2, HAMZA) and req(rad3, "ل")
end
-- Form-I verb كان
local function kaan_radicals(rad1, rad2, rad3)
return req(rad1, "ك") and req(rad2, W) and req(rad3, N)
end
-------------------------------------------------------------------------------
-- Sets of past endings --
-------------------------------------------------------------------------------
-- The 13 endings of the sound/hollow/geminate past tense.
local past_endings = {
-- singular
SK .. TU, SK .. TA, SK .. "تِ", A, A .. "تْ",
--dual
SK .. "تُمَا", AA, A .. "تَا",
-- plural
SK .. "نَا", SK .. "تُمْ",
-- shadda + vowel diacritic ends up in the wrong order due to Unicode
-- bug, so keep them separate to avoid this
SK .. "تُن" .. SH .. A, UU .. ALIF, SK .. "نَ"
}
-- Make endings for final-weak past in -aytu or -awtu. AYAW is AY or AW as appropriate. Note that AA and AW are
-- global variables.
local function make_past_endings_ay_aw(ayaw, third_sg_masc)
return {
-- singular
ayaw .. SK .. TU, ayaw .. SK .. TA, ayaw .. SK .. "تِ",
third_sg_masc, A .. "تْ",
--dual
ayaw .. SK .. "تُمَا", ayaw .. AA, A .. "تَا",
-- plural
ayaw .. SK .. "نَا", ayaw .. SK .. "تُمْ",
-- shadda + vowel diacritic ends up in the wrong order due to Unicode
-- bug, so keep them separate to avoid this
ayaw .. SK .. "تُن" .. SH .. A, AW .. SK .. ALIF, ayaw .. SK .. "نَ"
}
end
-- past final-weak -aytu endings
local past_endings_ay = make_past_endings_ay_aw(AY, AAMAQ)
-- past final-weak -awtu endings
local past_endings_aw = make_past_endings_ay_aw(AW, AA)
-- used for alternative endings for form-X geminate verbs like اِسْتَمَرَّ
local past_endings_ay_12_person_only = {
-- singular
AY .. SK .. TU, AY .. SK .. TA, AY .. SK .. "تِ",
{}, {},
--dual
AY .. SK .. "تُمَا", {}, {},
-- plural
AY .. SK .. "نَا", AY .. SK .. "تُمْ",
-- shadda + vowel diacritic ends up in the wrong order due to Unicode
-- bug, so keep them separate to avoid this
AY .. SK .. "تُن" .. SH .. A, {}, {},
}
-- Make endings for final-weak past in -ītu or -ūtu. IIUU is ī or ū as appropriate. Note that AA and UU are global
-- variables.
local function make_past_endings_ii_uu(iiuu)
return {
-- singular
iiuu .. TU, iiuu .. TA, iiuu .. "تِ", iiuu .. A, iiuu .. A .. "تْ",
--dual
iiuu .. "تُمَا", iiuu .. AA, iiuu .. A .. "تَا",
-- plural
iiuu .. "نَا", iiuu .. "تُمْ",
-- shadda + vowel diacritic ends up in the wrong order due to Unicode
-- bug, so keep them separate to avoid this
iiuu .. "تُن" .. SH .. A, UU .. ALIF, iiuu .. "نَ"
}
end
-- past final-weak -ītu endings
local past_endings_ii = make_past_endings_ii_uu(II)
-- past final-weak -ūtu endings
local past_endings_uu = make_past_endings_ii_uu(UU)
-------------------------------------------------------------------------------
-- Sets of non-past prefixes and endings --
-------------------------------------------------------------------------------
local nonpast_prefix_consonants = {
-- singular
HAMZA, T, T, Y, T,
-- dual
T, Y, T,
-- plural
N, T, T, Y, Y
}
-- There are only five distinct endings in all non-past verbs. Make any set of non-past endings given these five
-- distinct endings.
local function make_nonpast_endings(null, fem, dual, pl, fempl)
return {
-- singular
null, null, fem, null, null,
-- dual
dual, dual, dual,
-- plural
null, pl, fempl, pl, fempl
}
end
-- endings for non-past indicative
local ind_endings = make_nonpast_endings(
U,
II .. NA,
AANI,
UU .. NA,
SK .. NA
)
-- Make the endings for non-past subjunctive/jussive, given the vowel diacritic used in "null" endings
-- (1s/2ms/3ms/3fs/1p).
local function make_sub_juss_endings(dia_null)
return make_nonpast_endings(
dia_null,
II,
AA,
UU .. ALIF,
SK .. NA
)
end
-- endings for non-past subjunctive
local sub_endings = make_sub_juss_endings(A)
-- endings for non-past jussive
local juss_endings = make_sub_juss_endings(SK)
-- endings for alternative geminate non-past jussive in -a; same as subjunctive
local juss_endings_alt_a = sub_endings
-- endings for alternative geminate non-past jussive in -i
local juss_endings_alt_i = make_sub_juss_endings(I)
-- Endings for final-weak non-past indicative in -ā. Note that AY, AW and AAMAQ are global variables.
local ind_endings_aa = make_nonpast_endings(
AAMAQ,
AYSK .. NA,
AY .. AANI,
AWSK .. NA,
AYSK .. NA
)
-- Make endings for final-weak non-past indicative in -ī or -ū; IIUU is ī or ū as appropriate. Note that II and UU
-- are global variables.
local function make_ind_endings_ii_uu(iiuu)
return make_nonpast_endings(
iiuu,
II .. NA,
iiuu .. AANI,
UU .. NA,
iiuu .. NA
)
end
-- endings for final-weak non-past indicative in -ī
local ind_endings_ii = make_ind_endings_ii_uu(II)
-- endings for final-weak non-past indicative in -ū
local ind_endings_uu = make_ind_endings_ii_uu(UU)
-- Endings for final-weak non-past subjunctive in -ā. Note that AY, AW, ALIF, AAMAQ are global variables.
local sub_endings_aa = make_nonpast_endings(
AAMAQ,
AYSK,
AY .. AA,
AWSK .. ALIF,
AYSK .. NA
)
-- Make endings for final-weak non-past subjunctive in -ī or -ū. IIUU is ī or ū as appropriate. Note that AA, II,
-- UU, ALIF are global variables.
local function make_sub_endings_ii_uu(iiuu)
return make_nonpast_endings(
iiuu .. A,
II,
iiuu .. AA,
UU .. ALIF,
iiuu .. NA
)
end
-- endings for final-weak non-past subjunctive in -ī
local sub_endings_ii = make_sub_endings_ii_uu(II)
-- endings for final-weak non-past subjunctive in -ū
local sub_endings_uu = make_sub_endings_ii_uu(UU)
-- endings for final-weak non-past jussive in -ā
local juss_endings_aa = make_nonpast_endings(
A,
AYSK,
AY .. AA,
AWSK .. ALIF,
AYSK .. NA
)
-- Make endings for final-weak non-past jussive in -ī or -ū. IU is short i or u, IIUU is long ī or ū as appropriate.
-- Note that AA, II, UU, ALIF are global variables.
local function make_juss_endings_ii_uu(iu, iiuu)
return make_nonpast_endings(
iu,
II,
iiuu .. AA,
UU .. ALIF,
iiuu .. NA
)
end
-- endings for final-weak non-past jussive in -ī
local juss_endings_ii = make_juss_endings_ii_uu(I, II)
-- endings for final-weak non-past jussive in -ū
local juss_endings_uu = make_juss_endings_ii_uu(U, UU)
-------------------------------------------------------------------------------
-- Sets of imperative endings --
-------------------------------------------------------------------------------
-- Extract the second person jussive endings to get corresponding imperative endings.
local function imperative_endings_from_jussive(endings)
return {endings[2], endings[3], endings[6], endings[10], endings[11]}
end
-- normal imperative endings
local imp_endings = imperative_endings_from_jussive(juss_endings)
-- alternative geminate imperative endings in -a
local imp_endings_alt_a = imperative_endings_from_jussive(juss_endings_alt_a)
-- alternative geminate imperative endings in -i
local imp_endings_alt_i = imperative_endings_from_jussive(juss_endings_alt_i)
-- final-weak imperative endings in -ā
local imp_endings_aa = imperative_endings_from_jussive(juss_endings_aa)
-- final-weak imperative endings in -ī
local imp_endings_ii = imperative_endings_from_jussive(juss_endings_ii)
-- final-weak imperative endings in -ū
local imp_endings_uu = imperative_endings_from_jussive(juss_endings_uu)
-------------------------------------------------------------------------------
-- Basic functions to inflect tenses --
-------------------------------------------------------------------------------
-- Add to `base` the inflections for the tense indicated by `tense` (the prefix in the slot names, e.g. 'past'
-- or 'juss_pass'), formed by combining the `prefixes`, `stems` and `endings`. Each of `prefixes`, `stems` and
-- `endings` is either a sequence of 5 (for the imperative) or 13 (for other tenses) abbreviated form lists (each of
-- which is either a string, a form object, or a list of strings and/or form objects; see
-- [[Module:inflection utilities]] for more info). Alternatively, any of `prefixes`, `stems` or `endings` can be a
-- single-element list containing an abbreviated form list, with an additional key `all_same` set to true, or (as a
-- special case) a single string; in the latter cases, the same value is used for all 5 or 13 slots. If existing
-- inflections already exist, they will be added to, not overridden. `pnums` is the list of person/number slot name
-- suffixes, which must match up with the elements in `prefixes`, `stems` and `endings` (i.e. 5 for imperative, 13
-- otherwise).
local function inflect_tense_1(base, tense, prefixes, stems, endings, pnums)
if not prefixes or not stems or not endings then
return
end
local function verify_affixes(affixname, affixes)
local function interr(msg)
error(("Internal error: For tense '%s', '%s' %s: %s"):format(tense, affixname, msg, dump(affixes)))
end
if type(affixes) == "string" then
-- do nothing
elseif type(affixes) ~= "table" then
interr("is not a table or string")
elseif affixes.all_same then
if #affixes ~= 1 then
interr(("with all_same = true should have length 1 but has length %s"):format(#affixes))
end
else
if #affixes ~= #pnums then
interr(("should have length %s but has length %s"):format(#pnums, #affixes))
end
end
end
verify_affixes("prefixes", prefixes)
verify_affixes("stems", stems)
verify_affixes("endings", endings)
local function get_affix(affixes, i)
if type(affixes) == "string" then
return affixes
elseif affixes.all_same then
return affixes[1]
else
return affixes[i]
end
end
for i, pnum in ipairs(pnums) do
local prefix = get_affix(prefixes, i)
local stem = get_affix(stems, i)
local ending = get_affix(endings, i)
local slot = tense .. "_" .. pnum
add3(base, slot, prefix, stem, ending)
end
end
-- Add to `base` the inflections for the tense indicated by `tense` (the prefix in the slot names, e.g. 'past'
-- or 'juss_pass'), formed by combining the `prefixes`, `stems` and `endings`. This is a simple wrapper around
-- inflect_tense_1() that applies to all tenses other than the imperative; see inflect_tense_1() for more
-- information about the parameters.
local function inflect_tense(base, tense, prefixes, stems, endings)
inflect_tense_1(base, tense, prefixes, stems, endings, all_person_number_list)
end
-- Like inflect_tense() but for the imperative, which has only five parts instead of 13 and no prefixes.
local function inflect_tense_imp(base, stems, endings)
inflect_tense_1(base, "imp", "", stems, endings, imp_person_number_list)
end
-------------------------------------------------------------------------------
-- Functions to inflect the past tense --
-------------------------------------------------------------------------------
-- Generate past verbs using specified vowel and consonant stems; works for sound, assimilated, hollow, and geminate
-- verbs, active and passive.
local function past_2stem_conj(base, tense, v_stem, c_stem, footnote_12)
local passive = tense:find("_pass") and "_pass" or ""
-- Override stems with user-specified stems if available.
v_stem = override_stem_if_needed(base, "past" .. passive .. "_v", v_stem)
local c_stem_12 = c_stem
if footnote_12 then
c_stem_12 = iut.combine_form_and_footnotes(c_stem_12, footnote_12)
end
c_stem_12 = override_stem_if_needed(base, "past" .. passive .. "_c", c_stem_12)
local c_stem_3 = override_stem_if_needed(base, "past" .. passive .. "_c", c_stem)
inflect_tense(base, tense, "", {
-- singular
c_stem_12, c_stem_12, c_stem_12, v_stem, v_stem,
--dual
c_stem_12, v_stem, v_stem,
-- plural
c_stem_12, c_stem_12, c_stem_12, v_stem, c_stem_3
}, past_endings)
end
-- Generate past verbs using single specified stem; works for sound and assimilated verbs, active and passive.
local function past_1stem_conj(base, tense, stem)
past_2stem_conj(base, tense, stem, stem)
end
-------------------------------------------------------------------------------
-- Functions to inflect non-past tenses --
-------------------------------------------------------------------------------
-- Generate non-past conjugation, with two stems, for vowel-initial and consonant-initial endings, respectively.
-- Useful for active and passive; for all forms; for all weaknesses (sound, assimilated, hollow, final-weak and
-- geminate) and for all types of non-past (indicative, subjunctive, jussive) except for the imperative. (There is a
-- separate wrapper function below for geminate jussives because they have three alternants.) Both stems may be the
-- same, e.g. for sound verbs.
-- `prefix_vowel` will be either "a" or "u". `endings` should be an array of 13 items. If `endings` is nil or
-- omitted, infer the endings from the tense. If `jussive` is true, or `endings` is nil and `tense` indicatives
-- jussive, use the jussive pattern of vowel/consonant stems (different from the normal ones).
local function nonpast_2stem_conj(base, tense, prefix_vowel, v_stem, c_stem, endings, jussive)
local passive = tense:find("_pass") and "_pass" or ""
-- Override stems with user-specified stems if available.
v_stem = override_stem_if_needed(base, "nonpast" .. passive .. "_v",
v_stem and q(dia[prefix_vowel], v_stem) or nil)
c_stem = override_stem_if_needed(base, "nonpast" .. passive .. "_c",
c_stem and q(dia[prefix_vowel], c_stem) or nil)
if not endings then
if tense:find("^ind") then
endings = ind_endings
elseif tense:find("^sub") then
endings = sub_endings
elseif tense:find("^juss") then
jussive = true
endings = juss_endings
else
error("Internal error: Unrecognized tense '" .. tense .."'")
end
end
if not jussive then
inflect_tense(base, tense, nonpast_prefix_consonants, {
-- singular
v_stem, v_stem, v_stem, v_stem, v_stem,
--dual
v_stem, v_stem, v_stem,
-- plural
v_stem, v_stem, c_stem, v_stem, c_stem
}, endings)
else
inflect_tense(base, tense, nonpast_prefix_consonants, {
-- singular
-- 'adlul, tadlul, tadullī, yadlul, tadlul
c_stem, c_stem, v_stem, c_stem, c_stem,
--dual
-- tadullā, yadullā, tadullā
v_stem, v_stem, v_stem,
-- plural
-- nadlul, tadullū, tadlulna, yadullū, yadlulna
c_stem, v_stem, c_stem, v_stem, c_stem
}, endings)
end
end
-- Generate non-past conjugation with one stem (no distinct stems for vowel-initial and consonant-initial endings).
-- See nonpast_2stem_conj().
local function nonpast_1stem_conj(base, tense, prefix_vowel, stem, endings, jussive)
nonpast_2stem_conj(base, tense, prefix_vowel, stem, stem, endings, jussive)
end
-- Generate active/passive jussive geminative. There are three alternants, two with terminations -a and -i and one
-- in a null termination with a distinct pattern of vowel/consonant stem usage. See nonpast_2stem_conj() for a
-- description of the arguments.
local function jussive_gem_conj(base, tense, prefix_vowel, v_stem, c_stem)
-- alternative in -a
nonpast_2stem_conj(base, tense, prefix_vowel, v_stem, c_stem, juss_endings_alt_a)
-- alternative in -i
nonpast_2stem_conj(base, tense, prefix_vowel, v_stem, c_stem, juss_endings_alt_i)
-- alternative in -null; requires different combination of v_stem and
-- c_stem since the null endings require the c_stem (e.g. "tadlul" here)
-- whereas the corresponding endings above in -a or -i require the v_stem
-- (e.g. "tadulla, tadulli" above)
nonpast_2stem_conj(base, tense, prefix_vowel, v_stem, c_stem, juss_endings, "jussive")
end
-------------------------------------------------------------------------------
-- Functions to inflect the imperative --
-------------------------------------------------------------------------------
-- Generate imperative conjugation, with two stems, for vowel-initial and consonant-initial endings, respectively.
-- Useful for all forms, and for all weaknesses other than final-weak. Note that the two stems may be the same
-- (specifically for sound and assimilated verbs). If `endings` is nil or omitted, use `imp_endings`. If `alt_gem`
-- is specified, use the pattern of vowel and consonant stems appropriate for the alternative geminate imperatives
-- that use a null ending of -a or -i instead of an empty ending.
local function make_2stem_imperative(base, v_stem, c_stem, endings, alt_gem)
endings = endings or imp_endings
-- Override stems with user-specified stems if available.
v_stem = override_stem_if_needed(base, "imp_v", v_stem)
c_stem = override_stem_if_needed(base, "imp_c", c_stem)
if alt_gem then
inflect_tense_imp(base, {v_stem, v_stem, v_stem, v_stem, c_stem}, endings)
else
inflect_tense_imp(base, {c_stem, v_stem, v_stem, v_stem, c_stem}, endings)
end
end
-- Generate imperative parts for sound or assimilated verbs.
local function make_1stem_imperative(base, stem)
make_2stem_imperative(base, stem, stem)
end
-- Generate imperative parts for geminate verbs form I (also IV, VII, VIII, X).
local function make_gem_imperative(base, v_stem, c_stem)
make_2stem_imperative(base, v_stem, c_stem, imp_endings_alt_a, "alt gem")
make_2stem_imperative(base, v_stem, c_stem, imp_endings_alt_i, "alt gem")
make_2stem_imperative(base, v_stem, c_stem)
end
-------------------------------------------------------------------------------
-- Functions to inflect entire verbs --
-------------------------------------------------------------------------------
-- Generate finite parts of a sound verb (also works for assimilated verbs) from five stems (past and non-past,
-- active and passive, plus imperative) plus the prefix vowel in the active non-past ("a" or "u").
local function make_sound_verb(base, past_stem, past_pass_stem, nonpast_stem, nonpast_pass_stem, imp_stem,
prefix_vowel)
past_1stem_conj(base, "past", past_stem)
past_1stem_conj(base, "past_pass", past_pass_stem)
nonpast_1stem_conj(base, "ind", prefix_vowel, nonpast_stem)
nonpast_1stem_conj(base, "sub", prefix_vowel, nonpast_stem)
nonpast_1stem_conj(base, "juss", prefix_vowel, nonpast_stem)
nonpast_1stem_conj(base, "ind_pass", "u", nonpast_pass_stem)
nonpast_1stem_conj(base, "sub_pass", "u", nonpast_pass_stem)
nonpast_1stem_conj(base, "juss_pass", "u", nonpast_pass_stem)
make_1stem_imperative(base, imp_stem)
end
local function past_final_weak_endings_from_vowel(vowel)
if vowel == "ay" then
return past_endings_ay
elseif vowel == "aw" then
return past_endings_aw
elseif vowel == "ī" then
return past_endings_ii
elseif vowel == "ū" then
return past_endings_uu
elseif not vowel then
return nil
else
error(("Internal error: Unrecognized past final-weak vowel spec '%s'"):format(vowel))
end
end
local function nonpast_final_weak_endings_from_vowel(vowel)
if vowel == "ā" then
return ind_endings_aa, sub_endings_aa, juss_endings_aa, imp_endings_aa
elseif vowel == "ī" then
return ind_endings_ii, sub_endings_ii, juss_endings_ii, imp_endings_ii
elseif vowel == "ū" then
return ind_endings_uu, sub_endings_uu, juss_endings_uu, imp_endings_uu
elseif not vowel then
return nil
else
error(("Internal error: Unrecognized non-past final-weak vowel spec '%s'"):format(vowel))
end
end
-- Generate finite parts of a final-weak verb from five stems (past and non-past, active and passive, plus
-- imperative), the past active ending vowel (ay, aw, ī or ū), the non-past active ending vowel (ā, ī or ū) and the
-- prefix vowel in the active non-past (a or u).
local function make_final_weak_verb(base, past_stem, past_pass_stem, nonpast_stem, nonpast_pass_stem, imp_stem,
past_ending_vowel, nonpast_ending_vowel, prefix_vowel)
past_stem = override_stem_if_needed(base, "past", past_stem)
past_pass_stem = override_stem_if_needed(base, "past_pass", past_pass_stem)
-- Don't call override_stem_if_needed() here for non-past stems; it's called in nonpast_2stem_conj().
imp_stem = override_stem_if_needed(base, "imp", imp_stem)
-- + not supported for ending vowel overrides
past_ending_vowel = base.stem_overrides.past_final_weak_vowel or past_ending_vowel
local past_pass_ending_vowel = base.stem_overrides.past_pass_final_weak_vowel or "ī"
nonpast_ending_vowel = base.stem_overrides.nonpast_final_weak_vowel or nonpast_ending_vowel
local nonpast_pass_ending_vowel = base.stem_overrides.nonpast_pass_final_weak_vowel or "ā"
local past_endings = past_final_weak_endings_from_vowel(past_ending_vowel)
local past_pass_endings = past_final_weak_endings_from_vowel(past_pass_ending_vowel)
local ind_endings, sub_endings, juss_endings, imp_endings =
nonpast_final_weak_endings_from_vowel(nonpast_ending_vowel)
local ind_pass_endings, sub_pass_endings, juss_pass_endings =
nonpast_final_weak_endings_from_vowel(nonpast_pass_ending_vowel)
inflect_tense(base, "past", "", {past_stem, all_same = 1}, past_endings)
inflect_tense(base, "past_pass", "", {past_pass_stem, all_same = 1}, past_pass_endings)
nonpast_1stem_conj(base, "ind", prefix_vowel, nonpast_stem, ind_endings)
nonpast_1stem_conj(base, "sub", prefix_vowel, nonpast_stem, sub_endings)
nonpast_1stem_conj(base, "juss", prefix_vowel, nonpast_stem, juss_endings)
nonpast_1stem_conj(base, "ind_pass", "u", nonpast_pass_stem, ind_pass_endings)
nonpast_1stem_conj(base, "sub_pass", "u", nonpast_pass_stem, sub_pass_endings)
nonpast_1stem_conj(base, "juss_pass", "u", nonpast_pass_stem, juss_pass_endings)
inflect_tense_imp(base, {imp_stem, all_same = 1}, imp_endings)
end
-- Generate finite parts of an augmented (form II+) final-weak verb from five stems (past and non-past, active and
-- passive, plus imperative) plus the prefix vowel in the active non-past ("a" or "u") and a flag indicating if it
-- behaves like a form V/VI verb in taking non-past endings in -ā instead of -ī.
local function make_augmented_final_weak_verb(base, past_stem, past_pass_stem, nonpast_stem, nonpast_pass_stem,
imp_stem, prefix_vowel, form56)
make_final_weak_verb(base, past_stem, past_pass_stem, nonpast_stem, nonpast_pass_stem, imp_stem, "ay",
form56 and "ā" or "ī", prefix_vowel)
end
-- Generate finite parts of an augmented (form II+) sound or final-weak verb, given:
-- * `base` (conjugation data structure);
-- * `vowel_spec` (radicals, weakness);
-- * `past_stem_base` (active past stem minus last syllable (= -al or -ā));
-- * `nonpast_stem_base` (non-past stem minus last syllable (= -al/-il or -ā/-ī);
-- * `past_pass_stem_base` (passive past stem minus last syllable (= -il or -ī));
-- * `vn` (verbal noun).
local function make_augmented_sound_final_weak_verb(base, vowel_spec, past_stem_base, nonpast_stem_base,
past_pass_stem_base, vn)
insert_form_or_forms(base, "vn", vn)
local lastrad = base.quadlit and vowel_spec.rad4 or vowel_spec.rad3
local final_weak = is_final_weak(base, vowel_spec)
local prefix_vowel = prefix_vowel_from_vform(base.verb_form)
local form56 = vform_nonpast_a_vowel(base.verb_form)
local a_base_suffix = final_weak and "" or q(A, lastrad)
local i_base_suffix = final_weak and "" or q(I, lastrad)
-- past and non-past stems, active and passive
local past_stem = q(past_stem_base, a_base_suffix)
-- In forms 5 and 6, non-past has /a/ as last stem vowel in the non-past
-- in both active and passive, but /i/ in the active participle and /a/
-- in the passive participle. Elsewhere, consistent /i/ in active non-past
-- and participle, consistent /a/ in passive non-past and participle.
