Categorial Features ↔ Category-Changing Morphology #
[Pan15] [Mar97]Connects the theory-side predictions of [Pan15] — substantive categorial features [N] and [V] hosted on categorizer heads — to the empirical data on category-changing morphology in English.
What this bridge proves #
Categorizer–LexCat correspondence: Each theory-side categorizer (v, n, a) maps to exactly one empirical lexical category (verb, noun, adjective).
Feature predictions: The categorial features [N]/[V] on each categorizer correctly predict the interpretive perspective of the resulting category — nouns have sortal perspective ([N]), verbs have temporal perspective ([V]), adjectives have both ([N, V]).
EP well-formedness: Each categorizer extends its lexical anchor into a well-formed EP (A→a, N→n, V→v).
Categorizer parallelism: All three categorizers sit at the same F-level (F1 in Grimshaw's system), formalizing Panagiotidis's claim that categorization is a uniform operation across category families.
Derivational chain #
ExtendedProjection/Basic.lean (CategorialFeatures, isCategorizer, categorialFeatures)
↓
THIS BRIDGE FILE
↓
Morphology/RootFamily.lean (RootFamily, LexCat)
Map a Minimalist categorizer to the empirical lexical category of the word it produces. This is the core link between the theory (Cat.v, Cat.n, Cat.a) and the data (LexCat).
Equations
- Panagiotidis2015.categorizerToLexCat Minimalist.Cat.v = some Morphology.LexCat.verb
- Panagiotidis2015.categorizerToLexCat Minimalist.Cat.n = some Morphology.LexCat.noun
- Panagiotidis2015.categorizerToLexCat Minimalist.Cat.a = some Morphology.LexCat.adjective
- Panagiotidis2015.categorizerToLexCat x✝ = none
Instances For
Map an empirical lexical category to its theory-side categorizer.
Equations
Instances For
The mapping is a partial bijection: lexCat → categorizer → lexCat roundtrips.
Every categorizer maps to some LexCat.
Non-categorizers don't map to any LexCat.
Does a categorizer produce a category with sortal perspective? Panagiotidis §4.3: [N] = sortal perspective / referentiality. Items bearing [N] have the capacity to introduce discourse referents (nouns, adjectives) — items lacking [N] do not (verbs).
Equations
Instances For
Does a categorizer produce a category with temporal perspective? Panagiotidis §4.3: [V] = temporal perspective / eventivity. Items bearing [V] can anchor to time/events (verbs, adjectives) — items lacking [V] do not have temporal anchoring (nouns).
Equations
Instances For
Nouns have sortal but not temporal perspective: n bears [N] only.
Verbs have temporal but not sortal perspective: v bears [V] only.
Adjectives have both sortal and temporal perspective: a bears [N, V].
The noun–verb asymmetry: nouns have sortal but not temporal perspective; verbs have temporal but not sortal perspective. Adjectives have both. This follows from the [N]/[V] feature distribution on categorizers.
Each categorizer forms a well-formed EP with its lexical anchor: V→v, N→n, A→a are all category-consistent and F-monotone.
The F-level jump from lexical head to categorizer is exactly 1 in all cases. The uniformity of categorization is Panagiotidis's prediction (§4.4–§4.5); the F-value encoding is [Gri05]'s EP architecture.
All categorizers sit at exactly F1 (in Grimshaw's system), parallel across families. Panagiotidis's core claim: v, n, a are structurally parallel — they differ only in which interpretable features they bear.
The categorizers are in their respective families.
Category-changing morphology = changing the categorizer. The same root under different categorizers yields items in different EP families — this is what it means to "change category."
English root families ([Pan15] §5.2, [Mar97]) #
Standard examples from the Distributed Morphology literature: roots that surface as nouns, verbs, and adjectives via different morphological processes.
√DESTROY: destroy (V), destruction (N), destructive (A)
Equations
- One or more equations did not get rendered due to their size.
Instances For
√BEAUTY: beautify (V), beauty (N), beautiful (A)
Equations
- One or more equations did not get rendered due to their size.
