Indexed Constraints #
@cite{pater-2010}
The conservative frequency-conditioning theory: frequency does not enter the grammar at all; instead, lexically-indexed constraints partition the lexicon into fixed sublexica (typically two: a high-frequency or "core" stratum and a low-frequency or "peripheral" stratum) with their own constraint rankings.
The interface to ItemSpecificity.HasTokenFreq is one bit per item:
isCore thresholds log-frequency at a fixed cutoff. Within a stratum,
no further frequency dependence — the strongest contrast with
ScaledWeights, where every log-frequency unit shifts the weight
linearly.
Empirical signature #
Indexed-constraint theories predict discontinuous frequency effects:
two distinct distributions, with no in-stratum gradience. Where the
empirical distribution is gradient and continuous in log-frequency
(e.g., the Breiss-Katsuda-Kawahara compounds in
Phenomena/Phonology/Studies/BreissKatsudaKawahara2026.lean), this
theory under-fits relative to ScaledWeights /
RepresentationStrength.
A two-stratum lexical partition: an item is "core" iff its
log-frequency reaches cutoff. Discrete by construction.
Implemented as an alias of ItemSpecificity.isAboveThreshold so the
shared decidability instance carries through; the abbrev form
means unfold isCore reduces directly to the inequality.
Equations
- Phonology.ItemSpecificity.Indexed.isCore cutoff a = Phonology.ItemSpecificity.isAboveThreshold cutoff a
Instances For
An indexed constraint fires only on items in a particular stratum.
mkCoreOnly cutoff base returns a constraint that evaluates to
base.eval c when c is in the core stratum, 0 otherwise.
This is the formalization of @cite{pater-2010}'s lexically-indexed markedness/faithfulness: the same structural penalty applies, but only to a sublexicon.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The peripheral counterpart: fires on items below the cutoff.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Indexed constraints produce the same violation count for any two items on the same side of the cutoff, regardless of their log-frequency. This is the discontinuity claim that distinguishes indexed-constraint theories from scaled-weights theories.