Major 2024: Re-analyzing 'say' complementation #
Uyghur dep clauses look like complementizer-headed CP complements
but are converbial adjunct clauses headed by the verb de 'say' plus
the converb -(I)p, merging at VP or TP (his 4, 9). Because the linker
contains the verb 'say', main-verb properties of de- persist inside
dep clauses (his 39–41), and because the linker is a converb, dep
clauses are banned from argument positions: they cannot be grammatical
subjects (his 49b) and "are never internal arguments" (§3.2). The
paper's methodological point — "taking the morphology at face value
(i.e. dep is 'say' + CNV)" (§1) — is rendered literally here:
mergeMode reads a morpheme's merge behavior off its recorded
morphology (verbForm, noonanType), and the ban is derived, not
stipulated per position. The Washo parallel is [BH21]'s
modifier analysis of non-factive embedding, which Major extends with
the verb 'say' inside the linker.
The case-theoretic payoff (his §§4–6) is out of scope here: the same decomposition holds of Sakha dien = die 'say' + converb -(E)n, so [BV10]'s accusative subjects under dien reduce to ECM by the v of 'say', resurrecting Case-by-Agree where B&V argued for Dependent Case Theory.
Main declarations #
ClausePosition,MergeMode,mergeMode,licensedIn— merge behavior read off the fragment entries' morphologySayConverbAnalysis— witness record for the re-analysis (cf.Bondarenko2022.ContAnalysisfor the rival carving);depAnalysisis the Uyghur witness, andSayConverbAnalysis.argument_banderives the ban for any witnessdep_never_argument,participial_licensed_iff_argument— the argument-position asymmetry (his 49, 59, 61)coerced_speech_reading— unergative 'scream' + dep (his 38–41)dep_nci_diagnoses_height— the VP/TP height contrast (his 54)dep_never_feeds_factivity— the factivity alternation as a corollary of the ban (his 61–65)
The rival carving (docstring-only, per the chronology rule) #
[Bon22] assigns the structurally parallel Buryat say-complex
gɘ-žɘ to functional heads: the say-root expones Cont and the converb
is a Comp allomorph (Bondarenko2022.buryatAnalysis), so the complex
heads a selected complement. Major's carving keeps both pieces
lexical — verb plus adjunct-forming converb — and denies complement
status altogether. Major does not engage that analysis (he cites only
[Bon20] among factivity-alternation accounts, §3.2), so no
divergence theorem is stated here; the two witnesses coexist as rival
records over their respective fragment inventories.
Typed paradigm sentences (his 2, 38–41, 49) live in Major2024.Examples,
generated from Data/Examples/Major2024.json.
Merge modes read off the morphology (§2) #
Converbial -(I)p clauses adjoin at two heights (his 4): VP-level -(I)p is a manner modifier interpreted under matrix aspect (his 10–13), answers qandaq 'how' (his 15–16), and sits below the matrix accusative position (his 19); TP-level -(I)p precedes the whole matrix clause (his 29), tolerates aspect and voice mismatches (his 12, 30), and merges at (at least) TP (his 31). Dep clauses replicate both profiles (his 37–38, 52–56).
Structural positions at issue for an embedded clause: the two converb adjunction heights (his 4) and the two argument positions the paper tests — complement of V (his 59a, 61a, 73) and grammatical subject (his 49).
- complementOfV : ClausePosition
- subject : ClausePosition
- vpAdjunct : ClausePosition
- tpAdjunct : ClausePosition
Instances For
Equations
- Major2024.instDecidableEqClausePosition x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Major2024.instReprClausePosition = { reprPrec := Major2024.instReprClausePosition.repr }
Argument positions: those saturating a θ-position of the matrix predicate.