Instances For
√CLEAR: clear (V), clarity (N), clear (A)
Equations
- Panagiotidis2015.clear = { rootLabel := "CLEAR", forms := [("clear", Morphology.LexCat.verb), ("clarity", Morphology.LexCat.noun), ("clear", Morphology.LexCat.adjective)] }
Instances For
√PRODUCE: produce (V), product/production (N), productive (A)
Equations
- One or more equations did not get rendered due to their size.
Instances For
√CREATE: create (V), creation (N), creative (A)
Equations
- Panagiotidis2015.create = { rootLabel := "CREATE", forms := [("create", Morphology.LexCat.verb), ("creation", Morphology.LexCat.noun), ("creative", Morphology.LexCat.adjective)] }
Instances For
√ACT: act (V), action (N), active (A)
Equations
- Panagiotidis2015.act = { rootLabel := "ACT", forms := [("act", Morphology.LexCat.verb), ("action", Morphology.LexCat.noun), ("active", Morphology.LexCat.adjective)] }
Instances For
All sample root families.
Equations
Instances For
A root family is predicted to be tricategorial iff categorization by each of v, n, a is possible. Since all three categorizers are available in English, any root can in principle surface in all three categories.
The √DESTROY family's three categories correspond to three categorizers.
Every root family in the sample has a form for each categorizer's category.
Bridge: §6.7.1 modifier-distribution diagnostic ↔ M&deS §2.3 (13) #
[Pan15] §6.7.1 (35)–(36) deploys a modifier-distribution diagnostic for SWITCH placement in mixed projections, with Dutch examples adapted from Ackema & Neeleman (2004:173):
- (35)
dat [stiekem succesvolle liedjes jatten]— adverbstiekemsits BELOW the SWITCH (in the verbal/adjectival subtree) - (36)
dat stiekeme [succesvolle liedjes jatten]— adjectivestiekemesits ABOVE the SWITCH (in the nominal subtree)
Per Panagiotidis p. 146, the SWITCH's complement is recategorised by its [N] feature. So a constituent dominated by a SWITCH projects nominally and takes adjectival modifiers; a constituent below the SWITCH retains its verbal/adjectival categorial identity and takes adverbial modifiers. The diagnostic gives SWITCH placement: where the modifier-category transition occurs is where the SWITCH sits.
[McNdS11] §2.3 (13) makes a similar modifier-
distribution observation about the inflected adjective in het rode van X: M&deS observe het intens/*intense rode (adverbial-only) and
conclude that rode remains adjectival, with het carrying the
type-shift.
Methodological lineage, not independent rediscovery. Both M&deS and Panagiotidis cite Ackema & Neeleman 2004 (Beyond Morphology) as the source of the modifier-as-domain diagnostic. The convergence below is a shared-source consequence, not two independent frameworks landing on the same test. The bridge formalises that the two frameworks make predictions of the same shape on the same data.
Caveat. Panagiotidis nowhere specifically analyses Dutch het as a
SWITCH; §6.6 covers V→N SWITCHes only (Korean -um, Basque -te/tze,
Turkish -dIk and -AcAk) and §6.9 covers Dutch nominalised infinitives.
Mapping het to a Panagiotidis-style SWITCH on the inflected adjective
is the formaliser's extrapolation. The bridge below identifies the
M&deS rivals with SWITCH-position commitments (low/high) and reads off
predictions geometrically; it does not claim Panagiotidis himself
analyses M&deS's data.
The structural commitment each InflectedAnalysis rival makes about
SWITCH placement, modelling Panagiotidis §6.7.1's geometric reasoning
over the rivals' defining proposals. This is the substantive content
each rival commits to: where in the structure of het rode van X
is the categorising head sitting?
- low : SwitchPosition
SWITCH is at the inflected-form level (the
-emorpheme is the SWITCH; the inflectedrodeis the categorised constituent). - none : SwitchPosition
No SWITCH; regular adjectival projection (e.g., normal AP modifying a noun, where the noun is elided).
- high : SwitchPosition
SWITCH is at the DP edge (
hetis the SWITCH; the AProdeis the SWITCH's complement).
