Bhadra 2024: Verb roots encode outcomes #
@cite{bhadra-2024}
Bhadra, D. (2024). Verb roots encode outcomes: argument structure and lexical semantics of reversal and restitution. Linguistics and Philosophy 47: 557–610.
Summary #
The reversative prefix un- and the restitutive prefix re- are sensitive to the outcome set of the base verb root. All verb roots come equipped with sets of outcomes (possible states of the object after the action). The cardinality of this set determines affix compatibility:
- PFC verbs (fold, wrap, coil): multi-membered outcomes → un- ✓, re- ✓
- IE verbs (hit, scrub, kick): singleton outcomes, surface only → un- ✗, re- ✗
- COS verbs (break, destroy, paint): singleton outcomes → un- ✗, re- partial
Key formalizations #
ForceTransmissionClassclassifies verbs by impact type (§§2, 3, 4)BoundaryStatesformalizesres/preoperators for state equivalence (§5.2)LevinClass.forceTransmissionClassbridges Levin classes to outcome classes- Per-verb un- and re- predictions verified against empirical data from (12), (45)
Only surface contact verbs (PFC class) allow un-: - (d) Surface contact: unpin, unwrap, untwist, unpack, unplug ✓ - (a) Physical property: *unpaint, *unclean, *unfix, *unbreak ✗ - (b) Transforms: *unturn, *uncarve ✗ - (e) Creation: *unbuild, *unconstruct ✗ - (f) Consumption: *undestroy, *uneat ✗ - (g) Degree achievements: *unfill, *unwarm ✗ - (h) No change: *unswim, *unwalk ✗
PFC classes allow un-.
COS classes disallow un-.
IE classes disallow un-.
No-change classes disallow un-.
re- is more permissive than un-: - (a) Physical property: repaint ✓, reclean ✓, refix ✓, rebreak ✓ - (d) Surface contact: repin ✓, rewrap ✓, retwist ✓ - (e) Creation: rebuild ✓, reconstruct ✓, recreate ✓ - (g) Degree achievements: refill ⊳, rewarm ⊳ But re- is blocked when the object is eliminated: - (f) Consumption: *redestroy, *reeat ✗ - (b) Transforms: *retransform ✗ (mostly) - IE: *rehit, *rescrub ✗ - No change: *reswim, *rewalk ✗
PFC classes allow re-.
Physical property and creation COS classes allow re-.
Consumption, destruction, killing COS classes block re-.
IE classes block re-.
No-change classes block re- (paper's (45h): *reswim, *rewalk).
PFC verbs have multi-membered outcome sets.
IE verbs have singleton outcome sets (impingement only).
COS verbs have singleton outcome sets (specific result).
No-change verbs have empty outcome sets.
Possible states of a parchment under folding. Illustrates the multi-membered outcome set of a PFC verb: folding can yield any of these states depending on the force and manner of folding.
- flat : ParchmentState
- slightlyCreased : ParchmentState
- folded : ParchmentState
- tightlyFolded : ParchmentState
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
fold(parchment): flat → folded
Equations
- One or more equations did not get rendered due to their size.
Instances For
unfold(parchment): folded → flat (reverses fold)
Equations
- One or more equations did not get rendered due to their size.
Instances For
Reversibility: fold and unfold satisfy the inverse equivalence condition.
res(fold) = pre(unfold) and res(unfold) = pre(fold).
Restitution: refold achieves the same result as fold.
res(refold) = res(fold).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
repaint satisfies restitution: res(repaint) = res(paint).
*unpaint is blocked: color (24) is COS with singleton outcomes. un- requires multi-membered outcomes (PFC only).
PFC is the unique class compatible with BOTH un- and re-. This is the central distributional generalization of the paper.
@cite{bhadra-2024} reclassifies the bend class (45.2) from COS to PFC.
@cite{levin-1993}: bend has changeOfState = true (diagnosed by
causative/inchoative alternation: "the paper folded" / "she folded the paper").
@cite{bhadra-2024}: fold has multi-membered outcomes (slightly creased, halfway bent, tightly folded, etc.) → PFC, not COS. The change IS possible but not to a SPECIFIC fixed state.
This is the paper's central theoretical move: outcome set structure is a finer-grained classification than the traditional COS label.
