Meinhardt, Mai, Baković & McCollum (2024): ATR Harmony Subregular Classification #
Worked example using the function-level subregular substrate (the subregular function classes) to classify a fragment of Eastern Nilotic ATR harmony per [MMBMcC24].
The paper #
[MMBMcC24] (NLLT 42:1191-1232) propose a tightening of the [HL13] Weakly Deterministic (WD) function class. Their thesis:
- Bidirectional iterative ATR harmony in Maasai (Eastern Nilotic) is attested and WD — unbounded semiambient: every target depends on one side at a time (though the side may vary), so its two contradirectional subsequential passes do not interact in the technical sense the paper formalises (§4).
- Turkana, identical to Maasai but for a set of exceptionally dominant [-ATR] suffix vowels, is attested and non-deterministic — unbounded circumambient: some targets depend on both sides at once, so the passes must interact. "The Maasai pattern is weakly deterministic, and the Turkana pattern is not" (end of §2.3): the paper's ND case is attested, not merely the hypothetical sour grapes ([Wil03b]; [Wil06]) it also places above WD.
- The original Heinz-Lai 2013 WD definition was too permissive, admitting some unattested patterns; Meinhardt et al. patch it by formalising the interaction condition explicitly (§4).
Maasai dominant/recessive harmony (paper §3.1, p. 1203) #
The empirically critical fact (paper p. 1203): "(non-exceptional) dominant vowels are underlyingly specified for the spreading value of the harmonic feature, [+ATR]" — dominance is an underlying featural specification on vowels, carried by roots and by suffixes alike (the paper shows both root- and suffix-controlled spreading), not a diacritic property of roots. A dominant vowel triggers spreading; recessive vowels harmonise.
The bidirectional Maasai pattern (paper p. 1193, ex 1a):
- (i) Full bidirectional spreading from a dominant root: /kɪ-√ñorr-ʊ/ → [ki-√ñorr-u] '1pl-love-pres' — [+ATR] spreads from the (dominant) root in both directions, raising recessive prefix /ɪ/ to [i] and recessive suffix /ʊ/ to [u].
- (ii) Leftward spread blocked by /a/: /ɪ-√as-ɪʃɔ-re/ → [ɪ-√as-iʃo-re]
'2sg-do-intr-appl' — the dominant applied suffix
-retriggers leftward [+ATR] spread, raising /ɪʃɔ/ to [iʃo]; the opaque /a/ of the root blocks it, leaving the prefix /ɪ/ unchanged.
What this file does (audit-corrected) #
Per CLAUDE.md "stimulus contrasts" discipline for Studies files:
- Defines a 4-symbol alphabet
Segcapturing the dominant/recessive distinction the paper insists is empirically load-bearing — the formaliser-invented alphabet from the previous version of this file (which conflated [+ATR] with the spreading trigger) is replaced. - Encodes the rightward [+ATR] spreading half of Maasai harmony
as both a
SubsequentialTransducerand anOSLRule. Single-direction iterative spreading is Output-Strictly-Local ([CEH15], the result [MMBMcC24] builds on) — the output decision at each position depends on whether the immediately preceding output symbol is +ATR. The OSL classification is tighter than Left-Subsequential and is what the paper actually predicts for the unidirectional pass. - Decide-checks input/output examples corresponding to ex 1a-i and ex 1a-ii from the paper (encoded in the toy alphabet).
- Proves the bidirectional dominant-recessive map
maasaiis weakly deterministic (maasai_weaklyDeterministic) via a non-interacting bimachine, with two-sided unbounded dependence (maasai_twoSidedUnboundedDependence) — a predicate Tutrugbu also satisfies — while notRequiresBothSides(maasai_not_requiresBothSides). That asymmetry is the WD/ND boundary [MMBMcC24] draws.