Instances For
Equations
- Panagiotidis2015.MdSBridge.instDecidableEqSwitchPosition 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
Each rival's defining commitment about SWITCH placement.
nominalisation: -e itself is the SWITCH/categoriser. SWITCH = low.ellipsis: regular AP-modifying-N structure with elided N; no SWITCH/categoriser intervenes between modifier androde(the adjectival projection is intact pre-ellipsis). SWITCH = none.hetAsCap: het carries the categorising operation. SWITCH = high.
Equations
- Panagiotidis2015.MdSBridge.switchPosition McNallyDeSwart2011.InflectedAnalysis.nominalisation = Panagiotidis2015.MdSBridge.SwitchPosition.low
- Panagiotidis2015.MdSBridge.switchPosition McNallyDeSwart2011.InflectedAnalysis.ellipsis = Panagiotidis2015.MdSBridge.SwitchPosition.none
- Panagiotidis2015.MdSBridge.switchPosition McNallyDeSwart2011.InflectedAnalysis.hetAsCap = Panagiotidis2015.MdSBridge.SwitchPosition.high
Instances For
Per [Pan15] p. 146 + §6.7.1: the SWITCH's complement
is recategorised by [N], so a constituent dominated by a SWITCH
projects nominally (takes adjectival modifiers) while a constituent
below the SWITCH retains its adjectival identity (takes adverbial
modifiers). For the inflected form rode, the diagnostic is read
by asking where is rode relative to the SWITCH:
- SWITCH = low (
-eIS the SWITCH):rodeIS the SWITCH-headed constituent → projects nominally → predicts ADJECTIVAL modification ofrode. - SWITCH = high (
hetis the SWITCH,rodeis its AP-complement):rodeis BELOW the SWITCH → retains adjectival identity → predicts ADVERBIAL modification ofrode.
For ellipsis (no SWITCH), the surface AP is intact, so adverbial
modification of rode is licensed just as any adjective licenses it.
panagiotidisPredictsAdverbialMod a is now derived from
switchPosition a: the geometric prediction is "no low SWITCH
dominating rode", i.e. the modifier-attachment site is below or
independent of any SWITCH.
Equations
Instances For
Equations
The Panagiotidis prediction matches the [McNdS11]
prediction on every rival. Both predicates encode the same modifier-
distribution diagnostic (which they both inherit from Ackema &
Neeleman 2004); the agreement is shared-methodology consequence, not
independent rediscovery. The substance of the bridge: the geometric
SWITCH-placement reasoning derives the same predictions as M&deS's
case-by-case PredictsAdverbialModOnly.
The nominalisation rival fails the joint prediction:
Panagiotidis's geometric diagnostic over its low-SWITCH commitment
predicts the inflected form should admit adjectival modification
(because rode would be SWITCH-dominated and project nominally);
[McNdS11] (13) shows the inflected form REJECTS
adjectival modification. The combined refutation routes through
switchPosition .nominalisation = .low → ¬panagiotidisPredictsAdverbialMod
and the M&deS data point.
Conversely, the hetAsCap rival passes: high-SWITCH commitment +
rode below SWITCH → adverbial modification predicted, matching
M&deS data.
The ellipsis rival also passes: no-SWITCH commitment means
surface AP is intact and adverbial modification is licensed in the
standard way.
Categoriser identification at the surface head level. Under each
rival, what is the lexical category of the inflected form rode as
it is projected at the surface?
nominalisation:-ecategorisesrodeas a noun →Cat.n.ellipsis: surfacerodeis an adjective; the n is the elided null noun, structurally elsewhere →Cat.a(the visible head).hetAsCap:roderemains adjectival;hetis the SWITCH →Cat.aat the surface head.
The frameworks-divergence is captured: only nominalisation
promotes the surface category to nominal. The other two leave the
surface adjectival.
Equations
- Panagiotidis2015.MdSBridge.surfaceCategorizer McNallyDeSwart2011.InflectedAnalysis.nominalisation = Minimalist.Cat.n
- Panagiotidis2015.MdSBridge.surfaceCategorizer McNallyDeSwart2011.InflectedAnalysis.ellipsis = Minimalist.Cat.a
- Panagiotidis2015.MdSBridge.surfaceCategorizer McNallyDeSwart2011.InflectedAnalysis.hetAsCap = Minimalist.Cat.a
Instances For
The surface categoriser distinguishes nominalisation from the other two rivals — exactly the M&deS §2.3 distinction of "is the inflected form a noun?". This is a real per-rival commitment, not a constant.
