Rose & Walker (2004) @cite{rose-walker-2004} #

A Typology of Consonant Agreement as Correspondence. Language 80(3): 475–531.

@cite{rose-walker-2004} present a typology of long-distance consonant agreement (LDCA) — covering nasal, laryngeal, liquid, coronal, and dorsal harmonies — and analyze it within Optimality Theory using a correspondence-theoretic apparatus: pairs of similar consonants in the output stand in a CORR-CC correspondence relation, and IDENT-CC[F] constraints require corresponding consonants to share the value of feature [F]. Their architecture predicts that agreement is more likely between more similar consonants and that intervening dissimilar material is "neutral" rather than blocking.

What this file formalizes (and what it does not) #

We do not formalize Rose & Walker's correspondence apparatus (CORR-CC, IDENT-CC[F], the similarity-based ranking) — that is a constraint architecture in OT and lives outside the subregular layer. What we formalize here is the surface stringset of one of their core case studies — Kikongo nasal harmony, the paper's leading example — as a tier-based strictly 2-local language, following the Heinz/Rogers/Hansson subregular tradition (@cite{hansson-2010}, @cite{mcmullin-2016}).

The framing is: Rose & Walker's correspondence analysis derives the surface generalization; the TSL_2 description characterizes the surface stringset that their derivation produces. The two analyses operate at different levels and are not in competition for the same explanatory work.

The function-level subregular substrate at Core/Computability/Subregular/Function/ provides a separate axis of classification for the map (UR → SR) rather than the stringset; LDCA maps are typically Tier-Subsequential (specifically Right-Tier-Subsequential per @cite{hansson-2010}'s anticipatory directionality argument). We keep the language-level TSL_2 statement here because the directionality argument is upstream of the surface filter formalised in this file.

Kikongo nasal harmony is asymmetric: the forbidden tier-adjacent pair is (nasal, voiced-stop), not the reverse. This is a strict instance of the generic TSLGrammar.ofForbiddenPairs constructor and does not factor through either TSLGrammar.ocp (R := (· = ·)) or TSLGrammar.agree (R := (· ≠ ·)); the AGREE specialization captures symmetric harmony (e.g. Navajo sibilant harmony in Hansson2010.lean). The asymmetry here tracks the morphological geometry (stem→suffix direction), not a feature-symmetric agreement relation.

Design boundary #

Things this formalization is silent on, by design:

Similarity-scaled correspondence: in ABC, harmony probability depends gradiently on featural similarity. The TSL_2 stringset is sharp — a pair either is or is not forbidden.
Trigger/target asymmetry: the OT derivation distinguishes the feature source from its target; the surface phonotactic does not.
Transparency vs blocking by similarity: ABC predicts that intervening segments behave differently depending on how similar they are to the harmonizing pair. Single-tier TSL with a fixed tier predicate cannot express this — see the design-boundary docstring on tierProject non-monotonicity in ForbiddenPairs.lean. Cases where transparency is similarity-graded (e.g. the Sanskrit n-retroflexion pattern Rose & Walker discuss) require richer formalisms (multi-tier strictly local; ITSL).

Kikongo nasal harmony — the paper's leading case #

@cite{rose-walker-2004} open with Kikongo (Bantu): a voiced oral stop in the suffix is realized as a nasal stop when the stem contains a preceding nasal consonant. For example, the suffix -idi alternates to -ini after a nasal. Equivalently, the surface phonotactic is: no nasal consonant may be followed at any consonantal distance by a voiced oral stop within the relevant morphological domain.

The TSL_2 description: project the harmonizing-class tier (only nasal consonants and voiced oral stops project; neutral consonants and vowels are transparent), then forbid any tier-adjacent pair (C₁, C₂) where C₁ is [+nasal] and C₂ is a voiced oral stop.