-- Hence, forms 5 and 6 differ only in the non-past active (but not
-- active participle), so we have to split the finite non-past stem and
-- active participle stem.
local nonpast_stem = q(nonpast_stem_base, form56 and a_base_suffix or i_base_suffix)
local ap_stem = q(nonpast_stem_base, i_base_suffix)
local past_pass_stem = q(past_pass_stem_base, i_base_suffix)
local nonpast_pass_stem = q(nonpast_stem_base, a_base_suffix)
-- imperative stem
local imp_stem = q(past_stem_base, form56 and a_base_suffix or i_base_suffix)
-- make parts
if final_weak then
make_augmented_final_weak_verb(base, past_stem, past_pass_stem, nonpast_stem, nonpast_pass_stem, imp_stem,
prefix_vowel, form56)
else
make_sound_verb(base, past_stem, past_pass_stem, nonpast_stem, nonpast_pass_stem, imp_stem, prefix_vowel)
end
-- active and passive participle
if final_weak then
insert_form_or_forms(base, "ap", q(MU, ap_stem, IN))
insert_form_or_forms(base, "pp", q(MU, nonpast_pass_stem, AN, AMAQ))
else
insert_form_or_forms(base, "ap", q(MU, ap_stem))
insert_form_or_forms(base, "pp", q(MU, nonpast_pass_stem))
end
end
-- Generate finite parts of a hollow or geminate verb from ten stems (vowel and consonant stems for each of past and
-- non-past, active and passive, plus imperative) plus the prefix vowel in the active non-past ("a" or "u"), plus a
-- flag indicating if we are a geminate verb.
local function make_hollow_geminate_verb(base, geminate, past_v_stem, past_c_stem, past_pass_v_stem,
past_pass_c_stem, nonpast_v_stem, nonpast_c_stem, nonpast_pass_v_stem, nonpast_pass_c_stem, imp_v_stem,
imp_c_stem, prefix_vowel, altgem_note)
past_2stem_conj(base, "past", past_v_stem, past_c_stem, altgem_note)
past_2stem_conj(base, "past_pass", past_pass_v_stem, past_pass_c_stem)
nonpast_2stem_conj(base, "ind", prefix_vowel, nonpast_v_stem, nonpast_c_stem)
nonpast_2stem_conj(base, "sub", prefix_vowel, nonpast_v_stem, nonpast_c_stem)
nonpast_2stem_conj(base, "ind_pass", "u", nonpast_pass_v_stem, nonpast_pass_c_stem)
nonpast_2stem_conj(base, "sub_pass", "u", nonpast_pass_v_stem, nonpast_pass_c_stem)
if geminate then
jussive_gem_conj(base, "juss", prefix_vowel, nonpast_v_stem, nonpast_c_stem)
jussive_gem_conj(base, "juss_pass", "u", nonpast_pass_v_stem, nonpast_pass_c_stem)
make_gem_imperative(base, imp_v_stem, imp_c_stem)
else
nonpast_2stem_conj(base, "juss", prefix_vowel, nonpast_v_stem, nonpast_c_stem)
nonpast_2stem_conj(base, "juss_pass", "u", nonpast_pass_v_stem, nonpast_pass_c_stem)
make_2stem_imperative(base, imp_v_stem, imp_c_stem)
end
end
-- Generate finite parts of an augmented (form II+) hollow verb, given:
-- * `base` (conjugation data structure);
-- * `vowel_spec` (radicals, weakness);
-- * `past_stem_base` (invariable part of active past stem);
-- * `nonpast_stem_base` (invariable part of nonpast stem);
-- * `past_pass_stem_base` (invariable part of passive past stem);
-- * `vn` (verbal noun).
local function make_augmented_hollow_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base,
vn)
insert_form_or_forms(base, "vn", vn)
local lastrad = base.quadlit and vowel_spec.rad4 or vowel_spec.rad3
local form410 = base.verb_form == "IV" or base.verb_form == "X"
local prefix_vowel = prefix_vowel_from_vform(base.verb_form)
local a_base_suffix_v, a_base_suffix_c
local i_base_suffix_v, i_base_suffix_c
a_base_suffix_v = q(AA, lastrad) -- 'af-āl-a, inf-āl-a
a_base_suffix_c = q(A, lastrad) -- 'af-al-tu, inf-al-tu
i_base_suffix_v = q(II, lastrad) -- 'uf-īl-a, unf-īl-a
i_base_suffix_c = q(I, lastrad) -- 'uf-il-tu, unf-il-tu
-- past and non-past stems, active and passive, for vowel-initial and
-- consonant-initial endings
local past_v_stem = q(past_stem_base, a_base_suffix_v)
local past_c_stem = q(past_stem_base, a_base_suffix_c)
-- yu-f-īl-u, ya-staf-īl-u but yanf-āl-u, yaft-āl-u
local nonpast_v_stem = q(nonpast_stem_base, form410 and i_base_suffix_v or a_base_suffix_v)
local nonpast_c_stem = q(nonpast_stem_base, form410 and i_base_suffix_c or a_base_suffix_c)
local past_pass_v_stem = q(past_pass_stem_base, i_base_suffix_v)
local past_pass_c_stem = q(past_pass_stem_base, i_base_suffix_c)
local nonpast_pass_v_stem = q(nonpast_stem_base, a_base_suffix_v)
local nonpast_pass_c_stem = q(nonpast_stem_base, a_base_suffix_c)
-- imperative stem
local imp_v_stem = q(past_stem_base, form410 and i_base_suffix_v or a_base_suffix_v)
local imp_c_stem = q(past_stem_base, form410 and i_base_suffix_c or a_base_suffix_c)
-- make parts
make_hollow_geminate_verb(base, false, past_v_stem, past_c_stem, past_pass_v_stem,
past_pass_c_stem, nonpast_v_stem, nonpast_c_stem, nonpast_pass_v_stem,
nonpast_pass_c_stem, imp_v_stem, imp_c_stem, prefix_vowel)
-- active participle
insert_form_or_forms(base, "ap", q(MU, nonpast_v_stem))
-- passive participle
insert_form_or_forms(base, "pp", q(MU, nonpast_pass_v_stem))
end
-- Generate finite parts of an augmented (form II+) geminate verb, given:
-- * `base` (conjugation data structure);
-- * `vowel_spec` (radicals, weakness);
-- * `past_stem_base` (invariable part of active past stem; this and the stem bases below will end with a consonant
-- for forms IV, X, IVq, and a short vowel for the others);
-- * `nonpast_stem_base` (invariable part of nonpast stem);
-- * `past_pass_stem_base` (invariable part of passive past stem);
-- * `vn` (verbal noun);
-- * `altgem_note` (footnote to add to active past 1/2-person forms, when alternative forms are supplied [form X]).
local function make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base,
past_pass_stem_base, vn, altgem_note)
insert_form_or_forms(base, "vn", vn)
local vform = base.verb_form
local lastrad = base.quadlit and vowel_spec.rad4 or vowel_spec.rad3
local prefix_vowel = prefix_vowel_from_vform(vform)
local a_base_suffix_v, a_base_suffix_c
local i_base_suffix_v, i_base_suffix_c
if vform == "IV" or vform == "X" or vform == "IVq" then
a_base_suffix_v = q(A, lastrad, SH) -- 'af-all
a_base_suffix_c = q(SK, lastrad, A, lastrad) -- 'af-lal
i_base_suffix_v = q(I, lastrad, SH) -- yuf-ill
i_base_suffix_c = q(SK, lastrad, I, lastrad) -- yuf-lil
else
a_base_suffix_v = q(lastrad, SH) -- fā-ll, infa-ll
a_base_suffix_c = q(lastrad, A, lastrad) -- fā-lal, infa-lal
i_base_suffix_v = q(lastrad, SH) -- yufā-ll, yanfa-ll
i_base_suffix_c = q(lastrad, I, lastrad) -- yufā-lil, yanfa-lil
end
-- past and non-past stems, active and passive, for vowel-initial and
-- consonant-initial endings
local past_v_stem = q(past_stem_base, a_base_suffix_v)
local past_c_stem = q(past_stem_base, a_base_suffix_c)
local nonpast_v_stem = q(nonpast_stem_base, vform_nonpast_a_vowel(vform) and a_base_suffix_v or i_base_suffix_v)
local nonpast_c_stem = q(nonpast_stem_base, vform_nonpast_a_vowel(vform) and a_base_suffix_c or i_base_suffix_c)
-- NOTE: Formerly had a comment that "vform III and VI passive past do not have contracted parts, only
-- uncontracted parts, which are added separately by those functions". This is based on Mace
-- "Arabic Verbs and Essential Grammar" (1999) entry 63 (continued), which shows passive ḥūjija but no ḥūjja;
-- but that is apparently a mistake, as (1) verb tables in other books do show contracted passive parts for
-- these forms; (2) there is no mention of such an exception on p. 99, which explains how geminate ("doubled")
-- verbs work (on the contrary, it says "The contracted and uncontracted pairs (see above) are found all
-- over Forms III and VI of the doubled verbs").
local past_pass_v_stem = q(past_pass_stem_base, i_base_suffix_v)
local past_pass_c_stem = q(past_pass_stem_base, i_base_suffix_c)
local nonpast_pass_v_stem = q(nonpast_stem_base, a_base_suffix_v)
local nonpast_pass_c_stem = q(nonpast_stem_base, a_base_suffix_c)
-- imperative stem
local imp_v_stem = q(past_stem_base, vform_nonpast_a_vowel(vform) and a_base_suffix_v or i_base_suffix_v)
local imp_c_stem = q(past_stem_base, vform_nonpast_a_vowel(vform) and a_base_suffix_c or i_base_suffix_c)
-- make parts
make_hollow_geminate_verb(base, "geminate", past_v_stem, past_c_stem, past_pass_v_stem,
past_pass_c_stem, nonpast_v_stem, nonpast_c_stem, nonpast_pass_v_stem,
nonpast_pass_c_stem, imp_v_stem, imp_c_stem, prefix_vowel, altgem_note)
-- active participle
insert_form_or_forms(base, "ap", q(MU, nonpast_v_stem))
-- passive participle
insert_form_or_forms(base, "pp", q(MU, nonpast_pass_v_stem))
end
-------------------------------------------------------------------------------
-- Conjugation functions for specific conjugation types --
-------------------------------------------------------------------------------
local function form_i_imp_stem_through_rad1(base, nonpast_vowel, rad1)
local imp_vowel = map_vowel(nonpast_vowel, function(vow)
if vow == A or vow == I then
return I
elseif vow == U then
return U
elseif not skip_slot(base, "imp_2ms") then
error(("Internal error: Non-past vowel %s isn't a, i, or u, should have been caught earlier"):format(
dump(nonpast_vowel)))
else
-- Passive-only; imperative won't ever be displayed so it doesn't matter.
return I
end
end)
-- Mace ("Arabic Verbs and Essentials of Grammar" p. 63: [https://archive.org/details/arabicverbsessen00john/page/62/mode/2up])
-- claims that initial hamza is assimilated/elided into a long vowel in the form-I imperative, but apparently
-- this isn't corrrect.
local vowel_on_alif = map_vowel(imp_vowel, function(vow)
return ALIF .. vow
end)
return q(vowel_on_alif, rad1, SK)
end
-- Implement form-I sound or assimilated verb. ASSIMILATED is true for assimilated verbs.
local function make_form_i_sound_assimilated_verb(base, vowel_spec, assimilated)
local rad1, rad2, rad3, past_vowel, nonpast_vowel = get_radicals_3(vowel_spec)
-- Verbal nouns (maṣādir) for form I are unpredictable and have to be supplied
-- past and non-past stems, active and passive
local past_stem = q(rad1, A, rad2, past_vowel, rad3)
local nonpast_stem = assimilated and q(rad2, nonpast_vowel, rad3) or
q(rad1, SK, rad2, nonpast_vowel, rad3)
local past_pass_stem = q(rad1, U, rad2, I, rad3)
local nonpast_pass_stem = q(rad1, SK, rad2, A, rad3)
-- imperative stem
-- check for irregular verb with reduced imperative (أَخَذَ or أَكَلَ or أَمَرَ)
local reducedimp = reduced_imperative_verb(rad1, rad2, rad3)
if reducedimp then
base.irregular = true
end
local imp_stem_suffix = q(rad2, nonpast_vowel, rad3)
local long_imp_stem_base = form_i_imp_stem_through_rad1(base, nonpast_vowel, rad1)
local short_imp_stem_base = ""
local imp_stem = q((assimilated or reducedimp) and "" or long_imp_stem_base, imp_stem_suffix)
-- make parts
make_sound_verb(base, past_stem, past_pass_stem, nonpast_stem, nonpast_pass_stem, imp_stem, "a")
if reducedimp == "shortlong" then
make_1stem_imperative(base, iut.combine_form_and_footnotes(q(long_imp_stem_base, imp_stem_suffix),
mw.getCurrentFrame():preprocess("[used especially with a clitic such as {{m|ar|فَ}} or {{m|ar|وَ}}]")))
end
-- Check for irregular verb سَأَلَ with alternative jussive and imperative. Calling this after make_sound_verb()
-- adds additional entries to the paradigm parts.
if saal_radicals(rad1, rad2, rad3) then
base.irregular = true
nonpast_1stem_conj(base, "juss", "a", "سَل")
nonpast_1stem_conj(base, "juss_pass", "u", "سَل")
make_1stem_imperative(base, "سَل")
end
-- Active participle.
insert_form_or_forms(base, "ap", q(rad1, AA, rad2, I, rad3))
-- Insert alternative active participle (stative type I) فَعِيل. Since not all verbs have this, we require that
-- verbs that do have it specify it explicitly; a shortcut ++ is provided to make this easier (e.g. <ap:++> to
-- indicate that the alternative form should be used for the active participle, <ap:+,++> to indicate that both
-- forms can be used, and <ap:-> to indicate that there is no active participle). The same form is used for
-- secondary default passive participle.
insert_ap2_pp2(base, q(rad1, A, rad2, II, rad3))
-- Active participle, stative type II فَعِل (+++).
insert_form_or_forms(base, "ap3", q(rad1, A, rad2, I, rad3))
-- Active participle, color/defect أَفْعَل (+cd).
insert_form_or_forms(base, "apcd", q(HAMZA, A, rad1, SK, rad2, A, rad3))
-- Active participle, -ān فَعْلَان (+an).
insert_form_or_forms(base, "apan", q(rad1, A, rad2, SK, rad3, AAN))
-- Passive participle.
insert_form_or_forms(base, "pp", q(MA, rad1, SK, rad2, UU, rad3))
end
conjugations["I-sound"] = function(base, vowel_spec)
make_form_i_sound_assimilated_verb(base, vowel_spec, false)
end
conjugations["none-sound"] = function(base, vowel_spec)
-- All default stems are nil.
make_sound_verb(base)
end
conjugations["none-hollow"] = function(base, vowel_spec)
-- All default stems are nil.
make_hollow_geminate_verb(base, false)
end
conjugations["none-geminate"] = function(base, vowel_spec)
-- All default stems are nil.
make_hollow_geminate_verb(base, "geminate")
end
conjugations["none-final-weak"] = function(base, vowel_spec)
-- All default stems are nil.
make_final_weak_verb(base)
end
conjugations["I-assimilated"] = function(base, vowel_spec)
make_form_i_sound_assimilated_verb(base, vowel_spec, "assimilated")
end
local function make_form_i_hayy_verb(base, vowel_spec)
-- Verbal nouns (maṣādir) for form I are unpredictable and have to be supplied
base.irregular = true
-- past and non-past stems, active and passive, and imperative stem
local past_c_stem = "حَيِي"
local past_v_stem_long = past_c_stem
local past_v_stem_short = "حَيّ"
local past_pass_c_stem = "حُيِي"
local past_pass_v_stem_long = past_pass_c_stem
local past_pass_v_stem_short = "حُيّ"
local nonpast_stem = "حْي"
local nonpast_pass_stem = nonpast_stem
local imp_stem = _I .. nonpast_stem
-- make parts
past_2stem_conj(base, "past", {}, past_c_stem)
past_2stem_conj(base, "past_pass", {}, past_pass_c_stem)
local variant = vowel_spec.variant or "both"
if variant == "short" or variant == "both" then
past_2stem_conj(base, "past", past_v_stem_short, {})
past_2stem_conj(base, "past_pass", past_pass_v_stem_short, {})
end
function inflect_long_variant(tense, long_stem, short_stem)
inflect_tense_1(base, tense, "",
{long_stem, long_stem, long_stem, long_stem, short_stem},
{past_endings[4], past_endings[5], past_endings[7], past_endings[8],
past_endings[12]},
{"3ms", "3fs", "3md", "3fd", "3mp"})
end
if variant == "long" or variant == "both" then
inflect_long_variant("past", past_v_stem_long, past_v_stem_short)
inflect_long_variant("past_pass", past_pass_v_stem_long, past_pass_v_stem_short)
end
nonpast_1stem_conj(base, "ind", "a", nonpast_stem, ind_endings_aa)
nonpast_1stem_conj(base, "sub", "a", nonpast_stem, sub_endings_aa)
nonpast_1stem_conj(base, "juss", "a", nonpast_stem, juss_endings_aa)
nonpast_1stem_conj(base, "ind_pass", "u", nonpast_pass_stem, ind_endings_aa)
nonpast_1stem_conj(base, "sub_pass", "u", nonpast_pass_stem, sub_endings_aa)
nonpast_1stem_conj(base, "juss_pass", "u", nonpast_pass_stem, juss_endings_aa)
inflect_tense_imp(base, {imp_stem, all_same = 1}, imp_endings_aa)
-- active and passive participles apparently do not exist for this verb
end
-- Implement form-I final-weak assimilated+final-weak verb. ASSIMILATED is true for assimilated verbs.
local function make_form_i_final_weak_verb(base, vowel_spec, assimilated)
local rad1, rad2, rad3, past_vowel, nonpast_vowel = get_radicals_3(vowel_spec)
-- حَيَّ or حَيِيَ is weird enough that we handle it as a separate function.
if hayy_radicals(rad1, rad2, rad3) then
make_form_i_hayy_verb(base, vowel_spec)
return
end
-- Verbal nouns (maṣādir) for form I are unpredictable and have to be supplied.
-- Past and non-past stems, active and passive, and imperative stem.
local past_stem = q(rad1, A, rad2)
local past_pass_stem = q(rad1, U, rad2)
local nonpast_stem, nonpast_pass_stem, imp_stem
if raa_radicals(rad1, rad2, rad3) then
base.irregular = true
nonpast_stem = rad1
nonpast_pass_stem = rad1
imp_stem = rad1
else
nonpast_pass_stem = q(rad1, SK, rad2)
if assimilated then
nonpast_stem = rad2
imp_stem = rad2
else
nonpast_stem = nonpast_pass_stem
imp_stem = q(form_i_imp_stem_through_rad1(base, nonpast_vowel, rad1), rad2)
end
end
-- Make parts.
local past_ending_vowel =
req(rad3, Y) and req(past_vowel, A) and "ay" or
req(rad3, W) and req(past_vowel, A) and "aw" or
req(past_vowel, I) and "ī" or "ū"
-- Try to preserve footnotes attached to the third radical and/or past and/or non-past vowels.
local past_footnotes = iut.combine_footnotes(rget_footnotes(rad3), rget_footnotes(past_vowel))
local nonpast_ending_vowel = req(nonpast_vowel, A) and "ā" or req(nonpast_vowel, I) and "ī" or "ū"
local nonpast_footnotes = iut.combine_footnotes(rget_footnotes(rad3), rget_footnotes(nonpast_vowel))
make_final_weak_verb(base,
iut.combine_form_and_footnotes(past_stem, past_footnotes),
iut.combine_form_and_footnotes(past_pass_stem, past_footnotes),
iut.combine_form_and_footnotes(nonpast_stem, nonpast_footnotes),
iut.combine_form_and_footnotes(nonpast_pass_stem, nonpast_footnotes),
iut.combine_form_and_footnotes(imp_stem, nonpast_footnotes),
past_ending_vowel, nonpast_ending_vowel, "a")
-- Active participle.
insert_form_or_forms(base, "ap", q(rad1, AA, rad2, IN))
-- Active participle, stative type I فَعِيّ (++). FIXME: Is this correct when rad3 is W?
insert_ap2_pp2(base, q(rad1, A, rad2, II, SH))
-- Active participle, stative type II فَعٍ (+++). FIXME: Any examples of this to verify it's correct?
insert_form_or_forms(base, "ap3", q(rad1, A, rad2, IN))
-- Active participle, color/defect أَفْعَى (+cd).
insert_form_or_forms(base, "apcd", q(HAMZA, A, rad1, SK, rad2, AAMAQ))
-- Active participle, -ān فَعْيَان or فَعْوَان (+an).
-- FIXME: Any examples of this for both rad3 = W and y to verify it's correct?
insert_form_or_forms(base, "apan", q(rad1, A, rad2, SK, rad3, AAN))
-- Passive participle.
insert_form_or_forms(base, "pp", q(MA, rad1, SK, rad2, req(rad3, Y) and II or UU, SH))
end
conjugations["I-final-weak"] = function(base, vowel_spec)
make_form_i_final_weak_verb(base, vowel_spec, false)
end
conjugations["I-assimilated+final-weak"] = function(base, vowel_spec)
make_form_i_final_weak_verb(base, vowel_spec, "assimilated")
end
conjugations["I-hollow"] = function(base, vowel_spec)
local rad1, rad2, rad3, past_vowel, nonpast_vowel = get_radicals_3(vowel_spec)
-- Formerly we signaled an error when past_vowel is "a" but that seems too harsh. We can interpret a past vowel
-- of "a" as meaning to use the non-past vowel in forms requiring a short vowel. If the non-past vowel is "a"
-- then the past vowel can only be "i" (e.g. in nāma yanāmu with first singular past of nimtu).
if req(past_vowel, A) then
-- Enable this when we've fixed the upstream code
-- error("Internal error: For form I hollow, past vowel 'a' should have been converted to 'i' or 'u' already")
past_vowel = map_vowel(past_vowel, function(vow)
return req(nonpast_vowel, A) and I or rget(nonpast_vowel)
end)
end
local lengthened_nonpast = map_vowel(nonpast_vowel, function(vow)
return vow == U and UU or vow == I and II or AA
end)
-- Verbal nouns (maṣādir) for form I are unpredictable and have to be supplied.
