Cross-Linguistic Typology of Change-of-State Verbs #
[Dix82] [Lev93] [BEJ+21] [Coo19]
Empirical data from Beavers, Everdell, Jerro, Kauhanen, [BEJ+21] "States and changes of state: A crosslinguistic study of the roots of verbal meaning." Language 97(3), 439–484.
88-language balanced sample (WALS 100 + additions). For each of 36 property concept (PC) and 36 result root meanings, the authors collected five-item paradigms (underlying root, simple stative, inchoative, causative, result stative) and coded morphological relationships.
Key findings (theory-neutral) #
Simple statives: PC roots overwhelmingly have simple stative forms (median = 95.67% of languages); result roots overwhelmingly lack them (median = 1.59%). Mann-Whitney U = 1266.5, p < 0.001.
Verbal markedness: PC root verbs tend to be marked (median = 56.01%); result root verbs tend to be unmarked (median = 15.20%). Mann-Whitney U = 1291, p < 0.001.
Subclass clustering: PC subclasses (dimension, color, value, physical property, speed) cluster near 100% simple statives. Result subclasses (breaking, cooking, killing, destroying, directed motion) cluster near 0%.
Theory apparatus #
The change-entailment theory (§§3–14 below) is stated over the substrate
Verb.Root.ChangeType; this paper is its sole consumer. The "Markedness
Generalization" label is the formaliser's — the content (ex. (44)) is faithful
to the paper.
Two classes of change-of-state verb roots, defined by morphological and semantic diagnostics ([BEJ+21] §3.1).
Classification criteria:
- PC roots: the root of deadjectival CoS verbs ([Lev93]:245); describe [Dix82]'s basic property types
- Result roots: the root of non-deadjectival CoS verbs; describe specific result states (physical damage, cooking, etc.)
- propertyConcept : CoSRootClass
- result : CoSRootClass
Instances For
Equations
- BeaversEtAl2021.instDecidableEqCoSRootClass x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
PC subclasses ([Dix82]; [BEJ+21] ex. 5).
- dimension : PCSubclass
- age : PCSubclass
- value : PCSubclass
- color : PCSubclass
- physicalProperty : PCSubclass
- speed : PCSubclass
Instances For
Equations
- BeaversEtAl2021.instDecidableEqPCSubclass x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- BeaversEtAl2021.instReprPCSubclass = { reprPrec := BeaversEtAl2021.instReprPCSubclass.repr }
Equations
- One or more equations did not get rendered due to their size.
Instances For
Result root subclasses ([Lev93]; [BEJ+21] ex. 6).
- entitySpecificCoS : ResultSubclass
- cooking : ResultSubclass
- breaking : ResultSubclass
- bending : ResultSubclass
- killing : ResultSubclass
- destroying : ResultSubclass
- calibratableCoS : ResultSubclass
- inherentlyDirectedMotion : ResultSubclass
Instances For
Equations
- BeaversEtAl2021.instDecidableEqResultSubclass 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.
Instances For
Equations
The five positions in a CoS verb paradigm ([BEJ+21] eq. 40). Every root meaning is associated with (up to) five forms.
- underlyingRoot : ParadigmPosition
- simpleStative : ParadigmPosition
- inchoative : ParadigmPosition
- causative : ParadigmPosition
- resultStative : ParadigmPosition
Instances For
Equations
- BeaversEtAl2021.instDecidableEqParadigmPosition x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
Morphological relationship codes between forms ([BEJ+21] eq. 41, generalizing [Has93]:90–92).
- input : MorphRelation
- derived : MorphRelation
- transitive : MorphRelation
- labile : MorphRelation
- equipollent : MorphRelation
- unattested : MorphRelation
- suppletive : MorphRelation
Instances For
Equations
- BeaversEtAl2021.instDecidableEqMorphRelation 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.
Instances For
Equations
A root meaning with its crosslinguistic attestation.
- gloss : String
English gloss(es)
- rootClass : CoSRootClass
PC or result root
- subclass : Option (PCSubclass ⊕ ResultSubclass)
Subclass, if applicable
- nSimpleStative : ℕ
Number of languages with a simple stative for this root (Table A1)
- nLanguages : ℕ
Number of languages with any data for this root (Table A1)
- nMarkedVerbal : ℕ
Number of languages with a marked verbal paradigm (Table A2)
- nVerbalLanguages : ℕ
Number of languages with verbal paradigm data (Table A2)
Instances For
Equations
- BeaversEtAl2021.instReprRootMeaning = { reprPrec := BeaversEtAl2021.instReprRootMeaning.repr }
Equations
- One or more equations did not get rendered due to their size.
