Documentation

Studies.MaldonadoCulbertson2022

Maldonado and Culbertson (2022): The learnability of person systems #

[MC22] test in artificial language learning whether the constraints proposed for person partitions are hard or soft. Without a clusivity contrast languages assimilate the inclusive into the first person, and neither the second-inclusive nor the third-inclusive tripartition is attested. For [Har16a] both are underivable, while for Harley and Ritter's features, which include an addressee feature, the second-inclusive one is a natural class and derivable though marked, and only the third-inclusive one is underivable (pp. 313–314). In Experiment 2 the third-inclusive system was learned significantly worse than both others, but the second-inclusive one only marginally worse than the first-inclusive one (p. 318, the coefficients of contrasts): the inclusive can form a natural class with the second person, and the cost of the third-inclusive system is a bias against homophony outside natural classes (pp. 318–319). That cost falls on the one condition no feature set generates.

Main results #

Implementation notes #

The plural categories are their participant sets, first exclusive {speaker}, inclusive {speaker, addressee}, second {addressee} and third ∅, and the features those of Harbour2016.ThreeFeature, [±hearer] standing for Harley and Ritter's addressee feature. The model coefficients are kept as printed data; the study does not restate the paper's reading of them. Experiments 1 and 3 are not formalized.

References #

The conditions as partitions (Table 7, p. 315) #

source
def MaldonadoCulbertson2022.Condition.host :
Condition → Finset Discourse.Role

The participant set of the category whose form the inclusive shares.

Equations
Instances For
    source
    def MaldonadoCulbertson2022.Condition.graph (c : Condition) :
    Finset (Finset Discourse.Role × Finset Discourse.Role)

    The partition a condition teaches: every category its own form, except that the inclusive shares the form of the condition's host category.

    Equations
    Instances For
      source
      theorem MaldonadoCulbertson2022.graph_firstInclusive :
      Condition.firstInclusive.graph = Harbour2016.graph {Harbour2016.ThreeFeature.participant, Harbour2016.ThreeFeature.author}

      The first-inclusive condition is the tripartition, generated by [±participant] and [±author].

      source
      theorem MaldonadoCulbertson2022.graph_secondInclusive :
      Condition.secondInclusive.graph = Harbour2016.graph {Harbour2016.ThreeFeature.participant, Harbour2016.ThreeFeature.hearer}

      The second-inclusive condition is generated by [±participant] and [±hearer].

      source
      theorem MaldonadoCulbertson2022.generated_iff (c : Condition) :
      (∃ (F : Finset Harbour2016.ThreeFeature), Harbour2016.graph F = c.graph) ↔ c ≠ Condition.thirdInclusive

      A condition's partition is generated by some feature set iff it is not the third-inclusive one, whose inclusive and third person share no feature value.

      source
      theorem MaldonadoCulbertson2022.admissible_iff (c : Condition) :
      (∃ (F : Finset Harbour2016.ThreeFeature), Harbour2016.Admissible F ∧ Harbour2016.graph F = c.graph) ↔ c = Condition.firstInclusive

      Of the three, only the first-inclusive partition is among Harbour's five, those his bivalent geometry (9) generates.