Lucy 1994: The role of semantic value in lexical comparison #
@cite{lucy-1994}
@cite{lucy-1994} argues that the right way to identify lexical classes is morpho-distributional, not denotational. Yucatek's "spatial" verbs are his test case: a notional set assembled by English intuition fails to coincide with any morphologically defined Yucatek class. What the morphology does deliver is a different, three-way classification of roots by salience profile, derivable from which transitiviser the root requires:
| Required transitiviser | Salience class | Examples |
|---|---|---|
=t (affective) | agent-salient | síit' "jump", ¢'iib "write" |
=∅ (zero) | agent-patient salient | kuč "carry", p'is "measure", lo'š "punch" |
=s (causative) | patient-salient | kíim "die", háan "cease", lúub' "fall", 'ok "enter" |
Crucially, Lucy's "motion" roots (luub, ok, etc.) do not form a
unified class — they pattern with other state-change verbs as
patient-salient. What does form a unified class is the positional
roots (čin "bend", kul "sit", etc.), distinguished by their derivation
via -tal (allomorph -lah) for the positional inchoative.
Here we recast the salience cut as a derived equivalence on roots
under the operator inventory in Fragments/Mayan/Yukatek/Operators.lean:
two roots have the same salience class iff the same operator(s) apply
to them. The 3-way classification then falls out of (B&K-G feature
signature) × (operator applicability conditions) — it is not stipulated.
The structural theorem salienceClass_from_signature makes this
precise: salience class depends only on the feature signature, not on
the specific root identity. The per-root checks then degenerate to
finite signature lookups.
SalienceClass and classOfSignature live in
Theories/Semantics/Lexical/Roots/SalienceClass.lean. This file
provides the full @cite{lucy-1994} analysis on top of them:
operator-orbit characterizations and per-root sanity checks.
Local short alias `predictedClass = Root.predictedSalience`.
A root's predicted salience class. Alias for
Root.predictedSalience (SalienceClass.lean).
Instances For
Each of the four orbit characterisations follows the same
pattern: unfold the orbit and the named class predicates, then
case-split on the four B&K-G feature bits and let simp_all
discharge each cell of the 16-row truth table. The local macro
lucy_orbit packages this so the four theorems differ only in
which class label appears on the LHS.
The =t-only orbit characterises agent-salient roots.
The =∅-only orbit characterises agent-patient salient roots.
The =s-only orbit characterises patient-salient roots.
The -tal-only orbit characterises positional roots.
An empty orbit characterises roots outside @cite{lucy-1994}'s
diagnostic gap ((¬manner, ¬result) rows that lack the positional
configuration state ∧ ¬cause).
Orbit-as-classifier. Two roots have the same operator orbit under @cite{lucy-1994}'s diagnostic inventory iff they have the same predicted salience class. The 4 named-class iff-theorems are special cases.
Agent-salient.
Agent-patient salient.
Patient-salient.
Motion roots — pattern as patient-salient (Lucy's central point).
Positional.
@cite{lucy-1994}'s central typological point: "motion" verbs
(luub "fall", ok "enter") are not in their own salience class
— they pattern with other patient-salient state-change roots.
Concretely: their predicted class is the same as kiim "die".
Conversely, positional roots do form a unified class distinguishable from any "motion" or state-change root: their predicted class differs from every patient-salient root's.
The class predicted from a root's closed feature signature
(bkgRules applied to the base entailments).
Equations
Instances For
Closure under bkgRules does not change the Lucy 1994 salience
classification: agent / agentPatient / patient classes ignore the
state field, and the positional class requires result = false,
which closure does not affect (closedFeatureSignature_result).
The only way closure can flip state from false to true is if
the base set has a becomesState atom (so result = true) — in
which case the positional arm is already excluded by result.
@cite{bohnemeyer-2004} refines @cite{lucy-1994}'s 4-way salience
cut into a 5-way stem classification (active, inactive,
inchoative, positional, transitiveActive). The mapping is
VerbStemClass.toSalienceClass in VerbClasses.lean. The agent /
patient / agent-patient classes correspond one-to-one; Lucy's
positional covers both Bohnemeyer's inchoative and positional.
The per-root theorems below check that for each Yukatek root in
`Roots.lean`, Lucy's predicted class agrees with the Bohnemeyer
stem-class label converted via `toSalienceClass`.
Agent-salient Lucy roots map to Bohnemeyer's active stem class.
Agent-patient salient Lucy roots map to Bohnemeyer's
transitiveActive.
Patient-salient Lucy roots map to Bohnemeyer's inactive.
Positional Lucy roots map to Bohnemeyer's positional (and would
equally map to inchoative per inchoative_positional_collapse_under_lucy).