-- active past stems - vowel (v) and consonant (c)
local past_v_stem = q(rad1, AA, rad3)
local past_c_stem = q(rad1, past_vowel, rad3)
-- active non-past stems - vowel (v) and consonant (c)
local nonpast_v_stem = q(rad1, lengthened_nonpast, rad3)
local nonpast_c_stem = q(rad1, nonpast_vowel, rad3)
-- passive past stems - vowel (v) and consonant (c)
-- 'ufīla, 'ufiltu
local past_pass_v_stem = q(rad1, II, rad3)
local past_pass_c_stem = q(rad1, I, rad3)
-- passive non-past stems - vowel (v) and consonant (c)
-- yufāla/yufalna
-- stem is built differently but conjugation is identical to sound verbs
local nonpast_pass_v_stem = q(rad1, AA, rad3)
local nonpast_pass_c_stem = q(rad1, A, rad3)
-- imperative stem
local imp_v_stem = nonpast_v_stem
local imp_c_stem = nonpast_c_stem
-- make parts
make_hollow_geminate_verb(base, false, past_v_stem, past_c_stem, past_pass_v_stem,
past_pass_c_stem, nonpast_v_stem, nonpast_c_stem, nonpast_pass_v_stem,
nonpast_pass_c_stem, imp_v_stem, imp_c_stem, "a")
if kaan_radicals(rad1, rad2, rad3) then
local endings = make_nonpast_endings(U, {}, {}, {}, {})
inflect_tense(base, "juss", nonpast_prefix_consonants, q(A, rad1), endings)
end
-- Active participle.
insert_form_or_forms(base, "ap", req(rad3, HAMZA) and q(rad1, AA, HAMZA, IN) or
q(rad1, AA, HAMZA, I, rad3))
-- Active participle, stative type I فَيِّد (++). FIXME: Any examples of this to verify it's correct?
insert_ap2_pp2(base, q(rad1, A, Y, SH, I, rad3))
-- Active participle, stative type II فَيِد (+++). FIXME: Any examples of this to verify it's correct?
insert_form_or_forms(base, "ap3", q(rad1, A, Y, I, rad3))
-- Active participle, color/defect أَفّيَد or أَفّوَد (+cd). FIXME: Any examples of this to verify it's correct?
insert_form_or_forms(base, "apcd", q(HAMZA, A, rad1, SK, rad2, A, rad3))
-- Active participle, -ān فَيْدَان or فَوْدَان (+an). Example: جَاعَ "to be hungry", act part جَوْعَان
insert_form_or_forms(base, "apan", q(rad1, A, rad2, SK, rad3, AAN))
-- Passive participle.
insert_form_or_forms(base, "pp", q(MA, rad1, req(rad2, Y) and II or UU, rad3))
end
conjugations["I-geminate"] = function(base, vowel_spec)
local rad1, rad2, rad3, past_vowel, nonpast_vowel = get_radicals_3(vowel_spec)
-- Verbal nouns (maṣādir) for form I are unpredictable and have to be supplied.
-- active past stems - vowel (v) and consonant (c)
local past_v_stem = q(rad1, A, rad2, SH)
local past_c_stem = q(rad1, A, rad2, past_vowel, rad2)
-- active non-past stems - vowel (v) and consonant (c)
local nonpast_v_stem = q(rad1, nonpast_vowel, rad2, SH)
local nonpast_c_stem = q(rad1, SK, rad2, nonpast_vowel, rad2)
-- passive past stems - vowel (v) and consonant (c)
-- dulla/dulilta
local past_pass_v_stem = q(rad1, U, rad2, SH)
local past_pass_c_stem = q(rad1, U, rad2, I, rad2)
-- passive non-past stems - vowel (v) and consonant (c)
--yudallu/yudlalna
-- stem is built differently but conjugation is identical to sound verbs
local nonpast_pass_v_stem = q(rad1, A, rad2, SH)
local nonpast_pass_c_stem = q(rad1, SK, rad2, A, rad2)
-- imperative stem
local imp_v_stem = q(rad1, nonpast_vowel, rad2, SH)
local imp_c_stem = q(form_i_imp_stem_through_rad1(base, nonpast_vowel, rad1), rad2, nonpast_vowel, rad2)
-- make parts
make_hollow_geminate_verb(base, "geminate", past_v_stem, past_c_stem, past_pass_v_stem,
past_pass_c_stem, nonpast_v_stem, nonpast_c_stem, nonpast_pass_v_stem,
nonpast_pass_c_stem, imp_v_stem, imp_c_stem, "a")
-- Active participle.
insert_form_or_forms(base, "ap", q(rad1, AA, rad2, SH))
-- Active participle, stative type I فَعِيع (++). FIXME: Any examples of this to verify it's correct?
insert_ap2_pp2(base, q(rad1, A, rad2, II, rad2))
-- Active participle, stative type II فَعّ (+++). Example: بَرَّ "to be pious", active participle بَرّ
insert_form_or_forms(base, "ap3", q(rad1, A, rad2, SH))
-- Active participle, color/defect أَفَعّ (+cd).
-- Example: لَصَّ "to be thievish, to steal repeatedly", active participle أَلَصّ.
insert_form_or_forms(base, "apcd", q(HAMZA, A, rad1, A, rad2, SH))
-- Active participle, -ān فَعَّان (+an). FIXME: Any examples of this to verify it's correct?
insert_form_or_forms(base, "apan", q(rad1, A, rad2, SH, AAN))
-- Passive participle.
insert_form_or_forms(base, "pp", q(MA, rad1, SK, rad2, UU, rad2))
end
-- Return the ta- (active, past and non-past) and tu- (passive past) prefixes for a form II/III/V/VI verb.
-- Form V and VI verbs normally use ta- and tu-, but reduced (base.reduced) verbs use different prefixes. Form II
-- and III verbs have no prefix.
local function form_ii_iii_v_vi_ta_tu_prefix(base, rad1)
local vform = base.verb_form
if vform == "V" or vform == "VI" then
if base.reduced then
-- To simplify the code, we generate two rad1's with a sukūn between them, which is cleaned up in
-- postprocessing.
return q(_I, rad1, SK), q(rad1, SK), q(_U, rad1, SK)
else
return TA, TA, TU
end
else
return "", "", ""
end
end
-- Make form II or V sound or final-weak verb.
local function make_form_ii_v_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local final_weak = is_final_weak(base, vowel_spec)
local vform = base.verb_form
local ta_past_prefix, ta_nonpast_prefix, tu_past_prefix = form_ii_iii_v_vi_ta_tu_prefix(base, rad1)
local vn = vform == "V" and
q(ta_past_prefix, rad1, A, rad2, SH, final_weak and IN or q(U, rad3)) or
q(TA, rad1, SK, rad2, II, final_weak and AH or rad3)
-- various stem bases
local past_stem_base = q(ta_past_prefix, rad1, A, rad2, SH)
local nonpast_stem_base = q(ta_nonpast_prefix, rad1, A, rad2, SH)
local past_pass_stem_base = q(tu_past_prefix, rad1, U, rad2, SH)
-- make parts
make_augmented_sound_final_weak_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base,
vn)
end
conjugations["II-sound"] = function(base, vowel_spec)
make_form_ii_v_sound_final_weak_verb(base, vowel_spec)
end
conjugations["II-final-weak"] = function(base, vowel_spec)
make_form_ii_v_sound_final_weak_verb(base, vowel_spec)
end
local function make_form_iii_alt_vn(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local final_weak = is_final_weak(base, vowel_spec)
-- Insert alternative verbal noun فِعَال. Since not all verbs have this, we require that verbs that do have it
-- specify it explicitly; a shortcut ++ is provided to make this easier (e.g. <vn:+,++> to indicate that
-- both the normal verbal noun مُفَاعَلَة and secondary verbal noun فِعَال are available).
insert_form_or_forms(base, "vn2", q(rad1, I, rad2, AA, final_weak and HAMZA or rad3))
end
-- Make form III or VI sound or final-weak verb.
local function make_form_iii_vi_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local final_weak = is_final_weak(base, vowel_spec)
local vform = base.verb_form
local ta_past_prefix, ta_nonpast_prefix, tu_past_prefix = form_ii_iii_v_vi_ta_tu_prefix(base, rad1)
local vn = vform == "VI" and
q(ta_past_prefix, rad1, AA, rad2, final_weak and IN or q(U, rad3)) or
q(MU, rad1, AA, rad2, final_weak and AAH or q(A, rad3, AH))
-- various stem bases
local past_stem_base = q(ta_past_prefix, rad1, AA, rad2)
local nonpast_stem_base = q(ta_nonpast_prefix, rad1, AA, rad2)
local past_pass_stem_base = q(tu_past_prefix, rad1, UU, rad2)
-- make parts
make_augmented_sound_final_weak_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base,
vn)
if vform == "III" then
make_form_iii_alt_vn(base, vowel_spec)
end
end
conjugations["III-sound"] = function(base, vowel_spec)
make_form_iii_vi_sound_final_weak_verb(base, vowel_spec)
end
conjugations["III-final-weak"] = function(base, vowel_spec)
make_form_iii_vi_sound_final_weak_verb(base, vowel_spec)
end
-- Make form III or VI geminate verb.
local function make_form_iii_vi_geminate_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vform = base.verb_form
local ta_past_prefix, ta_nonpast_prefix, tu_past_prefix = form_ii_iii_v_vi_ta_tu_prefix(base, rad1)
-- Alternative verbal noun فِعَال will be inserted when we add sound parts below.
local vn = vform == "VI" and q(ta_past_prefix, rad1, AA, rad2, SH) or q(MU, rad1, AA, rad2, SH, AH)
-- Various stem bases.
local past_stem_base = q(ta_past_prefix, rad1, AA)
local nonpast_stem_base = q(ta_nonpast_prefix, rad1, AA)
local past_pass_stem_base = q(tu_past_prefix, rad1, UU)
-- Make parts.
local variant = vowel_spec.variant or "short"
if variant == "short" or variant == "both" then
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
-- Also add alternative sound (non-compressed) parts. This will lead to some duplicate entries, but they are
-- removed during addition.
if variant == "long" or variant == "both" then
make_form_iii_vi_sound_final_weak_verb(base, vowel_spec)
elseif vform == "III" then
-- Still need to add the alternative form-III verbal noun.
make_form_iii_alt_vn(base, vowel_spec)
end
end
conjugations["III-geminate"] = function(base, vowel_spec)
make_form_iii_vi_geminate_verb(base, vowel_spec)
end
-- Make form IV sound or final-weak verb.
local function make_form_iv_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local final_weak = is_final_weak(base, vowel_spec)
-- core of stem base, minus stem prefixes
local stem_core
-- check for irregular verb أَرَى
local is_raa = raa_radicals(rad1, rad2, rad3)
if is_raa then
base.irregular = true
stem_core = rad1
else
stem_core = q(rad1, SK, rad2)
end
-- verbal noun
local vn = is_raa and
q(HAMZA, I, stem_core, AA, HAMZA, AH) or
q(HAMZA, I, stem_core, AA, final_weak and HAMZA or rad3)
-- various stem bases
local past_stem_base = q(HAMZA, A, stem_core)
local nonpast_stem_base = stem_core
local past_pass_stem_base = q(HAMZA, U, stem_core)
-- make parts
make_augmented_sound_final_weak_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base,
vn)
end
conjugations["IV-sound"] = function(base, vowel_spec)
make_form_iv_sound_final_weak_verb(base, vowel_spec)
end
conjugations["IV-final-weak"] = function(base, vowel_spec)
make_form_iv_sound_final_weak_verb(base, vowel_spec)
end
conjugations["IV-hollow"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
-- verbal noun
local vn = q(HAMZA, I, rad1, AA, rad3, AH)
-- various stem bases
local past_stem_base = q(HAMZA, A, rad1)
local nonpast_stem_base = rad1
local past_pass_stem_base = q(HAMZA, U, rad1)
-- make parts
make_augmented_hollow_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
conjugations["IV-geminate"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vn = q(HAMZA, I, rad1, SK, rad2, AA, rad2)
-- various stem bases
local past_stem_base = q(HAMZA, A, rad1)
local nonpast_stem_base = rad1
local past_pass_stem_base = q(HAMZA, U, rad1)
-- make parts
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
conjugations["V-sound"] = function(base, vowel_spec)
make_form_ii_v_sound_final_weak_verb(base, vowel_spec)
end
conjugations["V-final-weak"] = function(base, vowel_spec)
make_form_ii_v_sound_final_weak_verb(base, vowel_spec)
end
conjugations["VI-sound"] = function(base, vowel_spec)
make_form_iii_vi_sound_final_weak_verb(base, vowel_spec)
end
conjugations["VI-final-weak"] = function(base, vowel_spec)
make_form_iii_vi_sound_final_weak_verb(base, vowel_spec)
end
conjugations["VI-geminate"] = function(base, vowel_spec)
make_form_iii_vi_geminate_verb(base, vowel_spec)
end
-- Make a verbal noun of the general form that applies to forms VII and above. RAD12 is the first consonant cluster
-- (after initial اِ) and RAD34 is the second consonant cluster. RAD5 is the final consonant.
local function high_form_verbal_noun(rad12, rad34, rad5)
return q(_I, rad12, I, rad34, AA, rad5)
end
-- Populate a sound or final-weak verb for any of the various high-numbered augmented forms (form VII and up) that
-- have up to 5 consonants in two clusters in the stem and the same pattern of vowels between. Some of these
-- consonants in certain verb parts are w's, which leads to apparent anomalies in certain stems of these parts, but
-- these anomalies are handled automatically in postprocessing, where we resolve sequences of iwC -> īC, uwC -> ūC,
-- w + sukūn + w -> w + shadda.
-- RAD12 is the first consonant cluster (after initial اِ) and RAD34 is the second consonant cluster. RAD5 is the
-- final consonant.
local function make_high_form_sound_final_weak_verb(base, vowel_spec, rad12, rad34, rad5)
local final_weak = is_final_weak(base, vowel_spec)
local vn = high_form_verbal_noun(rad12, rad34, final_weak and HAMZA or rad5)
-- various stem bases
local nonpast_stem_base = q(rad12, A, rad34)
local past_stem_base = q(_I, nonpast_stem_base)
local past_pass_stem_base = q(_U, rad12, U, rad34)
-- make parts
make_augmented_sound_final_weak_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base,
vn)
end
local function form_vii_nrad1(base, rad1)
if base.reduced then
if not req(rad1, M) then
error(("Internal error: Form VII first radical %s is not م but .reduced specified; should have been caught earlier"):
format(rget(rad1)))
end
return M .. SH
else
return q("نْ", rad1)
end
end
-- Make form VII sound or final-weak verb.
local function make_form_vii_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
make_high_form_sound_final_weak_verb(base, vowel_spec, form_vii_nrad1(base, rad1), rad2, rad3)
end
conjugations["VII-sound"] = function(base, vowel_spec)
make_form_vii_sound_final_weak_verb(base, vowel_spec)
end
conjugations["VII-final-weak"] = function(base, vowel_spec)
make_form_vii_sound_final_weak_verb(base, vowel_spec)
end
conjugations["VII-hollow"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local nrad1 = form_vii_nrad1(base, rad1)
local vn = high_form_verbal_noun(nrad1, Y, rad3)
-- various stem bases
local nonpast_stem_base = nrad1
local past_stem_base = q(_I, nonpast_stem_base)
local past_pass_stem_base = q(_U, nrad1)
-- make parts
make_augmented_hollow_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
conjugations["VII-geminate"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local nrad1 = form_vii_nrad1(base, rad1)
local vn = high_form_verbal_noun(nrad1, rad2, rad2)
-- various stem bases
local nonpast_stem_base = q(nrad1, A)
local past_stem_base = q(_I, nonpast_stem_base)
local past_pass_stem_base = q(_U, nrad1, U)
-- make parts
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
-- Return Form VIII verbal noun.
local function form_viii_verbal_noun(base, vowel_spec, rad1, rad2, rad3)
local final_weak = is_final_weak(base, vowel_spec)
rad3 = final_weak and HAMZA or rad3
return {high_form_verbal_noun(vowel_spec.form_viii_assim, rad2, rad3)}
end
-- Make form VIII sound or final-weak verb.
local function make_form_viii_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
-- check for irregular verb اِتَّخَذَ
if axadh_radicals(rad1, rad2, rad3) then
base.irregular = true
rad1 = T
end
make_high_form_sound_final_weak_verb(base, vowel_spec, vowel_spec.form_viii_assim, rad2, rad3)
end
conjugations["VIII-sound"] = function(base, vowel_spec)
make_form_viii_sound_final_weak_verb(base, vowel_spec)
end
conjugations["VIII-final-weak"] = function(base, vowel_spec)
make_form_viii_sound_final_weak_verb(base, vowel_spec)
end
conjugations["VIII-hollow"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vn = form_viii_verbal_noun(base, vowel_spec, rad1, Y, rad3)
-- various stem bases
local nonpast_stem_base = vowel_spec.form_viii_assim
local past_stem_base = q(_I, nonpast_stem_base)
local past_pass_stem_base = q(_U, nonpast_stem_base)
-- make parts
make_augmented_hollow_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
conjugations["VIII-geminate"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vn = form_viii_verbal_noun(base, vowel_spec, rad1, rad2, rad2)
-- various stem bases
local nonpast_stem_base = q(vowel_spec.form_viii_assim, A)
local past_stem_base = q(_I, nonpast_stem_base)
local past_pass_stem_base = q(_U, vowel_spec.form_viii_assim, U)
-- make parts
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
conjugations["IX-sound"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vn = q(_I, rad1, SK, rad2, I, rad3, AA, rad3)
-- various stem bases
local nonpast_stem_base = q(rad1, SK, rad2, A)
local past_stem_base = q(_I, nonpast_stem_base)
local past_pass_stem_base = q(_U, rad1, SK, rad2, U)
-- make parts
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
conjugations["IX-final-weak"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
make_high_form_sound_final_weak_verb(base, vowel_spec, q(rad1, SK, rad2), rad3, rad3)
end
-- Populate a sound or final-weak verb for any of the various high-numbered
-- augmented forms that have 5 consonants in the stem and the same pattern of
-- vowels. Some of these consonants in certain verb parts are w's, which leads to
-- apparent anomalies in certain stems of these parts, but these anomalies
-- are handled automatically in postprocessing, where we resolve sequences of
-- iwC -> īC, uwC -> ūC, w + sukūn + w -> w + shadda.
local function make_high5_form_sound_final_weak_verb(base, vowel_spec, rad1, rad2, rad3, rad4, rad5)
make_high_form_sound_final_weak_verb(base, vowel_spec, q(rad1, SK, rad2), q(rad3, SK, rad4), rad5)
end
-- Make form X sound or final-weak verb.
local function make_form_x_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
-- check for irregular verb اِسْتَحْيَا (also اِسْتَحَى)
local is_hayy = hayy_radicals(rad1, rad2, rad3)
local variant = vowel_spec.variant or "both"
if not is_hayy or variant == "long" or variant == "both" then
make_high5_form_sound_final_weak_verb(base, vowel_spec, S, T, rad1, rad2, rad3)
end
if is_hayy and (variant == "short" or variant == "both") then
base.irregular = true
-- Add alternative entries to the verbal paradigms. Any duplicates are removed during addition.
make_high_form_sound_final_weak_verb(base, vowel_spec, S .. SK .. T, rad1, rad3)
end
end
conjugations["X-sound"] = function(base, vowel_spec)
make_form_x_sound_final_weak_verb(base, vowel_spec)
end
conjugations["X-final-weak"] = function(base, vowel_spec)
make_form_x_sound_final_weak_verb(base, vowel_spec)
end
conjugations["X-hollow"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vn = q(base.reduced and "اِسْ" or "اِسْتِ", rad1, AA, rad3, AH)
-- various stem bases
local past_stem_base = q(base.reduced and "اِسْ" or "اِسْتَ", rad1)
local nonpast_stem_base = q(base.reduced and "سْ" or "سْتَ", rad1)
local past_pass_stem_base = q(base.reduced and "اُسْ" or "اُسْتُ", rad1)
-- make parts
make_augmented_hollow_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
conjugations["X-geminate"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vn = q("اِسْتِ", rad1, SK, rad2, AA, rad2)
-- various stem bases
local past_stem_base = q("اِسْتَ", rad1)
local nonpast_stem_base = q("سْتَ", rad1)
local past_pass_stem_base = q("اُسْتُ", rad1)
-- make parts
if base.altgem then
inflect_tense(base, "past", "", {q(past_stem_base, A, rad2, SH), all_same = 1},
past_endings_ay_12_person_only)
end
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn,
base.altgem and "[uncommon]" or nil)
end
conjugations["XI-sound"] = function(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local vn = q(_I, rad1, SK, rad2, II, rad3, AA, rad3)
-- various stem bases
local nonpast_stem_base = q(rad1, SK, rad2, AA)
local past_stem_base = q(_I, nonpast_stem_base)
local past_pass_stem_base = q(_U, rad1, SK, rad2, UU)
-- make parts
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
-- Probably no form XI final-weak, since already geminate in form; would behave as XI-sound.
-- Make form XII sound or final-weak verb.
local function make_form_xii_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
make_high5_form_sound_final_weak_verb(base, vowel_spec, rad1, rad2, W, rad2, rad3)
end
conjugations["XII-sound"] = function(base, vowel_spec)
make_form_xii_sound_final_weak_verb(base, vowel_spec)
end
conjugations["XII-final-weak"] = function(base, vowel_spec)
make_form_xii_sound_final_weak_verb(base, vowel_spec)
end
-- Make form XIII sound or final-weak verb.
local function make_form_xiii_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
make_high5_form_sound_final_weak_verb(base, vowel_spec, rad1, rad2, W, W, rad3)
end
conjugations["XIII-sound"] = function(base, vowel_spec)
make_form_xiii_sound_final_weak_verb(base, vowel_spec)
end
conjugations["XIII-final-weak"] = function(base, vowel_spec)
make_form_xiii_sound_final_weak_verb(base, vowel_spec)
end
-- Make a form XIV or XV sound or final-weak verb. Last radical appears twice (if`anlala / yaf`anlilu) so if it were
-- w or y you'd get if`anwā / yaf`anwī or if`anyā / yaf`anyī, i.e. unlike for most augmented verbs, the identity of
-- the radical matters.
local function make_form_xiv_xv_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3 = get_radicals_3(vowel_spec)
local lastrad = base.verb_form == "XV" and Y or rad3
make_high5_form_sound_final_weak_verb(base, vowel_spec, rad1, rad2, N, rad3, lastrad)
end
conjugations["XIV-sound"] = function(base, vowel_spec)
make_form_xiv_xv_sound_final_weak_verb(base, vowel_spec)
end
conjugations["XIV-final-weak"] = function(base, vowel_spec)
make_form_xiv_xv_sound_final_weak_verb(base, vowel_spec)
end
conjugations["XV-sound"] = function(base, vowel_spec)
make_form_xiv_xv_sound_final_weak_verb(base, vowel_spec)
end
-- Probably no form XV final-weak, since already final-weak in form; would behave as XV-sound.
