Major (2024): Re-analyzing 'say' complementation: Implications for case theory and beyond

Uyghur dep follows a finite clause where English has that, and Major argues that it is what its morphology shows, the converb in -(I)p of the verb de- 'say'. A dep clause then has the outside of a converb clause and the inside of a clause of saying. Outside, it adjoins to the matrix VP or TP and is never an argument: no verb takes it, it is no subject, a factive verb does not presuppose it, and the diagnostics of the two heights sort it as they sort any converb clause. Bochnak and Hanink's modifier clauses of Washo are the same kind of clause. Inside, de- keeps its obligatory complement, and its v values the subject of that complement accusative as the v of a transitive verb values an object.

The accusative subject bears on case theory. Baker and Vinokurova take Sakha dien, the same converb of 'say', for a complementizer, so that an accusative subject beside an unaccusative matrix verb has no v above it and calls for a dependent case rule. With 'say' in the clause there is a v, and the dependent rule and Agree then value the same cases, while with a complementizer each goes wrong on the Uyghur examples.

Main results

• subcategorization: a verb needs the complement its frames give it, finite or under the converb suffix alike, so de- keeps its obligatory complement inside a dep clause.
• not_takes_ip, factive_denial: no verb takes a converb clause, and denying a clause is contradictory exactly when a factive verb takes it.
• case_below_v: under either modality an object, or the subject below 'say', is accusative exactly where a θ-assigning v bears the feature that attracts it.
• complementizer_agree, complementizer_dependent: with dep a complementizer, Agree leaves the raised subject caseless beside an unaccusative matrix verb, and the dependent rule values it accusative beside an inanimate matrix subject, where it cannot be.

Implementation notes

• The feature [+ACC/SPEC] of v is one Boolean, which makes v a probe and shifts its goal to the edge of the verb phrase. Case is assigned by Baker and Vinokurova's two grammars, theirs with the dependent accusative and the one with Agree alone.
• A dep clause is its own domain of case, the subject of 'say' above the subject of its complement; the complementizer analysis instead puts the raised subject among the matrix NPs.
• The voice of 'say' follows a row's sayer: an agent's saying is an activity, and a saying whose source is inanimate or missing is a state with no external argument.
• A row's height is the paper's classification of its converb clause, and a diagnostic is read as scope under a matrix head, negation for a negative concord item and v for the pro-form of a manner.

TODO

• Raising against prolepsis, by default agreement on the embedded verb, and the complementary distribution of accusative subjects and indexical shift.
• The nominative of a subject promoted out of a dep clause under a passive matrix verb.

References

• [Maj24]
• [BV10]
• [BH21]

def Major2024.verb? (key : String) (r : Datum) : Option Verb

verb? key r is the fragment entry that the feature key of the row r names.

Equations

• Major2024.verb? key r = (r.feature? key).bind fun (x : String) => Verb.find? Uyghur.verbs x

def Major2024.linker? (r : Datum) : Option Complementizer

The linker of a row's embedded clause is the converb suffix on a dep clause, as on any converb clause, and -lik on a participial one.

Equations

• Major2024.linker? r = r.parse? "clause" [("dep", Uyghur.ip), ("converb", Uyghur.ip), ("participial", Uyghur.lik)]

'Say' inside the converb clause

theorem Major2024.subcategorization (r : Datum) : r ∈ Examples.all → ∀ fr ∈ r.parse? "complement" [("none", ArgumentFrame.intransitive), ("dp", ArgumentFrame.np), ("participial", ArgumentFrame.typedBy Uyghur.lik)], ∃ v ∈ verb? "verb" r, r.judgment = Judgment.acceptable ↔ ∃ g ∈ v.valency.frames, fr ≤ g

A verb is acceptable with or without a complement exactly when it has a frame for it, whether it is finite or under the converb suffix: de- needs its complement in a dep clause as oyla- 'think' does in its converb clause, and warqira- 'scream' takes none.