Equations
Instances For
Equations
- Major2024.instDecidablePredClausePositionIsArgument Major2024.ClausePosition.complementOfV = isTrue trivial
- Major2024.instDecidablePredClausePositionIsArgument Major2024.ClausePosition.subject = isTrue trivial
- Major2024.instDecidablePredClausePositionIsArgument Major2024.ClausePosition.vpAdjunct = isFalse Major2024.instDecidablePredClausePositionIsArgument._proof_1
- Major2024.instDecidablePredClausePositionIsArgument Major2024.ClausePosition.tpAdjunct = isFalse Major2024.instDecidablePredClausePositionIsArgument._proof_2
How a clause-forming morpheme merges the clause it heads: a converb builds an adjunct to a verbal projection (his 4); a nominalizer builds a case-bearing nominal that saturates an argument position (his 49a, 59a).
Instances For
Equations
- Major2024.instDecidableEqMergeMode x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- Major2024.instReprMergeMode = { reprPrec := Major2024.instReprMergeMode.repr }
Equations
- One or more equations did not get rendered due to their size.
Instances For
Converbial adjuncts modify, never saturate ("dep clauses are never internal arguments", §3.2; subject ban his 49b); nominalized clauses saturate. Oblique case-marked participial adjuncts (his 50b) go through case morphology, outside this position set.
Equations
Instances For
The merge mode of a clause-forming morpheme, read off its recorded
morphology — the paper's "morphology at face value" (§1): a converbial
suffix (verbForm = .Conv) builds adjuncts; a nominalizing
clause-typer (noonanType = .nominalized) builds arguments. none
for morphemes heading no clause of their own.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A clause headed by c is licensed in pos iff c's merge mode
admits it. No matrix-verb parameter: dep clauses appear regardless of
matrix transitivity (his 44a söz qil- vs. unaccusative 44b söz
bol-) and with unaccusative 'be surprised' (his 50c).
Equations
- Major2024.licensedIn c pos = match Major2024.mergeMode c with | some m => m.admits pos | none => False
Instances For
Equations
- Major2024.instDecidableLicensedIn c pos = match hm : Major2024.mergeMode c with | some m => decidable_of_iff (m.admits pos) ⋯ | none => decidable_of_iff False ⋯
The say-root heads no clause of its own — the linker is the converb, not 'say' (his 2–3).
The say-converb witness (his 2–3, 9) #
A say-converb re-analysis of an apparent complementizer
([Maj24]; cf. Bondarenko2022.ContAnalysis for the rival
Cont-exponence carving): the linker decomposes into a say-root and a
converbial suffix drawn from the language's inventory, and the
say-root is the independently attested main verb 'say' — one lexical
item, so main-verb properties persist inside the adjunct by
construction (his 39–41). A structure, not a class: rival frameworks
construct rival witnesses.
- inventory : List Complementizer
The fragment inventory analyzed.
- surface : String
The apparent complementizer being decomposed (dep; Sakha dien).
- sayRoot : Complementizer
The say-root inside the complex linker.
- converb : Complementizer
The converbial suffix heading the adjunct clause.
- say : Verb
The main verb 'say' — the same lexical item as
sayRoot. - converb_conv : self.converb.verbForm = some UD.VerbForm.Conv
Morphology at face value: the linker is a converb.
- say_transitive : self.say.complementType ≠ ComplementType.none ∧ self.say.implicitObj = none
'say' is transitive with an obligatory internal argument (his 39a, 40a), a requirement that persists inside the adjunct (his 41:
*(birnémi-ler-ni) de-p warqiri-di).
Instances For
Any say-converb analysis fixes adjunction as the linker's merge mode.
Any say-converb analysis licenses the say-clause at both adjunction heights (his 4, 51, 56).
The argument-position ban, derived for any witness: the converb morphology fixes adjunction, and adjuncts never saturate.
The Uyghur witness: dep = de 'say' + -(I)p (his 2–3), over the fragment inventory.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Dep clauses adjoin at VP and TP (his 4, 37, 51, 56).
The argument ban (subject: his 49b; internal argument: §3.2): derived from -(I)p's converb morphology, and independent of the matrix verb — dep clauses occur in clearly unselected environments, as reasons or excuses (his 5, 53).
