Documentation

Linglib.Studies.Levin1993

Diathesis Alternation Bridge [Lev93] #

Connects the alternation judgment rows in Data/Examples/Levin1993.json to the LevinClass.participatesIn prediction function, the Fragment verb entries, and [Dow91]'s proto-role account of the three verb classes.

The verbs break, cut, hit, touch form Levin's §0.4 diagnostic quadruple (pp. 5–10): each participates in a distinct subset of the causative/inchoative, middle, conative, and body-part possessor ascension alternations.

VerbClassCIMidConBPPA
break45.1
cut21.1
hit18.1
touch20

Main declarations #

Map a row's levin_class section string to the curated LevinClass enum. Subclasses collapse to their enum representative (51.3.2 run verbs → mannerOfMotion; 47.5.1 swarm verbs → the §47 existence family's exist; 48.1.1 → appear; 41.1.1 dress verbs → dress; 36.3 → socialInteraction; 54.1 register verbs → measure). Classes outside the enum's grain (40.2 nonverbal expression, 40.3.2 wink, 40.4 snooze, 37.5 talk) map to none.

Equations
Instances For

    Map a row's alternation tag to the curated enum.

    Equations
    Instances For

      Levin class recorded on a row.

      Equations
      Instances For

        Alternation recorded on a row.

        Equations
        Instances For

          Categorical participation recorded on a row; none for marginal judgments (the participatesIn profile is Boolean).

          Equations
          Instances For

            A row agrees with LevinClass.participatesIn (vacuously, if its class or alternation is unrepresentable or its judgment is marginal).

            Equations
            Instances For

              Every categorical row with a representable class and alternation matches the participatesIn prediction. In particular the CI rule blocks cut via instrument specification (Levin pp. 9–10: an inherently specified instrument requires an agent, blocking the agentless inchoative).

              Every row's alternation tag is one of the curated alternations.

              theorem Levin1993.unrepresentable_classes :
              List.map (fun (x : Data.Examples.LinguisticExample) => x.id) (List.filter (fun (e : Data.Examples.LinguisticExample) => (classOf e).isNone) Examples.all) = ["levin1993_ubpo_wave", "levin1993_co_laugh", "levin1993_pp_sleep", "levin1993_pp_talk"]

              Exactly four rows carry Levin classes outside the curated enum's grain: wave (40.3.2 wink), laugh (40.2 nonverbal expression), sleep (40.4 snooze), talk (37.5). These are excluded from participation_matches_prediction by classOf.

              def Levin1993.quadruplePattern (rows : List Data.Examples.LinguisticExample) :
              List (Option Bool)

              Participation pattern of a verb across the four §0.4 alternations (CI, middle, conative, BPPA).

              Equations
              Instances For

                Each verb of the quadruple shows a pairwise distinct pattern across the four alternations — Levin's §0.4 table has no repeated rows, so the four verbs instantiate four distinct verb classes.

                Each Fragment verb entry's levinClass matches the class recorded on its alternation rows.

                [Dow91] §9.3 derives the alternation behavior of the spray/load, break, and hit classes from the distribution of the change-of-state entailment across non-subject arguments: symmetric CoS permits alternation, asymmetric CoS fixes the CoS argument as direct object. Levin's judgment rows confirm each prediction.

                Spray/load: CoS is symmetric across theme and location, predicting the locative alternation — attested for both spray and load.

                Break: CoS is asymmetric (direct object vs. instrument), fixing the CoS argument as direct object — break lacks the locative alternation.

                Hit: both arguments symmetrically lack CoS — the conative is attested while the CoS-sensitive causative/inchoative and middle are blocked.