What this file does NOT do #
- Does not encode the full Maasai/Turkana paradigms — only the minimal pair sufficient to illustrate substrate use. Paper §§3–4 add the re-paired low vowel (raised /A/ surfaces as [o]), opaque /A/ blocking, glides disrupting harmony, and Turkana's exceptionally dominant [-ATR] suffix vowels.
- Does not formalize Turkana — the paper's attested non-deterministic
half. Its exceptionally dominant [-ATR] suffixes make the pattern
unbounded circumambient, so the
RequiresBothSidesmachinery applies directly; this is the natural next arc. - Models the bidirectional map directly (
maasai: raise a recessive iff a dominant occurs anywhere) rather than as an explicit two-pass composition of a rightward and a leftward transducer, and omits the opaque /a/ blocking — the non-opaque core already exhibits weak determinism and the WD/ND covariation contrast.
Scope note: cross-construction extrapolation #
The OSL framing here characterises ATR harmony within a single
spell-out domain (root + affixes). [SCD26]
argue that ATR harmony in Guébie particle-verb focus-fronting
constructions is also local at the moment of spell-out but
surface-discontinuous after subsequent A′-movement. Their pattern
is not a counterexample to the OSL classification of root-internal
ATR harmony — the two analyses describe disjoint construction types
— but it does refute the broader extrapolation that "ATR harmony is
strictly local on the surface" universally. See
Studies/SandeClemDabkowski2026.lean for the
discontinuous case.
Minimal alphabet capturing the dominance-vs-recessive distinction that drives Maasai ATR harmony per [MMBMcC24] p. 1203. Four symbols stand in for the relevant phonological contrasts:
recL— a recessive [-ATR] vowel (e.g., /ɪ/, /ʊ/). Surfaces as [-ATR] absent harmony; raises to [+ATR] under spread.recH— a recessive vowel surfacing [+ATR] (e.g., /i/, /u/): the raised form ofrecL, and transparent to further spread.dom— a dominant vowel: underlyingly specified [+ATR], the trigger of spreading (the paper's load-bearing distinction).a— the opaque /a/. Blocks spread.
Consonants are omitted as they are transparent to the harmony.
Instances For
Equations
- MeinhardtEtAl2024.instDecidableEqSeg x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- One or more equations did not get rendered due to their size.
- MeinhardtEtAl2024.instReprSeg.repr MeinhardtEtAl2024.Seg.dom prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "MeinhardtEtAl2024.Seg.dom")).group prec✝
- MeinhardtEtAl2024.instReprSeg.repr MeinhardtEtAl2024.Seg.a prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "MeinhardtEtAl2024.Seg.a")).group prec✝
Instances For
Equations
- MeinhardtEtAl2024.instReprSeg = { reprPrec := MeinhardtEtAl2024.instReprSeg.repr }
Equations
- One or more equations did not get rendered due to their size.
Whether the segment is a [+ATR] vowel (after surface realisation).
A dom segment surfaces as [+ATR] by definition; recH is underlyingly
[+ATR]; recL is [-ATR] (until raised by spread); a is neither.
Equations
- MeinhardtEtAl2024.Seg.recH.isPlusATR = true
- MeinhardtEtAl2024.Seg.dom.isPlusATR = true
- x✝.isPlusATR = false
Instances For
Surface form of a segment under spreading [+ATR]: recessive [-ATR] vowels raise; everything else passes through unchanged (including the opaque /a/, which blocks the spread that would have reached it).
Equations
Instances For
OSL rule encoding rightward [+ATR] spreading from a dominant root.
The rule's k = 2: the output decision at each position depends on the single immediately preceding output symbol (per [CEH15] the canonical OSL fragment of phonological maps). Rule logic:
- Current input is
dom→ emitrecH(dominant always surfaces as [+ATR]). - Current input is
a→ emita(opaque, blocks spread; the next position's output context will bea, not a +ATR vowel). - Current input is
recH→ emitrecH(already +ATR, passes through). - Current input is
recLand previous output wasrecH→ emitrecH(spread continues). - Current input is
recLotherwise → emitrecL(no spread to here).