No argument

theorem Major2024.not_takes_ip (v : Verb) : ¬v.Takes Uyghur.ip

No verb takes a converb clause.

theorem Major2024.dep_unselected : ∃ r ∈ Examples.all, r.judgment = Judgment.acceptable ∧ r.feature? "clause" = some "dep" ∧ ∃ v ∈ verb? "matrixVerb" r, ¬v.TakesClausal

A dep clause is acceptable with a verb that takes no clause at all.

theorem Major2024.clausal_subject (r : Datum) : r ∈ Examples.all → r.feature? "diagnostic" = some "subject" → ∃ z ∈ linker? r, r.judgment = Judgment.acceptable ↔ ∃ (p : ArgumentFrame.Position), p.Takes z

A clause is acceptable as a subject exactly when some argument position takes its linker, as one takes a participial clause and none a dep clause.

theorem Major2024.factive_denial (r : Datum) : r ∈ Examples.all → r.feature? "diagnostic" = some "denial" → ∃ v ∈ verb? "matrixVerb" r, ∃ z ∈ linker? r, r.judgment = Judgment.acceptable ↔ ¬(v.IsFactive ∧ v.Takes z)

Denying the content of a clause is contradictory exactly when a factive verb takes the clause: under bil- 'know' a participial clause is presupposed and a dep clause is not.

theorem Major2024.forget_that (r : Datum) : r ∈ Examples.all → ∀ j ∈ List.lookup "forgot that" r.readings, ∃ v ∈ verb? "matrixVerb" r, ∃ z ∈ linker? r, j = Judgment.acceptable ↔ v.Takes z

unut- 'forget' is read 'forget that' exactly with a clause it takes.

theorem Major2024.washo_parallel : (∀ (v : Verb), ¬v.Takes Washo.aq) ∧ (∃ (p : ArgumentFrame.Position), p.Takes Washo.ge) ∧ ∃ (p : ArgumentFrame.Position), p.Takes Uyghur.lik

The dependent mood of Washo, the linker of a modifier clause, is likewise a linker no verb takes, and the nominalizer of either language one that some argument position takes.

Two heights

theorem Major2024.adjoined_below (h : Minimalist.Cat) : h ∈ [Minimalist.Cat.v, Minimalist.Cat.Asp, Minimalist.Cat.Neg, Minimalist.Cat.T] → Minimalist.Cat.V.fValue < h.fValue ∧ ¬Minimalist.Cat.T.fValue < h.fValue

A clause adjoined to VP is below v, aspect, negation and tense, and one adjoined to TP is below none of them.

theorem Major2024.height_diagnostics (r : Datum) : r ∈ Examples.all → ∀ s ∈ r.parse? "height" [("VP", Minimalist.Cat.V), ("TP", Minimalist.Cat.T)], ∃ h ∈ r.parse? "diagnostic" [("nci", Minimalist.Cat.Neg), ("proform", Minimalist.Cat.v)], r.judgment = Judgment.acceptable ↔ s.fValue < h.fValue

A converb clause passes a diagnostic exactly when it adjoins below the head the diagnostic turns on, and a dep clause does as any other: negation licenses a negative concord item in its clause, and the pro-form shundaq ~ undaq 'like this' stands in the verb phrase of v.

Accusative from v

def Major2024.vProbe (active : Bool) : List (Minimalist.Cat × Minimalist.Cat)

The verb phrase has v as a probe when v bears [+ACC/SPEC].

Equations

• Major2024.vProbe active = if active = true then [(Minimalist.Cat.v, Minimalist.Cat.v)] else []

def Major2024.vP (g : Minimalist.CaseAssigners) (tensed active : Bool) : List (Option Case)

vP g tensed active is the cases g gives a subject and the NP below v, which [+ACC/SPEC] attracts to the edge of the verb phrase, under finite T or, as in a converb clause, under none.

Equations

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

theorem Major2024.vP_eq (tensed active : Bool) (g : Minimalist.CaseAssigners) : g ∈ [BakerVinokurova2010.grammar, BakerVinokurova2010.pureChomsky] → vP g tensed active = [if tensed = true then some Case.nom else none, if active = true then some Case.acc else none]

Whether case is valued by the dependent rule or by v, the lower NP is accusative exactly when v bears the feature.

theorem Major2024.shifted_without_v : BakerVinokurova2010.cases BakerVinokurova2010.grammar BakerVinokurova2010.finite [BakerVinokurova2010.subject, BakerVinokurova2010.internal true] = [some Case.nom, some Case.acc] ∧ BakerVinokurova2010.cases BakerVinokurova2010.pureChomsky BakerVinokurova2010.finite [BakerVinokurova2010.subject, BakerVinokurova2010.internal true] = [some Case.nom, none]

The modalities come apart on an NP that has shifted with no v to attract it, which the dependent rule values accusative below another NP and Agree leaves caseless. The subject of a dep clause beside an unaccusative verb would be such an NP if dep were a complementizer, and a predicate nominal would be one if shifting were free.

def Major2024.case? (r : Datum) : Option (Option Case)

case? r is the case the gloss of the row r shows on the NP at issue.

