Documentation

Studies.Anttila1997

Anttila 1997: deriving variation from grammar #

Finnish genitive plurals come in a strong variant (heavy penult, naa.pu.rei.den) and a weak one (light penult, naa.pu.ri.en). Stems of one or two syllables choose categorically, longer CV-final stems vary, and the variation is biased by the stem-final vowel and by the weight of the antepenult. One partially ordered grammar covers all of it. Its constraints are the harmonic alignments of stress, weight, and sonority together with bans on adjacent syllables of equal stress or weight; Finnish stratifies them by promoting No Clash and the two stress–weight mismatches above the rest. Every total ranking consistent with the grammar picks a winner, and a variant's probability is the share of rankings under which it wins. The stress constraints decide short stems and stems with a heavy antepenult outright; where they tie, secondary stress being optional, the two internally unranked intermediate sets resolve the competition at rates that the corpus frequencies track.

Item and page numbers follow the ROA-63 manuscript.

Main definitions #

References #

Syllable prominence (25) #

Nuclear sonority, ordered i < o < a.

Instances For
    @[instance_reducible]
    Equations
    def Anttila1997.instReprSonority.repr :
    Sonority → ℕ → Std.Format
    Equations
    Instances For

      Position on the sonority hierarchy.

      Equations
      Instances For

        Syllable weight.

        Instances For
          @[instance_reducible]
          Equations
          def Anttila1997.instReprWeight.repr :
          Weight → ℕ → Std.Format
          Equations
          Instances For
            @[instance_reducible]
            Equations

            Whether a weight is the prominent value of its scale, the one stress aligns with.

            Equations
            Instances For

              A syllable of a candidate: its stress, and its weight and nuclear sonority where the tableau specifies them (X leaves both open).

              Instances For
                def Anttila1997.instDecidableEqSyllable.decEq (x✝ x✝¹ : Syllable) :
                Decidable (x✝ = x✝¹)
                Equations
                • One or more equations did not get rendered due to their size.
                Instances For
                  def Anttila1997.instReprSyllable.repr :
                  Syllable → ℕ → Std.Format
                  Equations
                  • One or more equations did not get rendered due to their size.
                  Instances For

                    The constraints of (50): a starred stress–weight, weight–sonority, or stress–sonority combination within a syllable (28), or a starred adjacency of equal stress or weight (29).

                    Instances For
                      def Anttila1997.instDecidableEqConstraint.decEq (x✝ x✝¹ : Constraint) :
                      Decidable (x✝ = x✝¹)
                      Equations
                      Instances For
                        Equations
                        • One or more equations did not get rendered due to their size.
                        Instances For
                          @[instance_reducible]
                          Equations
                          • One or more equations did not get rendered due to their size.

                          Violations on a syllable string: the syllables showing the starred combination, or the adjacent pairs sharing the starred value.

                          Equations
                          Instances For

                            The grammar for Finnish (28), (50) #

                            def Anttila1997.alignedAbove (prominent : Bool) (s s' : Sonority) :

                            Harmonic alignment (27) of sonority with a binary scale: on the scale's prominent value the constraint against the less sonorous nucleus ranks higher; on the other value, the more.

                            Equations
                            Instances For
                              @[instance_reducible]
                              instance Anttila1997.instDecidableAlignedAbove (p : Bool) (s s' : Sonority) :
                              Decidable (alignedAbove p s s')
                              Equations

                              The universal rankings (28): the stress–weight mismatches outrank the matches, and sonority aligns with weight and with stress.

                              Equations
                              Instances For
                                @[instance_reducible]
                                Equations
                                • One or more equations did not get rendered due to their size.
                                theorem Anttila1997.alignedAbove_trans {p : Bool} {s₁ s₂ s₃ : Sonority} (h : alignedAbove p s₁ s₂) (h' : alignedAbove p s₂ s₃) :
                                alignedAbove p s₁ s₃
                                theorem Anttila1997.alignedAbove_asymm {p : Bool} {s s' : Sonority} (h : alignedAbove p s s') (h' : alignedAbove p s' s) :
                                False
                                theorem Anttila1997.Universal.trans {a b c : Constraint} (hab : Universal a b) (hbc : Universal b c) :
                                theorem Anttila1997.Universal.asymm {a b : Constraint} (hab : Universal a b) (hba : Universal b a) :
                                False
                                theorem Anttila1997.set_mono (a b : Constraint) :
                                Universal a b → a.set ≤ b.set

                                Finnish only adds rankings to (28): no universal ranking is reversed across strata.

                                The roster in the column order of (50).

                                Equations
                                • One or more equations did not get rendered due to their size.
                                Instances For
                                  def Anttila1997.stratumOf (c : Fin 20) :
                                  Fin 5

                                  Stratum of a roster position.

                                  Equations
                                  Instances For
                                    def Anttila1997.universalOn (a b : Fin 20) :

                                    The universal rankings on the roster, reflexively.

