Documentation

Linglib.Phenomena.Gradability.Studies.Tham2025

@cite{tham-2025} #

Shiao Wei Tham (2025). Multidimensionality and the scalar components of physical disturbance predicates. Glossa: a journal of general linguistics 10(1): 1–30.

Key Claims #

Physical disturbance predicates (crack/cracked, scratch/scratched, dent/dented) are associated with a totally closed, multi-point scale:

  1. Contra @cite{rappaport-hovav-2014}: crack/cracked is NOT two-point (like die/dead). The verb allows durative for adverbials and the adjective accepts comparatives (more cracked).

  2. Contra @cite{rotstein-winter-2004}: cracked is NOT a "partial" adjective with a lower-bounded, upper-open scale. It accepts completely (upper bound) and partially (lower bound).

  3. The scale is multidimensional (@cite{sassoon-2013}, @cite{solt-2018}): the gradable interval comes from multiple dimensions — quantity of disturbances, qualitative severity (depth, length), and spatial positioning.

  4. Lower bound = physical instantiation of disturbance (objective: no faultless disagreement in simple predication). Upper bound = spatial extent of host entity (structural integrity limit).

  5. Adjective vs verb asymmetry: the adjective allows grammatical access to individual dimensions via respect PPs and quantificational operators; the verb allows only conceptual access to dimensions.

Formalization #

A physical disturbance predicate entry, encoding the scalar and distributional properties argued for in the paper.

  • root : String

    Root form (shared by verb, adjective, and count noun).

  • hasCountNoun : Bool

    Whether the root has a count noun form (a crack, a dent). Distinguishes disturbance predicates from other CoS predicates like shatter (no *a shatter) and damage (mass only).

  • adjGradable : Bool

    Whether the deverbal adjective is gradable (more cracked).

  • adjCompletely : Bool

    Compatible with completely (tests upper closure).

  • adjPartially : Bool

    Compatible with partially (tests lower closure).

  • adjSlightly : Bool

    Compatible with slightly (indicates lower bound).

  • adjBadly : Bool

    Compatible with badly/well (degree modifier for closed scales).

  • adjMuch : Bool

    Compatible with much (selects quantity dimension specifically).

  • verbForX : Bool

    Whether the verb allows durative for X adverbials (atelic reading).

  • verbInX : Bool

    Whether the verb allows in X adverbials (telic reading).

  • verbCompletely : Bool

    Whether the verb is compatible with completely.

  • verbBadly : Bool

    Whether the verb is compatible with badly.

  • adjRespectPP : Bool

    Whether the adjective allows respect PPs (dented with respect to dent size).

  • verbRespectPP : Bool

    Whether the verb allows respect PPs (degraded for intransitive disturbance CoS verbs, but accepted for causative uses).

  • verbInXEntailsResult : Bool

    Whether verb + in X entails the result state (cracked in a minute |= is cracked).

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

              Data for predicates that contrast with disturbance predicates, demonstrating that the disturbance pattern is class-specific.

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

                  die/dead: two-point scale, non-gradable (#more dead), punctual. @cite{tham-2025} §2.3: crack ≠ die despite both being Levin 45.1. @cite{rappaport-hovav-2014} Table 1 groups both as "two-valued" — Tham shows this is wrong for crack.

                  Equations
                  Instances For

                    shatter/shattered: punctual, non-gradable. Like die/dead: describes physical objects but NOT a disturbance predicate. Crucially shows that gradability of disturbance predicates is NOT simply a consequence of applying to physical objects.

                    Equations
                    Instances For

                      damage/damaged: has mass noun (damage to X) but NOT count noun (*a damage in X). Not a disturbance predicate per (4).

                      Equations
                      Instances For

                        All disturbance predicates accept both completely (upper bound) and partially (lower bound), demonstrating total closure.

                        All disturbance predicates are gradable (contra two-point classification in @cite{rappaport-hovav-2014} Table 1).

                        All disturbance predicates have a count noun form, distinguishing them from other CoS predicates (§2.1, ex. 3–5).

                        Contrast: non-disturbance predicates are NOT gradable.

                        Contrast: non-disturbance verbs lack durative readings.

                        Physical disturbance verbs allow BOTH telic and atelic readings, unlike standard degree achievements which have only one.

                        §2.4 ex. (13)–(14): Both crack and cool accept in X and for X adverbials (variable telicity). But they differ in entailment: - cracked in a minute |= is cracked (14a) - cooled in ten minutes |= is cool (13a telic reading) - cooled for ten minutes |≠ is cool (13b — process, no endpoint reached) - cracked for a while |= is cracked (14b — process, but entails result!)

                        Disturbance predicates entail their result state regardless of aspect.
                        This follows from the lower bound being physical instantiation:
                        once any cracking exists, the host IS cracked. 
                        

                        All disturbance verbs entail their result state under both in X and for X readings.

                        Disturbance adjectives have three dimensions (§3.1): - Quantity: number of disturbances (23a, 24a, 25a) - Quality: severity — depth, length, width (23b, 24b, 25b) - Positioning: centrality, array pattern, functionality impact

                        The adjective allows **quantificational** access to individual dimensions
                        via *respect* PPs and quantificational operators (*every respect*, *at
                        least with respect to*). The verb allows only **conceptual** access:
                        dimensions are interpretively available but resist grammatical
                        individuation (§4.2).
                        
                        This quantificational/conceptual distinction aligns with
                        @cite{ruiz-faroldi-2022}'s typology of multidimensionality: disturbance
                        adjectives exhibit "quantificational" multidimensionality (dimensions
                        are grammatically individuable), while disturbance verbs have only
                        "conceptual" multidimensionality (dimensions available for interpretation
                        but not grammatical access). 
                        