Equations

• Major2024.case? r = r.parse? "case" [("ACC", some Case.acc), ("GEN", some Case.gen), ("bare", none)]

def Major2024.voice? (r : Datum) : Option VoiceType

The voice of the verb whose v is above the NP at issue is the verb's own above an object, and above the subject of the complement of de- it is the voice of an agent's saying or of the state of a source that is no agent.

Equations

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

theorem Major2024.case_below_v (g : Minimalist.CaseAssigners) : g ∈ [BakerVinokurova2010.grammar, BakerVinokurova2010.pureChomsky] → ∀ r ∈ Examples.all, ∀ c ∈ case? r, r.feature? "clause" ≠ some "participial" → ∃ vt ∈ voice? r, r.judgment = Judgment.acceptable ↔ ∃ (active : Bool), (active = true → vt.AssignsTheta) ∧ (∀ b ∈ r.parse? "objectPosition" [("beforeAdverb", true), ("afterAdverb", false)], active = b) ∧ (vP g (r.feature? "clause" != some "dep") active)[1]? = some c

An object, or the subject of the complement of de-, is acceptable in a case exactly when some setting of [+ACC/SPEC] gives it that case, the feature being borne only by a θ-assigning v and shown by the NP's place before an adverb: so a predicate nominal under bol- 'become' and the subject below a saying with no agent stay bare, under either modality.

theorem Major2024.participial_subject_genitive (g : Minimalist.CaseAssigners) : g ∈ [BakerVinokurova2010.grammar, BakerVinokurova2010.pureChomsky] → ∀ r ∈ Examples.all, ∀ c ∈ case? r, r.feature? "clause" = some "participial" → (r.judgment = Judgment.acceptable ↔ BakerVinokurova2010.cases g [(Minimalist.Cat.D, Minimalist.Cat.D)] [BakerVinokurova2010.possessor] = [c])

The subject of a participial clause is genitive, valued by the D of the nominalization under either modality, and never accusative.

Were dep a complementizer

def Major2024.complementizerCases (g : Minimalist.CaseAssigners) (r : Datum) : List (Option Case)

If dep is a complementizer, the subject of its clause raises to the clause edge and so into the matrix clause, below a matrix subject or implied agent and an accusative matrix object, where finite T probes and the v of an active θ-assigning verb; these are the cases the row then has.

Equations

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

theorem Major2024.complementizer_agree (r : Datum) : r ∈ [Examples.ex_123, Examples.ex_122, Examples.ex_124] → r.judgment = Judgment.acceptable ∧ case? r = some (some Case.acc) ∧ (complementizerCases BakerVinokurova2010.pureChomsky r).getLast? = some none

With a complementizer and Agree alone the raised subject is caseless where the examples have it accusative: beside unaccusative ket- 'leave', beside a passive, and beside éyt- 'tell', whose one v has an object of its own. By pureChomsky_v_of_acc nothing but a v could value it, which is the argument for the dependent rule that the verb inside dep answers.

theorem Major2024.complementizer_dependent : Examples.ex_128_acc.judgment = Judgment.ungrammatical ∧ (complementizerCases BakerVinokurova2010.grammar Examples.ex_128_acc).getLast? = some (some Case.acc) ∧ Examples.ex_129.judgment = Judgment.acceptable ∧ case? Examples.ex_129 = some (some Case.acc) ∧ (complementizerCases BakerVinokurova2010.grammar Examples.ex_129).getLast? = some (some Case.nom)

With a complementizer the dependent rule values the raised subject accusative wherever the matrix clause has an argument: so beside the inanimate subject 'this news', where the accusative is out, and not when the news is a locative whose source is inside a relative clause, where the accusative is in.