                                    Equations
                                    Instances For
                                      @[instance_reducible]
                                      Equations
                                      theorem Anttila1997.set_eq_of_universal (a b : Constraint) :
                                      Universal a b → a.set = b.set → a.set = 4

                                      Only Set 5 ranks constraints internally.

                                      theorem Anttila1997.stratum_triv {k : Fin 5} (hk : k ≠ 4) (a b : Fin 20) :
                                      stratumOf a = k → stratumOf b = k → universalOn a b → a = b

                                      Below Set 5 the inner order is trivial.

                                      def Anttila1997.finnishGrammar :
                                      Fin 20 → Fin 20 → Prop

                                      The grammar for Finnish, final version (50).

                                      Equations
                                      Instances For

                                        Categorical predictions (§5.1) #

                                        Tableau syllables: stressed or unstressed, of unspecified, heavy, or light weight.

                                        Equations
                                        Instances For
                                          Equations
                                          Instances For
                                            Equations
                                            Instances For
                                              Equations
                                              Instances For
                                                Equations
                                                Instances For
                                                  Equations
                                                  Instances For

                                                    The stem shapes of the categorical tableaux, by their examples: (33) maa, (35) kala, (37) maailma, (39) ministeri, (40) margariini, (42) Aleksanteri, (43) koordinaatisto, (44) italiaano.

                                                    Instances For
                                                      @[instance_reducible]
                                                      Equations
                                                      def Anttila1997.instReprShape.repr :
                                                      Shape → ℕ → Std.Format
                                                      Equations
                                                      Instances For
                                                        @[instance_reducible]
                                                        Equations
                                                        def Anttila1997.Shape.cands :
                                                        Shape → Finset (List Syllable)

                                                        The candidates of the tableaux (32), (34), (36), (38), (41): the weak variant, the strong one, and after a heavy antepenult the strong one with its stress shifted.

                                                        Equations
                                                        Instances For

                                                          The roster's constraints on the candidates of a shape's tableau.

                                                          Equations
                                                          Instances For

                                                            The total rankings consistent with the Finnish grammar.

                                                            Equations
                                                            Instances For

                                                              A tableau candidate that every consistent ranking picks.

                                                              Equations
                                                              Instances For

                                                                Monosyllabic stems take the strong variant (32): mai.den, not ma.jen (33).

                                                                Disyllabic CV-final stems take the weak variant (34): ka.lo.jen, not ka.loi.den (35).

                                                                A heavy antepenult in a 4-syllabic stem is stressed and forces a light penult (38): mar.ga.rii.ni.en only (40).

                                                                A light ante-antepenult before a heavy antepenult likewise (41): i.ta.li.aa.no.jen only (44).

                                                                theorem Anttila1997.stress_tie (s : Shape) :
                                                                s ∈ [Shape.maailma, Shape.ministeri, Shape.aleksanteri, Shape.koordinaatisto] → ∀ o ∈ s.cands, ∀ o' ∈ s.cands, ∀ (c : Fin 20), stratumOf c ≤ 1 → tableauCon c (s, o) = tableauCon c (s, o')

                                                                After a light antepenult, or a heavy one following a heavy ante-antepenult, the stress constraints tie (36), (38), (41): secondary stress is optional, and both variants survive Sets 1 and 2.

                                                                Quantitative predictions (§5.3) #

                                                                The two genitive-plural variants (1).

                                                                Instances For
                                                                  @[instance_reducible]
                                                                  Equations
                                                                  @[instance_reducible]
                                                                  Equations
                                                                  def Anttila1997.instReprVariant.repr :
                                                                  Variant → ℕ → Std.Format
                                                                  Equations
                                                                  Instances For
                                                                    theorem Anttila1997.Variant.univ_eq_pair (v : Variant) :
                                                                    Finset.univ = {v, v.other}
                                                                    @[reducible, inline]

                                                                    A motif of (52): the antepenult's weight and the stem-final vowel.

                                                                    Equations
                                                                    Instances For

                                                                      The penult of a variant with nucleus s: stressed and heavy, or unstressed and light.

                                                                      Equations
                                                                      Instances For

                                                                        A trisyllabic candidate of (52): initial stress, an unstressed antepenult of the motif's weight, the variant's penult, and an unstressed heavy final.

                                                                        Equations
                                                                        Instances For

                                                                          The roster's constraints on the variants of a motif.

                                                                          Equations
                                                                          Instances For
                                                                            def Anttila1997.wins (m : Motif) (v : Variant) :
                                                                            Finset (OptimalityTheory.Ranking (Fin 20) 20)

                                                                            The rankings under which variant v wins motif m.

                                                                            Equations
                                                                            Instances For
                                                                              theorem Anttila1997.card_wins_mul (m : Motif) (v : Variant) (k : Fin 5) (h_triv : ∀ (a b : Fin 20), stratumOf a = k → stratumOf b = k → universalOn a b → a = b) (h_tie : ∀ (c : Fin 20), stratumOf c < k → con c (m, v) = con c (m, v.other)) (h_dec : {x ∈ OptimalityTheory.active con m v v.other | stratumOf x = k}.Nonempty) :
                                                                              (wins m v).card * {x ∈ OptimalityTheory.active con m v v.other | stratumOf x = k}.card = rankings.card * (OptimalityTheory.favoring con m v v.other ∩ {x ∈ OptimalityTheory.active con m v v.other | stratumOf x = k}).card

                                                                              The deciding-stratum count: once the strata above k tie, the rankings won by v stand to all rankings as stratum k's constraints favoring v stand to its active ones.