                        much selects only the quantity dimension, providing distributional evidence that dimensions are independently accessible for adjectives. (§3.3, ex. 30): "a much cracked dish" = many cracks (not one deep crack).

                        Verify that the Fragment adjective entries classify disturbance adjectives with the correct Boundedness value.

                        dead also has .closed — but is non-gradable (two-point). Same Boundedness, different gradability. Boundedness alone does not distinguish two-point from multi-point closed scales.

                        Disturbance adjectives are licensed for degree modification by the Kennedy pipeline, just like full and clean.

                        Interpretive Economy predicts a max-endpoint standard for disturbance adjectives (closed → maxEndpoint). This interacts non-trivially with Tham's analysis: simple predication (is cracked) requires only the minimum physical instantiation (lower bound), but Interpretive Economy selects the maximum as the positive standard. The resolution is that the positive standard determines the degree needed for the positive form to apply, while the lower bound determines the threshold for being on the scale at all.

                        The @cite{kennedy-levin-2008} pipeline predicts that closed-scale CoS verbs are telic (accomplishments). But disturbance verbs allow both telic and atelic readings: cracked in a minute AND cracked for two days.

                        The pipeline gives `defaultVendlerClass = .accomplishment` for crack,
                        but the fragment entry stipulates `.achievement` (punctual). This is a
                        genuine divergence: disturbance verbs are NOT standard degree achievements.
                        Their variable telicity does not reduce to scale boundedness alone. 
                        

                        But the fragment stipulates achievement (punctual telic), capturing the in X = "after" reading (ex. 9): "The mirror will crack in five minutes."

                        The pipeline and stipulation DIVERGE for disturbance verbs. This is the formal expression of Tham's challenge to the Kennedy-Levin mapping: closed scale does not uniformly predict accomplishment.

                        break (also Levin 45.1) IS an accomplishment — the standard pipeline works for it. So the divergence is specific to disturbance predicates within the same Levin class.

                        crack and break share LevinClass but differ in VendlerClass. Both are Levin 45.1 Break verbs, but crack is achievement (punctual + optional durative extension) while break is accomplishment (durative). This within-class split is exactly what Tham's analysis predicts: disturbance predicates have distinctive aspectual behavior.

                        shatter is also achievement — but unlike crack, shatter ONLY has the punctual reading. Same Vendler class, different aspectual flexibility. VendlerClass is too coarse for disturbance predicates.

                        Boundedness convergence: both pipelines agree crack is .closed. Achievement and accomplishment are both telic → .closed, so the disturbance-specific VendlerClass divergence (§7 above) is invisible at the Boundedness granularity. This is exactly the gap Tham identifies: the Kennedy-Levin pipeline is correct about scale closure but wrong about what that closure implies for aspectual class.

                        Contrast: @cite{kennedy-levin-2008} study (KennedyLevin2008.lean) proves convergence for 12 standard DAs — there, convergence at Boundedness ALSO means convergence at VendlerClass. For crack, only Boundedness converges.

                        Disturbance adjectives are multidimensional like @cite{sassoon-2013}'s adjectives (healthy, sick), but their dimensions compose differently.

                        Sassoon 2013 dimensions bind via quantifiers (∀ = conjunctive, ∃ = disjunctive).
                        Disturbance adjective dimensions compose via **weighted aggregation**
                        (eq. 47b): μ(x) = Σᵢ kᵢ · μ_EXTENT(distᵢ(x)) / μ_SPATIAL_EXTENT(x).
                        
                        This is a new composition mode not captured by `DimensionBindingType`. 
                        

                        Disturbance adjectives are NOT conjunctive: an entity can be badly cracked along one dimension (quality: one deep crack) while scoring low on another (quantity: few cracks). Conjunctive binding would require ALL dimensions to be high, but speakers accept (24b).

                        Under disjunctive binding, the same entity DOES satisfy the predicate — one high dimension suffices. This is closer to simple predication (is cracked), but still wrong for completely cracked which requires spatial coverage (more like conjunctive over spatial parts).

                        The key insight: disturbance adjectives behave like BOTH conjunctive (under completely) and disjunctive (under simple predication). Neither Sassoon 2013 binding type captures this. The binding mode shifts with degree modification — a property not shared by Sassoon's healthy/sick adjectives, which are stably conjunctive/disjunctive.

                        @cite{dambrosio-hedden-2024} show that @cite{sassoon-2013}'s binding types (conjunctive, disjunctive, mixed) are all counting aggregation — a single escape route from Arrow's impossibility. Disturbance adjectives require utilitarian aggregation (weighted sum), which is a categorically different mechanism.

                        The `sassoon_binding_insufficient` theorem above proves the gap.
                        D&H's framework explains WHY: counting cannot capture dimension-weight
                        asymmetries. Utilitarian aggregation resolves this by assigning
                        different weights to quantity, quality, and positioning. 
                        
                        def Phenomena.Gradability.Studies.Tham2025.disturbanceDims (quantity quality positioning : Bool) :
                        List (UnitBool)

                        Disturbance adjective dimensions as Bool predicates.

                        Equations
                        Instances For

                          Counting (k=2) accepts a vase with many shallow scratches (high quantity, low quality, high positioning): 2 of 3 dims.

                          Quality-weighted aggregation [1, 3, 1] with θ=3 REJECTS the same vase: severity matters more than count. Score = 1·1 + 3·0 + 1·1 = 2 < 3.

                          Counting and weighted aggregation DIVERGE on high-quantity low-quality configurations. This is the formal expression of eq. 47b's advantage: weighted aggregation captures dimension asymmetry that counting misses.