The Panagiotidis-side referential prediction follows the surface
categoriser: nominalisation predicts the inflected form is
referential (per noun_referential_not_predicative above);
ellipsis and hetAsCap predict it remains predicative-bearing
(per adjective_both).
Categorial features: [Cho70] vs [Pan15] (relocated from Minimalist/CategorialFeatures.lean) #
Two theories of what makes a noun a noun and a verb a verb:
[Cho70]: [±V, ±N] as arbitrary binary diacritics that cross-classify the four lexical categories. Adopted by [Gri05] for Extended Projections. Implemented in
CatFeatures.[Pan15]: [N] and [V] as substantive, LF-interpretable features:
- [N] = sortal perspective / referentiality
- [V] = temporal perspective / eventivity (§4.3)
On categorizers (v, n, a), these are interpretable; on functional heads
(T, C, D, etc.), they are uninterpretable copies (§5.8).
Implemented in
CategorialFeatures.
Key Agreement #
The two systems produce the same four equivalence classes — {verbal}, {nominal},
{adjectival}, {adpositional} — and therefore agree on all EP-consistency judgments
(chomsky_panagiotidis_agree).
Key Disagreement #
The status of P (prepositions/adpositions):
- Chomsky: P actively bears [-V, -N] — two negative feature values
- Panagiotidis: P is the default categorizer, lacking both [N] and [V]
This matters for the theory of features: in Chomsky's system, [-V] and [-N] are just as "real" as [+V] and [+N]. In Panagiotidis's system, the absence of [N] and [V] is genuinely the absence of categorial content — P is the elsewhere case. This predicts that P should be the most promiscuous category (appearing in the most diverse syntactic environments), which Panagiotidis argues is borne out.
Adjectives #
Both systems agree that A shares properties with both N and V:
- Chomsky: A = [+V, +N], the only category with both positive values
- Panagiotidis: A = [N, V], bearing both substantive features
The difference: for Panagiotidis, this is explanatory — adjectives have temporal anchoring (because [V]) and sortal perspective (because [N]). For Chomsky, the co-presence of [+V] and [+N] is a notational fact without semantic content.
In Chomsky's system, every category has at least one positive feature except P (which has [-V, -N] = ⟨false, false⟩).
In Panagiotidis's system, P is the default: no categorial features.
These are extensionally identical — but the theories interpret ⟨false, false⟩ differently: Chomsky reads it as "actively [-V, -N]"; Panagiotidis reads it as "absence of both [N] and [V]" (the elsewhere case).
Adjectives bear both features in both systems.
[N] = sortal perspective / referentiality. Every category in the nominal EP bears [N] (interpretable on n, uninterpretable on Num/Q/D; §5.8).
[V] = temporal perspective / eventivity. Every category in the verbal EP bears [V] (interpretable on v, uninterpretable on T/C; §5.8).
Adjectives bear both [N] (sortal perspective) and [V] (temporal perspective). This explains why adjectives can be nominalized (via [N]) and have temporal anchoring (via [V]).
The adjectival categorizer a inherits [N, V] from A.
P bears neither [N] nor [V] — the default/elsewhere categorizer. This predicts P is the most syntactically promiscuous category.
The three categorizers (v, n, a) form a natural class at F1 (Grimshaw's system). Each bears the interpretable categorial features of its EP family. Note: Panagiotidis (§4.5) argues categorizers are lexical, not functional, despite being placed at F1 in the EP.
Lexical heads (V, N, A, P) are not categorizers. In Panagiotidis's system, these represent categorized items (√+categorizer), not the categorizers themselves.
Each categorizer bears the interpretable features of its family: v bears [V] (temporal), n bears [N] (sortal), a bears [N, V] (both).
CatFamily is the theory-neutral representation: it records which categories group together without committing to the mechanism. Both Chomsky and Panagiotidis produce the same CatFamily partition.