                                                                              theorem Anttila1997.card_wins_of_counts (m : Motif) (v : Variant) (k : Fin 5) (n t : ℕ) (h_triv : ∀ (a b : Fin 20), stratumOf a = k → stratumOf b = k → universalOn a b → a = b) (h_tie : ∀ (c : Fin 20), stratumOf c < k → con c (m, v) = con c (m, v.other)) (h_dec : {x ∈ OptimalityTheory.active con m v v.other | stratumOf x = k}.Nonempty) (hn : (OptimalityTheory.favoring con m v v.other ∩ {x ∈ OptimalityTheory.active con m v v.other | stratumOf x = k}).card = n) (ht : {x ∈ OptimalityTheory.active con m v v.other | stratumOf x = k}.card = t) :
                                                                              (wins m v).card * t = rankings.card * n

                                                                              card_wins_mul with the two counts evaluated.

                                                                              theorem Anttila1997.card_wins_strong_of {m : Motif} {n t : ℕ} (hp : predictedStrong m = (n, t)) (h : (wins m Variant.strong).card * t = rankings.card * n) :

                                                                              A motif's strong share, from the counts behind the paper's fraction.

                                                                              The shares of (52): decided in Set 3 after a light antepenult and for h.tii, in Set 4 for h.taa and h.too; the strata below never matter.

                                                                              theorem Anttila1997.card_wins_add (m : Motif) :
                                                                              (wins m Variant.strong).card + (wins m Variant.weak).card = rankings.card

                                                                              Every consistent ranking picks one of the two variants.

                                                                              The weak variant's share is the complement of the strong one's.

                                                                              theorem Anttila1997.card_wins (m : Motif) (v : Variant) :
                                                                              (wins m v).card * (predicted m v).2 = rankings.card * (predicted m v).1

                                                                              Each variant wins the share of rankings (52) predicts.

                                                                              The paper's data #

                                                                              def Anttila1997.Weight.ofChar? :
                                                                              Char → Option Weight

                                                                              Antepenult weight from a motif label.

                                                                              Equations
                                                                              Instances For
                                                                                def Anttila1997.Motif.parse? (s : String) :
                                                                                Option (Motif × Variant)

                                                                                A motif label of (52)–(53): antepenult weight, t, and the stem-final vowel, doubled for the strong variant.

                                                                                Equations
                                                                                • One or more equations did not get rendered due to their size.
                                                                                Instances For
                                                                                  def Anttila1997.digits (l : List Char) :
                                                                                  ℕ

                                                                                  The value of a digit string.

                                                                                  Equations
                                                                                  • Anttila1997.digits l = List.foldl (fun (n : ℕ) (c : Char) => 10 * n + (c.toNat - '0'.toNat)) 0 l
                                                                                  Instances For
                                                                                    def Anttila1997.percent? (s : String) :
                                                                                    Option ℕ

                                                                                    A printed whole percentage.

                                                                                    Equations
                                                                                    Instances For
                                                                                      def Anttila1997.tenths? (s : String) :
                                                                                      Option ℕ

                                                                                      A printed percentage in tenths: 99.4 as 994, 100 as 1000.

                                                                                      Equations
                                                                                      • One or more equations did not get rendered due to their size.
                                                                                      Instances For
                                                                                        theorem Anttila1997.rows_predicted (r : Datum) :
                                                                                        r ∈ Examples.all → r.feature? "syllables" = some "3" → ∀ mv ∈ (r.feature? "motif" >>= Motif.parse?).toList, ∀ p ∈ (r.feature? "pred_pct" >>= percent?).toList, 2 * p * (predicted mv.1 mv.2).2 ≤ 200 * (predicted mv.1 mv.2).1 + (predicted mv.1 mv.2).2 ∧ 200 * (predicted mv.1 mv.2).1 ≤ (2 * p + 1) * (predicted mv.1 mv.2).2

                                                                                        Table (53): the paper's predicted percentage is the share of (52) to the nearest percent.

                                                                                        theorem Anttila1997.rows_frequency (r : Datum) :
                                                                                        r ∈ Examples.all → r.feature? "syllables" = some "3" → ∀ mv ∈ (r.feature? "motif" >>= Motif.parse?).toList, ∀ q ∈ (r.feature? "obs_pct" >>= tenths?).toList, ((predicted mv.1 mv.2).2 < 2 * (predicted mv.1 mv.2).1 → 500 < q) ∧ (2 * (predicted mv.1 mv.2).1 < (predicted mv.1 mv.2).2 → q < 500)

                                                                                        Tables (48) and (53): the variant the grammar favors is the more frequent one.