Lexical Conservatism — Steriade 1997 #
Lexical Conservatism (LC) is paradigm uniformity anchored on attestation: a candidate surfacing form preferentially aligns with an attested wordform of the lexeme, and the strength of the alignment scales with how well-attested the anchor is. LC predicts that paradigms with no attested anchor (e.g. bound stems with no free wordform) impose no LC pressure — the candidate is free to satisfy markedness alone — while a strongly-attested anchor imposes strong pressure preserving the anchor's segments.
This file holds the LC substrate — the paper's anchored-paradigm primitive,
anchored to its originating paper and consumed by later study files.
Studies/BreissKatsudaKawahara2026.lean instantiates it for Japanese velar
nasalisation: bound-N2 compounds nasalise categorically (no anchor), while in
free-N2 compounds nasalisation is suppressed in proportion to the N2's free-form
attestation strength.
Architecture #
LC reuses the generic symmetric lift Constraints.liftPairwise; what
distinguishes it from Optimal Paradigms ([McC05], Studies/McCarthy2005.lean)
is which forms enter the paradigm, not the lift. lcParadigm takes a
candidate plus an optional attested anchor, returning a singleton when the
anchor is absent (no LC pressure) and a pair when present. OP sums over every
member with no distinguished anchor and so cannot model the bound/free split
without auxiliary stipulation; LC handles it by paradigm membership alone
(lc_unanchored_zero).
The strength of the LC pressure is a downstream frequency-conditioned weight on the constraint, supplied by the consuming study.
Main definitions #
lcParadigm— build an LC paradigm from a candidate and an optional anchor.mkLCFaith— an LC-FAITH constraint: per-pair mismatch summed over the LC paradigm.lc_unanchored_zero— an unanchored (singleton) paradigm has zero LC-FAITH violations; the structural source of "bound stems show no LC effect."
Not the 2000 paper #
[Ste00b] ("Paradigm Uniformity and the Phonetics–Phonology Boundary") is a different Steriade contribution, arguing PU effects are gradient and phonetically grounded. This file formalises only the categorical 1997 Lexical Conservatism account; a gradient-correspondence treatment along the lines of [Ste00b] is not yet modelled.
Out of scope #
- The specific shape of attestation strength (sigmoid, linear, step) — supplied externally by the consuming study. LC only commits to the qualitative claim that strength is monotone in attestation.
- Specific segment-level mismatch metrics — passed in as
mismatch, with the only requirement thatmismatch f f = 0(well-formedness on the diagonal).
Anchored paradigm #
Build an LC paradigm from a candidate form and an optional attested anchor.
When the anchor is present, the paradigm is [candidate, anchor]; when
absent, it is [candidate]. Anchor presence is the LC channel:
bound/unattested → singleton, free/attested → pair.
Equations
- Steriade1997.lcParadigm candidate none = [candidate]
- Steriade1997.lcParadigm candidate (some a) = [candidate, a]
Instances For
LC-FAITH constraint #
Build an LC-FAITH constraint: per-pair faithfulness mismatch summed over the
LC paradigm. The same liftPairwise is reused; what distinguishes LC from OP
is the paradigm-construction discipline (lcParadigm), not the lift.
Equations
- Steriade1997.mkLCFaith _name mismatch = Constraints.liftPairwise mismatch
Instances For
Unanchored paradigms have zero LC-FAITH violations #
A paradigm with no attested anchor (singleton) has zero LC-FAITH violations:
the only ordered pair is the diagonal, and a well-formed mismatch returns 0
on the diagonal. This is the structural source of the prediction that bound
stems exhibit no LC effect — a derived consequence of paradigm
construction, not a stipulation.