-- Make form Iq or IIq sound or final-weak verb.
local function make_form_iq_iiq_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3, rad4 = get_radicals_4(vowel_spec)
local final_weak = is_final_weak(base, vowel_spec)
local vform = base.verb_form
local vn = vform == "IIq" and
q(TA, rad1, A, rad2, SK, rad3, (final_weak and IN or q(U, rad4))) or
q(rad1, A, rad2, SK, rad3, (final_weak and AAH or q(A, rad4, AH)))
local ta_pref = vform == "IIq" and TA or ""
local tu_pref = vform == "IIq" and TU or ""
-- various stem bases
local past_stem_base = q(ta_pref, rad1, A, rad2, SK, rad3)
local nonpast_stem_base = past_stem_base
local past_pass_stem_base = q(tu_pref, rad1, U, rad2, SK, rad3)
-- make parts
make_augmented_sound_final_weak_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base,
vn)
end
conjugations["Iq-sound"] = function(base, vowel_spec)
make_form_iq_iiq_sound_final_weak_verb(base, vowel_spec)
end
conjugations["Iq-final-weak"] = function(base, vowel_spec)
make_form_iq_iiq_sound_final_weak_verb(base, vowel_spec)
end
conjugations["IIq-sound"] = function(base, vowel_spec)
make_form_iq_iiq_sound_final_weak_verb(base, vowel_spec)
end
conjugations["IIq-final-weak"] = function(base, vowel_spec)
make_form_iq_iiq_sound_final_weak_verb(base, vowel_spec)
end
-- Make form IIIq sound or final-weak verb.
local function make_form_iiiq_sound_final_weak_verb(base, vowel_spec)
local rad1, rad2, rad3, rad4 = get_radicals_4(vowel_spec)
make_high5_form_sound_final_weak_verb(base, vowel_spec, rad1, rad2, N, rad3, rad4)
end
conjugations["IIIq-sound"] = function(base, vowel_spec)
make_form_iiiq_sound_final_weak_verb(base, vowel_spec)
end
conjugations["IIIq-final-weak"] = function(base, vowel_spec)
make_form_iiiq_sound_final_weak_verb(base, vowel_spec)
end
conjugations["IVq-sound"] = function(base, vowel_spec)
local rad1, rad2, rad3, rad4 = get_radicals_4(vowel_spec)
local vn = q(_I, rad1, SK, rad2, I, rad3, SK, rad4, AA, rad4)
-- various stem bases
local past_stem_base = q(_I, rad1, SK, rad2, A, rad3)
local nonpast_stem_base = q(rad1, SK, rad2, A, rad3)
local past_pass_stem_base = q(_U, rad1, SK, rad2, U, rad3)
-- make parts
make_augmented_geminate_verb(base, vowel_spec, past_stem_base, nonpast_stem_base, past_pass_stem_base, vn)
end
-- Probably no form IVq final-weak, since already geminate in form; would behave as IVq-sound.
end
create_conjugations()
-------------------------------------------------------------------------------
-- Guts of main conjugation function --
-------------------------------------------------------------------------------
-- Given form, weakness and radicals, check to make sure the radicals present are allowable for the weakness. Hamzas on
-- alif/wāw/yāʾ seats are never allowed (should always appear as hamza-on-the-line), and various weaknesses have various
-- strictures on allowable consonants.
local function check_radicals(form, weakness, rad1, rad2, rad3, rad4)
local function hamza_check(index, rad)
if rad == HAMZA_ON_ALIF or rad == HAMZA_UNDER_ALIF or
rad == HAMZA_ON_W or rad == HAMZA_ON_Y then
error("Radical " .. index .. " is " .. rad .. " but should be ء (hamza on the line)")
end
end
local function check_waw_ya(index, rad)
if not is_waw_ya(rad) then
error("Radical " .. index .. " is " .. rad .. " but should be و or ي")
end
end
local function check_not_waw_ya(index, rad)
if is_waw_ya(rad) then
error("In a sound verb, radical " .. index .. " should not be و or ي")
end
end
hamza_check(rad1)
hamza_check(rad2)
hamza_check(rad3)
hamza_check(rad4)
if weakness == "assimilated" or weakness == "assimilated+final-weak" then
if rad1 ~= W then
error("Radical 1 is " .. rad1 .. " but should be و")
end
-- don't check that non-assimilated form I verbs don't have wāw as their
-- first radical because some form-I verbs exist where a first-radical wāw
-- behaves as sound, e.g. wajuha yawjuhu "to be distinguished".
end
if weakness == "final-weak" or weakness == "assimilated+final-weak" then
if rad4 then
check_waw_ya(4, rad4)
else
check_waw_ya(3, rad3)
end
elseif vform_supports_final_weak(form) then
-- non-final-weak verbs cannot have weak final radical if there's a corresponding
-- final-weak verb category. I think this is safe. We may have problems with
-- ḥayya/ḥayiya yaḥyā if we treat it as a geminate verb.
if rad4 then
check_not_waw_ya(4, rad4)
else
check_not_waw_ya(3, rad3)
end
end
if weakness == "hollow" then
check_waw_ya(2, rad2)
-- don't check that non-hollow verbs in forms that support hollow verbs
-- don't have wāw or yāʾ as their second radical because some verbs exist
-- where a middle-radical wāw/yāʾ behaves as sound, e.g. form-VIII izdawaja
-- "to be in pairs".
end
if weakness == "geminate" then
if rad4 then
error("Internal error: No geminate quadrilaterals, should not be seen")
end
if rad2 ~= rad3 then
error("Weakness is geminate; radical 3 is " .. rad3 .. " but should be same as radical 2 " .. rad2)
end
elseif vform_supports_geminate(form) then
-- non-geminate verbs cannot have second and third radical same if there's
-- a corresponding geminate verb category. I think this is safe. We
-- don't fuss over double wāw or double yāʾ because this could legitimately
-- be a final-weak verb with middle wāw/yāʾ, treated as sound.
if rad4 then
error("Internal error: No quadrilaterals should support geminate verbs")
end
if rad2 == rad3 and not is_waw_ya(rad2) then
error("Weakness is '" .. weakness .. "'; radical 2 and 3 are same at " .. rad2 .. " but should not be; consider making weakness 'geminate'")
end
end
end
-- array of substitutions; each element is a 2-entry array FROM, TO; do it
-- this way so the concatenations only get evaluated once
local postprocess_subs = {
-- reorder short-vowel + shadda -> shadda + short-vowel for easier processing
{"(" .. AIU .. ")" .. SH, SH .. "%1"},
----------same letter separated by sukūn should instead use shadda---------
------------happens e.g. in kun-nā "we were".-----------------
{"(.)" .. SK .. "%1", "%1" .. SH},
---------------------------- assimilated verbs ----------------------------
-- iw, iy -> ī (assimilated verbs)
{I .. W .. SK, II},
{I .. Y .. SK, II},
-- uw, uy -> ū (assimilated verbs)
{U .. W .. SK, UU},
{U .. Y .. SK, UU},
-------------- final -yā uses tall alif not alif maqṣūra ------------------
{"(" .. Y .. SH .. "?" .. A .. ")" .. AMAQ, "%1" .. ALIF},
----------------------- handle hamza assimilation -------------------------
-- initial hamza + short-vowel + hamza + sukūn -> hamza + long vowel
{HAMZA .. A .. HAMZA .. SK, HAMZA .. A .. ALIF},
{HAMZA .. I .. HAMZA .. SK, HAMZA .. I .. Y},
{HAMZA .. U .. HAMZA .. SK, HAMZA .. U .. W}
}
local postprocess_tr_subs = {
{"ī([" .. vowels .. "y*])", "iy%1"},
{"ū([" .. vowels .. "w*])", "uw%1"},
{"(.)%*", "%1%1"}, -- implement shadda
---------------------------- assimilated verbs ----------------------------
-- iw, iy -> ī (assimilated verbs)
{"iw([^" .. vowels .. "w])", "ī%1"},
{"iy([^" .. vowels .. "y])", "ī%1"},
-- uw, uy -> ū (assimilated verbs)
{"uw([^" .. vowels .. "w])", "ū%1"},
{"uy([^" .. vowels .. "y])", "ū%1"},
----------------------- handle hamza assimilation -------------------------
-- initial hamza + short-vowel + hamza + sukūn -> hamza + long vowel
{"ʔaʔ(" .. NV .. ")", "ʔā%1"},
{"ʔiʔ(" .. NV .. ")", "ʔī%1"},
{"ʔuʔ(" .. NV .. ")", "ʔū%1"},
}
-- Post-process verb parts to eliminate phonological anomalies. Many of the changes, particularly the tricky ones,
-- involve converting hamza to have the proper seat. The rules for this are complicated and are documented on the
-- [[w:Hamza]] Wikipedia page. In some cases there are alternatives allowed, and we handle them below by returning
-- multiple possibilities.
local function postprocess_term(term)
if term == "?" then
return "?"
end
-- Add BORDER at text boundaries.
term = BORDER .. term .. BORDER
-- Do the main post-processing, based on the pattern substitutions in postprocess_subs.
for _, sub in ipairs(postprocess_subs) do
term = rsub(term, sub[1], sub[2])
end
term = term:gsub(BORDER, "")
if not rfind(term, HAMZA) then
return term
end
term = term:gsub(HAMZA, HAMZA_PH)
term = ar_utilities.process_hamza(term)
if #term == 1 then
term = term[1]
end
return term
end
local function postprocess_translit(translit)
if translit == "?" then
return "?"
end
-- Add BORDER at text boundaries.
translit = BORDER .. translit .. BORDER
-- Do the main post-processing, based on the pattern substitutions in postprocess_tr_subs.
for _, sub in ipairs(postprocess_tr_subs) do
translit = rsub(translit, sub[1], sub[2])
end
translit = translit:gsub(BORDER, "")
return translit
end
local function postprocess_forms(base)
local converted_values = {}
for slot, forms in pairs(base.forms) do
local need_dedup = false
for i, form in ipairs(forms) do
local term = postprocess_term(form.form)
local translit = form.translit and postprocess_translit(form.translit) or nil
if term ~= form.form or translit ~= form.translit then
need_dedup = true
end
converted_values[i] = {term, translit}
end
if need_dedup then
local temp_dedup = {}
for i = 1, #forms do
local new_term, new_translit = unpack(converted_values[i])
if type(new_term) == "table" then
for _, nt in ipairs(new_term) do
local new_formobj = {
form = nt,
translit = new_translit,
footnotes = forms[i].footnotes,
}
iut.insert_form(temp_dedup, "temp", new_formobj)
end
else
local new_formobj = {
form = new_term,
translit = new_translit,
footnotes = forms[i].footnotes,
}
iut.insert_form(temp_dedup, "temp", new_formobj)
end
end
base.forms[slot] = temp_dedup.temp
end
end
end
local function process_slot_overrides(base)
for slot, forms in pairs(base.slot_overrides) do
local existing_values = base.forms[slot]
base.forms[slot] = nil
for _, form in ipairs(forms) do
if form.form == "+" then
if not existing_values then
error(("Slot '%s' requested the default value but no such value available"):format(slot))
end
-- We maintain an invariant that no two slots share a form object (although they may share the footnote
-- lists inside the form objects). However, there is no need to copy the form objects here because there
-- is a one-to-one correspondence between slots and slot overrides, i.e. you can't have a default value
-- go into two slots.
insert_form_or_forms(base, slot, existing_values, "allow overrides", form.uncertain)
elseif default_indicator_to_active_participle_slot[form.form] then
if form.form == "++" then
if slot ~= "vn" and slot ~= "ap" and slot ~= "pp" then
error(("Secondary default value request '++' only applicable to verbal nouns and pariciples, but found in slot '%s'"):
format(slot))
end
else
if slot ~= "ap" then
error(("Secondary default value request '%s' only applicable to active pariciples, but found in slot '%s'"):
format(form.form, slot))
end
end
local secondary_default_slot =
slot == "vn" and "vn2" or slot == "pp" and "pp2" or
default_indicator_to_active_participle_slot[form.form]
local existing_values = base.forms[secondary_default_slot]
if not existing_values then
error(("Slot '%s' requested a secondary default value using '%s' but no such value available"):
format(slot, form.form))
end
-- See comment above about the lack of need to copy the form objects.
insert_form_or_forms(base, slot, existing_values, "allow overrides", form.uncertain)
-- To make sure there aren't shared form objects.
base.forms[secondary_default_slot] = nil
else
insert_form_or_forms(base, slot, form, "allow overrides", form.uncertain)
end
end
end
end
local function handle_lemma_linked(base)
-- Compute linked versions of potential lemma slots, for use in {{ar-verb}}. We substitute the original lemma
-- (before removing links) for forms that are the same as the lemma, if the original lemma has links.
for _, slot in ipairs(export.potential_lemma_slots) do
if base.forms[slot] then
insert_form_or_forms(base, slot .. "_linked", iut.map_forms(base.forms[slot], function(form)
if form == base.lemma and rfind(base.linked_lemma, "%[%[") then
return base.linked_lemma
else
return form
end
end))
end
end
end
-- Process specs given by the user using 'addnote[SLOTSPEC][FOOTNOTE][FOOTNOTE][...]'.
local function process_addnote_specs(base)
for _, spec in ipairs(base.addnote_specs) do
for _, slot_spec in ipairs(spec.slot_specs) do
slot_spec = "^" .. slot_spec .. "$"
for slot, forms in pairs(base.forms) do
if rfind(slot, slot_spec) then
-- To save on memory, side-effect the existing forms.
for _, form in ipairs(forms) do
form.footnotes = iut.combine_footnotes(form.footnotes, spec.footnotes)
end
end
end
end
end
end
local function add_missing_links_to_forms(base)
-- Any forms without links should get them now. Redundant ones will be stripped later.
for slot, forms in pairs(base.forms) do
for _, form in ipairs(forms) do
if not form.form:find("%[%[") then
form.form = "[[" .. form.form .. "]]"
end
end
end
end
local function conjugate_verb(base)
construct_stems(base)
for _, vowel_spec in ipairs(base.conj_vowels) do
-- Reconstruct conjugation type from verb form and (possibly inferred) weakness.
conj_type = base.verb_form .. "-" .. vowel_spec.weakness
-- Check that the conjugation type is recognized.
if not conjugations[conj_type] then
error("Unknown conjugation type '" .. conj_type .. "'")
end
-- The way the conjugation functions work is they always add entries to the appropriate parts of the paradigm
-- (each of which is an array), rather than setting the values. This makes it possible to call more than one
-- conjugation function and essentially get a paradigm of the "either A or B" kind. Doing this may insert
-- duplicate entries into a particular paradigm part, but this is not a problem because we check for duplicate
-- entries when adding them, and don't insert in that case.
conjugations[conj_type](base, vowel_spec)
end
postprocess_forms(base)
process_slot_overrides(base)
-- This should happen before add_missing_links_to_forms() so that the comparison `form == base.lemma` in
-- handle_lemma_linked() works correctly and compares unlinked forms to unlinked forms.
handle_lemma_linked(base)
process_addnote_specs(base)
if not base.alternant_multiword_spec.args.noautolinkverb then
add_missing_links_to_forms(base)
end
end
local function parse_indicator_spec(angle_bracket_spec)
-- Store the original angle bracket spec so we can reconstruct the overall conj spec with the lemma(s) in them.
local base = {
angle_bracket_spec = angle_bracket_spec,
conj_vowels = {},
root_consonants = {},
user_stem_overrides = {},
user_slot_overrides = {},
slot_explicitly_missing = {},
slot_uncertain = {},
slot_override_uses_default = {},
addnote_specs = {},
}
local function parse_err(msg)
error(msg .. ": " .. angle_bracket_spec)
end
local function fetch_footnotes(separated_group)
local footnotes
for j = 2, #separated_group - 1, 2 do
if separated_group[j + 1] ~= "" then
parse_err("Extraneous text after bracketed footnotes: '" .. table.concat(separated_group) .. "'")
end
if not footnotes then
footnotes = {}
end
table.insert(footnotes, separated_group[j])
end
return footnotes
end
local inside = angle_bracket_spec:match("^<(.*)>$")
assert(inside)
local segments = iut.parse_multi_delimiter_balanced_segment_run(inside, {{"[", "]"}, {"<", ">"}})
local dot_separated_groups = iut.split_alternating_runs_and_strip_spaces(segments, "%.")
-- The first dot-separated element must specify the verb form, e.g. IV or IIq. If the form is I, it needs to include
-- the the past and non-past vowels, e.g. I/a~u for kataba ~ yaktubu. More than one vowel can be given,
-- comma-separated, and more than one past~non-past pair can be given, slash-separated, e.g. I/a,u~u/i~a for form I
-- كمل, which can be conjugated as kamala/kamula ~ yakmulu or kamila ~ yakmalu. An individual vowel spec must be one
-- of a, i or u and in general (a) at least one past~non-past pair most be given, and (b) both past and non-past
-- vowels must be given even though sometimes the vowel can be determined from the unvocalized form. An exception is
-- passive-only verbs, where the vowels can't in general be determined (except indirectly in some cases by looking
-- at an associated non-passive verb); in that case, the vowel~vowel spec can left out.
local slash_separated_groups = iut.split_alternating_runs_and_strip_spaces(dot_separated_groups[1], "/")
local form_spec = slash_separated_groups[1]
base.form_footnotes = fetch_footnotes(form_spec)
if form_spec[1] == "" then
parse_err("Missing verb form")
end
if not allowed_vforms_with_weakness_set[form_spec[1]] then
parse_err(("Unrecognized verb form '%s', should be one of %s"):format(
form_spec[1], m_table.serialCommaJoin(allowed_vforms, {conj = "or", dontTag = true})))
end
if form_spec[1]:find("%-") then
base.verb_form, base.explicit_weakness = form_spec[1]:match("^(.-)%-(.*)$")
else
base.verb_form = form_spec[1]
end
if #slash_separated_groups > 1 then
if base.verb_form ~= "I" then
parse_err(("Past~non-past vowels can only be specified when verb form is I, but saw form '%s'"):format(
base.verb_form))
end
for i = 2, #slash_separated_groups do
local slash_separated_group = slash_separated_groups[i]
local tilde_separated_groups = iut.split_alternating_runs_and_strip_spaces(slash_separated_group, "~")
if #tilde_separated_groups ~= 2 then
parse_err(("Expected two tilde-separated vowel specs: %s"):format(table.concat(slash_separated_group)))
end
local function parse_conj_vowels(tilde_separated_group, vtype)
local conj_vowel_objects = {}
local comma_separated_groups = iut.split_alternating_runs_and_strip_spaces(tilde_separated_group, ",")
for _, comma_separated_group in ipairs(comma_separated_groups) do
local conj_vowel = comma_separated_group[1]
if conj_vowel ~= "a" and conj_vowel ~= "i" and conj_vowel ~= "u" then
parse_err(("Expected %s conjugation vowel '%s' to be one of a, i or u in %s"):format(
vtype, conj_vowel, table.concat(slash_separated_group)))
end
conj_vowel = dia[conj_vowel]
local conj_vowel_footnotes = fetch_footnotes(comma_separated_group)
-- Try to use strings when possible as it makes q() significantly more efficient.
if conj_vowel_footnotes then
table.insert(conj_vowel_objects, {form = conj_vowel, footnotes = conj_vowel_footnotes})
else
table.insert(conj_vowel_objects, conj_vowel)
end
end
return conj_vowel_objects
end
local conj_vowel_spec = {
past = parse_conj_vowels(tilde_separated_groups[1], "past"),
nonpast = parse_conj_vowels(tilde_separated_groups[2], "non-past"),
}
table.insert(base.conj_vowels, conj_vowel_spec)
end
end
for i = 2, #dot_separated_groups do
local dot_separated_group = dot_separated_groups[i]
local first_element = dot_separated_group[1]
if first_element == "addnote" then
local spec_and_footnotes = fetch_footnotes(dot_separated_group)
if #spec_and_footnotes < 2 then
parse_err("Spec with 'addnote' should be of the form 'addnote[SLOTSPEC][FOOTNOTE][FOOTNOTE][...]'")
end
local slot_spec = table.remove(spec_and_footnotes, 1)
local slot_spec_inside = rmatch(slot_spec, "^%[(.*)%]$")
if not slot_spec_inside then
parse_err("Internal error: slot_spec " .. slot_spec .. " should be surrounded with brackets")
end
local slot_specs = rsplit(slot_spec_inside, ",")
-- FIXME: Here, [[Module:it-verb]] called strip_spaces(). Generally we don't do this. Should we?
table.insert(base.addnote_specs, {slot_specs = slot_specs, footnotes = spec_and_footnotes})
elseif first_element:find("^var:") then
if #dot_separated_group > 1 then
parse_err(("Can't attach footnotes to 'var:' spec '%s'"):format(first_element))
end
base.variant = first_element:match("^var:(.*)$")
elseif first_element:find("^I+V?:") then
local root_cons, root_cons_value = first_element:match("^(I+V?):(.*)$")
local root_index
if root_cons == "I" then
root_index = 1
elseif root_cons == "II" then
root_index = 2
elseif root_cons == "III" then
root_index = 3
elseif root_cons == "IV" then
root_index = 4
if not base.verb_form:find("q$") then
parse_err(("Can't specify root consonant IV for non-quadriliteral verb form '%s': %s"):format(
base.verb_form, first_element))
end
end
local cons, translit = root_cons_value:match("^(.*)//(.*)$")
if not cons then
cons = root_cons_value
end
local root_footnotes = fetch_footnotes(dot_separated_group)
if not translit and not root_footnotes then
base.root_consonants[root_index] = cons
else
base.root_consonants[root_index] = {form = cons, translit = translit, footnotes = root_footnotes}
end
elseif first_element:find("^[a-z][a-z0-9_]*:") then
local slot_or_stem, remainder = first_element:match("^(.-):(.*)$")
dot_separated_group[1] = remainder
local comma_separated_groups = iut.split_alternating_runs_and_strip_spaces(dot_separated_group, "[,،]")
if overridable_stems[slot_or_stem] then
if base.user_stem_overrides[slot_or_stem] then
parse_err("Overridable stem '" .. slot_or_stem .. "' specified twice")
end
base.user_stem_overrides[slot_or_stem] = overridable_stems[slot_or_stem](comma_separated_groups,
{prefix = slot_or_stem, base = base, parse_err = parse_err, fetch_footnotes = fetch_footnotes})
else -- assume a form override; we validate further later when the possible slots are available
if base.user_slot_overrides[slot_or_stem] then
parse_err("Form override '" .. slot_or_stem .. "' specified twice")
end
base.user_slot_overrides[slot_or_stem] = allow_multiple_values_for_override(comma_separated_groups,
{prefix = slot_or_stem, base = base, parse_err = parse_err, fetch_footnotes = fetch_footnotes},
"is form override")
end
elseif indicator_flags[first_element] then
if #dot_separated_group > 1 then
parse_err("No footnotes allowed with '" .. first_element .. "' spec")
end
if base[first_element] then
parse_err("Spec '" .. first_element .. "' specified twice")
end
base[first_element] = true
else
local passive, uncertain = first_element:match("^(.*)(%?)$")
passive = passive or first_element
uncertain = not not uncertain
if passive_types[passive] then
if #dot_separated_group > 1 then
parse_err("No footnotes allowed with '" .. passive .. "' spec")
end
if base.passive then
parse_err("Value for passive type specified twice")
end
base.passive = passive
base.passive_uncertain = uncertain
else
parse_err("Unrecognized spec '" .. first_element .. "'")
end
end
end
return base
end
-- Normalize all lemmas, substituting the pagename for blank lemmas and adding links to multiword lemmas.