Single-direction iterative spreading patterns are OSL but not ISL ([CEH15]), because the output decision genuinely depends on the output history (how spread has propagated) rather than the input history alone.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Rightward [+ATR] spreading is Left-Output-Strictly-Local
([CEH15], the result
[MMBMcC24] builds on). Witness: the OSL rule
rightwardATR_osl defined above.
This is the tighter classification per the paper — single-direction iterative spreading patterns are properly contained in OSL, strictly above the ISL class but strictly below the (Left-)Subsequential class.
Rightward [+ATR] spreading is also Left-Subsequential — the umbrella class,
lifted from the OSL classification via OSL ⊆ Left-Subsequential
(isLeftOutputStrictlyLocal_left_subsequential).
Bidirectional Maasai harmony — weakly deterministic (faithful) #
Maasai dominant-recessive ATR harmony spreads [+ATR] from a dominant root to recessive
vowels on both sides. Modelled here in its non-opaque core: a recessive recL raises
to recH iff the word contains a dominant vowel anywhere (maasai). This is a union
of two independent spreading passes, so it is weakly deterministic
([MMBMcC24], unbounded semiambient): the bimachine maasaiBM
tracks a dominant seen on each side and its output is literally a unite of one-sided
rules. A recessive's surface ATR still co-varies with information unboundedly far on
either side, so maasai satisfies TwoSidedUnboundedDependence — as does Tutrugbu. The
contrast is exactly the WD/ND boundary the paper draws: only Tutrugbu is circumambient,
needing both sides at once (RequiresBothSides), which Maasai never does.
The word contains a dominant trigger.
Equations
- MeinhardtEtAl2024.hasDom xs = xs.any fun (x : MeinhardtEtAl2024.Seg) => x == MeinhardtEtAl2024.Seg.dom
Instances For
Bidirectional dominant-recessive harmony (non-opaque core): a recessive raises iff the word has a dominant trigger anywhere.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The non-interacting bimachine: each side's state tracks a dominant seen on that side; a recessive raises if either side has one — a union of one-sided rules.
Equations
- One or more equations did not get rendered due to their size.
Instances For
maasaiBM's cell output is a unite of one-sided raise-rules.
Maasai ATR harmony is weakly deterministic ([MMBMcC24]): the bidirectional dominant-recessive spread is a non-interacting bimachine.
Maasai has two-sided unbounded dependence — at every distance, a medial
recessive's ATR flips under a dominant placed far to the left or far to the right,
each side alone sufficing. Tutrugbu satisfies this too
(tutrugbu_twoSidedUnboundedDependence); the difference is that Maasai does not
RequiresBothSides, so it stays weakly deterministic. The paper's positive
classification of Maasai as unbounded semiambient — every target fixed by information
from at most one side — is the stronger claim, maasai_semiambient.
Maasai is semiambient — the paper's positive classification: every harmonised cell is licensed by one side alone, the far dominant that triggers it.
Hence Maasai does not require both sides — it escapes the teeth, unlike Tutrugbu. Covariation (both languages) and interaction (Tutrugbu only) come apart.
Strictness witness synchronous ⊊ WD: Maasai is weakly deterministic yet not
Mealy-computable — the length-preserving-stratum reading of [HL13]'s
LSF, RSF ⊆ WD corollary being strict, with maasai as their own dominant-recessive
witness (their Thms. 6 and 7). A Mealy-computable map is right-myopic
(IsMealyComputable.boundedDependence_right), but Maasai's bidirectional spread is not
(maasai_twoSidedUnboundedDependence); the block-class exclusion is
maasai_not_leftSubsequential.
Maasai is not left-subsequential — the block class is excluded too: maasai
is length-preserving, so a left-subsequential computer's delay bound would cap its
right dependence (IsLeftSubsequential.boundedDependence_right), but the spread's
right dependence is unbounded.