@cite{rappaport-hovav-levin-2024}'s motionContact template corresponds exactly to @cite{bhadra-2024}'s IE class. The wipe class (10.4) has the motionContact template and is classified as IE.
End-to-end chain: the Fragment entry for "kick" (Levin 18.1 hit class) derives IE classification and correctly predicts both un- and re- blocking. kick.levinClass → .hit → .impingementEffecting → unCompatible=false, reCompatible=false
End-to-end chain: the Fragment entry for "bend" (Levin 45.2) derives PFC classification and correctly predicts both un- and re- compatibility. bend.levinClass → .bend → .potentialForChange → unCompatible=true, reCompatible=true
End-to-end chain: the Fragment entry for "break" (Levin 45.1) derives COS classification → un- blocked, re- allowed. break.levinClass → .break_ → .integralChange → unCompatible=false, reCompatible=true
VRO for "fold": PFC verb with multi-membered outcome set. The parchment can end up in any of several states after folding. Outcome set = {slightlyCreased, folded, tightlyFolded} (3 members). Threshold set = {flat, slightlyCreased} (contextual pre-states).
Equations
- One or more equations did not get rendered due to their size.
Instances For
fold's outcome set is multi-membered: slightlyCreased ≠ folded.
VRO for "break": COS verb with singleton outcome set. Breaking yields exactly one lexically specified result: broken.
Instances For
Equations
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
break's singleton outcome set blocks un-: *unbreak is predicted to fail because |O| = 1, so the multi-membered presupposition of un- (eq. 66) cannot be satisfied.
VRO for "hit": IE verb with singleton outcome set (surface alteration). The object's surface is altered in exactly one way.
- unaltered : SurfaceState
- surfaceAltered : SurfaceState
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
hit's singleton outcome set blocks un-: *unhit is predicted to fail for the same reason as *unbreak.
VRO for "destroy": COS consumption verb with singleton outcome set. The outcome (ceased to exist) is also a blocking threshold for re-, since the object cannot be acted on again after being destroyed.
- exists_ : ObjectExistence
- ceasedToExist : ObjectExistence
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
destroy blocks un- (singleton outcomes).
The three-way distributional split is derived from outcome set structure:
- PFC (fold): multi-membered → un- possible
- COS (break): singleton → un- blocked
- IE (hit): singleton → un- blocked
The theorems in § 10 prove that un- is BLOCKED for certain verb classes. But blocking theorems alone don't guarantee the compositional semantics is non-vacuous. We also need to show un- and re- are SATISFIABLE for the classes that should allow them. The following construct concrete event witnesses and state functions to demonstrate positive satisfiability.
Positive existence: un- IS satisfiable for PFC verbs.
Constructs a concrete witness showing unSem holds for fold.
Positive existence: re- IS satisfiable for PFC verbs. Three-event scenario: fold(ev₁), unfold(ev₂), refold(ev₃). The rePresupposition of ev₃ is witnessed by ev₁ (prior fold with matching result state).
Positive existence: re- IS satisfiable for COS verbs. COS verbs (break) block un- but allow re-. This demonstrates that reSem is satisfiable for break despite singleton outcomes.
Cross-layer agreement: Boolean and compositional predictions align. The Boolean layer (ForceTransmissionClass) and compositional layer (VRO) agree on the un- prediction for fold (both allow) and break (both block).
With APPLIES in the semantics, we can now compositionally derive that
*redestroy is blocked — not just via the Boolean layer, but because
APPLIES fails when the object has ceased to exist. This closes the
gap between the Boolean prediction (destroy_no_re) and the
compositional semantics (reSem).
VRO for "destroy" with state-aware APPLIES: force can only be exerted
on an object that exists at the start of the event.
The applies predicate is parameterized by a state function, capturing
the fact that you can't destroy what doesn't exist.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Compositional re- blocking for destroy.
With state-aware APPLIES, reSem is unsatisfiable for destroy because
the re-event requires APPLIES(e)(x), but the object has ceased to exist
after the first destruction. The proof shows the assertion's APPLIES
condition directly contradicts the post-destruction state.
Cross-layer re- agreement for destroy.
The Boolean layer (destroy_no_re) and the compositional layer
(destroy_re_blocked_compositionally) now agree: *redestroy is blocked
at both levels.
For break, both layers agree that re- IS allowed:
Boolean (break_re) and compositional (break_re_satisfiable).