local function normalize_all_lemmas(alternant_multiword_spec, head)
-- (1) Add links to all before and after text. Remember the original text so we can reconstruct the verb spec later.
if not alternant_multiword_spec.args.noautolinktext then
iut.add_links_to_before_and_after_text(alternant_multiword_spec, "remember original")
end
-- (2) Remove any links from the lemma, but remember the original form so we can use it below in the 'lemma_linked'
-- form.
iut.map_word_specs(alternant_multiword_spec, function(base)
if base.lemma == "" then
base.lemma = head
end
base.user_specified_lemma = base.lemma
base.lemma = m_links.remove_links(base.lemma)
base.user_specified_verb = base.lemma
base.verb = base.user_specified_verb
local linked_lemma
if alternant_multiword_spec.args.noautolinkverb or base.user_specified_lemma:find("%[%[") then
linked_lemma = base.user_specified_lemma
else
-- Add links to the lemma so the user doesn't specifically need to, since we preserve
-- links in multiword lemmas and include links in non-lemma forms rather than allowing
-- the entire form to be a link.
linked_lemma = iut.add_links(base.user_specified_lemma)
end
base.linked_lemma = linked_lemma
end)
end
-- Determine weakness from radicals. Used when root given in place of lemma (e.g. for {{ar-verb forms}}).
local function weakness_from_radicals(form, rad1, rad2, rad3, rad4)
local weakness = nil
local quadlit = form:find("q$")
-- If weakness unspecified, derive from radicals.
if not quadlit then
if is_waw_ya(rad3) and rad1 == W and form == "I" then
weakness = "assimilated+final-weak"
elseif is_waw_ya(rad3) and vform_supports_final_weak(form) then
weakness = "final-weak"
elseif rad2 == rad3 and vform_supports_geminate(form) then
weakness = "geminate"
elseif is_waw_ya(rad2) and vform_supports_hollow(form) then
weakness = "hollow"
elseif rad1 == W and form == "I" then
weakness = "assimilated"
else
weakness = "sound"
end
else
if is_waw_ya(rad4) then
weakness = "final-weak"
else
weakness = "sound"
end
end
return weakness
end
-- Join the infixed tāʔ (ت) to the first radical in form VIII verbs. This may cause assimilation of the tāʔ to the
-- radical or in some cases the radical to the tāʔ. Used when a root is supplied instead of a lemma (which already has
-- the appropriate assimilation in it).
local function form_viii_join_ta(rad)
if rad == W or rad == Y or rad == "ت" then return "تّ"
elseif rad == "د" then return "دّ"
elseif rad == "ث" then return "ثّ"
elseif rad == "ذ" then return "ذّ"
elseif rad == "ز" then return "زْد"
elseif rad == "ص" then return "صْط"
elseif rad == "ض" then return "ضْط"
elseif rad == "ط" then return "طّ"
elseif rad == "ظ" then return "ظّ"
else return rad .. SK .. "ت"
end
end
local function detect_indicator_spec(base)
base.forms = {}
base.stem_overrides = {}
base.slot_overrides = {}
if not base.conj_vowels[1] then
-- These may be converted to inferred vowels. If not, we throw an error if form I and not passive-only.
base.conj_vowels = {{
past = "-",
nonpast = "-",
}}
else
-- If multiple vowels specified for a given vowel type (e.g. a,u~u), expand so that each spec in
local expansion = {}
for _, spec in ipairs(base.conj_vowels) do
for _, past in ipairs(spec.past) do
for _, nonpast in ipairs(spec.nonpast) do
table.insert(expansion, {past = past, nonpast = nonpast})
end
end
end
base.conj_vowels = expansion
end
local vform = base.verb_form
-- check for quadriliteral form (Iq, IIq, IIIq, IVq)
base.quadlit = not not vform:find("q$")
-- Infer radicals as necessary. We infer a separate set of radicals for each past~non-past vowel combination because
-- they may be different (particularly with form-I hollow verbs).
for _, vowel_spec in ipairs(base.conj_vowels) do
-- NOTE: rad1, rad2, etc. refer to user-specified radicals, which are formobj tables that optionally specify an
-- explicit manual translit, whereas ir1, ir2, etc. refer to inferred radicals, which are either strings or
-- lists of possible radicals.
local rads = base.root_consonants
local rad1, rad2, rad3, rad4 = rads[1], rads[2], rads[3], rads[4]
-- Default any unspecified radicals to radicals determined from the headword. The returned radicals may be
-- lists of possible radicals, where the first radical should be chosen if the user didn't explicitly specify a
-- radical but all are allowed. If `ambig = true` is set in the table, the radical is considered ambiguous and
-- categories won't be created for weak radicals.
local weakness, ir1, ir2, ir3, ir4
if vform ~= "none" then
ir1, ir2, ir3 = rmatch(base.lemma, "^([^_])_([^_])_([^_])$")
if not ir1 then
ir1, ir2, ir3, ir4 = rmatch(base.lemma, "^([^_])_([^_])_([^_])_([^_])$")
end
if ir1 then
-- root given instead of lemma
weakness = weakness_from_radicals(vform, ir1, ir2, ir3, ir4)
if vform == "VIII" then
vowel_spec.form_viii_assim = form_viii_join_ta(ir1)
end
else
local ret = export.infer_radicals {
headword = base.lemma,
vform = vform,
passive = base.passive,
past_vowel = vowel_spec.past,
nonpast_vowel = vowel_spec.nonpast,
is_reduced = base.reduced,
}
weakness, ir1, ir2, ir3, ir4 = ret.weakness, ret.rad1, ret.rad2, ret.rad3, ret.rad4
vowel_spec.form_viii_assim = ret.form_viii_assim
vowel_spec.past = ret.past_vowel
vowel_spec.nonpast = ret.nonpast_vowel
vowel_spec.variant = base.variant or ret.variant
end
end
-- For most ambiguous radicals, the choice of radical doesn't matter because it doesn't affect the conjugation
-- one way or another. For form I hollow verbs, however, it definitely does. In fact, the choice of radical is
-- critical even beyond the past and non-past vowels because it affects the form of the passive participle. So,
-- check for this and signal an error if the radical could not be inferred and is not given explicitly.
if vform == "I" and type(ir2) == "table" and ir2.need_radical and not rad2 then
error("Unable to guess middle radical of hollow form I verb; need to specify radical explicitly")
end
if vform == "I" and not is_passive_only(base.passive) and (
rget(vowel_spec.past) == "-" or rget(vowel_spec.nonpast) == "-") then
error("Form I verb that isn't passive-only or final-weak must have past~non-past vowels specified")
end
-- Convert ambiguous radicals.
local function regularize_inferred_radical(rad)
if type(rad) == "table" then
if rad.ambig then
return {form = rad[1], ambig = true}
else
return rad[1]
end
else
return rad
end
end
-- Return the appropriate radical at index `index` (1 through 4), based either on the user-specified radical
-- `user_radical` or (if unspecified) `inferred_radical`, inferred from the unvocalized lemma. Two values are
-- returned, the "regularized" version of the radical (where ambiguous inferred radicals are converted to their
-- most likely actual radical) and the non-regularized version. The returned values are form objects rather than
-- strings.
local function fetch_radical(user_radical, inferred_radical, index)
if not user_radical then
return regularize_inferred_radical(inferred_radical), inferred_radical
else
local rad_formval = rget(user_radical)
if type(inferred_radical) == "table" then
local allowed_radical_set = m_table.listToSet(inferred_radical)
if not allowed_radical_set[rad_formval] then
error(("For lemma %s, radical %s ambiguously inferred as %s but user radical incompatibly given as %s"):
format(base.lemma, index,
m_table.serialCommaJoin(inferred_radical, {conj = "or", dontTag = true}), rad_formval))
end
elseif rad_formval ~= inferred_radical then
error(("For lemma %s, radical %s inferred as %s but user radical incompatibly given as %s"):
format(base.lemma, index, inferred_radical, rad_formval))
end
return user_radical, user_radical
end
end
if vform ~= "none" then
vowel_spec.rad1, vowel_spec.unreg_rad1 = fetch_radical(rad1, ir1, 1)
vowel_spec.rad2, vowel_spec.unreg_rad2 = fetch_radical(rad2, ir2, 2)
vowel_spec.rad3, vowel_spec.unreg_rad3 = fetch_radical(rad3, ir3, 3)
if base.quadlit then
vowel_spec.rad4, vowel_spec.unreg_rad4 = fetch_radical(rad4, ir4, 4)
end
end
if vform == "I" then
-- If explicit weakness given using 'I-sound' or 'I-assimilated', we may need to adjust the inferred weakness.
if base.explicit_weakness == "sound" then
if weakness == "assimilated" then
weakness = "sound"
elseif weakness == "assimilated+final-weak" then
-- Verbs like waniya~yawnā "to be faint; to languish" (although the defaults should handle this
-- correctly)
weakness = "final-weak"
else
error(("Can't specify form 'I-sound' when inferred weakness is '%s' for lemma %s"):format(
weakness, base.lemma))
end
elseif base.explicit_weakness == "assimilated" then
if weakness == "sound" then
-- i~a verbs like waṭiʔa~yaṭaʔu "to tread, to trample"; wasiʕa~yasaʕu "to be spacious; to be well-off";
-- waṯiʔa~yaṯaʔu "to get bruised, to be sprained", which would default to sound.
weakness = "assimilated"
elseif weakness == "final-weak" then
-- For completeness; not clear if any verbs occur where this is needed. (There are plenty of
-- assimilated+final-weak verbs but the defaults should take care of them.)
weakness = "assimilated+final-weak"
else
error(("Can't specify form 'I-assimilated' when inferred weakness is '%s' for lemma %s"):format(
weakness, base.lemma))
end
elseif base.explicit_weakness then
error(("Internal error: Unrecognized value '%s' for base.explicit_weakness"):format(base.explicit_weakness))
end
elseif vform == "none" then
weakness = base.explicit_weakness
elseif base.explicit_weakness then
error(("Internal error: Explicit weakness should not be specifiable except with forms I and none, but saw explicit weakness '%s' with verb form '%s'"):
format(base.explicit_weakness, vform))
end
vowel_spec.weakness = weakness
if vform ~= "none" then
-- Error if radicals are wrong given the weakness. More likely to happen if the weakness is explicitly given
-- rather than inferred. Will also happen if certain incorrect letters are included as radicals e.g. hamza on
-- top of various letters, alif maqṣūra, tā' marbūṭa.
check_radicals(vform, weakness, rget(vowel_spec.rad1), rget(vowel_spec.rad2), rget(vowel_spec.rad3),
base.quadlit and rget(vowel_spec.rad4) or nil)
end
-- Check the variant value.
local form_iii_vi_geminate = (vform == "III" or vform == "VI") and rget(vowel_spec.rad2) == rget(vowel_spec.rad3) and
not req(vowel_spec.rad2, Y)
local hayy_i_x = hayy_radicals(vowel_spec.rad1, vowel_spec.rad2, vowel_spec.rad3) and (vform == "I" or vform == "X")
if form_iii_vi_geminate or hayy_i_x then
if vowel_spec.variant and vowel_spec.variant ~= "long" and vowel_spec.variant ~= "short" and vowel_spec.variant ~= "both" then
error(("For form-III/VI geminate verb or form-I/X verb with ح-ي-ي radicals, saw unrecognized 'var:%s' value; should be 'var:long', 'var:short' or 'var:both'"):format(
vowel_spec.variant))
end
elseif vowel_spec.variant then
error(("Variant value 'var:%s' not allowed in this context"):format(vowel_spec.variant))
end
end
-- If form I, regroup expanded vowels for display purposes.
if vform == "I" then
local group_by_past = {}
for _, vowel_spec in ipairs(base.conj_vowels) do
m_table.insertIfNot(group_by_past, {
past = undia[rget(vowel_spec.past)],
nonpasts = {undia[rget(vowel_spec.nonpast)]},
}, {
key = function(obj) return obj.past end,
combine = function(obj1, obj2)
for _, nonpast in ipairs(obj2.nonpasts) do
m_table.insertIfNot(obj1.nonpasts, nonpast)
end
end,
})
end
local group_by_nonpast = {}
for _, vowel_spec in ipairs(group_by_past) do
m_table.insertIfNot(group_by_nonpast, {
pasts = {vowel_spec.past},
nonpasts = vowel_spec.nonpasts,
}, {
key = function(obj) return obj.nonpasts end,
combine = function(obj1, obj2)
for _, past in ipairs(obj2.pasts) do
m_table.insertIfNot(obj1.pasts, past)
end
end,
})
end
base.grouped_conj_vowels = group_by_nonpast
end
-- Set value of passive. If not specified, default is yes, but no for forms VII, IX,
-- XI - XV and IIq - IVq, and impersonal for form VI.
local passive = base.passive
if not passive then
if vform_probably_impersonal_passive(vform) then
base.passive = "ipass"
else
local has_passive = false
for _, vowel_spec in ipairs(base.conj_vowels) do
if not vform_probably_no_passive(vform, vowel_spec.weakness, vowel_spec.past, vowel_spec.nonpast) then
has_passive = true
break
end
end
if has_passive then
base.passive = "pass"
base.passive_uncertain = true
else
base.passive = "nopass"
end
end
end
-- NOTE: Currently there are no built-in stems or form overrides for Arabic; this code is inherited from
-- [[Module:ca-verb]], where such things do exist, and is kept for generality in case we decide in the future to
-- implement such things.
-- Override built-in verb stems and overrides with user-specified ones.
for stem, values in pairs(base.user_stem_overrides) do
base.stem_overrides[stem] = values
end
for slot, values in pairs(base.user_slot_overrides) do
if not base.alternant_multiword_spec.verb_slots_map[slot] then
error("Unrecognized override slot '" .. slot .. "': " .. base.angle_bracket_spec)
end
if export.unsettable_slots_set[slot] then
error("Slot '" .. slot .. "' cannot be set using an override: " .. base.angle_bracket_spec)
end
if skip_slot(base, slot, "allow overrides") then
error("Override slot '" .. slot ..
"' would be skipped based on the passive, 'noimp' and/or 'no_nonpast' settings: " ..
base.angle_bracket_spec)
end
base.slot_overrides[slot] = values
end
if base.verb_form == "none-final-weak" then
for _, stem_type in ipairs { "past", "past_pass", "nonpast", "nonpast_pass" } do
if base.stem_overrides[stem_type .. "_c"] or base.stem_overrides[stem_type .. "_v"] then
error(("Specify past stem for verb type 'none-final-weak' using '%s:...' not '%s_c:...' or '%s_v:...'"):
format(stem_type, stem_type, stem_type))
end
end
for _, stem_type in ipairs { "past", "nonpast" } do
if base.stem_overrides[stem_type] or not base.stem_overrides[stem_type .. "_final_weak_vowel"] then
error(("For verb type 'none-final-weak', if '%s:...' specified, so must '%s_final_weak_vowel:...'"):
format(stem_type, stem_type))
end
end
end
end
local function detect_all_indicator_specs(alternant_multiword_spec)
add_slots(alternant_multiword_spec)
alternant_multiword_spec.verb_forms = {}
-- This means at least one individual base had the slot marked as explicitly missing. Another base (e.g. when
-- there are multiple alternants) might have a value for the slot. In practice, we only respect this when there are
-- no overall values in the slot and `slot_uncertain` isn't set; in this case, we display "no ..." for the slot
-- instead of simply not displaying anything for the slot.
alternant_multiword_spec.slot_explicitly_missing = {}
-- This means at least one individual base had no values for the slot and the slot marked as explicitly uncertain.
-- Note that this is different from a value being present but marked as uncertain (e.g. if an override was given
-- with a ? after it); this causes the form object for the value to have `uncertain = true` set. If there are no
-- overall values in the slot and `slot_uncertain` is set, we display this in the headword.
alternant_multiword_spec.slot_uncertain = {}
iut.map_word_specs(alternant_multiword_spec, function(base)
-- So arguments, etc. can be accessed. WARNING: Creates circular reference.
base.alternant_multiword_spec = alternant_multiword_spec
detect_indicator_spec(base)
if not base.nocat then
m_table.insertIfNot(alternant_multiword_spec.verb_forms, base.verb_form)
end
if base.passive_uncertain then
alternant_multiword_spec.passive_uncertain = true
end
for slot, _ in pairs(base.slot_explicitly_missing) do
alternant_multiword_spec.slot_explicitly_missing[slot] = true
end
end)
end
local function determine_slot_uncertainty_from_forms(alternant_multiword_spec)
iut.map_word_specs(alternant_multiword_spec, function(base)
-- If no verbal noun and verb form is not 'none' (manually-specified stems) — which currently only happens for
-- form I — and the verbal noun wasn't explicitly indicated as missing using <vn:->, we assume it's just
-- unknown/unspecified rather than missing.
if not base.forms.vn and vform ~= "none" and not base.slot_explicitly_missing.vn then
base.slot_uncertain.vn = true
end
-- Propagate slot uncertainty up. Currently only the verbal noun can have this set but we write the code
-- generally.
for slot, _ in pairs(base.slot_uncertain) do
alternant_multiword_spec.slot_uncertain[slot] = true
end
end)
end
-- Determine certain properties of the verb from the overall forms, such as whether the verb is active-only or
-- passive-only, is impersonal, lacks an imperative, etc.
local function determine_verb_properties_from_forms(alternant_multiword_spec)
alternant_multiword_spec.has_active = false
alternant_multiword_spec.has_passive = false
alternant_multiword_spec.has_non_impers_active = false
alternant_multiword_spec.has_non_impers_passive = false
alternant_multiword_spec.has_imp = false
alternant_multiword_spec.has_past = false
alternant_multiword_spec.has_nonpast = false
for slot, _ in pairs(alternant_multiword_spec.forms) do
if slot == "ap" or slot:find("[123]") and not slot:find("_pass") then
alternant_multiword_spec.has_active = true
end
if slot == "pp" or slot:find("[123]") and slot:find("_pass") then
alternant_multiword_spec.has_passive = true
end
if slot:find("[123]") and not slot:find("pass_[123]") and not slot:find("3ms") then
alternant_multiword_spec.has_non_impers_active = true
end
if slot:find("pass_[123]") and not slot:find("3ms") then
alternant_multiword_spec.has_non_impers_passive = true
end
if slot:find("^imp_") then
alternant_multiword_spec.has_imp = true
end
if slot:find("^past_") then
alternant_multiword_spec.has_past = true
end
if slot:find("^ind_") or slot:find("^sub_") or slot:find("^juss_") then
alternant_multiword_spec.has_nonpast = true
end
end
end
local function add_categories_and_annotation(alternant_multiword_spec, base, multiword_lemma, insert_ann, insert_cat)
-- Useful e.g. in constructing suppletive verbs out of parts. For a verb like جاء or أتى whose imperative comes
-- from the unrelated verb تعالى, we don't want the latter verb showing up in categories or annotations.
if base.nocat then
return
end
local vform = base.verb_form
if vform ~= "none" then
insert_ann("form", vform)
insert_cat("form-" .. vform .. " verbs")
end
if base.reduced then
insert_ann("reduced", "reduced")
if vform ~= "none" then
insert_cat("form-" .. vform .. " reduced verbs")
end
end
if base.quadlit then
insert_cat("verbs with quadriliteral roots")
end
for _, vowel_spec in ipairs(base.conj_vowels) do
local rad1, rad2, rad3, rad4 = get_radicals_4(vowel_spec)
local final_weak = is_final_weak(base, vowel_spec)
local weakness = vowel_spec.weakness
-- We have to distinguish weakness by form and weakness by conjugation. Weakness by form merely indicates the
-- presence of weak letters in certain positions in the radicals. Weakness by conjugation is related to how the
-- verbs are conjugated. For example, form-II verbs that are "hollow by form" (middle radical is wāw or yāʾ) are
-- conjugated as sound verbs. Another example: form-I verbs with initial wāw are "assimilated by form" and most
-- are assimilated by conjugation as well, but a few are sound by conjugation, e.g. wajuha yawjuhu "to be
-- distinguished" (rather than wajuha yajuhu); similarly for some hollow-by-form verbs in various forms, e.g.
-- form VIII izdawaja yazdawiju "to be in pairs" (rather than izdāja yazdāju). Categories referring to weakness
-- always refer to weakness by conjugation; weakness by form is distinguished only by categories such as
-- [[:Category:Arabic form-III verbs with و as second radical]].
insert_ann("weakness", weakness)
if vform ~= "none" then
insert_cat(("%s form-%s verbs"):format(weakness, vform))
end
local function radical_is_ambiguous(rad)
return type(rad) == "table" and rad.ambig
end
local function radical_is_unambiguous_weak(rad)
return not radical_is_ambiguous(rad) and (is_waw_ya(rad) or req(rad, HAMZA))
end
if vform ~= "none" then
local ur1, ur2, ur3, ur4 =
vowel_spec.unreg_rad1, vowel_spec.unreg_rad2, vowel_spec.unreg_rad3, vowel_spec.unreg_rad4
-- Create headword categories based on the radicals. Do the following before
-- converting the Latin radicals into Arabic ones so we distinguish
-- between ambiguous and non-ambiguous radicals.
if radical_is_ambiguous(ur1) or radical_is_ambiguous(ur2) or radical_is_ambiguous(ur3) or
ur4 and radical_is_ambiguous(ur4) then
insert_cat("verbs with ambiguous radicals")
end
if radical_is_unambiguous_weak(ur1) then
insert_cat("form-" .. vform .. " verbs with " .. rget(ur1) .. " as first radical")
end
if radical_is_unambiguous_weak(ur2) then
insert_cat("form-" .. vform .. " verbs with " .. rget(ur2) .. " as second radical")
end
if radical_is_unambiguous_weak(ur3) then
insert_cat("form-" .. vform .. " verbs with " .. rget(ur3) .. " as third radical")
end
if ur4 and radical_is_unambiguous_weak(ur4) then
insert_cat("form-" .. vform .. " verbs with " .. rget(ur4) .. " as fourth radical")
end
end
end
if vform == "I" and not is_passive_only(base.passive) then
for _, vowel_spec in ipairs(base.grouped_conj_vowels) do
insert_ann("vowels",
("%s ~ %s"):format(table.concat(vowel_spec.pasts, "/"), table.concat(vowel_spec.nonpasts, "/")))
for _, past in ipairs(vowel_spec.pasts) do
for _, nonpast in ipairs(vowel_spec.nonpasts) do
if past == "-" or nonpast == "-" then
error("Internal error: Saw form I past vowel %s and non-past vowel %s but - in place of vowel should have triggered an error earlier")
end
insert_cat(("form-I verbs with past vowel %s and non-past vowel %s"):format(past, nonpast))
end
end
end
end
if base.slot_uncertain.vn then
-- An unspecified and non-defaulted verbal noun (form I) is considered uncertain rather than explicitly missing.