His (49b): a dep clause cannot be the grammatical subject of 'make surprised' — unlike the participial clause (his 49a).
The participial strategy is the mirror image: nominalized clauses are licensed exactly in argument positions — subject (his 49a) and complement of V (his 59a, 61a) — never at the converb adjunction sites. Dep clauses also fail N-complement constituency: the head noun scrambles away from a dep clause (his 45–46) but never from a genuine N-complement (his 47–48).
Say-properties persist: coerced speech readings (§3.1) #
The persistence claim is carried by depAnalysis housing the single
lexical entry Uyghur.deVerb: whatever the fragment records of
main-verb 'say' holds of 'say' inside dep clauses. The same holds of
any converb-suffixed verb — 'think' imposes its own frame inside an
adjunct (his 42) — so the persistence is converbial, not dep-magic.
His (38)–(41): warqira- 'scream' has no complement frame (his 39b, 40b), yet 'scream' + dep reports propositional content — the content sits in the obligatory complement of de- inside the VP-adjoined say-clause, which "coerces it into a verb of speech" (§3.1). No hidden frame of 'scream' is needed.
The two heights diagnosed (his 54–55) #
Positions inside the matrix clausemate domain for NCI licensing: Uyghur héch- items need clausemate negation (his 20). Everything merged below matrix T is in the domain; the TP-adjunct, which precedes the entire matrix clause (his 29, 31), is outside.
Equations
Instances For
Equations
- Major2024.instDecidablePredClausePositionClausemateWithMatrixNeg Major2024.ClausePosition.complementOfV = isTrue trivial
- Major2024.instDecidablePredClausePositionClausemateWithMatrixNeg Major2024.ClausePosition.subject = isTrue trivial
- Major2024.instDecidablePredClausePositionClausemateWithMatrixNeg Major2024.ClausePosition.vpAdjunct = isTrue trivial
- Major2024.instDecidablePredClausePositionClausemateWithMatrixNeg Major2024.ClausePosition.tpAdjunct = isFalse Major2024.instDecidablePredClausePositionClausemateWithMatrixNeg._proof_1
His (54): matrix negation licenses an NCI inside a dep clause exactly at the VP site — replicating the bare -(I)p contrast (his 22–26) — so NCI licensing diagnoses a dep clause's attachment height. The shundaq-anaphora contrast (his 55) draws the same line.
The factivity alternation, derived (his 61–65) #
Positions feeding a factive predicate's presupposition: only its complement — "factive interpretations arise from predicates taking a nominal complement, while non-factive interpretations arise via adjunction", the adjunction site being "outside the scope of the factive predicate" (§3.2).
Equations
- Major2024.ClausePosition.complementOfV.feedsFactivity = True
- x✝.feedsFactivity = False
Instances For
Equations
- Major2024.instDecidablePredClausePositionFeedsFactivity Major2024.ClausePosition.complementOfV = isTrue trivial
- Major2024.instDecidablePredClausePositionFeedsFactivity Major2024.ClausePosition.subject = isFalse Major2024.instDecidablePredClausePositionFeedsFactivity._proof_1
- Major2024.instDecidablePredClausePositionFeedsFactivity Major2024.ClausePosition.vpAdjunct = isFalse Major2024.instDecidablePredClausePositionFeedsFactivity._proof_2
- Major2024.instDecidablePredClausePositionFeedsFactivity Major2024.ClausePosition.tpAdjunct = isFalse Major2024.instDecidablePredClausePositionFeedsFactivity._proof_3
Factive scope is a subrelation of argumenthood.
'know' + dep is non-factive (his 61b) across the whole factive class (his 63), and 'forget' + dep loses the forget-that reading (his 65): a dep clause can occupy no factivity-feeding position — a corollary of the argument ban, contra a per-verb homophony account the paper rejects (§3.2; 'know' stays presuppositional about its own object even with dep present, his 62).
The participial contrast (his 61a, 63a–b): the nominalized clause sits in complement position, where factivity is fed.