Instances For
Percentage of languages with simple stative (exact ℚ — kernel-decidable).
Equations
- r.pctSimpleStative = if r.nLanguages = 0 then 0 else ↑r.nSimpleStative * 100 / ↑r.nLanguages
Instances For
Percentage of languages with marked verbal paradigm (exact ℚ).
Equations
- r.pctMarkedVerbal = if r.nVerbalLanguages = 0 then 0 else ↑r.nMarkedVerbal * 100 / ↑r.nVerbalLanguages
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Summary of a crosslinguistic comparison between PC and result roots. Numeric fields are exact ℚ (mathlib idiom for kernel-decidable arithmetic; [BEJ+21] reports values to 2 decimal places).
- measure : String
What is being measured
- pcMedian : ℚ
PC root median percentage
- resultMedian : ℚ
Result root median percentage
- uStatistic : ℚ
Mann-Whitney U statistic
- pThreshold : ℚ
One-tailed p-value threshold
- nPC : ℕ
Sample sizes (PC roots, result roots)
- nResult : ℕ
Instances For
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
Simple stative form comparison (§6, Fig. 1). Medians: 95.67% and 1.59% encoded as exact rationals via OfScientific.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Verbal markedness comparison (§7, Fig. 5). Medians: 56.01% and 15.20% encoded as exact rationals via OfScientific.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Both comparisons are statistically significant.
Empirical diagnostics for classifying roots. Each diagnostic independently sorts roots into two classes that align with the PC vs result distinction.
- positive : DiagnosticResult
- negative : DiagnosticResult
Instances For
Equations
- BeaversEtAl2021.instDecidableEqDiagnosticResult 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.
Instances For
Equations
Stative form + change denial test (§3.3, ex. 10–11). "The bright photo has never brightened" → OK (PC) "#The shattered vase has never shattered" → contradictory (result)
Equations
Instances For
Restitutive 'again' test (§3.4, ex. 15–16). "John sharpened the knife again" → can be just one sharpening (PC) "#Chris thawed the meat again" → necessarily two thawings (result)
Equations
Instances For
The two diagnostics always agree.
A language's CoS verb typological profile.
- language : String
- family : String
- nPCParadigms : ℕ
Number of PC root verbal paradigms with data
- nResultParadigms : ℕ
Number of result root verbal paradigms with data
- pctPCMarked : ℚ
% of PC paradigms that are marked (exact ℚ)
- pctResultMarked : ℚ
% of result paradigms that are marked (exact ℚ)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
The six in-depth case study languages (§4).
Equations
- BeaversEtAl2021.kakataibo = { language := "Kakataibo", family := "Panoan", nPCParadigms := 59, nResultParadigms := 64, pctPCMarked := 23.73, pctResultMarked := 31.25 }
Instances For
Equations
- BeaversEtAl2021.kinyarwanda = { language := "Kinyarwanda", family := "Northeastern Bantu", nPCParadigms := 69, nResultParadigms := 33, pctPCMarked := 4.17, pctResultMarked := 9.09 }
Instances For
Equations
- BeaversEtAl2021.hebrew = { language := "Hebrew (Modern)", family := "Semitic", nPCParadigms := 35, nResultParadigms := 42, pctPCMarked := 97.62, pctResultMarked := 0.98 }
Instances For
Equations
- BeaversEtAl2021.marathi = { language := "Marathi", family := "Indic", nPCParadigms := 40, nResultParadigms := 35, pctPCMarked := 20.41, pctResultMarked := 21.33 }
Instances For
Equations
- BeaversEtAl2021.greek = { language := "Greek (Modern)", family := "Indo-European", nPCParadigms := 76, nResultParadigms := 57, pctPCMarked := 2.63, pctResultMarked := 2.15 }
Instances For
Equations
- BeaversEtAl2021.english = { language := "English", family := "Germanic", nPCParadigms := 43, nResultParadigms := 60, pctPCMarked := 0.0, pctResultMarked := 0.0 }
Instances For
Attested language types (§7.2, §8).
- asymmetric : LanguageType
- highMarking : LanguageType
- lowMarking : LanguageType
Instances For
Equations
- BeaversEtAl2021.instDecidableEqLanguageType x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
The fourth logically possible type (result marked, PC unmarked) is UNATTESTED — predicted by the markedness generalization.
Equations
Instances For
Three and only three types are attested.
All entries in our PC root sample are classified as PC.
All entries in our result root sample are classified as result.
Most PC roots in the sample have ≥ 50% simple stative attestation.
