%unimsg 0 lang/ar/broken-plurals/v0

-- A fragment of the Arabic broken plural system, modelled as a GENERATOR
-- plus RESIDUALS rather than as an enumeration of forms.
--
-- The distinction matters: a materialised paradigm is a photograph of a
-- theory at one moment, whereas rules plus exceptions ARE the theory, and a
-- correction propagates instead of needing thousands of edits.
--
-- Linguistic content is illustrative and needs review by a morphologist.
-- The shape of the modelling is the point, not the data.

uses { ling/ipa #sha256:6c82f1a9 }

-- Patterns are templates: a formula from a consonantal root to a surface
-- form. Inert data — nothing evaluates them at read time.
patterns {
  fiʿāl {
    arab     :فِعَال
    skeleton [ :C1, :I, :C2, :Ā, :C3 ]
    class    :singular
  }
  fuʿul {
    arab     :فُعُل
    skeleton [ :C1, :U, :C2, :U, :C3 ]
    class    :plural
    kind     :broken
    number   :multitude
    triptote true
  }
  ʾafʿāl {
    arab     :أَفْعَال
    skeleton [ :ʾ, :A, :C1, "∅", :C2, :Ā, :C3 ]  -- ∅ is a math symbol, so quoted
    class    :plural
    kind     :broken
    number   :paucity
    triptote true
    note     "paucity sense weakly observed in modern usage"
  }
  mafāʿil {
    arab     :مَفَاعِل
    class    :plural
    kind     :broken
    stem     :quadriconsonantal
    triptote false
  }
}

-- Rules are general. A rule and an exception are the same kind of object:
-- they differ only in how many lexemes they cover.
rules [
  | id        singular  plural   conditions              confidence
  | :r-001    -> @patterns.fiʿāl -> @patterns.fuʿul  [ :sound-root ]       0.82
  | :r-002    -> @faʿal -> @patterns.ʾafʿāl [ :sound-root ]       0.61
  | :r-003    -> @fāʿil -> @fuʿʿāl [ :human, :agentive ] 0.74
  ]

-- Dimensions are declared value spaces with their own provenance, not bare
-- strings. Without this, "formal" and "Formal" drift apart across thousands
-- of entries and nothing can be computed.
dimensions {
  register {
    values ordered [ :colloquial, :neutral, :formal, :literary ]
    note   "competing 3-way and 5-way schemes exist; see mappings"
  }
}

lexemes {
  -- The regular case: the rule generates the plural, nothing is stipulated.
  kitāb {
    root     [ :k, :t, :b ]
    arab     :كِتَاب
    gloss    en "book"
    template -> @patterns.fiʿāl
    plural   derived-by -> @r-001
    agree    :fem.sg
    attest   { src "Wright II", conf 0.99 }
  }

  -- A residual: two plurals differing by register, which no rule derives.
  ṭālib {
    root     [ :ṭ, :l, :b ]
    arab     :طَالِب
    gloss    en "student"
    template -> @fāʿil
    -- NOTE: docs/08 and docs/09 write this as `plurals discriminated-by
    -- [ :register ] [ … ]`, which the grammar cannot express — annotations
    -- are bare words and take no arguments. The nested form below says the
    -- same thing and is arguably better modelling.
    plurals {
      discriminated-by [ :register ]
      rows [
        | form      via         register  freq
        | :طُلَّاب     -> @fuʿʿāl  :neutral  0.71
        | :طَلَبَة      -> @faʿala  :formal   0.28
  ]
    }
  }

  -- The plural selects the SENSE, not the other way round. Modelling this
  -- forced plurals onto senses rather than onto the lexeme.
  bayt {
    root  [ :b, :y, :t ]
    arab  :بَيْت
    weak  :hollow-y
    senses {
      dwelling { gloss en "house", plural :بُيُوت,  via -> @fuʿūl }
      verse    { gloss en "verse of poetry", plural :أَبْيَات, via -> @patterns.ʾafʿāl }
    }
  }
}

-- Negative facts. Rules overgenerate, so a positive-only knowledge base
-- silently becomes an overgenerating one.
blocked [
  | lexeme        form       reason
  | -> @lexemes.kitāb     :كِتْبَان    "derivable by r-002 but unattested"
  | -> @lexemes.ṭālib     :طَوَالِب    "blocked by the existing plurals above"
  ]

-- Completeness must be stated. Open-world is usually right for language
-- data, but leaving it implicit makes "the rule is wrong" indistinguishable
-- from "we have not looked yet".
coverage {
  world     :open
  lexemes   3
  estimated 15000
  note      "fragment only; see docs/09 for sizing"
}