Documentation

Studies.Schwarzer2026

Schwarzer 2026: the law and order of selection-violating coordination #

Schwarzer tests three analyses of a clause coordinated with a noun phrase where only the noun phrase is selected on German, where complements follow the verb in a root clause and precede it in an embedded one. The bottom-up analyses (Sag, Gazdar, Wasow and Weisler; Munn) make the first conjunct prominent, so they check List.take 1; the linear closeness analysis (Bruening and Al Khalaf; Bruening) and the temporal one (Kim and Lu) check the conjunct next to the verb, checkedOnce. They part company before the verb, where only the bottom-up analyses admit the noun phrase first. Experiment 1 shows that German allows the construction, and in Experiment 2 the noun phrase first is preferred before the verb as much as after it, against both closeness analyses.

Main results #

exp1_predictions, exp1_rows: the grammar admits every condition of Experiment 1 but a bare clause after a verb whose fragment entry takes none, and decides the paper's samples.
accounts_diverge_embedded: the analyses differ before the verb and agree after it.
choices_refute_closeness, choices_match_structural: the order preferred before the verb is one the closeness analyses exclude and the bottom-up ones admit.

Implementation notes #

The verbs' c-selection is read off their entries in Fragments/German/Verbs. The results are Data/Experiments/Schwarzer2026; the mixed model and the logistic regression are not formalized, and the preferred order is the one chosen more often.
The bottom-up analyses are given the noun-phrase shell over the clause of the 2020 study; their variant that licenses only the first conjunct makes the same predictions here.

References #

The analyses #

@[reducible, inline]

abbrev Schwarzer2026.rootPosition :

In a German root declarative the verb in second position precedes its complements, the configuration of (17).

Equations

Schwarzer2026.rootPosition = HeadDirection.headInitial

Instances For

@[reducible, inline]

abbrev Schwarzer2026.embeddedPosition :

In an embedded finite clause the verb is clause-final, so the coordination precedes it, the configuration of (16).

Equations

Schwarzer2026.embeddedPosition = HeadDirection.headFinal

Instances For

def Schwarzer2026.Order.phrases :

Order → List PhraseStructure.Cat

o.phrases lists the categories of the conjuncts of a coordination in the order o.

Equations

Instances For

theorem Schwarzer2026.structural_admits_iff (o : Order) :

BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (List.take 1) {PhraseStructure.Cat.N}) o.phrases ↔ o = Order.dpFirst

The bottom-up analyses admit the selected noun phrase first, after a verb that does not select a clause, whatever the verb's position, (10b).

theorem Schwarzer2026.closeness_embedded_iff (o : Order) :

BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (BrueningAlKhalaf2020.checkedOnce embeddedPosition) {PhraseStructure.Cat.N}) o.phrases ↔ o = Order.cpFirst

The closeness analyses admit the clause first before the verb, (10a).

theorem Schwarzer2026.accounts_diverge_embedded :

(∃ (o : Order), ¬(BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (List.take 1) {PhraseStructure.Cat.N}) o.phrases ↔ BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (BrueningAlKhalaf2020.checkedOnce embeddedPosition) {PhraseStructure.Cat.N}) o.phrases)) ∧ BrueningAlKhalaf2020.checkedOnce rootPosition = List.take 1

The analyses differ before the verb only, which is what makes German the test case, since after it the closeness analyses check the first conjunct too.

Experiment 1 #

def Schwarzer2026.selects (v : German.Verb) :

Finset PhraseStructure.Cat

selects v is the set of categories the verb v c-selects for its object, read off its entry, which has a noun phrase if a frame takes one and a clause if a frame takes a dass-clause.

Equations

Schwarzer2026.selects v = (if v.TakesNominal then {PhraseStructure.Cat.N} else ∅) ∪ if v.TakesClausal then {PhraseStructure.Cat.C} else ∅

Instances For

def Schwarzer2026.Selection.verbs :

Selection → List German.Verb

s.verbs lists the four predicates of Experiment 1 that do or do not select a clause.