No result root in the sample exceeds 10% simple stative attestation.
PC roots have simple (unmarked) stative forms; result roots lack them. English: "bright" (PC, simple adj) vs *"shattered" requires prior change. Crosslinguistic evidence (§6, Fig. 1): PC median = 95.67%, result median = 1.59% (Mann-Whitney U = 1266.5, p < 0.001, n₁ = n₂ = 36).
Equations
Instances For
PC root verbs TEND to be morphologically marked (wid-en, flat-ten); result root verbs tend to be unmarked (break, crack). The cross-linguistic DEFAULT, not a universal — [HKG25] §4 notes deviating Wá·šiw Class 1 roots. Crosslinguistic evidence (§7, Fig. 5): PC median = 56.01%, result median = 15.20% (U = 1291, p < 0.001).
Equations
Instances For
The main theorem. A root's entailment of change determines all of its morphosyntactic behavior in a single biconditional: result roots lack simple statives (§6), have unmarked verbal forms (§7), and lack restitutive again (§3.4); PC roots are the reverse. Refutes the Bifurcation Thesis.
The converse: NOT entailing change determines the opposite package.
The Bifurcation Thesis for Roots ([Emb09]:1, [Ara05]:79; [BEJ+21] ex. 2): a component of meaning introduced by a templatic operator cannot be part of a root's meaning — so NO root should entail change.
Equations
- BeaversEtAl2021.bifurcationThesis rootEntailsChange = ∀ (ct : Verb.Root.ChangeType), rootEntailsChange ct = false
Instances For
[BEJ+21] main result: bifurcation does not hold. Result roots entail change, violating the thesis (§§3.3, 3.6, 9).
Result roots witness bifurcation failure.
PC roots are consistent with bifurcation (they don't entail change).
B&[BKG20] strengthened bifurcation failure via
Root.Kinds. [BEJ+21] show roots can entail CHANGE;
B&[BKG20] show roots can entail CHANGE, CAUSATION,
and MANNER simultaneously (√GUILLOTINE, √HAND) — a strictly stronger
refutation. Witness: Root.Kinds.fullSpec carries all four kinds.
Multiple Levin classes witness the stronger bifurcation failure.
Whether a form is morphologically marked (derived/complex) or unmarked (basic/simple).
- unmarked : Markedness
- marked : Markedness
Instances For
Equations
- BeaversEtAl2021.instDecidableEqMarkedness 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.
Instances For
Equations
- BeaversEtAl2021.instReprMarkedness = { reprPrec := BeaversEtAl2021.instReprMarkedness.repr }
The Markedness Generalization ([BEJ+21] ex. 44): morphological markedness reflects semantic mismatch between a functional head and its root complement. A verb is unmarked when v_become is redundant with the root's change entailment; a stative is unmarked when no change need be stripped. This yields the three attested language types (English-type asymmetric, equipollent, labile) and rules out the unattested fourth.
Equations
- BeaversEtAl2021.verbalMarkedness ct = if ct.entailsChange = true then BeaversEtAl2021.Markedness.unmarked else BeaversEtAl2021.Markedness.marked
Instances For
Stative markedness is the mirror image of verbal markedness.
Equations
- BeaversEtAl2021.stativeMarkedness ct = if ct.entailsChange = true then BeaversEtAl2021.Markedness.marked else BeaversEtAl2021.Markedness.unmarked
Instances For
Verbal and stative markedness are always complementary.
Result roots produce unmarked verbs.
PC roots produce marked verbs.
Result roots produce marked statives.
PC roots produce unmarked statives.
The markedness generalization is equivalent to the semantic distinction.
Dowty's P-Patient entailment "undergoes change of state" is precisely the
result root entailment: an object bearing a result root's state predicate
has changeOfState = true.
Equations
Instances For
A result verb's object (accomplishment template) has changeOfState = true,
so it patterns with result roots. Contact-verb objects (kick: CA+St, no
entailed change per [Bea11]) fall on the other side of the bridge.
Die subject undergoes change → result-type pattern.
Result roots MUST combine with a template containing BECOME (achievement or accomplishment): the root's change entailment is redundant with BECOME. PC roots CAN combine with any template.
Equations
Instances For
Result roots always get templates with BECOME.
Achievement and accomplishment templates contain BECOME.
Equations
Instances For
The templates result roots combine with always have BECOME.
State template lacks BECOME — only available to PC roots.
Templates with BECOME map to achievement/accomplishment Vendler classes, both telic.
Aspectual profile for root types in their stative use.
Equations
Instances For
Result root verbs pattern as achievements/accomplishments; PC roots in stative use are states.