-- Use <vn:-> to explicitly indicate the lack of verbal noun.
insert_cat("verbs needing verbal noun checked")
end
if base.slot_uncertain.ap then
-- FIXME: We should make form I stative verbs have uncertain active participles.
insert_cat("verbs needing active participle checked")
end
if base.irregular then
insert_ann("irreg", "irregular")
insert_cat("irregular verbs")
end
end
-- Compute the categories to add the verb to, as well as the annotation to display in the conjugation title bar. We
-- combine the code to do these functions as both categories and title bar contain similar information.
local function compute_categories_and_annotation(alternant_multiword_spec)
alternant_multiword_spec.categories = {}
local ann = {}
alternant_multiword_spec.annotation = ann
ann.form = {}
ann.weakness = {}
ann.vowels = {}
ann.passive = nil
ann.reduced = {}
ann.irreg = {}
ann.defective = {}
local multiword_lemma = false
for _, slot in ipairs(export.potential_lemma_slots) do
if alternant_multiword_spec.forms[slot] then
for _, formobj in ipairs(alternant_multiword_spec.forms[slot]) do
if formobj.form:find(" ") then
multiword_lemma = true
break
end
end
break
end
end
local function insert_ann(anntype, value)
m_table.insertIfNot(alternant_multiword_spec.annotation[anntype], value)
end
local function insert_cat(cat, also_when_multiword)
-- Don't place multiword terms in categories like 'Arabic form-II verbs' to avoid spamming the categories with
-- such terms.
if also_when_multiword or not multiword_lemma then
m_table.insertIfNot(alternant_multiword_spec.categories, "Arabic " .. cat)
end
end
iut.map_word_specs(alternant_multiword_spec, function(base)
add_categories_and_annotation(alternant_multiword_spec, base, multiword_lemma, insert_ann, insert_cat)
end)
for _, slot_with_name in ipairs {
{"vn", "verbal noun"},
{"ap", "active participle"},
} do
local slot, name = unpack(slot_with_name)
if alternant_multiword_spec.forms[slot] then
for _, form in ipairs(alternant_multiword_spec.forms[slot]) do
if form.uncertain then
insert_cat(("verbs needing %s checked"):format(name))
break
end
end
end
end
if alternant_multiword_spec.has_active then
if alternant_multiword_spec.has_passive and alternant_multiword_spec.has_non_impers_passive then
insert_cat("verbs with full passive")
ann.passive = "full passive"
elseif alternant_multiword_spec.has_passive then
insert_cat("verbs with impersonal passive")
ann.passive = "impersonal passive"
else
insert_cat("verbs lacking passive forms")
ann.passive = "no passive"
end
else
if alternant_multiword_spec.has_non_impers_passive then
insert_cat("passive verbs")
insert_cat("verbs with full passive")
ann.passive = "passive-only"
else
insert_cat("passive verbs")
insert_cat("impersonal verbs")
insert_cat("verbs with impersonal passive")
ann.passive = "impersonal (passive-only)"
end
end
if alternant_multiword_spec.passive_uncertain then
insert_cat("verbs needing passive checked")
ann.passive = ann.passive .. ' <abbr title="passive status uncertain">(?)</abbr>'
end
if alternant_multiword_spec.has_active and not alternant_multiword_spec.has_imp then
insert_ann("defective", "no imperative")
insert_cat("verbs lacking imperative forms")
end
if not alternant_multiword_spec.has_past then
insert_ann("defective", "no past")
insert_cat("verbs lacking past forms")
end
if not alternant_multiword_spec.has_nonpast then
insert_ann("defective", "no non-past")
insert_cat("verbs lacking non-past forms")
end
local ann_parts = {}
local function insert_ann_part(part, conj)
local val = table.concat(ann[part], conj or " or ")
if val ~= "" and val ~= "regular" then
table.insert(ann_parts, val)
end
end
insert_ann_part("form")
insert_ann_part("weakness")
insert_ann_part("reduced")
insert_ann_part("vowels")
if ann.passive then
table.insert(ann_parts, ann.passive)
end
insert_ann_part("irreg")
insert_ann_part("defective", ", ")
alternant_multiword_spec.annotation = table.concat(ann_parts, ", ")
end
local function show_forms(alternant_multiword_spec)
local lemmas = {}
for _, slot in ipairs(export.potential_lemma_slots) do
if alternant_multiword_spec.forms[slot] then
for _, formobj in ipairs(alternant_multiword_spec.forms[slot]) do
table.insert(lemmas, formobj)
end
break
end
end
alternant_multiword_spec.lemmas = lemmas -- save for later use in make_table()
alternant_multiword_spec.vn = alternant_multiword_spec.forms.vn -- save for later use in make_table()
-- Reconstruct the original verb spec without overrides for verbal nouns and participles, since those specific slots
-- are ignored by {{ar-verb form}}. Compute this once beforehand; `transform_accel_obj` is called repeatedly on each
-- form and we don't want to compute this repeatedly.
local reconstructed_verb_spec = iut.reconstruct_original_spec(alternant_multiword_spec, {
preprocess_angle_bracket_spec = function(spec)
spec = spec:match("^<(.*)>$")
assert(spec)
local segments = iut.parse_multi_delimiter_balanced_segment_run(spec, {{"[", "]"}, {"<", ">"}})
local dot_separated_groups = iut.split_alternating_runs_and_strip_spaces(segments, "%.")
-- Rejoin each dot-separated group into a single string, since we aren't actually going to do any parsing
-- of bracket-bounded textual runs; then filter out overrides for verbal nouns and participles.
local filtered_indicators = {}
for _, dot_separated_group in ipairs(dot_separated_groups) do
local indicator = table.concat(dot_separated_group)
-- FIXME: Do we want to filter out any other indicators?
if not (indicator:find("^vn:") or indicator:find("^[ap]p:")) then
table.insert(filtered_indicators, indicator)
end
end
return ("<%s>"):format(table.concat(filtered_indicators, "."))
end,
})
-- If we're dealing with a single word, no alternants and a single verb form, use the auto-conjugation-fetching
-- variant.
local reconstructed_lemma, inside = reconstructed_verb_spec:match("^([^ <>()]+)(%b<>)$")
if inside and alternant_multiword_spec.verb_forms[1] and not alternant_multiword_spec.verb_forms[2] then
reconstructed_verb_spec = ("+%s<%s>"):format(reconstructed_lemma, alternant_multiword_spec.verb_forms[1])
end
local function transform_accel_obj(slot, formobj, accel_obj)
if not accel_obj then
return accel_obj
end
if slot == "ap" or slot == "pp" or slot == "vn" then
-- FIXME: [[Module:accel]] can't correctly handle more than one verb form for participles and verbal nouns
accel_obj.form = slot .. "-" .. table.concat(alternant_multiword_spec.verb_forms, ",")
else
accel_obj.form = "verb-form-" .. reconstructed_verb_spec
end
return accel_obj
end
local function generate_link(data)
local form = data.form
local link = m_links.full_link {
lang = lang, term = form.formval_for_link, tr = "-", accel = form.accel_obj,
alt = form.alt, gloss = form.gloss, genders = form.genders, pos = form.pos, lit = form.lit, id = form.id,
} .. iut.get_footnote_text(form.footnotes, data.footnote_obj)
if form.q and form.q[1] or form.qq and form.qq[1] or form.l and form.l[1] or form.ll and form.ll[1] then
link = require(pron_qualifier_module).format_qualifiers {
lang = lang,
text = link,
q = form.q,
qq = form.qq,
l = form.l,
ll = form.ll,
}
end
return link
end
local props = {
lang = lang,
lemmas = lemmas,
transform_accel_obj = transform_accel_obj,
generate_link = generate_link,
slot_list = alternant_multiword_spec.verb_slots,
include_translit = true,
}
iut.show_forms(alternant_multiword_spec.forms, props)
end
-------------------------------------------------------------------------------
-- Functions to create inflection tables --
-------------------------------------------------------------------------------
-- Make the conjugation table. Called from export.show().
local function make_table(alternant_multiword_spec)
local notes_template = [=[
<div style="width:100%;text-align:left;background:#d9ebff">
<div style="display:inline-block;text-align:left;padding-left:1em;padding-right:1em">
{footnote}
</div></div>
]=]
local text = [=[
<div class="NavFrame ar-conj">
<div class="NavHead"> Conjugation of {title}</div>
<div class="NavContent">
{\op}| class="inflection-table"
|-
! colspan="6" class="nonfinite-header" | verbal noun<br /><<الْمَصْدَر>>
| colspan="7" | {vn}
]=]
if alternant_multiword_spec.has_active then
text = text .. [=[
|-
! colspan="6" class="nonfinite-header" | active participle<br /><<اِسْم الْفَاعِل>>
| colspan="7" | {ap}
]=]
end
if alternant_multiword_spec.has_passive then
text = text .. [=[
|-
! colspan="6" class="nonfinite-header" | passive participle<br /><<اِسْم الْمَفْعُول>>
| colspan="7" | {pp}
]=]
end
if alternant_multiword_spec.has_active then
text = text .. [=[
|-
! colspan="12" class="voice-header" | active voice<br /><<الْفِعْل الْمَعْلُوم>>
|-
! colspan="2" class="empty-header" |
! colspan="3" class="number-header" | singular<br /><<الْمُفْرَد>>
! rowspan="12" class="divider" |
! colspan="2" class="number-header" | dual<br /><<الْمُثَنَّى>>
! rowspan="12" class="divider" |
! colspan="3" class="number-header" | plural<br /><<الْجَمْع>>
|-
! colspan="2" class="empty-header" |
! class="person-header" | 1<sup>st</sup> person<br /><<الْمُتَكَلِّم>>
! class="person-header" | 2<sup>nd</sup> person<br /><<الْمُخَاطَب>>
! class="person-header" | 3<sup>rd</sup> person<br /><<الْغَائِب>>
! class="person-header" | 2<sup>nd</sup> person<br /><<الْمُخَاطَب>>
! class="person-header" | 3<sup>rd</sup> person<br /><<الْغَائِب>>
! class="person-header" | 1<sup>st</sup> person<br /><<الْمُتَكَلِّم>>
! class="person-header" | 2<sup>nd</sup> person<br /><<الْمُخَاطَب>>
! class="person-header" | 3<sup>rd</sup> person<br /><<الْغَائِب>>
|-
! rowspan="2" class="tam-header" | past (perfect) indicative<br /><<الْمَاضِي>>
! class="gender-header" | m
| rowspan="2" | {past_1s}
| {past_2ms}
| {past_3ms}
| rowspan="2" | {past_2d}
| {past_3md}
| rowspan="2" | {past_1p}
| {past_2mp}
| {past_3mp}
|-
! class="gender-header" | f
| {past_2fs}
| {past_3fs}
| {past_3fd}
| {past_2fp}
| {past_3fp}
|-
! rowspan="2" class="tam-header" | non-past (imperfect) indicative<br /><<الْمُضَارِع الْمَرْفُوع>>
! class="gender-header" | m
| rowspan="2" | {ind_1s}
| {ind_2ms}
| {ind_3ms}
| rowspan="2" | {ind_2d}
| {ind_3md}
| rowspan="2" | {ind_1p}
| {ind_2mp}
| {ind_3mp}
|-
! class="gender-header" | f
| {ind_2fs}
| {ind_3fs}
| {ind_3fd}
| {ind_2fp}
| {ind_3fp}
|-
! rowspan="2" class="tam-header" | subjunctive<br /><<الْمُضَارِع الْمَنْصُوب>>
! class="gender-header" | m
| rowspan="2" | {sub_1s}
| {sub_2ms}
| {sub_3ms}
| rowspan="2" | {sub_2d}
| {sub_3md}
| rowspan="2" | {sub_1p}
| {sub_2mp}
| {sub_3mp}
|-
! class="gender-header" | f
| {sub_2fs}
| {sub_3fs}
| {sub_3fd}
| {sub_2fp}
| {sub_3fp}
|-
! rowspan="2" class="tam-header" | jussive<br /><<الْمُضَارِع الْمَجْزُوم>>
! class="gender-header" | m
| rowspan="2" | {juss_1s}
| {juss_2ms}
| {juss_3ms}
| rowspan="2" | {juss_2d}
| {juss_3md}
| rowspan="2" | {juss_1p}
| {juss_2mp}
| {juss_3mp}
|-
! class="gender-header" | f
| {juss_2fs}
| {juss_3fs}
| {juss_3fd}
| {juss_2fp}
| {juss_3fp}
|-
! rowspan="2" class="tam-header" | imperative<br /><<الْأَمْر>>
! class="gender-header" | m
| rowspan="2" |
| {imp_2ms}
| rowspan="2" |
| rowspan="2" | {imp_2d}
| rowspan="2" |
| rowspan="2" |
| {imp_2mp}
| rowspan="2" |
|-
! class="gender-header" | f
| {imp_2fs}
| {imp_2fp}
]=]
end
if alternant_multiword_spec.has_passive then
text = text .. [=[
|-
! colspan="12" class="voice-header" | passive voice<br /><<الْفِعْل الْمَجْهُول>>
|-
| colspan="2" class="empty-header" |
! colspan="3" class="number-header" | singular<br /><<الْمُفْرَد>>
| rowspan="10" class="divider" |
! colspan="2" class="number-header" | dual<br /><<الْمُثَنَّى>>
| rowspan="10" class="divider" |
! colspan="3" class="number-header" | plural<br /><<الْجَمْع>>
|-
| colspan="2" class="empty-header" |
! class="person-header" | 1<sup>st</sup> person<br /><<الْمُتَكَلِّم>>
! class="person-header" | 2<sup>nd</sup> person<br /><<الْمُخَاطَب>>
! class="person-header" | 3<sup>rd</sup> person<br /><<الْغَائِب>>
! class="person-header" | 2<sup>nd</sup> person<br /><<الْمُخَاطَب>>
! class="person-header" | 3<sup>rd</sup> person<br /><<الْغَائِب>>
! class="person-header" | 1<sup>st</sup> person<br /><<الْمُتَكَلِّم>>
! class="person-header" | 2<sup>nd</sup> person<br /><<الْمُخَاطَب>>
! class="person-header" | 3<sup>rd</sup> person<br /><<الْغَائِب>>
|-
! rowspan="2" class="tam-header" | past (perfect) indicative<br /><<الْمَاضِي>>
! class="gender-header" | m
| rowspan="2" | {past_pass_1s}
| {past_pass_2ms}
| {past_pass_3ms}
| rowspan="2" | {past_pass_2d}
| {past_pass_3md}
| rowspan="2" | {past_pass_1p}
| {past_pass_2mp}
| {past_pass_3mp}
|-
! class="gender-header" | f
| {past_pass_2fs}
| {past_pass_3fs}
| {past_pass_3fd}
| {past_pass_2fp}
| {past_pass_3fp}
|-
! rowspan="2" class="tam-header" | non-past (imperfect) indicative<br /><<الْمُضَارِع الْمَرْفُوع>>
! class="gender-header" | m
| rowspan="2" | {ind_pass_1s}
| {ind_pass_2ms}
| {ind_pass_3ms}
| rowspan="2" | {ind_pass_2d}
| {ind_pass_3md}
| rowspan="2" | {ind_pass_1p}
| {ind_pass_2mp}
| {ind_pass_3mp}
|-
! class="gender-header" | f
| {ind_pass_2fs}
| {ind_pass_3fs}
| {ind_pass_3fd}
| {ind_pass_2fp}
| {ind_pass_3fp}
|-
! rowspan="2" class="tam-header" | subjunctive<br /><<الْمُضَارِع الْمَنْصُوب>>
! class="gender-header" | m
| rowspan="2" | {sub_pass_1s}
| {sub_pass_2ms}
| {sub_pass_3ms}
| rowspan="2" | {sub_pass_2d}
| {sub_pass_3md}
| rowspan="2" | {sub_pass_1p}
| {sub_pass_2mp}
| {sub_pass_3mp}
|-
! class="gender-header" | f
| {sub_pass_2fs}
| {sub_pass_3fs}
| {sub_pass_3fd}
| {sub_pass_2fp}
| {sub_pass_3fp}
|-
! rowspan="2" class="tam-header" | jussive<br /><<الْمُضَارِع الْمَجْزُوم>>
! class="gender-header" | m
| rowspan="2" | {juss_pass_1s}
| {juss_pass_2ms}
| {juss_pass_3ms}
| rowspan="2" | {juss_pass_2d}
| {juss_pass_3md}
| rowspan="2" | {juss_pass_1p}
| {juss_pass_2mp}
| {juss_pass_3mp}
|-
! class="gender-header" | f
| {juss_pass_2fs}
| {juss_pass_3fs}
| {juss_pass_3fd}
| {juss_pass_2fp}
| {juss_pass_3fp}
]=]
end
text = text .. [=[
|{\cl}{notes_clause}</div></div>]=]
local forms = alternant_multiword_spec.forms
if not alternant_multiword_spec.lemmas then
forms.title = "—"
else
local linked_lemmas = {}
for _, form in ipairs(alternant_multiword_spec.lemmas) do
table.insert(linked_lemmas, link_term(form.form, "term"))
end
forms.title = table.concat(linked_lemmas, ", ")
end
local ann_parts = {}
if alternant_multiword_spec.annotation ~= "" then
table.insert(ann_parts, alternant_multiword_spec.annotation)
end
if alternant_multiword_spec.vn then
local linked_vns = {}
for _, form in ipairs(alternant_multiword_spec.vn) do
table.insert(linked_vns, link_term(form.form, "term"))
end
table.insert(ann_parts, (#linked_vns > 1 and "verbal nouns" or "verbal noun") .. " " ..
table.concat(linked_vns, ", "))
end
local annotation = table.concat(ann_parts, ", ")
if annotation ~= "" then
forms.title = forms.title .. " (" .. annotation .. ")"
end
-- Format the table.
forms.notes_clause = forms.footnote ~= "" and m_string_utilities.format(notes_template, forms) or ""
local tagged_table = rsub(text, "<<(.-)>>", tag_text)
return m_string_utilities.format(tagged_table, forms) .. mw.getCurrentFrame():extensionTag{
name = "templatestyles", args = { src = "Template:ar-conj/style.css" }
}
end
-------------------------------------------------------------------------------
-- External entry points --
-------------------------------------------------------------------------------
-- Append two lists `l1` and `l2`, removing duplicates. If either is {nil}, just return the other.
local function combine_lists(l1, l2)
-- combine_footnotes() does exactly what we want.
return iut.combine_footnotes(l1, l2)
end
local function combine_metadata(data)
local src1 = data.form1
local src2 = data.form2
local dest = data.dest_form
dest.uncertain = src1.uncertain or src2.uncertain
if src1.genders and src2.genders and not m_table.deepEquals(src1.genders, src2.genders) then
-- do nothing
else
dest.genders = src1.genders or src2.genders
end
if src1.pos and src2.pos and src1.pos ~= src2.pos then
-- do nothing
else
dest.pos = src1.pos or src2.pos
end
-- Don't copy .alt, .gloss, .lit, .id, which describe a single term and don't extend to multiword terms.
dest.q = combine_lists(src1.q, src2.q)
dest.qq = combine_lists(src1.qq, src2.qq)
dest.l = combine_lists(src1.l, src2.l)
dest.ll = combine_lists(src1.ll, src2.ll)
end
-- Externally callable function to parse and conjugate a verb given user-specified arguments.
-- Return value is WORD_SPEC, an object where the conjugated forms are in `WORD_SPEC.forms`
-- for each slot. If there are no values for a slot, the slot key will be missing. The value
-- for a given slot is a list of objects {form=FORM, footnotes=FOOTNOTES}.
function export.do_generate_forms(args, source_template, headword_head)
local PAGENAME = mw.loadData("Module:headword/data").pagename
local function in_template_space()
return mw.title.getCurrentTitle().nsText == "Template"
end
-- Determine the verb spec we're being asked to generate the conjugation of. This may be taken from the current page
-- title or the value of |pagename=; but not when called from {{ar-verb form}}, where the page title is a
-- non-lemma form. Note that the verb spec may omit the lemma; e.g. it may be "<II>". For this reason, we use the
-- value of `pagename` computed here down below, when calling normalize_all_lemmas().
local pagename = source_template ~= "ar-verb form" and args.pagename or PAGENAME
local head = headword_head or pagename
local arg1 = args[1]
if not arg1 then
if (pagename == "ar-conj" or pagename == "ar-verb" or pagename == "ar-verb form") and in_template_space() then
arg1 = "كتب<I/a~u.pass>"
else
arg1 = "<>"
end
end
-- When called from {{ar-verb form}}, determine the non-lemma form whose inflections we're being asked to
-- determine. This normally comes from the page title or the value of |pagename=.
local verb_form_of_form
if source_template == "ar-verb form" then
verb_form_of_form = args.pagename
if not verb_form_of_form then
if PAGENAME == "ar-verb form" and in_template_space() then
verb_form_of_form = "كتبت"
else
verb_form_of_form = PAGENAME
end
end
end
local incorporated_headword_head_into_lemma = false
if arg1:find("^<.*>$") then -- missing lemma
if head:find(" ") then
-- If multiword lemma, try to add arg spec after the first word.
-- Try to preserve the brackets in the part after the verb, but don't do it
-- if there aren't the same number of left and right brackets in the verb
-- (which means the verb was linked as part of a larger expression).
local first_word, post = rmatch(head, "^(.-)( .*)$")
local left_brackets = rsub(first_word, "[^%[]", "")
local right_brackets = rsub(first_word, "[^%]]", "")
if #left_brackets == #right_brackets then
arg1 = iut.remove_redundant_links(first_word) .. arg1 .. post
incorporated_headword_head_into_lemma = true
else
-- Try again using the form without links.
local linkless_head = m_links.remove_links(head)
if linkless_head:find(" ") then
first_word, post = rmatch(linkless_head, "^(.-)( .*)$")
arg1 = first_word .. arg1 .. post
else
error("Unable to incorporate <...> spec into explicit head due to a multiword linked verb or " ..