Equations

Instances For

theorem Schwarzer2026.selects_verbs (s : Selection) (v : German.Verb) :

v ∈ s.verbs → selects v = if s = Selection.yes then {PhraseStructure.Cat.N, PhraseStructure.Cat.C} else {PhraseStructure.Cat.N}

Each predicate's entry takes a noun phrase, and a clause exactly when the paper classes it as selecting one (p. 7).

def Schwarzer2026.Complement.phrases :

Complement → List PhraseStructure.Cat

c.phrases lists the categories of the conjuncts of a complement of Experiment 1, a bare clause or a coordination with the noun phrase first.

Equations

Instances For

theorem Schwarzer2026.exp1_predictions (s : Selection) (c : Complement) (v : German.Verb) :

v ∈ s.verbs → (BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (BrueningAlKhalaf2020.checkedOnce rootPosition) (selects v)) c.phrases ↔ c = Complement.coord ∨ s = Selection.yes)

The grammar admits every condition of Experiment 1, after the verb, except a bare clause after a verb that does not select one, where the coordination is admitted through the null N. All the analyses agree, since the verb precedes. Coordinations are accordingly rated above bare clauses after those verbs, and lose less from the absence of selection (table (14), p. 9).

def Schwarzer2026.entryOf (form : String) :

Option German.Verb

entryOf form is the fragment entry of the verb a row names.

Equations

Schwarzer2026.entryOf form = List.find? (fun (x : German.Verb) => decide (x.form = form)) German.Verbs.allVerbs

Instances For

def Schwarzer2026.complementOf? (e : Datum) :

Option Complement

complementOf? e is the complement of a row of Experiment 1.

Equations

Schwarzer2026.complementOf? e = e.parse? "complement" [("dass", Schwarzer2026.Complement.dass), ("coord", Schwarzer2026.Complement.coord)]

Instances For

theorem Schwarzer2026.exp1_rows (e : Datum) :

e ∈ Examples.all → ∃ v ∈ (e.feature? "verb").bind entryOf, ∃ c ∈ complementOf? e, BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (BrueningAlKhalaf2020.checkedOnce rootPosition) (selects v)) c.phrases ↔ e.judgment = Judgment.acceptable

The grammar, with each verb's c-selection read off its entry, decides the paper's samples of Experiment 1, (11) and (12).

theorem Schwarzer2026.rows_selectsCP (e : Datum) :

e ∈ Examples.all → ∀ v ∈ (e.feature? "verb").bind entryOf, e.feature? "selectsCP" = some "yes" ↔ v.TakesClausal

Each row's selection value is that of its predicate's entry.

Experiment 2 #

def Schwarzer2026.Position.direction :

Position → HeadDirection

The head direction of a position of Experiment 2.

Equations

Instances For

def Schwarzer2026.preferred (p : Position) :

The order chosen more often in a position of Experiment 2.

Equations

One or more equations did not get rendered due to their size.

Instances For

theorem Schwarzer2026.preferred_eq (p : Position) :

preferred p = Order.dpFirst

The noun phrase first is preferred in both positions, chosen 23 times of 30 (p. 13).

theorem Schwarzer2026.choices_refute_closeness :

¬BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (BrueningAlKhalaf2020.checkedOnce Position.preverbal.direction) {PhraseStructure.Cat.N}) (preferred Position.preverbal).phrases

Experiment 2 refutes the linear and temporal closeness analyses, since the order preferred before the verb is one they exclude.

theorem Schwarzer2026.choices_match_structural (p : Position) (o : Order) :

BrueningAlKhalaf2020.Admits (BrueningAlKhalaf2020.Licensed (List.take 1) {PhraseStructure.Cat.N}) o.phrases ↔ o = preferred p

The bottom-up analyses admit exactly the preferred order in both positions, which the squib takes to support them only indirectly (p. 15).