Equations
Instances For
PC roots in stative use are states; result roots pattern as achievements.
[Emb04] posits basic statives ([AspP AspS √ROOT], PC roots only) and result statives ([AspP AspR [vP DP v_become √ROOT]], deverbal; result roots always, PC roots optionally).
- basicStative : AdjectivalStructure
- resultStative : AdjectivalStructure
Instances For
Equations
- BeaversEtAl2021.instDecidableEqAdjectivalStructure x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
PC roots admit both structures; result roots only admit resultStative.
Equations
Instances For
This is equivalent to NOT entailing change.
Sublexical again attaches low (to the root, restitutive) or high (over
vbecome, repetitive). Restitutive is available iff the root is change-free;
a result root's state itself entails change, collapsing its low attachment
into the repetitive reading.
- restitutive : AgainReading
- repetitive : AgainReading
Instances For
Equations
- BeaversEtAl2021.instDecidableEqAgainReading 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.
Instances For
Equations
Which readings of 'again' are available for each root type.
Equations
Instances For
PC roots have strictly more 'again' readings than result roots.
Result roots lack the restitutive reading.
PC roots have the restitutive reading.
If a root entails change, its verb is associated with a template containing BECOME (§9) — even when the change comes from the ROOT, not the template.
Conversely: a root not requiring BECOME does not entail change.
The full correlation package. From the single Boolean entailsChange,
all of the paper's morphosyntactic predictions follow: one semantic property
is the sole determinant of six independently observable properties.
Change entailment determines markedness in the unified Classification.
Roots with the same change type have identical morphosyntactic behavior regardless of valency — markedness, stative forms, and again readings are orthogonal to internal argument selection.
The §§8–14 sections below were originally housed in
Studies/Coon2019.lean as bridge content
between Coon's Chuj data, the present empirical typology, and the root
typology substrate. They are relocated here per the chronological-
dependency rule (Coon 2019 < Beavers et al. 2021 — only the later
paper may reference the earlier).
Map the theory's change type to the empirical root class. These are parallel enums — the bridge makes the correspondence explicit.
Equations
Instances For
Map back from empirical to theory.
Equations
Instances For
The mapping is a bijection (left inverse).
The mapping is a bijection (right inverse).
The empirical changeDenialTest agrees with the theory's entailsChange.
Theory: ChangeType.entailsChange .result = true (result roots entail change)
Empirical: changeDenialTest.result =.negative ("#The shattered vase
has never shattered" is contradictory — the state entails prior change)
The relationship is: entailsChange = true ↔ changeDenial = negative.
The empirical restitutiveAgainTest agrees with the theory's
allowsRestitutiveAgain.
Both diagnostics jointly align with the full semantic correlate package.
The bridge form of semantic_determines_morphosyntax.
Theory predicts: PC roots have simple statives. Data confirms: 7 of 8 PC sample roots have ≥ 50% attestation. The one exception (oldRoot, age class) has 0 — noted by the present paper as a crosslinguistic outlier.
Theory predicts: result roots LACK simple statives. Data confirms: all 10 result sample roots have ≤ 10% attestation.
Theory predicts: PC verbs are morphologically marked; result verbs are unmarked (Markedness Generalization, [BEJ+21]). Data confirms: PC median marked % (56.01) > result median (15.20).
The theory's markedness complementarity predicts that if a language
marks PC verbs, it should NOT also show result verbs as more marked
than PC verbs. The fourth logically possible language type (result
marked, PC unmarked) is unattested — exactly 3 types are attested.
This matches the theory: markedness_complementarity says verbal and
stative markedness are always opposite.
The property-concept row of the √TV subdivision (no change entailment).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Coon's √ITV coordinates, reused unchanged.
Instances For
Coon's √POS coordinates, reused unchanged.
Instances For
Chuj √TV result roots instantiate the theory's result root predictions: entails change, no simple stative, unmarked verb.
Chuj √TV PC roots instantiate the theory's PC root predictions: no change entailment, has simple stative, marked verb.
The Chuj fragment witnesses cells of the (valency × changeType) matrix: theme-taking roots with and without change entailment, and valency-free property-concept roots (√POS — √ITV is unaccusative and theme-taking per [Coo19] §3.3).
Per-root class verification: each Chuj root's change entailment matches
its predicted morphosyntactic correlates via grand_unification.
Every PC root in the empirical sample is classified as PC, and the theory predicts PC roots should have simple statives — they do.
Every result root in the empirical sample is classified as result, and the theory predicts result roots lack simple statives — they do.