"unbalanced brackets; please include <> explicitly: " .. arg1)
end
end
else
-- Will be incorporated through `head` below in the call to normalize_all_lemmas().
incorporated_headword_head_into_lemma = true
end
end
local parse_props = {
parse_indicator_spec = parse_indicator_spec,
angle_brackets_omittable = true,
allow_blank_lemma = true,
}
local alternant_multiword_spec = iut.parse_inflected_text(arg1, parse_props)
alternant_multiword_spec.pos = pos or "verbs"
alternant_multiword_spec.args = args
alternant_multiword_spec.source_template = source_template
alternant_multiword_spec.verb_form_of_form = verb_form_of_form
alternant_multiword_spec.incorporated_headword_head_into_lemma = incorporated_headword_head_into_lemma
normalize_all_lemmas(alternant_multiword_spec, head)
detect_all_indicator_specs(alternant_multiword_spec)
local inflect_props = {
lang = lang,
slot_list = alternant_multiword_spec.verb_slots,
inflect_word_spec = conjugate_verb,
combine_metadata = combine_metadata,
-- We add links around the generated verbal forms rather than allow the entire multiword
-- expression to be a link, so ensure that user-specified links get included as well.
include_user_specified_links = true,
}
iut.inflect_multiword_or_alternant_multiword_spec(alternant_multiword_spec, inflect_props)
if debug_translit then
for slot, forms in pairs(alternant_multiword_spec.forms) do
for _, form in ipairs(forms) do
if form.translit then
local full_form_translit = (lang:transliterate(m_links.remove_links(form.form)))
if full_form_translit ~= form.translit then
error(("Internal error: For slot '%s', form '%s' incremental translit '%s' not same as full translit '%s'"):
format(slot, form.form, form.translit, full_form_translit))
end
end
form.form = iut.remove_redundant_links(form.form)
end
end
end
-- Remove redundant brackets around entire forms.
for slot, forms in pairs(alternant_multiword_spec.forms) do
for _, form in ipairs(forms) do
form.form = iut.remove_redundant_links(form.form)
end
end
determine_slot_uncertainty_from_forms(alternant_multiword_spec)
determine_verb_properties_from_forms(alternant_multiword_spec)
compute_categories_and_annotation(alternant_multiword_spec)
if args.json and source_template == "ar-conj" then
-- There is a circular reference in `base.alternant_multiword_spec`, which points back to top level.
iut.map_word_specs(alternant_multiword_spec, function(base)
base.alternant_multiword_spec = nil
end)
return require("Module:JSON").toJSON(alternant_multiword_spec)
end
return alternant_multiword_spec
end
-- Entry point for {{ar-conj}}. Template-callable function to parse and conjugate a verb given
-- user-specified arguments and generate a displayable table of the conjugated forms.
function export.show(frame)
local parent_args = frame:getParent().args
local params = {
[1] = {},
["noautolinktext"] = {type = "boolean"},
["noautolinkverb"] = {type = "boolean"},
["t"] = {}, -- for use by {{ar-verb form}}; otherwise ignored
["id"] = {}, -- for use by {{ar-verb form}}; otherwise ignored
["pagename"] = {}, -- for testing/documentation pages
["json"] = {type = "boolean"}, -- for bot use
}
local args = require("Module:parameters").process(parent_args, params)
local alternant_multiword_spec = export.do_generate_forms(args, "ar-conj")
if type(alternant_multiword_spec) == "string" then
-- JSON return value
return alternant_multiword_spec
end
show_forms(alternant_multiword_spec)
return make_table(alternant_multiword_spec) ..
require("Module:utilities").format_categories(alternant_multiword_spec.categories, lang, nil, nil, force_cat)
end
function export.verb_forms(frame)
local parargs = frame:getParent().args
local params = {
[1] = {},
[2] = {},
[3] = {},
[4] = {},
[5] = {},
pagename = {},
}
for _, form in ipairs(allowed_vforms) do
-- FIXME: We go up to 5 here. The code supports unlimited variants but it's unlikely we will ever see more than
-- 2.
for index = 1, 5 do
local prefix = index == 1 and form or form .. index
params[prefix .. "-pv"] = {}
for _, extn in ipairs { "", "-vn", "-ap", "-pp" } do
params[prefix .. extn] = {}
params[prefix .. extn .. "-head"] = {}
-- FIXME: No -tr?
params[prefix .. extn .. "-gloss"] = {}
end
end
end
local args = require("Module:parameters").process(parargs, params)
local i = 1
local past_vowel_re = "^[aui,]*$"
local combined_root = nil
if not args[i] or rfind(args[i], past_vowel_re) then
combined_root = args.pagename or mw.loadData("Module:headword/data").pagename
if not rfind(combined_root, "^([^ ]) ([^ ]) ([^ ])$") and not
rfind(combined_root, "^([^ ]) ([^ ]) ([^ ]) ([^ ])$") then
error("When inferring roots from page title, need three or four space-separated radicals: " .. combined_root)
end
elseif rfind(args[i], " ") then
combined_root = args[i]
i = i + 1
else
local separate_roots = {}
while args[i] and not rfind(args[i], past_vowel_re) do
table.insert(separate_roots, args[i])
i = i + 1
end
combined_root = table.concat(separate_roots, " ")
end
local past_vowel = args[i]
i = i + 1
if past_vowel and not rfind(past_vowel, past_vowel_re) then
error("Unrecognized past vowel, should be 'a', 'i', 'u', 'a,u', etc. or empty: " .. past_vowel)
end
-- Spaces interfere with parsing as a unit in [[Module:inflection utilities]], so replace with underscore.
combined_root = combined_root:gsub(" ", "_")
local split_root = rsplit(combined_root, "_")
-- Map from verb forms (I, II, etc.) to a table of verb properties,
-- which has entries e.g. for "verb" (either true to autogenerate the verb
-- head, or an explicitly specified verb head using e.g. argument "I-head"),
-- and for "verb-gloss" (which comes from e.g. the argument "I" or "I-gloss"),
-- and for "vn" and "vn-gloss", "ap" and "ap-gloss", "pp" and "pp-gloss".
local verb_properties = {}
for _, form in ipairs(allowed_vforms) do
local formpropslist = {}
local derivs = {{"verb", ""}, {"vn", "-vn"}, {"ap", "-ap"}, {"pp", "-pp"}}
local index = 1
while true do
local formprops = {}
local prefix = index == 1 and form or form .. index
if prefix == "I" then
formprops.pv = past_vowel
end
if args[prefix .. "-pv"] then
formprops.pv = args[prefix .. "-pv"]
end
for _, deriv in ipairs(derivs) do
local prop = deriv[1]
local extn = deriv[2]
if args[prefix .. extn] == "+" then
formprops[prop] = true
elseif args[prefix .. extn] == "-" then
formprops[prop] = false
elseif args[prefix .. extn] then
formprops[prop] = true
formprops[prop .. "-gloss"] = args[prefix .. extn]
end
if args[prefix .. extn .. "-head"] then
if formprops[prop] == nil then
formprops[prop] = true
end
formprops[prop] = args[prefix .. extn .. "-head"]
end
if args[prefix .. extn .. "-gloss"] then
if formprops[prop] == nil then
formprops[prop] = true
end
formprops[prop .. "-gloss"] = args[prefix .. extn .. "-gloss"]
end
end
if formprops.verb then
-- If a verb form specified, also turn on vn (unless form I, with
-- unpredictable vn) and ap, and maybe pp, according to form,
-- weakness and past vowel. But don't turn these on if there's
-- an explicit on/off specification for them (e.g. I-pp=-).
if form ~= "I" and formprops.vn == nil then
formprops.vn = true
end
if formprops.ap == nil then
formprops.ap = true
end
local weakness = weakness_from_radicals(form, split_root[1], split_root[2], split_root[3],
split_root[4])
if formprops.pp == nil and not vform_probably_no_passive(form,
weakness, rsplit(formprops.pv or "", ","), {}) then
formprops.pp = true
end
if formprops.verb == true or formprops.vn == true or formprops.ap == true or formprops.pp == true then
formprops.need_autogen = true
end
table.insert(formpropslist, formprops)
index = index + 1
else
break
end
end
table.insert(verb_properties, {form, formpropslist})
end
-- Go through and create the verb form derivations as necessary, when they haven't been explicitly given.
for _, vplist in ipairs(verb_properties) do
local vform = vplist[1]
for _, props in ipairs(vplist[2]) do
if props.need_autogen then
local form_with_vowels
if vform == "I" then
local pv = props.pv
if not pv then
-- Make up likely past vowels based on weakness and actual radical.
if split_root[3] == W then -- final-weak
form_with_vowels = "I/a~u"
elseif split_root[3] == Y then
form_with_vowels = "I/a~i"
elseif split_root[2] == W then --hollow
form_with_vowels = "I/u~u"
elseif split_root[2] == Y then
form_with_vowels = "I/i~i"
else
-- most common; doesn't matter so much since we're not displaying the non-past
form_with_vowels = "I/a~u"
end
else
local pvs = rsplit(pv, ",")
local vowel_sufs = {}
for _, pv in ipairs(pvs) do
local vowel_spec
if pv == "a" then
-- Make up likely past vowels based on weakness and actual radical.
if split_root[3] == W then -- final-weak
vowel_spec = "a~u"
elseif split_root[3] == Y then
vowel_spec = "a~i"
elseif split_root[2] == W then --hollow
vowel_spec = "a~u"
elseif split_root[2] == Y then
vowel_spec = "a~i"
else
-- most common; doesn't matter so much since we're not displaying the non-past
vowel_spec = "a~u"
end
elseif pv == "i" then
-- most common; doesn't matter so much since we're not displaying the non-past
vowel_spec = "i~a"
elseif pv == "u" then
-- most common; doesn't matter so much since we're not displaying the non-past
vowel_spec = "u~u"
else
error(("Internal error: Bad past vowel '%s' in {{ar-verb forms}}"):format(pv))
end
table.insert(vowel_sufs, vowel_spec)
end
form_with_vowels = "I/" .. table.concat(vowel_sufs, "/")
end
else
form_with_vowels = vform
end
local angle_bracket_spec = ("%s<%s.pass>"):format(combined_root, form_with_vowels)
local alternant_multiword_spec = export.do_generate_forms({angle_bracket_spec}, "ar-verb forms")
local function format_forms(forms)
if not forms then
return "-" -- FIXME: Throw an error?
end
local formatted = {}
for _, form in ipairs(forms) do
if form.translit then
table.insert(formatted, ("%s//%s"):format(form.form, form.translit))
else
table.insert(formatted, form.form)
end
end
return table.concat(formatted, ",")
end
if props.verb == true then
props.verb = format_forms(alternant_multiword_spec.forms.past_3ms)
end
for _, deriv in ipairs({"vn", "ap", "pp"}) do
if props[deriv] == true then
props[deriv] = format_forms(alternant_multiword_spec.forms[deriv])
end
end
end
end
end
-- Go through and output the result
local formtextarr = {}
for _, vplist in ipairs(verb_properties) do
local form = vplist[1]
for _, props in ipairs(vplist[2]) do
local textarr = {}
if props.verb then
local text = "* '''[[Appendix:Arabic verbs#Form " .. form .. "|Form " .. form .. "]]''': "
local linktext = {}
local splitheads = rsplit(props.verb, "[,،]")
for _, head in ipairs(splitheads) do
table.insert(linktext, m_links.full_link({lang = lang, term = head, gloss = props["verb-gloss"]}))
end
text = text .. table.concat(linktext, ", ")
table.insert(textarr, text)
for _, derivengl in ipairs({{"vn", "Verbal noun"}, {"ap", "Active participle"}, {"pp", "Passive participle"}}) do
local deriv = derivengl[1]
local engl = derivengl[2]
if props[deriv] then
local text = "** " .. engl .. ": "
local linktext = {}
local splitheads = rsplit(props[deriv], "[,،]")
for _, head in ipairs(splitheads) do
local ar, translit = head:match("^(.*)//(.-)$")
if not ar then
ar = head
end
table.insert(linktext, m_links.full_link {lang = lang, term = ar, tr = translit,
gloss = props[deriv .. "-gloss"]} )
end
text = text .. table.concat(linktext, ", ")
table.insert(textarr, text)
end
end
table.insert(formtextarr, table.concat(textarr, "\n"))
end
end
end
return table.concat(formtextarr, "\n")
end
-- Infer radicals from lemma headword (i.e. 3rd masculine singular past) and verb form (I, II, etc.). Throw an error if
-- headword is malformed. A given returned radical may be actually be a list of possible radicals, where the first one
-- should be used if the user didn't explicitly give the radical. If the list contains a field `ambig = true`, the
-- radical is considered ambiguous and should not be categorized. `is_reduced` indicates that the user specified
-- `.reduced` to indicate that the verb form is reduced by assimilation and/or haplology (typically archaic Koranic
-- forms such as اِدَّارَأَ instead of تَدَارَأَ; or اِسْطَاعَ instead of اِسْتِطَاعَ; etc.
function export.infer_radicals(data)
local headword, vform, passive, past_vowel, nonpast_vowel, is_reduced =
data.headword, data.vform, data.passive, data.past_vowel, data.nonpast_vowel, data.is_reduced
past_vowel = past_vowel or "-"
nonpast_vowel = nonpast_vowel or "-"
local function verify_vowel(vowel, param)
if vowel ~= A and vowel ~= I and vowel ~= U and vowel ~= "-" then
error(("Internal error: Bad value for %s: %s (should be Arabic diacritic vowel or '-')"):format(
param, vowel))
end
end
verify_vowel(past_vowel, "past_vowel")
verify_vowel(nonpast_vowel, "nonpast_vowel")
local ch = {}
local form_viii_assim, variant
-- sub out alif-madda for easier processing
headword = rsub(headword, AMAD, HAMZA .. ALIF)
local function infer_err(msg, noann)
local anns = {}
local nohead, novform
if noann == "nohead" then
nohead = true
elseif noann == "novform" then
novform = true
elseif noann == "nohead-vform" then
nohead = true
novform = true
elseif noann then
error(("Internal error: Unrecognized value for 'noann': %s"):format(dump(noann)))
end
if not nohead then
table.insert(anns, ("headword=%s"):format(data.headword))
end
if not novform then
table.insert(anns, ("verb form=%s"):format(data.vform))
end
anns = table.concat(anns, ", ")
if anns ~= "" then
anns = ": " .. anns
end
error(msg .. anns)
end
local len = ulen(headword)
local expected_length
-- extract the headword letters into an array
for i = 1, len do
table.insert(ch, usub(headword, i, i))
end
-- check that the letter at the given index is the given string, or
-- is one of the members of the given array
local function check(index, must)
local letter = ch[index]
if type(must) == "string" then
if not letter then
infer_err("Letter " .. index .. " is nil")
end
if letter ~= must then
infer_err(("For verb form %s, letter %s must be %s, not %s"):format(vform, index, must, letter),
"novform")
end
elseif not m_table.contains(must, letter) then
infer_err("For verb form " .. vform .. ", radical " .. index ..
" must be one of " .. table.concat(must, " ") .. ", not " .. letter, "novform")
end
end
-- Check that length of headword is within [min, max]
local function check_len(min, max)
if min and len < min then
infer_err(("Not enough letters for verb form %s, expected at least %s"):format(vform, min), "novform")
end
if max and len > max then
infer_err(("Too many letters for verb form %s, expected at most %s"):format(vform, max), "novform")
end
end
-- If the vowels are i~a or u~u, a form I verb beginning with w- normally keeps the w in the non-past. Otherwise it
-- loses it (i.e. it is "assimilated").
local function form_I_w_non_assimilated()
return req(past_vowel, I) and req(nonpast_vowel, A) or req(past_vowel, U) and req(nonpast_vowel, U)
end
-- Convert radicals to canonical form (handle various hamza varieties and check for misplaced alif or alif maqṣūra;
-- legitimate cases of these letters are handled above).
local function convert(rad, index)
if type(rad) == "table" then
for i, r in ipairs(rad) do
rad[i] = convert(r, index)
end
return rad
elseif rad == HAMZA_ON_ALIF or rad == HAMZA_UNDER_ALIF or
rad == HAMZA_ON_W or rad == HAMZA_ON_Y then
return HAMZA
elseif rad == AMAQ then
infer_err("Radical " .. index .. " must not be alif maqṣūra")
elseif rad == ALIF then
infer_err("Radical " .. index .. " must not be alif")
else
return rad
end
end
local quadlit = vform:find("q$")
-- find first radical, start of second/third radicals, check for
-- required letters
local radstart, rad1, rad2, rad3, rad4
local weakness
if vform == "I" or vform == "II" then
rad1 = ch[1]
radstart = 2
elseif vform == "III" then
rad1 = ch[1]
check(2, {ALIF, W}) -- W occurs in passive-only verbs
radstart = 3
elseif vform == "IV" then
-- this would be alif-madda but we replaced it with hamza-alif above.
if ch[1] == HAMZA and ch[2] == ALIF then
rad1 = HAMZA
else
check(1, HAMZA_ON_ALIF)
rad1 = ch[2]
end
radstart = 3
elseif vform == "V" then
check(1, is_reduced and ALIF or T)
rad1 = ch[2]
radstart = 3
elseif vform == "VI" then
check(1, is_reduced and ALIF or T)
if ch[2] == AMAD then
rad1 = HAMZA
radstart = 3
else
rad1 = ch[2]
check(3, {ALIF, W}) -- W occurs in passive-only verbs
radstart = 4
end
elseif vform == "VII" then
check(1, ALIF)
if is_reduced then
check(2, M)
rad1 = M
radstart = 3
else
check(2, N)
rad1 = ch[3]
radstart = 4
end
elseif vform == "VIII" then
check(1, ALIF)
rad1 = ch[2]
if rad1 == "د" then
rad1 = {"د", "ذ"} -- not considered ambiguous since it's usually د
radstart = 3
form_viii_assim = "دّ"
elseif rad1 == "ظ" and ch[3] == "ط" and len >= 5 then
-- [[اظطلم]], variant of [[اظلم]]
radstart = 4
form_viii_assim = "ظْط"
elseif rad1 == "ذ" and ch[3] == "د" and len >= 5 then
-- [[اذدكر]], variant of [[اذكر]]
radstart = 4
form_viii_assim = "ذْد"
elseif rad1 == T or rad1 == "ث" or rad1 == "ذ" or rad1 == "ط" or rad1 == "ظ" then
radstart = 3
form_viii_assim = rad1 .. SH
elseif rad1 == "ز" then
check(3, "د")
radstart = 4
form_viii_assim = "زْد"
elseif rad1 == "ص" or rad1 == "ض" then
check(3, "ط")
radstart = 4
form_viii_assim = rad1 .. SK .. "ط"
else
check(3, T)
radstart = 4
rad1 = convert(rad1, 1)
form_viii_assim = rad1 .. SK .. "ت"
end
if rad1 == T then
-- Radical is ambiguous, might be ت or و or ي but doesn't affect conjugation. Note that there are no
-- form-VIII verbs with initial radical ي given in Hans Wehr but Lane mentions at least:
-- - (page 2973) اِتَّأَسَ, with assimilation of the ي to ت, from root ي ء س;
-- - (page 2975) اِتَّبَسَ non-past يَتَّبِسُ and alternative اِيتَبَسَ non-past يَاتَبِسُ from the root ي ب س;
-- - (page 2976) اِتَّسَرَ non-past يَتَّسِرُ or alternatively يَأْتَسِرُ with hamza preserved from the root ي س ر.
-- These alternative forms seem very rare and probably not worth worrying about, but if we want to handle
-- them, we can do it when the time comes.
rad1 = {T, W, Y, ambig = true}
-- اِتَّخَذَ irregularly has hamza as the radical but assimilates like و
if ch[3] == "خ" and ch[4] == "ذ" then
rad1[4] = HAMZA
end
end
elseif vform == "IX" then
check(1, ALIF)
rad1 = ch[2]
radstart = 3
elseif vform == "X" then
check(1, ALIF)
check(2, S)
if is_reduced then
rad1 = ch[3]
radstart = 4
else
check(3, T)
rad1 = ch[4]
radstart = 5
end
elseif vform == "Iq" then
rad1 = ch[1]
rad2 = ch[2]
radstart = 3
elseif vform == "IIq" then
check(1, T)
rad1 = ch[2]
rad2 = ch[3]
radstart = 4
elseif vform == "IIIq" then
check(1, ALIF)
rad1 = ch[2]
rad2 = ch[3]
check(4, N)
radstart = 5
elseif vform == "IVq" then
check(1, ALIF)
rad1 = ch[2]
rad2 = ch[3]
radstart = 4
elseif vform == "XI" then
check_len(5, 5)
check(1, ALIF)
rad1 = ch[2]
rad2 = ch[3]
check(4, ALIF)
rad3 = ch[5]
weakness = "sound"
elseif vform == "XII" then
check(1, ALIF)
rad1 = ch[2]
if ch[3] ~= ch[5] then
infer_err("For verb form XII, letters 3 and 5 should be the same", "novform")
end
check(4, W)
radstart = 5
elseif vform == "XIII" then
check_len(5, 5)
check(1, ALIF)
rad1 = ch[2]
rad2 = ch[3]
check(4, W)
rad3 = ch[5]
if rad3 == AMAQ then
weakness = "final-weak"
else
weakness = "sound"
end
elseif vform == "XIV" then
check_len(6, 6)
check(1, ALIF)
rad1 = ch[2]
rad2 = ch[3]
check(4, N)
rad3 = ch[5]
if ch[6] == AMAQ then
check_waw_ya(rad3)
weakness = "final-weak"
else
if ch[5] ~= ch[6] then
infer_err("For verb form XIV, letters 5 and 6 should be the same", "novform")
end
weakness = "sound"
end
elseif vform == "XV" then
check_len(6, 6)
check(1, ALIF)
rad1 = ch[2]
rad2 = ch[3]
check(4, N)
rad3 = ch[5]
if rad3 == Y then
check(6, ALIF)
else
check(6, AMAQ)
end
weakness = "sound"
else
error("Internal error: Unrecognized verb form " .. vform)
end
-- Process the last two radicals. RADSTART is the index of the first of the two. If it's nil then all radicals have
-- already been processed above, and we don't do anything.
if radstart then
-- There must (normally) be one or two letters left.
if len == radstart then
if vform == "I" and ch[len] == Y then
-- short form حَيَّ
weakness = "final-weak"
rad2 = Y
rad3 = Y
variant = "short"
elseif vform == "IV" and rad1 == "ر" and ch[len] == AMAQ then
-- irregular verb أَرَى
weakness = "final-weak"
rad2 = HAMZA
rad3 = Y
elseif vform == "X" and rad1 == "ح" and ch[len] == AMAQ then
-- irregular verb اِسْتَحَى
weakness = "final-weak"
rad2 = Y
rad3 = Y
variant = "short"
else
-- If one letter left, then it's a geminate verb. If the letter is alif or alif maqṣūra, it will trigger
-- an error down the line.
if vform_supports_geminate(vform) then
weakness = "geminate"
rad2 = ch[len]
rad3 = ch[len]
if vform == "III" or vform == "VI" then
variant = "short"
end
else
infer_err("Apparent geminate verb, but geminate verbs not allowed for this verb form")
end
end
elseif quadlit then
-- Process last two radicals of a quadriliteral verb form.
rad3 = ch[radstart]
rad4 = ch[radstart + 1]
expected_length = radstart + 1
check_len(expected_length)
if rad4 == AMAQ or rad4 == ALIF and rad3 == Y or rad4 == Y then
-- rad4 can be Y in passive-only verbs.
if vform_supports_final_weak(vform) then
weakness = "final-weak"
-- Ambiguous radical; randomly pick wāw as radical (but avoid two wāws in a row); it could be wāw or
-- yāʾ, but doesn't affect the conjugation.
rad4 = rad3 == W and {Y, W, ambig = true} or {W, Y, ambig = true}
else
infer_err("Last radical is " .. rad4 .. " but verb form " .. vform ..
" doesn't support final-weak verbs", "novform")
end
else
weakness = "sound"
end
else
-- Process last two radicals of a triliteral verb form.
rad2 = ch[radstart]
rad3 = ch[radstart + 1]
expected_length = radstart + 1
check_len(expected_length)
if vform == "I" and (is_waw_ya(rad3) or rad3 == ALIF or rad3 == AMAQ) then
local inferred_past_vowel, inferred_nonpast_vowel
-- Check for final-weak form I verb. It can end in tall alif (rad3 = wāw) or alif maqṣūra (rad3 = yāʾ)
-- or a wāw or yāʾ (with a past vowel of i or u, e.g. nasiya/yansā "forget" or with a passive-only
-- verb).
if rad1 == W and not form_I_w_non_assimilated() then
weakness = "assimilated+final-weak"
else
weakness = "final-weak"
end
if rad3 == ALIF then
rad3 = W
inferred_past_vowel = A
inferred_nonpast_vowel = U
if is_passive_only(passive) then
infer_err("Final-weak form-I passive verbs should end in yāʔ (ي), not tall alif (ا)", "novform")
end
elseif rad3 == AMAQ then
rad3 = Y
inferred_past_vowel = A
inferred_nonpast_vowel = I
if is_passive_only(passive) then
infer_err("Final-weak form-I passive verbs should end in yāʔ (ي), not alif maqṣūra (ى)",
"novform")
end
elseif rad1 == "ح" and rad2 == Y and rad3 == Y then
-- Long variant حَيِيَ.
inferred_past_vowel = I
inferred_nonpast_vowel = A
variant = "long"
else
if not is_passive_only(passive) then
-- does a non-passive final-weak verb in -uwa ever happen?
inferred_past_vowel = rad3 == Y and I or U
inferred_nonpast_vowel = A
end
-- Ambiguous radical; randomly pick wāw as radical (but avoid two wāws); it could be wāw or yāʾ, but
-- doesn't affect the conjugation.
rad3 = (rad1 == W or rad2 == W) and {Y, W, ambig = true} or {W, Y, ambig = true} -- ambiguous
end
if inferred_past_vowel then
local raw_past_vowel = rget(past_vowel)
local raw_nonpast_vowel = rget(nonpast_vowel)
if raw_past_vowel ~= "-" then
if raw_past_vowel ~= inferred_past_vowel then
infer_err(("Final-weak form-I verb inferred past vowel %s, which disagrees with " ..
"explicitly specified %s"):format(undia[inferred_past_vowel], undia[raw_past_vowel]), "novform")
else
-- in case of footnote in past_vowel
inferred_past_vowel = past_vowel
end
end
if raw_nonpast_vowel ~= "-" and raw_nonpast_vowel ~= A and inferred_nonpast_vowel == U then
-- if inferred as I or A, the reality can be the reverse; form-I final-weak verbs with a~a and
-- i~i exist, e.g. سَعَى/يَسْعَى, وَلِيَ/يَلِي. Weird verb [[صها]] (also written [[صهى]]) has non-past
-- يصهى so we can't throw an error in this situation.
if raw_nonpast_vowel ~= inferred_nonpast_vowel then
infer_err(("Final-weak form-I verb inferred non-past vowel %s, which disagrees with " ..
"explicitly specified %s"):format(undia[inferred_nonpast_vowel], undia[raw_nonpast_vowel]), "novform")
else
-- in case of footnote in nonpast_vowel
inferred_nonpast_vowel = nonpast_vowel
end
end
end
if not is_passive_only(passive) then
if rget(past_vowel) == "-" then
past_vowel = inferred_past_vowel
end
if rget(nonpast_vowel) == "-" then
nonpast_vowel = inferred_nonpast_vowel
end
end
elseif vform == "IX" and is_waw_ya(rad3) and len == radstart + 2 and ch[len] == AMAQ then
-- Final-weak form IX verbs like اِرْعَوَى "to desist, to repent, to see the light".
weakness = "final-weak"
expected_length = radstart + 2
elseif vform == "X" and rad1 == "ح" and rad2 == Y and rad3 == ALIF then
-- Long variant اِسْتَحْيَا.
weakness = "final-weak"
rad3 = Y
variant = "long"
elseif rad3 == AMAQ or rad2 == Y and rad3 == ALIF or rad3 == Y then
-- rad3 == Y happens in passive-only verbs.
if vform_supports_final_weak(vform) then
weakness = "final-weak"
else
infer_err("Last radical is " .. rad3 .. " but verb form doesn't support final-weak verbs")
end
-- Ambiguous radical; randomly pick wāw as radical (but avoid two wāws); it could be wāw or yāʾ, but
-- doesn't affect the conjugation.
rad3 = (rad1 == W or rad2 == W) and {Y, W, ambig = true} or {W, Y, ambig = true}
elseif rad2 == ALIF then
if vform_supports_hollow(vform) then
weakness = "hollow"
local function set_past_to_nonpast()
if req(past_vowel, "-") or req(past_vowel, A) then
past_vowel = nonpast_vowel -- in case of footnote
end
end
if vform == "I" and req(nonpast_vowel, U) then
rad2 = W
set_past_to_nonpast()
elseif vform == "I" and req(nonpast_vowel, I) then
rad2 = Y
set_past_to_nonpast()
else
if req(nonpast_vowel, A) and not req(past_vowel, I) then
infer_err(("Form I hollow verb with nonpast vowel set to 'a' must have past vowel set to 'i', not %s"):
format(rget(past_vowel)), "novform")
end
-- Ambiguous radical; could be wāw or yāʾ; if verb form I, it's critical to get this right, and
-- the caller checks for this situation and throws an error if non-past vowel is "a" and second
-- radical isn't explicitly given.
rad2 = {W, Y, ambig = true, need_radical = true}
end
else
infer_err("Second radical is alif but verb form doesn't support hollow verbs")
end
elseif vform == "I" and rad1 == W and not form_I_w_non_assimilated() then
weakness = "assimilated"
elseif rad2 == rad3 and (vform == "III" or vform == "VI") then
weakness = "geminate"
variant = "long"
else
weakness = "sound"
end
end
if expected_length then
check_len(expected_length, expected_length)
end
end
rad1 = convert(rad1, 1)
rad2 = convert(rad2, 2)
rad3 = convert(rad3, 3)
rad4 = convert(rad4, 4)
if not weakness then
error("Internal error: Returned weakness from infer_radicals() is nil")
end
return {
weakness = weakness,
rad1 = rad1,
rad2 = rad2,
rad3 = rad3,
rad4 = rad4,
past_vowel = past_vowel,
nonpast_vowel = nonpast_vowel,
form_viii_assim = form_viii_assim,
variant = variant,
}
end
-- bot interface to infer_radicals()
function export.infer_radicals_json(frame)
local iparams = {
headword = {},
vform = {},
passive = {},
past_vowel = {},
nonpast_vowel = {},
is_reduced = {type = "boolean"},
}
local iargs = require("Module:parameters").process(frame.args, iparams)
return require("Module:JSON").toJSON(export.infer_radicals(iargs))
end
-- Infer vocalization from participle headword (active or passive), verb form (I, II, etc.) and whether the headword is
-- active or passive. Throw an error if headword is malformed. Returned radicals may contain Latin letters "t", "w" or "y"
-- indicating ambiguous radicals guessed to be tāʾ, wāw or yāʾ respectively.
function export.infer_participle_vocalization(headword, vform, weakness, is_active)
local chars = {}
local orig_headword = headword
-- Sub out alif-madda for easier processing.
headword = rsub(headword, AMAD, HAMZA .. ALIF)
local len = ulen(headword)
-- Extract the headword letters into an array.
for i = 1, len do
table.insert(chars, usub(headword, i, i))
end
local function form_intro_error_msg()
return ("For verb form %s %s%s participle %s, "):format(vform, orig_headword ~= headword and "normalized " or
"", is_active and "active" or "passive", headword)
end
local function err(msg)
error(form_intro_error_msg() .. msg, 1)
end
-- Check that length of headword is within [min, max].
local function check_len(min, max)
if min and len < min then
err(("expected at least %s letters but saw %s"):format(min, len))
elseif max and len > max then
err(("expected at most %s letters but saw %s"):format(max, len))
end
end
-- Get the character at `ind`, making sure it exists.
local function c(ind)
check_len(ind)
return chars[ind]
end
-- Check that the letter at the given index is the given string, or is one of the members of the given array
local function check(index, must)
local letter = chars[index]
local function make_possible_values()
if type(must) == "string" then
return must
else
return m_table.serialCommaJoin(must, {conj = "or", dontTag = true})
end
end
if not letter then
err(("expected a letter (specifically %s) at position %s, but participle is too short"):format(
make_possible_values(), index))
end
local matches
if type(must) == "string" then
matches = letter == must
else
matches = m_table.contains(must, letter)
end
if not matches then
err(("letter %s at index %s must be %s"):format(letter, index, make_possible_values()))
end
end
local function check_weakness(values, allow_missing, invert_condition)
local function make_possible_weaknesses()
for i, val in ipairs(values) do
values[i] = "'" .. val .. "'"
end
return m_table.serialCommaJoin(values, {conj = "or", dontTag = true})
end
if allow_missing and invert_condition then
error("Internal error: Can't specify both allow_missing and invert_condition")
end
if not weakness then
if allow_missing or invert_condition then
return
else
err(("weakness is unspecified but must be %s"):format(make_possible_weaknesses()))
end
else
local matches = m_table.contains(values, weakness)
if invert_condition and matches then
err(("weakness '%s' must not be %s"):format(weakness, make_possible_weaknesses()))
elseif not invert_condition and not matches then
err(("weakness '%s' must be %s"):format(weakness, make_possible_weaknesses()))
end
end
end
local vocalized
local function handle_possibly_final_weak(sound_prefix, expected_length)
check_len(expected_length, expected_length)
if c(expected_length) == AMAQ then
-- passive final-weak
if is_active then
err("participle in -ِى only allowed for passive participles")
end
check_weakness({"final-weak", "assimilated+final-weak"}, "allow missing")
vocalized = sound_prefix .. AN .. AMAQ
else
-- all others behave as if sound
check_weakness({"final-weak", "assimilated+final-weak"}, nil, "invert condition")
vocalized = sound_prefix .. (is_active and I or A) .. c(expected_length)
end
end
if not (vform == "I" and is_active) then
-- all participles except verb form I active begin in م-.
check(1, M)
end
if vform == "I" then
if is_active then
check(2, ALIF)
local sound_prefix = c(1) .. AA .. c(3)
if len == 3 then
if c(3) == HAMZA then
-- Either hollow with hamzated third radical, e.g. [[شاء]] active participle 'شَاءٍ', or final-weak
-- with hamzated second radical, e.g. [[رأى]] active participle 'رَاءٍ'. Theoretically (?), also
-- geminate with hamzated second/third radical, but I don't know if any such verbs exist.
if weakness == "geminate" then
vocalized = sound_prefix .. SH
else
check_weakness({"hollow", "final-weak"}, "allow missing")
vocalized = sound_prefix .. IN
end
else
check_weakness({"final-weak", "geminate"})
if weakness == "geminate" then
vocalized = sound_prefix .. SH
else
vocalized = sound_prefix .. IN
end
end
else
check_len(4, 4)
-- we will convert back to alif maqṣūra below as needed
vocalized = sound_prefix .. I .. c(4)
end
else
-- assimilated verbs: regular, e.g. مَوْزُون "weighed"
-- geminate verbs: regular, e.g. مَبْلُول "moistened"
-- third-hamzated verbs: مَبْرُوء
-- hollow verbs: مَقُود "led, driven"; مَزِيد "added, increased"
-- hollow first-hamzated verbs: مَئِيض "returned, reverted"; مَأْيُوس "despaired" (NOTE: formation is sound);
-- مَأُود or مَؤُود "bent; depleted"
-- hollow third-hamzated verbs: مَشِيء "willed, intended", مَضُوء "glittered?"
-- final-weak: مَلْقِيّ "found, encountered"; مَصْغُوّ "inclined"
-- hollow + final-weak: مَشْوِيّ "fried, grilled", مَهْوِيّ "loved"
-- first-hamzated + hollow + final-weak: مَأْوِيّ "received hospitably"
local sound_prefix = MA .. c(2) .. SK .. c(3)
if len == 5 then
-- sound, assimilated or geminate
check(4, W)
vocalized = sound_prefix .. UU .. c(5)
else
check_len(4, 4)
if c(4) == W then
-- final-weak third-wāw
vocalized = sound_prefix .. U .. W .. SH
elseif c(4) == Y then
-- final-weak third-yāʾ
vocalized = sound_prefix .. I .. Y .. SH
else
-- hollow
check(3, {W, Y})
if c(3) == W then
vocalized = MA .. c(2) .. UU .. c(4)
else
vocalized = MA .. c(2) .. II .. c(4)
end
end
end
end
elseif vform == "II" or vform == "V" or vform == "XII" or vform == "XIII" or vform == "Iq" or vform == "IIq" or
vform == "IIIq" then
local sound_prefix, expected_length
if vform == "II" then
sound_prefix = MU .. c(2) .. A .. c(3) .. SH
expected_length = 4
elseif vform == "V" then
check(2, T)
sound_prefix = MU .. T .. A .. c(3) .. A .. c(4) .. SH
expected_length = 5
elseif vform == "XII" then
-- e.g. [[احدودب]] "to be or become convex or humpbacked", مُحْدَوْدِب (active);
-- [[اثنونى]] "to be bent; to be doubled up", مُثْنَوْنٍ (active)
check(4, W)
if c(3) ~= c(5) then
err(("third letter %s should be the same as the fifth letter %s"):format(c(3), c(5)))
end
sound_prefix = MU .. c(2) .. SK .. c(3) .. A .. W .. SK .. c(5)
expected_length = 6
elseif vform == "XIII" then
-- e.g. [[اخروط]] "to get entangled; to extend", مُخْرَوِّط (active), مُخْرَوَّط (passive)
check(4, W)
sound_prefix = MU .. c(2) .. SK .. c(3) .. A .. W .. SH
expected_length = 5
elseif vform == "Iq" then
sound_prefix = MU .. c(2) .. A .. c(3) .. SK .. c(4)
expected_length = 5
elseif vform == "IIq" then
check(2, T)
sound_prefix = MU .. T .. A .. c(3) .. A .. c(4) .. SK .. c(5)
expected_length = 6
elseif vform == "IIIq" then
-- e.g. [[اخرنطم]] "to be proud and angry"
check(4, T)
sound_prefix = MU .. c(2) .. SK .. c(3) .. A .. N .. SK .. c(5)
expected_length = 6
else
error("Internal error: Unhandled verb form " .. vform)
end
if len == expected_length - 1 then
-- active final-weak
if not is_active then
err(("length-%s participle only allowed for active participles"):format(len))
end
check_weakness({"final-weak", "assimilated+final-weak"}, "allow missing")
vocalized = sound_prefix .. IN
else
handle_possibly_final_weak(sound_prefix, expected_length)
end
elseif vform == "III" or vform == "VI" then
local sound_prefix, expected_length
if vform == "VI" then
check(2, T)
check(4, ALIF)
sound_prefix = MU .. T .. A .. c(3) .. AA .. c(5)
expected_length = 6
else
sound_prefix = MU .. c(2) .. AA .. c(4)
expected_length = 5
end
if len == expected_length - 1 then
-- active final-weak or active or passive geminate
if is_active then
check_weakness({"geminate", "final-weak", "assimilated+final-weak"})
if weakness == "geminate" then
vocalized = sound_prefix .. SH
else
vocalized = sound_prefix .. IN
end
else
check_weakness({"geminate"}, "allow missing")
vocalized = sound_prefix .. SH
end
else
handle_possibly_final_weak(sound_prefix, expected_length)
end
elseif vform == "IV" or vform == "X" then
-- form IV:
-- sound: مُرْسِخ (active, "entrenching"), مُرْسَخ (passive, "entrenched")
-- first-hamzated (like sound): مُؤْيِس (active, "causing to despair"), مُؤْيَس (passive, "caused to despair")
-- final-weak: مُكْرٍ (active, "renting out"), مُكْرًى (passive, "rented out")
-- assimilated: مُورِد (active, "transferring"), مُورَد (passive, "transferred"); same when first-Y, e.g.
-- أَيْقَنَ "to be certain of": مُوقِن (active), مُوقَن (passive)
-- assimilated + final-weak: مُورٍ (active, "setting fire, kindling"), مُورًى (passive, "set fire, kindled")
-- geminate: مُمِدّ (active, "granting, helping"), مُمَدّ (passive, "granted, helped")
-- hollow: مُزِيل (active, "eliminating"), مُزَال (passive, "eliminated")
-- hollow + final-weak: مُعْيٍ (active, "tiring"), مُعْيًى (passive, "tired")
local sound_prefix, expected_length
if vform == "X" then
check(2, S)
check(3, T)
sound_prefix = MU .. S .. SK .. T .. A .. c(4)
expected_length = 6
else
sound_prefix = MU .. c(2)
expected_length = 4
end
if len == expected_length and c(len - 1) == Y and c(len) ~= AMAQ then
-- active hollow
if not is_active then
err("this shape only allowed for active participles")
end
check_weakness({"hollow"}, "allow missing")
vocalized = sound_prefix .. II .. c(len)
elseif len == expected_length and c(len - 1) == ALIF then
-- passive hollow
if is_active then
err("this shape only allowed for passive participles")
end
check_weakness({"hollow"}, "allow missing")
vocalized = sound_prefix .. AA .. c(len)
elseif len == expected_length - 1 then
-- active final-weak or active or passive geminate
if is_active then
check_weakness({"geminate", "final-weak", "assimilated+final-weak"})
if weakness == "geminate" then
vocalized = sound_prefix .. I .. c(len) .. SH
elseif vform == "IV" and c(2) == W then
-- assimilated final-weak
vocalized = sound_prefix .. c(len) .. IN
else
vocalized = sound_prefix .. SK .. c(len) .. IN
end
else
check_weakness({"geminate"}, "allow missing")
vocalized = sound_prefix .. A .. c(len) .. SH
end
else
if vform == "IV" and c(2) == W then
-- assimilated, possibly final-weak
sound_prefix = sound_prefix .. c(expected_length - 1)
else
sound_prefix = sound_prefix .. SK .. c(expected_length - 1)
end
handle_possibly_final_weak(sound_prefix, expected_length)
end
elseif vform == "VII" or vform == "VIII" then
-- form VII (passive participles are fairly rare but do exist):
-- sound: مُنْكَتِب (active "subscribing"), مُنْكَتَب (passive "subscribed")
-- geminate: مُنْضَمّ (both active "joining, containing" and passive "joined, contained")
-- final-weak: مُنْطَلٍ (active "fooling (someone)"), مُنْطَلًى (passive "fooled")
-- final-weak with medial wāw: مُنْطَوٍ (active "involving"), مُنْطَوًى (passive "involved")
-- hollow: مُنْقَاد (both active "complying with" and passive "complied with")
--
-- for form VIII, the same variants exist but things are complicated by assimilations involving the template T.
-- sound third-hamzated no assimilation: مُبْتَدِئ (active "beginning"), مُبْتَدَأ (passive "begun")
-- geminate no assimilation: مُبْتَزّ (both active "robbing" and passive "robbed")
-- final-weak no assimilation: مُبْتَنٍ (active "building"), مُبْتَنًى (passive "built")
-- final-weak with medial wāw no assimilation: مُحْتَوٍ (active "containing"), مُحْتَوًى (passive "contained")
-- hollow no assimilation: مُخْتَار (both active "choosing" and passive "chosen")
--
-- sound with total assimilation: مُتَّبِع (active "following"), مُتَّبَع (passive "followed")
-- sound with total assimilation, assimilating wāw: مُتَّعِد (active "threatening"), مُتَّعَد (passive "threatened")
-- sound with total assimilation, irregularly assimilating hamza: مُتَّخِذ (active "taking"), مُتَّخَذ (passive "taken")
-- sound with total assimilation (to ḏāl, producing dāl): مُدَّخِر (active "reserving"), مُدَّخَر (passive "reserved")
-- sound with total assimilation (to ḏāl): مُذَّكِر (active "remembering"), مُذَّكَر (passive "remembered")
-- sound with total assimilation (to ṭāʔ): مُطَّرِح (active "discarding"), مُطَّرَح (passive "discarded")
-- sound with total assimilation (to ẓāʔ): مُظَّلِم (active "tolerating"), مُظَّلَم (passive "tolerated")
-- final-weak with total assimilation, assimilating wāw: مُتَّقٍ (active "guarding against"), مُتَّقًى (passive "guarded against")
-- final-weak with total assimilation (to ṯāʔ): مُثَّنٍ (active "undulating"), مُثَّنًى (passive "undulated")
-- final-weak with total assimilation (to dāl): مُدَّعٍ (active "claiming"), مُدَّعًى (passive "claimed")
-- sound with partial assimilation (to zayn): مُزْدَهِر (active "thriving"), مُزْدَهَر (passive "thrived")
-- sound with medial wāw with partial assimilation (to zayn): مُزْدَوِج (active "appearing twice")
-- sound with partial assimilation (to ṣād): مُصْطَبِح (active "illuminating"), مُصْطَبَح (passive, "illuminated")
-- sound with partial assimilation (to ḍād): مُضْطَرِب (active "to be disturbed"; no passive)
-- geminate with partial assimilation (to ṣād): مُصْطَبّ (both active "effusing" and passive "effused")
-- geminate with partial assimilation (to ḍād): مُضْطَرّ (both active "forcing" and passive "forced")
-- final-weak with partial assimilation (to ṣād): مُصْطَلٍ (active "warming"), مُصْطَلًى (passive "warmed")
-- hollow with partial assimilation (to zayn): مُزْدَاد (both active "increasing" and passive "increased")
-- hollow with partial assimilation (to ṣad): مُصْطَاد (both active "hunting" and passive "hunted")
local sound_prefix, sufind
if vform == "VII" then
check(2, N)
sound_prefix = MU .. N .. SK .. c(3)
sufind = 4
else
local c2 = c(2)
if c2 == T or c2 == "د" or c2 == "ث" or c2 == "ذ" or c2 == "ط" or c2 == "ظ" then
-- full assimilation
sound_prefix = MU .. c2 .. SH
sufind = 3
else
-- partial or no assimilation
if c2 == "ز" then
check(3, "د")
elseif c2 == "ص" or c2 == "ض" then
check(3, "ط")
else
check(3, T)
end
sound_prefix = MU .. c2 .. SK .. c(3)
sufind = 4
end
end
if c(sufind) == ALIF then
-- hollow, active or passive
check_len(sufind + 1, sufind + 1)
check_weakness({"hollow"}, "allow missing")
vocalized = sound_prefix .. AA .. c(sufind + 1)
elseif len == sufind then
-- active final-weak or active or passive geminate
if is_active then
check_weakness({"geminate", "final-weak", "assimilated+final-weak"})
if weakness == "geminate" then
vocalized = sound_prefix .. A .. c(len) .. SH
else
vocalized = sound_prefix .. A .. c(len) .. IN
end
else
check_weakness({"geminate"}, "allow missing")
vocalized = sound_prefix .. A .. c(len) .. SH
end
else
sound_prefix = sound_prefix .. A .. c(sufind)
handle_possibly_final_weak(sound_prefix, sufind + 1)
end
elseif vform == "IX" then
check_len(4, 4)
vocalized = MU .. c(2) .. SK .. c(3) .. A .. c(4) .. SH
elseif vform == "IVq" then
-- e.g. [[اذلعب]] "to scamper away", مُذْلَعِبّ (active), مُذْلَعَبّ (passive);
-- [[اطمأن]] "to remain quietly; to be certain", مُطْمَئِنّ (active), مُطْمَأَنّ (passive)
check_len(5, 5)
local sound_prefix = MU .. c(2) .. SK .. c(3) .. A .. c(4)
if is_active then
vocalized = sound_prefix .. I .. c(5) .. SH
else
vocalized = sound_prefix .. A .. c(5) .. SH
end
elseif vform == "XI" then
check_len(5, 5)
check(4, ALIF)
vocalized = MU .. c(2) .. SK .. c(3) .. AA .. c(5) .. SH
-- e.g. [[احمار]] "to turn red, to blush", مُحْمَارّ (active)
elseif vform == "XIV" or vform == "XV" then
-- FIXME: Implement. No examples in Wiktionary currently; need to look up in a grammar.
error("Support for verb form " .. vform .. " not implemented yet")
else
error("Don't recognize verb form " .. vform)
end
vocalized = rsub(vocalized, HAMZA .. AA, AMAD)
local reconstructed_headword = lang:makeEntryName(vocalized)
if reconstructed_headword ~= orig_headword then
error(("Internal error: Vocalized participle %s doesn't match original participle %s"):format(
vocalized, orig_headword))
end
return vocalized
end
function export.infer_participle_vocalization_json(frame)
local iparams = {
[1] = {required = true},
[2] = {required = true},
["weakness"] = {},
["passive"] = {type = "boolean"}
}
local iargs = require("Module:parameters").process(frame.args, iparams)
return export.infer_participle_vocalization(iargs[1], iargs[2], iargs.weakness, not iargs.passive)
end
return export